Algorithmic Regulation and Personalized Law: A Handbook 9781509931767, 9781509931750

Citation preview

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/001_List of Authors 03.08.2020.3d from 30.10.2020 11:31:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

List of Authors Marietta Auer is Director at the Max Planck Institute for European Legal History in Frankfurt am Main and Professor of Private Law and international and interdisciplinary Foundations of Law at the University of Gießen, Germany. She received her doctorate in law and habilitation from the University of Munich (2003, 2012) as well as LL.M. and S.J.D. degrees from the Harvard Law School (2000, 2012). She has published extensively in the fields of private law, legal theory, and philosophy of law. Recent publications include Der privatrechtliche Diskurs der Moderne (Tübingen 2014) and Zum Erkenntnisziel der Rechtstheorie. Philosophische Grundlagen multidisziplinärer Rechtswissenschaft (Baden-Baden 2018). Philip Maximilian Bender, LL.M. (Yale Law School), Maître en droit (Paris II/Panthéon-Assas) currently is a Research Associate at the Max Planck Institute for Tax Law and Public Finance in Munich, as well as a Visiting Fellow at the Information Society Project at Yale Law School, and a Lecturer at Ludwig Maximilian University of Munich (LMU). He is qualified for the bar in both Germany and New York. His research focuses on contracts, law & tech, theory of law, and constitutional implications in private law. Omri Ben-Shahar is Leo and Eileen Herzel Professor of Law, Kearney Director of the CoaseSandor Institute for Law and Economics at University of Chicago. Christoph Busch, Maître en droit (Paris X), is Professor of Law at the University of Osnabrück and Visiting Fellow at the Yale Information Society Project at Yale Law School. He is co-chairman of the European Law Institute’s Digital Law Group and a member of the European Commission’s Expert Group to the EU Observatory on the Online Platform Economy. He is principal investigator of the research project “Granular Society – Granular Law? Individuality and Normative Models in the Data Society” funded by the Volkswagen Foundation under a Momentum Grant (2019–2024). Tony Casey is Professor of Law and the Faculty Director of the Center on Law and Finance at The University of Chicago Law School. His research examines the intersection of law and finance as well as the effects of technological innovation on legal systems. Before entering academics, Professor Casey was a partner at Kirkland & Ellis LLP. He received his JD from The University of Chicago Law School. After law school, Casey clerked for Chief Judge Joel M. Flaum of the United States Court of Appeals for the Seventh Circuit. Alberto De Franceschi is Professor of Private Law, Intellectual Property Law and Environmental Law at the University of Ferrara. He is member of the Italian National University Council, cochairman of the European Law Institute’s Digital Law Group, founding member and co-editor of the Journal of European Consumer and Market Law and of The Italian Law Journal. His current research focuses on issues related to the sale of goods, supply of digital content, privacy regulation and artificial intelligence. Francesco Denozza is Professor Emeritus of Commercial Law at the Faculty of Law of the University of Milan. He was Dean of the Faculty of Law, University of Milan (2012–2014) and President of the Italian Association of University Professors of Commercial Law (2009–2013). He is managing director of Orizzonti del diritto commerciale Online Journal of the Italian Association of University Professors of Commercial Law, co-editor of the Journal Giurisprudenza Commerciale and co-editor of the International Encyclopaedia of Laws – Competition Law/International Encyclopaedia for Competition Law. He is Member of the Milan Bar. Pasquale Femia is full Professor of Private Law and Dean of the Department of Political Sciences of the Università della Campania “Luigi Vanvitelli”. He is co-editor in chief of several journals – among which: The Italian Law Journal (Scopus indexed) and Tecnologie e diritto – and of the book series: Il diritto e l’Europa, Storie dal fondo, La cultura del diritto civile. His current research interests focus on philological method in private law theory, machine legal thinking, transsubjectivity in commons.

XI

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/001_List of Authors 03.08.2020.3d from 30.10.2020 11:31:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

List of Authors Hans Christoph Grigoleit holds the Chair for Private Law, Commercial Law, Corporate Law and Theory of Private Law at Ludwig Maximilian University of Munich (LMU). His research focuses on corporate governance and finance, information liability, fundamental principles of contract law, theory of legal reasoning and law & tech. Philipp Hacker, LL.M. (Yale), is Professor of Law and Ethics of the Digital Society at the European New School of Digital Studies, European University Viadrina. Before he was an AXA Postdoctoral Fellow at the Faculty of Law at Humboldt University of Berlin, a Max Weber Fellow at the European University Institute and an A.SK Fellow at WZB Berlin Social Science Center. His research focuses on the intersection of law and technology. In particular, he analyzes the impact of tracking technologies, Artificial Intelligence and the Internet of Things on consumer, privacy and anti-discrimination law. He often cooperates with computer scientists and mathematicians, especially on questions of explainable AI and algorithmic fairness. Marisaria Maugeri is Professor of Private Law at the University of Catania (Italy). She is on secondment to the Centro Linceo Interdisciplinare Beniamino Segre (Accademia dei Lincei – Roma). Her main research interests are in the field of Consumer, Contract and Business Law in their domestic and European dimensions. Her current work focuses on the implications of the new technologies for commercial and consumer transactions. She is Member of the board of the Italian Association of Law & Economics. She is Chair of the Palermo Panel of the Banking and Financial Ombudsman. Hans-W. Micklitz is Finland Distinguished Professor, University of Helsinki and Professor for Economic Law, Robert Schuman Centre for Advanced Studies, European University Institute, Florence. Anthony Niblett is an Associate Professor and Canada Research Chair in Law, Economics, & Innovation at the University of Toronto Faculty of Law. Professor Niblett researches law and economics, contract law, judicial behaviour, innovation, and competition policy. Professor Niblett holds a Ph.D. in economics from Harvard University as well as degrees in law and commerce from the University of Melbourne. Before joining the University of Toronto, he was a Bigelow Fellow at the University of Chicago Law School. In addition to his academic career, Professor Niblett is a cofounder of Blue J Legal, a start-up company bringing machine learning to tax and employment law. Francesco Paolo Patti is Associate Professor of Private Law at the Bocconi University, Milan. He received an LL.M. from the Westfälische Wilhelms-Universität Münster (2011) and a PhD from the Università degli Studi Sapienza Rome (2014). He was Research Assistant at the European University Institute (2016) and Senior Research Fellow at the Max Planck Institute for Comparative and International Private Law (2018). His research focuses on European and comparative contract law, comparative succession law and autonomous vehicles liability. Ariel Porat is the president of Tel Aviv University, a member of the Israel Academy of Sciences and the EMET Prize Laureate. In 2002–2006, he was the Dean of TAU Faculty of Law, and from 2003 until his appointment as president, in addition to being a professor at TAU, was the FischelNeil Distinguished Visiting Professor of Law at the University of Chicago. He was also a Visiting Professor at the Universities of Berkeley, Columbia, New York, Stanford, Toronto and Virginia. He is the author of several books including Tort Liability under Uncertainty (Oxford University Press, 2001) (with Alex Stein) and Getting Incentives Right (Princeton University Press, 2014) (with Robert Cooter). Pietro Sirena is full Professor of Civil Law, European Private Law, and Comparative Private Law at Bocconi University, Milan, where he currently serves as Dean of Bocconi Law School. He is furthermore: Member of the Executive Committee of the European Law Institute (ELI); Research Director of the Società Italiana per la Ricerca nel Diritto Comparato (SIRD); Member of the Executive Committee of the Society of European Contract Law (SECOLA); Membre Associé of the Académie Internationale de Droit Comparé (AIDC); President of the Arbitro Bancario Finanziario (ABF) of Rome. Jacob Lior Strahilevitz is Sidley Austin Professor of Law at University of Chicago.

XII

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/001_List of Authors 03.08.2020.3d from 30.10.2020 11:31:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

List of Authors Christian von Bar has been University Professor at Osnabrück since 1981. Founder and Director of the European Legal Studies Institute (ELSI) since 2003. Holder of the Leibniz Prize; Honorary Master of the Bench, Gray’s Inn; Corresponding Fellow of the British Academy. Doctor iuris honoris causa Czestochowa, Helsinki, Katowice, Leuven, Novi Sad, Tartu, Uppsala, Warmia and Mazury University. Chairman of the Study Group on a European Civil Code. Member of the Commission on European Contract Law. Head of various international research groups. Vincenzo Zeno-Zencovich is full Professor of Comparative Law at the University of Roma Tre. He holds degrees in Law and in Political Sciences from the university of Rome “La Sapienza”. He has authored or edited over forty volumes and published over 300 articles, many of which in the field of information and communication technologies. The full-text of his latest publications can be found on SSRN. Since 2013 he is the chairman of the Italian Association of Comparative Law (AIDC).

XIII

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/001_List of Authors 03.08.2020.3d from 30.10.2020 11:31:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/00_C. Busch - A. De Franceschi, Introduction.3d from 30.10.2020 11:32:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

INTRODUCTION Personalization and Granularity of Legal Norms in the Data Economy: A Transatlantic Debate* This handbook provides a collection of articles that explore the ways in which the use 1 of big data analytics and artificial intelligence could recalibrate the relationship between law and individuality and change the foundational structures of our legal systems. In this perspective, the volume contributes to the emerging literature on “personalized law”.1 Bringing together contributions by scholars from both sides of the Atlantic, it aims to serve both as a gateway to this emerging and promising field and as a catalyst for new scholarly research. The handbook is organized in four parts. Part 1 sets the scene by introducing the 2 concept of personalized law and explaining how the rise of big data could be instrumental for creating and administering personalized legal rules tailored to specific individuals or circumstances. This part presents two papers that are seminal to the jurisprudence on personalized law. Chapter A contains an article by Ariel Porat and Lior Jacob Strahilevitz, which provides the first comprehensive analysis of personalized default rules and personalized disclosures.2 It explores in detail the feasibility of personalized law and discusses a broad range of possible objections and limitations. Chapter B presents an article by Omri Ben-Shahar and Ariel Porat, which looks at the feasibility and desirability of personalization from the perspective of negligence law.3 The authors offer a detailed analysis of the efficiency gains provided by the use of personalized standards. Moreover, they discuss justice considerations and several practical aspects regarding the implementation of personalized standards. In Chapter C Anthony J. Casey and Anthony Niblett also explore the ways in which technological development could change the foundational structure of * The authors would like to gratefully acknowledge the support of Villa Vigoni and Deutsche Forschungsgemeinschaft (DFG) who provided funding for a workshop in Loveno di Menaggio in March 2017 which inspired this book. Christoph Busch also gratefully acknowledges the support of the Volkswagen Foundation which provides funding for the research project “Granular Society – Granular Law? Individuality and Normative Models in the Data Society” under a Momentum Grant (2019–2024). 1 See Porat/Strahilevitz, infra Part 1.A; Ben-Shahar/Porat, infra Part 1.B; Ben-Shahar/Porat, Personalizing Mandatory Rules in Contract Law, 86 U Chi L Rev (2019); see also Sunstein, Choosing not to Choose: Understanding the Value of Choice, Oxford 2015, 157–173; Busch, The Future of Pre-contractual Information Duties: From Behavioural Insights to Big Data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Edward Elgar 2016, 221; Casey/Niblett, infra Part 1.C; Hacker, Personalizing EU Private Law: From Disclosures to Nudges and Mandates, 25 European Review of Private Law 651 (2017); Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in: Mak et al. (eds.), Research Handbook in Data Science and Law, Cheltenham 2018, 408; Hacker, The Ambivalence of Algorithms: Gauging the Legitimacy of Personalized Law, in: Bakhoum et al. (eds.), Personal Data in Competition, Consumer Protection and Intellectual Property Law, Heidelberg 2018, 85; see also the contributions in Volume 86, Issue 2 (April 2019) of the University of Chicago Law Review which contains the papers presented at the Symposium on Personalized Law organized by Omri BenShahar, Anthony Casey, Ariel Porat, and Lior Strahilevitz at the University of Chicago in April 2018. 2 Porat/Strahilevitz, infra Part 1.A. 3 Ben-Shahar/Porat, infra Part 1.B.

Busch/De Franceschi

1

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/00_C. Busch - A. De Franceschi, Introduction.3d from 30.10.2020 11:32:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Introduction

the law.4 The authors argue that advances in predictive analysis and communication technology could give rise to a new form of legal norms – which they call “microdirectives” – that combines the benefits of both rules and standards without incurring the respective costs. Taken together, the two seminal papers on personalized law and the related piece on “microdirectives” serve as the starting point for a transatlantic debate on how the relationship between law and individuality could change in the emerging data society. 3 The contributions in Parts 2 to 4 discuss the concept of personalized law from a broad range of perspectives. Part 2 consists of four chapters which analyse the concept of personalized law mainly from a theoretical point of view. In chapter D, Hans Christoph Grigoleit and Philip Maximilian Bender venture to structure the discourse on personalized law around several fundamental pairs of dichotomic notions both with regard to the technological and the legal dimension of the personalization project. In their analysis they also draw a distinction between what they call the “evolutionary” approach to personalization, involving factual and normative changes within the existing legal system, and a “revolutionary” approach, involving changes that relate to the very structure of the legal system as a whole. While they see promising potential in the personalization of disclosures, they are critical regarding potential shifts within the system of separation of powers that could result in weakening the judiciary. Chapter E by Marietta Auer undertakes also a critical examination of the promises made by the proponents of personalized law. In her view, there is a structural connection between the concept of granular personalization and novel styles of liberal paternalist regulation. In this perspective, she sees personalized defaults mainly as a tool for modern consumerism. At the same time, she warns of the dangers of a resurgence and reinforcement of discriminatory stereotypes as a possible result of legal granularization. Finally, Auer links the debate about “microdirectives” with Wittgenstein’s treatment of rule-following5 and argues that the rise of algorithmic regulation will not lead to the “death of rules and standards” as predicted by Casey and Niblett, but rather reinforce the importance of flexible standards. In chapter F, Pasquale Femia starts his analysis by tracing the problem of legal complexity all the way back to the observation in Justinian’s Digest that there are “more business transactions than terms to designate them”.6 He offers a detailed analysis of how typifications are used as an instrument to “tame the chaos” and reduce complexity in the legal system. He also emphasises the idea of “learning law” and links the concept of granular law with the idea of “dynamic rules” that automatically respond to changes in future conditions. Chapter G by Francesco Denozza and Marisaria Maugeri takes a more economic perspective and contextualizes the personalization discourse with regard to the economic and political ideas of liberalism and neoliberalism. In addition, the authors explore potential cross-subsidies that may result from the application effects of personalized rules. 4 The contributions in Part 3 look at the idea of personalized law through the lens of specific fields of law, in particular contract law, consumer law and tort law. The chapters not only look at potential use cases for personalization, but also raise several fundamental issues. Chapter H by Pietro Sirena focuses on the field of financial services. Sirena analyses the shift towards more granular “images of the consumer” in recent EU legislation and highlights several examples of already existing personalized disclosures in EU financial services regulation. Looking beyond the personalization of information, he explores the personalization of banking and insurance contracts based on the “know4

Casey/Niblett, infra Part 1.C. Wittgenstein, Philosophical Investigations, (1958) §§ 82–87. 6 See D. 19, 5, 4 (Ulpianus, lib. XXX ad Sabinum). 5

2

Busch/De Franceschi

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/00_C. Busch - A. De Franceschi, Introduction.3d from 30.10.2020 11:32:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Personalization and Granularity of Legal Norms in the Data Economy

your-customer” rule. In chapter I, Hans Micklitz analyses various strands of granularization that have influenced European consumer law in the past. In his view, the rise of big data analytics could take this development to a new level. He describes two different scenarios how this could change the structure of existing consumer laws, one leading to an ever-increasing complexity of the regulatory landscape and another one which he refers to as “re-typification” of consumer laws. In chapter J, Francesco Paolo Patti explores the ways in which the personalization of default rules could change the unfairness control of standard contract terms in a European context. He argues that under a system of personalized default rules the fairness test on the basis of Article 3(1) of the Unfair Terms Directive could undergo a fundamental transformation if courts would have to apply a personalized fairness benchmark. While Patti has some reservations with regard to such a development, he sees more promising potential regarding personalized mandatory rules which could avoid cross-subsidies and enhance the efficiency of the legal system. Chapter K contains a brief but insightful comment by Christian von Bar on the personalization of tort law. He raises the question whether the very idea of negligence requires a comparison between the behaviour of a concrete person with the behaviour of another person, which only exists in theory. Thus, in his view, the concept of negligence becomes obsolete if the objective reference point for assessing the individual behaviour is abandoned. Moreover, he argues that the objective standard of negligence, which limits the amount of data that is relevant for assessing human behaviour, protects the personal freedom both of the potential wrongdoer and the potential victim. Finally, the contributions in Part 4 aim at creating links between the discourse on 5 personalized law and other fields of research, in particular behavioural law and economics, comparative law and cultural studies, and the scholarly debate on algorithmic decision-making. Thus, chapter L by Philipp Hacker looks at personalized law through the lens of behavioural law and economics. Hacker argues that one crucial, but often overlooked problem in the operationalization of behavioural insights for lawmaking is actor heterogeneity. Therefore, in his view behavioural law and economics has much to gain from the personalization of legal rules with the aim of matching regulatory strategies with the concrete needs, capacities and vulnerabilities of the addressees. In his chapter, Hacker maps out the potential advantages of behavioural personalization and highlights the inherent limit of personalized law. Chapter M by Vincenzo Zeno-Zencovich examines the notion of personalized or granular law from the perspective of cultural studies and comparative law. From this viewpoint, he identifies a cultural divide in the personalization discourse. Zeno-Zencovich argues that the cultural origin of personalized or granular law is closely linked to the methodology of common law systems that have grown incrementally through the stratification and consolidation of case-law. From this perspective, personalization of the law essentially involves some sort of retro-engineering and de-structuring of a general rule through the analysis of data concerning the parties involved. In contrast, for a continental European lawyer, a norm can only be générale because it is the result of the volonté générale. Finally, in chapter N Christoph Busch explores how algorithmic personalization of legal rules could be operationalized for tailoring disclosures on digital marketplaces, mitigating discrimination in the sharing economy and optimizing the flow of traffic in smart cities. The chapter also highlights issues and problems regarding the enforcement of personalized law. Busch underlines that the transition towards personalized law involves not only changes in the design of legal rules, but also necessitates modifications regarding compliance monitoring and enforcement. He argues that personalized law can be conceptualized as a form of algorithmic regulation Busch/De Franceschi

3

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/00_C. Busch - A. De Franceschi, Introduction.3d from 30.10.2020 11:32:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Introduction

or governance-by-data. Therefore, the implementation of personalized law requires setting up a regulatory framework for ensuring algorithmic accountability. In this perspective, the final chapter of the handbook aims to create a link between the scholarly debates on algorithmic decision-making, automated legal enforcement and the discourse on personalized law. 6 Summing up, it is our hope that the transatlantic debate which the editors and contributors initiated at the German-Italian Centre for the European Dialogue in Villa Vigoni, and which led to the publication of this volume, will continue and that the contributions in this handbook will spark further research on the feasibility and the desirability of personalized law.

4

Busch/De Franceschi

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

PART 1 THE CONCEPT OF PERSONALIZED LAW A. Personalizing Default Rules and Disclosure with Big Data* This Article provides the first comprehensive account of personalized default rules and 1 personalized disclosure in the law. Under a personalized approach to default rules, individuals are assigned default terms in contracts or wills that are tailored to their own personalities, characteristics, and past behaviors. Similarly, disclosures by firms or the state can be tailored so that only information likely to be relevant to an individual is disclosed and information likely to be irrelevant to her is omitted. The Article explains how the rise of Big Data makes the effective personalization of default rules and disclosure far easier than it would have been during earlier eras. The Article then shows how personalization might improve existing approaches to the law of consumer contracts, medical malpractice, organ donation, inheritance, landlord-tenant relations, and labor law. The paper makes several contributions to the literature. First, it shows how data 2 mining can be used to identify particular personality traits in individuals, and these traits may in turn predict preferences for particular packages of legal rights. Second, it proposes a regime whereby a subset of the population (“guinea pigs”) is given a lot of information about various contractual terms and plenty of time to evaluate their desirability, with the choices of particular guinea pigs becoming the default choices for those members of the general public who have similar personalities, demographic characteristics, and patterns of observed behavior. Third, we assess a lengthy list of drawbacks to the personalization of default rules and disclosure, including cross subsidization, strategic behavior, uncertainty, stereotyping, privacy, and institutionalcompetence concerns. Finally, we explain that the most trenchant critiques of the disclosure strategy for addressing social ills are really criticisms of impersonal disclosure. Personalized disclosure not only offers the potential to cure the ills associated with impersonal disclosure strategies, but it can also ameliorate many of the problems associated with the use of personalized default rules.

I. Introduction Law is impersonal. The state generally does not tailor the contents of the law to 3 people’s characteristics and traits. In this Article, we argue that in the era of Big Data, * The authors thank Michael Abramowicz, Omri Ben‐Shahar, Yitzhak Benbaji, Eyal Benvenisti, Michael Birnhack, Tony Casey, Hanoch Dagan, Giuseppe Dari-Mattiacci, Lee Fennell, Talia Fisher, Olga Frishman, James Grimmelmann, Sharon Hannes, Todd Henderson, Aziz Huq, Ehud Kamar, Amir Khoury, Shay Lavie, Saul Levmore, Yoram Margalioth, Jonathan Masur, William McGevern, Gideon Parchomovsky, Eric Posner, Adam Samaha, Max Stearns, Cass Sunstein, and Lauren Willis, as well as workshop participants at the University of Chicago, the Interdisciplinary Center at Herzliya, Tel Aviv University, Northwestern’s Kellogg School of Management, the University of Maryland, the Privacy Law Scholars Conference, and the annual meeting of the American Law and Economics Association for helpful conversations and comments. The authors would also like to thank Julian Dibbell, Jack Grein, and Omer Yehezkel for inspired research assistance, as well as the Russell J. Parsons Faculty Research Fund for generous research support. This chapter was originally published in 112 Michigan Law Review 1417 (2014).

Porat/Strahilevitz

5

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

law should become more personalized. Our main focuses are default rules (situations where people face a choice between sticking with a default option or specifying a different option instead) and disclosure (where the law mandates that individuals receive particular information). Our claim has important applications to contract law, consumer law, inheritance law, medical malpractice, property law, labor law, privacy law, organ donation law, and other fields. 4 Let us illustrate our approach with an example from inheritance law. Empirical research has shown that married fathers are more likely than married mothers to bequeath all their property to their spouse (55 % compared to 34 %).1 Moreover, according to these studies, men bequeath significantly larger shares of their estates to their spouses (80 % of estates are willed to widows versus 40 % to widowers).2 These data are consistent with rational-choice models of behavior: wives trust their husbands less than husbands trust their wives to use inherited resources in the best interests of their mutual children, since men are significantly more likely to remarry and devote resources to the children from their second marriage, which comes at the expense of children from their first marriage.3 5 If men’s testamentary preferences differ systematically from women’s, why should intestacy laws continue to be gender neutral?4 Why not have different default intestacy rules for men and women instead? We argue that as long as these preferences remain stable and gender correlated, a different set of default rules for women would lead in the long run to more estate resources being allocated to heirs according to decedents’ true preferences. We further posit that it may be desirable to use other readily observable characteristics (e.g., wealth, health, time of marriage, age of children, and occupation) that could predict default rules in intestacy for population subgroups. As with any default rules, individuals would be free to alter these defaults by executing a will.5 6 We also advocate a more ambitious version of personalization here, one that would let courts determine how an intestate’s estate should be allocated based on an analysis of his consumer behavior during his lifetime. In the era of Big Data,6 we suggest that it will be possible to find individuals whose observable behavior and characteristics closely match those of the intestate – we refer to these people as “guinea pigs” – to examine the kinds of choices that the guinea pigs made in their wills and then to use these choices as a template for determining what the intestate likely would have wanted.7 An upshot of widely employing this approach is that more estates would be allocated in a way that better approximates the true preferences of the decedent. Given the fact that most 1 Hacker, The Gendered Dimensions of Inheritance: Empirical Food for Legal Thought, 7 J. Empirical Legal Stud. 322, 334 (2010); Judge/Hrdy, Allocation of Accumulated Resources Among Close Kin: Inheritance in Sacramento, California, 1890–1984, 13 Ethology & Sociobiology 495 (1992). 2 Hacker, supra (fn. 1), at 334.rU. rporeated/rce readily available? the proposition to include a citation for these numbers. 3 Judge, American Legacies and the Variable Life Histories of Women and Men, 6 Hum. Nature 291, 307–08 (1995). 4 We will simplify the analysis by assuming that people of a particular gender who have wills and people of the same gender who die intestate have similar preferences – but this is an assumption that ought to be tested empirically. See generally Hacker, supra (fn. 1), at 329 (noting that intestates die at a younger age than testators on average); Hirsch, Default Rules in Inheritance Law: A Problem in Search of Its Context, 73 Fordham L. Rev. 1031, 1073 (2004) (noting that intestates are poorer than testators and that this factor may engender selection effects). 5 We take for granted the law’s assumption that ordinarily the decedent’s wishes should be the overriding factor in determining how her estate will be divided. 6 See supra text accompanying fn. 68. 7 For much more on guinea pigs, see infra Section III.3.

6

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

individuals leave no wills, this advantage could be significant. Furthermore, with detailed intestate defaults, many individuals who would have otherwise needed to incur the expenses of drafting wills now may no longer need to do so. After all, they will recognize that even in the absence of a written will, their intestacy rules will be personalized and hence will more closely approximate what they would have wanted than the status quo’s one-size-fits-all approach. We are not the first to raise the possibility of using personalized default rules. 7 Recently, Cass Sunstein offered a provocative assessment of existing, impersonal default rules and two alternatives to them: active choices and personalized default rules.8 Sunstein’s work continues a conversation begun by Ian Ayres, who first argued that default rules could be “tailored” to market conditions or the attributes of parties.9 This conversation was extended by George Geis, who modeled tailored and untailored default rules under particular sets of assumptions to analyze the welfare implications of trading off precision against complexity.10 Sunstein’s bottom line is that “personalized default rules are the wave of the future. 8 We should expect to see a significant increase in personalization as greater information becomes available about the informed choices of diverse people.”11 We agree wholeheartedly and regard his contribution to the literature as significant. He astutely notes that the appeal of personalized default rules depends on the heterogeneity among a given population, the state’s access to information about individuals’ preferences and its ability to create a structure conducive to rational choices, the richness of the data available about individual preferences, and the transaction and confusion costs associated with prompting parties to a transaction to make active choices about the parameters of a deal.12 He inventively envisions personalized default rules in contexts like the choice of retirement plans, cell phone plans, mortgages, and other settings.13 That said, Sunstein’s discussion of personalized default rules is truncated – it is a short part of a short essay. He has not addressed the question of how courts would apply personalized default rules. And the earlier work by Ayres and Geis explicitly lumps together default rules that are tailored based on both contracting parties’ characteristics and market conditions, focusing – in the abstract – on the costs of promulgating and adjudicating tailored default rules.14 No scholars have previously offered a comprehensive theory of personalized default 9 rules, nor has anyone explored in detail the feasibility of such an approach. In this Article, we will develop such a theory, show its feasibility in the real world, and point out what legislatures and courts should do in order to make a personalized default-rule regime implementable in many fields. In particular, we will show that with a bit of innovative tweaking, tools developed in the age of Big Data can facilitate providing 8 Sunstein, Deciding by Default, 162 U. Pa. L. Rev. 1 (2013). Under a regime of active choice, individuals are forced to decide among various options – the contract cannot be silent with respect to a particular term. 9 Ayres, Preliminary Thoughts on Optimal Tailoring of Contractual Rules, 3 S. Cal. Interdisc. L.J. 1, 4 & fn.15 (1993); see also Ayres/Gertner, Majoritarian vs. Minoritarian Defaults, 51 Stan. L. Rev. 1591, 1593, 1596–06 (1999) (identifying several types of contracting party heterogeneity and showing how they might affect the law’s choice among defaults preferred by the majority of contracting parties or those preferred only by a minority). 10 Geis, An Experiment in the Optimal Precision of Contract Default Rules, 80 Tul. L. Rev. 1109 (2006). 11 Sunstein, supra (fn. 8), at 57. 12 Id., at 9–10. 13 See, e.g., id., at 5–7, 37, 47, 56. 14 See Ayres, supra (fn. 9) (analogizing the tailored-versus-untailored default rule dilemma to the rulesversus-standards debate); Geis, supra (fn. 10), at 1124–29 (discussing the expected costs of having tailored default rules).

Porat/Strahilevitz

7

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

10

11

12

13

certainty around the meaning of default terms for heterogeneous individuals and firms. By mitigating the uncertainty associated with the development of personalized default rules, Big Data can make personalization far more appealing than it was in previous information environments. The Article proceeds as follows. Part II explores the existing conceptions of default rules and identifies the dominant strategies for supplying such rules: majoritarian default rules and minoritarian (penalty) default rules. It then shows how each type of default rule might be improved via personalization, such that the content of the rule in question will differ among heterogeneous individuals. In this Part, we illustrate our claims mostly through consumer contracts and point out the main considerations that could make the personalized default-rules approach a viable option. Part III examines the feasibility of personalizing default rules. It observes that crude default rules – which use one readily observable characteristic, such as gender or age, to sort individuals into appropriate default rules – are already feasible but are also imprecise and can be morally problematic. We show that granular default rules, which sort individuals into several or many different default terms based on the interactions of multiple factors, are becoming increasingly feasible in the era of Big Data. Part III also examines some of the potential gains from using both crude and granular default rules in inheritance law, consumer law, the law of medical malpractice, real property law, and potentially even in labor law. A key innovation in Part III is our proposed use of “guinea pigs” to personalize defaults. Under such an approach a small portion of the population is given a great deal of information and time to make decisions, and then the remaining members of the population are assigned the default terms chosen by the guinea pigs whose observed behavior and characteristics most closely match their own. Part IV considers a number of important objections to our proposal for personalizing default rules. These objections include concerns about unfair cross subsidies; strategic behavior by consumers; uncertainty and the fragmentation of case law interpreting contractual language; the use of statistics and stereotypes; the constitutional implications of a legal regime that provides different default rules to people based on immutable characteristics; the privacy tradeoffs associated with the collection and use of information about individuals; and the flexibility of personalized default rules to deal with people whose personalities, values, and behaviors change over time. In some cases, these objections have significant force and caution against a full-throated embrace of personalized default rules. In other instances, however, we show how personalized default rules can be structured so as to mitigate potential downsides. Part V shows how the same arguments for personalized default rules also buttress the case for personalized disclosure to consumers and citizens. The present regime uses distinctly twentieth-century technologies to disclose risks, side effects, and tradeoffs to consumers and citizens. In the modern era, there is little reason to rely on these antiquated, impersonal forms of disclosure. Instead, we propose a regime of “personalized disclosure” whereby data about individual preferences, characteristics, and predilections would be employed to improve the signal-to-noise ratio of disclosures concerning products and services. Under such a regime, pregnant women would be shown prominent warnings likely to be of greatest interest to them, and septuagenarian men would likewise see only the warnings of greatest interest to them. While this is how a family physician or small-town pharmacist has historically disclosed warnings to a well-known patient, it is not the way disclosure generally works for consumer products or medical services. Our insight is that the powerful existing critiques of 8

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

disclosure remedies are not critiques of disclosure as such but rather of impersonal disclosure. Personalized disclosure is becoming increasingly achievable in the modern era, and we provide some initial thoughts on how it might be accomplished. Indeed, we believe more broadly that personalized disclosures and personalized default rules – and even personalized law in general – will become essential tools in legal regulators’ quivers in the coming decades. We even posit that personalized disclosure can ameliorate some of the complexity problems associated with a shift toward personalized default rules. The ills of personalization, it turns out, may be countered by even more personalization.

II. Theories of personalized default rules Default rules regulate much of our lives. Any transaction in which consumers, 14 merchants, employees, employers, tenants, or landlords engage will be governed by default rules. Unsurprisingly, some commentators have suggested that one of the main goals of contract law is to reduce transaction costs by providing contracting parties with default rules that apply to their transactions unless they explicitly or implicitly reject them.15 Default rules also regulate what happens after people die. When people die intestate 15 (without a will), default rules prescribed by inheritance law allocate the estate among the heirs in a certain manner.16 An individual may opt out of the default intestacy rules by leaving a will that allocates the estate differently among the heirs, but as long as she does not do so, the default rules prevail. Since many people die intestate, the content of the default rules is of the utmost importance. Here, the default rules are particularly “sticky”17 because biases and cognitive constraints prevent people from contemplating their future death.18 In this situation, therefore, the transaction costs associated with creating a will can be high. Under the most influential default-rule theory, which we discuss in detail below,19 16 default rules are aimed at decreasing transaction costs. In order for default rules to achieve this goal, they should generally track most people’s preferences and desires. If default rules do not satisfy this condition, they would increase – rather than decrease – transaction costs since most parties would opt out, which is costly. Furthermore, sometimes parties would not opt out of undesirable default rules because opting out is too costly, and therefore they would be governed by rules they would have never chosen in the absence of transaction costs. Finally, sometimes transaction costs prevent deals from being struck where a meeting of the minds would have occurred but for those costs; thus, providing the parties with default rules they prefer would reduce transaction costs and potentially facilitate deals. 15 See, e.g., Cooter/Ulen, Law and Economics, 6th edn. 2012, 341 (“Default rules fill gaps in contracts in order to reduce transaction costs.”); Shavell, Foundations of Economic Analysis of Law, at 302, fns 13–14 (2004) (arguing that courts should complete gaps in contracts using rules that are most likely to be desired by the parties in order to reduce writing costs). 16 See, e.g., Cal. Prob. Code § 6400 (West 2009); Conn. Gen. Stat. § 45a-437 (2013); Me. Rev. Stat. tit. 18-A, § 2–101 (2012). 17 See Ben‐Shahar/Pottow, On the Stickiness of Default Rules, 33 Fla. St. U. L. Rev. 651 (2006) (using the term “sticky” to define default rules in settings where the default rule is rarely changed due to high transaction costs or for other reasons, such as fear of unknown contract provisions). 18 Hirsch, Text and Time: A Theory of Testamentary Obsolescence, 86 Wash. U. L. Rev. 609, 636 (2009). (“But a testator’s failure to consider the risk of premature death is entirely plausible, psychologically.”). 19 See infra Section II.2.a).

Porat/Strahilevitz

9

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law 17

Default rules governing specific types of transactions should be tailored until finer tailoring is not cost justified, i.e., when additional tailoring will increase transaction costs.20 Although default rules are everywhere, their prominent role in contracting is particularly well understood. The next sections focus on the personalization of contract default rules, while later in the Article we apply our insights to other default rules as well.

1. Contract law default rules If contracting parties were required to agree on all the terms of their contracts, negotiation would be endless, drafting costs would skyrocket, and many efficient contracts currently executed would never result in meetings of the mind. Contract law thus provides the parties with numerous default rules that become part of their contracts unless the parties implicitly or explicitly reject them.21 For instance, under section 2–308 of the Uniform Commercial Code (“U.C.C.”), “[u]nless otherwise agreed… the place for delivery of goods is the seller’s place of business or if none, the seller’s residence.”22 The parties hence do not need to agree beforehand on the place of delivery, since as long as they do not say otherwise, delivery would occur at the seller’s place of business. And section 2–314 of the U.C.C. maintains that “[u]nless excluded or modified… a warranty that the goods shall be merchantable is implied in a contract for their sale if the seller is a merchant with respect to goods of that kind.”23 The U.C.C. then clarifies in detail what merchantability means.24 As a result, parties to a sale contract need not explicitly agree that the goods sold should be merchantable if the seller is a merchant; they also do not need to define what merchantability means – the law does it for them. 19 Remedies for breach of contract can be understood as another important source of default rules. While expectation damages are the default rule, the parties may agree otherwise, for example, by excluding or limiting liability for consequential losses or by incorporating a liquidated damages clause into their contracts.25 Indeed, the parties’ power to opt out of the “full compensation” default rule is limited: courts can strike down a liquidated damages clause as a penalty26 or use the doctrine of unconscionability 18

20 This is a necessary implication of the economic argument that “[t]he fewer the terms requiring negotiation, the cheaper is the contracting process.” Cooter/Ulen, supra (fn. 15), at 293. But see Ayres/ Gertner, Filling Gaps in Incomplete Contracts: An Economic Theory of Default Rules, 99 Yale L.J. 87, 117–18 (1989) (arguing that adopting tailored rules to fill gaps in the contract creates costs of distinguishing different types of parties and transactions); Ayres, supra (fn. 9) (arguing that when a decisionmaker creates a tailored default rule, she should account for both precision and complexity). 21 Cooter/Ulen, supra (fn. 15), at 293 (“When a court imputes terms to fill in a contract, the implicit terms apply by default, which means ‘in the absence of explicit terms to the contrary.’”); Shavell, supra (fn. 15), at 302, fns 13–14 (arguing that when parties leave gaps in the contract, courts should fill these gaps by adopting an interpretation method that minimizes the sum of writing costs and the costs of errors in the interpretation). 22 U.C.C. § 2–308 (2013). 23 Id. § 2–314. 24 Id. § 2–314(2) (detailing the conditions under which goods are considered merchantable). 25 Restatement (Second) of Contracts § 356 (1981) (stating that the parties can decide in advance the amount of damages payable in case of breach and that such an agreement replaces the court’s inquiry about the correct level of damages); Farnsworth, Contracts, § 12.18 (4th edn. 2004) (stating that parties can agree on remedial rights different from the remedies usually supplied by the courts). 26 Restatement (Second) of Contracts § 356 (stating that the parties’ power to set liquidated damages is limited, and that the liquidated damages provision must regard the principle of compensation); Farnsworth, supra (fn. 25), § 12.18 (stating that parties’ power to bargain over remedial rights is limited by the principle of compensation, which means that the stipulated sum cannot be significantly larger than the amount required to compensate the injured party for its loss).

10

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:58 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

to refuse to enforce exclusionary clauses in consumer contracts, especially when such clauses exempt the merchant from liability for bodily injury.27

2. Majoritarian default rules a) In general. Under the majoritarian default-rules theory, which is the most 20 accepted and influential theory among law-and-economics theorists, a default rule should mimic the term that the majority of the parties to whom it applies would have agreed on had they considered it as an option when making their contract.28 Thus, if most contracting parties in a sales contract prefer delivery of the goods to occur at the seller’s place, section 2–308 of the U.C.C. is the appropriate default rule. The logic behind the majoritarian default-rules theory is simple: since default rules aim to decrease transaction costs, they should fit the parties’ preferences as closely as possible. There would always be some parties that prefer a rule different from the one preferred by the majority, and these parties would then have to opt out of the default rule and incur the attendant transaction costs. But the majority of parties would not opt out, thereby avoiding the transaction costs they would have incurred in the absence of the default rule.29 A central question for the majoritarian theory is how to predict most parties’ 21 preferences. Do most parties to sales contracts prefer delivery of the goods at the seller’s or the buyer’s place? Do they prefer expectation damages or maybe just reliance damages? Law-and-economics scholars contend that most contracting parties want their contracts to reduce costs and increase benefits, thereby increasing the surplus of their contract, which they can divide among themselves.30 The majoritarian default rule should therefore be efficient. Thus, according to this view, if in most cases the costs of delivery at the seller’s place of business are lower than at the buyer’s, section 2–308 of the U.C.C. is an efficient default rule. Similarly, if full expectation damages provide more efficient incentives to the parties to perform the contract and reduce expected losses compared to reliance damages, an expectation damages default rule is superior to a reliance damages default rule.31 Note that one need not be efficiency oriented to adopt the majoritarian default rule theory; this theory is committed to one notion only – the 27 U.C.C. § 2–719(3) (“Limitation of consequential damages for injury to the person in the case of consumer goods is prima facie unconscionable.”). 28 Cooter/Ulen, supra (fn. 15), at 293–94 (arguing that courts should impute to the parties the terms in the contract that the parties would have agreed on had they negotiated the term in advance); Charny, Hypothetical Bargains: The Normative Structure of Contract Interpretation, 89 Mich. L. Rev. 1815, 1820–23 (1991) (arguing that default rules should be the most likely result of a hypothetical bargaining between the parties). 29 Cooter/Ulen, supra (fn. 15), at 294 (arguing that the efficient default rule is preferable because most parties would not wish to opt out, which would save transaction costs). 30 Brooks/Stremitzer, Remedies On and Off Contract, 120 Yale L.J. 690 (2011) (arguing that the remedy of rescission followed by restitution is socially desirable and that the parties to the contract would want it ex ante, since it incentivizes them to invest in the contract to the level that maximizes the joint surplus); Posner, The Law and Economics of Contract Interpretation, 83 Tex. L. Rev. 1581, 1588 (2005) (“Each party [to the contract] wants to maximize his gain from the transaction, and that is usually best done by agreeing to terms that maximize the surplus created by the transaction – the excess of benefits over costs, the excess being divided between the parties.”). 31 See Cooter/Ulen, supra (fn. 15), at 287–89 (arguing that expectation damages usually give better incentives to the promisor and therefore are superior to reliance damages); Shavell, Damage Measures for Breach of Contract, 11 Bell J. Econ. 466 (1980) (arguing that full expectation damages provide efficient incentive to parties to perform and thus fill gaps in the contract that involve unlikely future contingencies); Shavell, Why Breach of Contract May Not Be Immoral Given the Incompleteness of Contracts, 107 Mich. L. Rev. 1569, 1573–74 (2009) (arguing that the promisor’s option to breach and pay expectation damages is a default rule incorporated into an incomplete contract).

Porat/Strahilevitz

11

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:59 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

default rule should mimic the majority of the parties’ preferences, whatever these preferences are.32 22 Default rules can be context sensitive, which is a nod in the direction of personalization.33 Thus, even if a damages default rule is better than a specific performance default rule in total – since most contracting parties would prefer the former remedy to the latter – this might not be true in specific situations or with certain types of contracts. While the more common remedy under American contract law is damages,34 when the contract is for the sale of a unique good, courts are often willing to grant a remedy of specific performance.35 Instead of having one default rule on the choice between damages and specific performance for all contracts, there are two different default rules: one for selling unique goods and another for other contracts. But the default rules could be – and indeed they are – even more specifically tailored. And at least from an economics perspective, they should be tailored until the point where additional tailoring is no longer cost justified.36 We discuss this issue in more detail below.37 23

b) Personalized Majoritarian Default Rules. Default rules are often tailored for different types of transactions or contexts. But as far as we can tell, they are usually not tailored to the personal characteristics of the parties.38 Consider the following example: Example 1. Place of delivery. Dan is a disabled consumer who uses a wheelchair for mobility. He purchases a large-screen television from an electronics store. Should the default place of delivery be the seller’s or the buyer’s place?

24

Even if for most consumer contracts the efficient rule is delivery at the seller’s place, this is not necessarily the case in Example 1. The personally tailored default rule for consumers who use wheelchairs, and who can be easily identified as using a wheelchair, would typically be delivery at the buyer’s place, since such delivery would reduce the parties’ total costs and would therefore be their preferred option.39 Indeed, 32 See Ben‐Shahar, A Bargaining Power Theory of Default Rules, 109 Colum. L. Rev. 396 (2009) (arguing that some default rules have a distributive, rather than an efficiency, effect, and proposing criteria for giving these default rules content); Craswell, Contract Law, Default Rules, and the Philosophy of Promising, 88 Mich. L. Rev. 489 (1989) (explaining how nonefficiency theories of contract law could be the source of default rules but arguing that efficiency is the much better source). 33 See Ayres, supra (fn. 9) (arguing that when a decisionmaker creates a tailored default rule, she should find the optimal point at which the rule is specific enough but not too complex); Geis, supra (fn. 10) (modeling the simplicity–complexity dimension of default rules and suggesting that under certain assumptions a simpler, although less accurate, default rule would better reduce transaction costs). 34 Farnsworth, supra (fn. 25), § 12.8 (stating that the award of damages, measured by the injured party’s expectation, is the common form of relief for breach of contract). 35 Kronman, Specific Performance, 45 U. Chi. L. Rev. 351, 355–56 (1978) (stating that courts typically grant specific performance in contracts for the sale of a “unique” item, such as the sale of land, antiques, and patent rights); Schwartz, The Case for Specific Performance, 89 Yale L.J. 271, 272–74 (1979) (same). 36 Ayres, supra (fn. 9), at 7–9 (arguing that since more tailoring creates complexity and uncertainty, the decisionmaker needs to tailor the rule up to the point where these costs outweigh the reduction in transaction costs). 37 See infra Section IV.3. 38 In the U.C.C, there is a distinction between merchants and nonmerchants (U.C.C. § 2–104(1) (2013) defines a merchant), and some of the Code’s terms offer customized rules for merchants. See, e.g., U.C.C. § 2–314 (imposing higher warranty standards by default on merchant sales). Some commentators have advocated for different rules of interpretation for sophisticated and nonsophisticated parties. See, e.g., Schwartz/Scott, Contract Theory and the Limits of Contract Law, 113 Yale L.J. 541, 569–70 (2003) (arguing that sophisticated parties prefer textualist interpretation of contracts). 39 See Shields v. Walt Disney Parks & Resorts US, 279 F.R.D. 529 (C.D. Cal. 2011). In a motion for class certification, plaintiffs, all visually impaired visitors of the Disney resorts in California, alleged that defendants discriminated against them. Id., at 540. One of the arguments was that the defendant’s audio

12

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:59 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

with such a default rule, a seller would probably be able to charge the disabled buyer a premium for delivery. And needless to say, either party should be able to opt out of the personalized default rule if he wishes. But as long as neither does so, delivery at the buyer’s place in Example 1 could be a better default rule than the rule commonly applied to buyers who are not disabled.40 Now consider a more complicated example: Example 2: Specific performance or damages. Steven is a classic rational actor. He feels no personal attachment to property and changes his residence quite often. Sarah holds Kantian moral values regarding keeping one’s promises, feels personal attachment to property, rarely changes her place of residence and when she does, she spends months searching for the perfect place. Both Steven and Sarah entered into (separate) contracts to purchase homes from John, who is a merchant in the business of selling homes. John breaches both contracts by failing to deliver possession and title, and the question of the adequate remedy arises. Assuming everything else about the contracts is equal, except the parties’ characteristics, should the court order the same remedy for Steven and Sarah? Under current law, the answer is typically yes. A possible qualification is that if John 25 could have reasonably understood while negotiating the contracts with Steven and Sarah that Steven preferred a damages remedy and Sarah preferred specific performance, the court may take John’s understanding into account in choosing the appropriate remedy. We argue that under the assumption that John and the courts can verify the parties’ characteristics, a court ought to award damages to Steven and grant specific performance to Sarah. Indeed, John may price the contract differently for Steven and Sarah, or at least offer them different contractual terms, which would balance the additional costs that specific performance entails for the seller.

3. Minoritarian (or penalty) default rules a) In general. In a seminal article published in 1989, Ian Ayres and Robert Gertner 26 identified a second type of default rule, which they called the “Penalty Default Rule.”41 Unlike the majoritarian default rule, the penalty default rule is not aimed at mimicking the contractual term most parties prefer but instead at penalizing the party who has private information that the other party does not have. Such a penalty is designed to incentivize the party with private information to reveal this information to the party without it, thereby facilitating an efficient contract.42 description devices were designed to shut off automatically after a given time and could not be reset by visually impaired users. Id. The court analyzed this argument in terms of the device’s potential design defects. Id., at 550. One could argue that most users preferred the automatic shutdown, thus making it the majoritarian default rule, while the plaintiffs were seeking to impose on the defendant a personalized default rule for visually impaired visitors. 40 Business practices in American grocery stores track this default rule to some extent. A grocery bagger is likely to ask an elderly customer with a large order whether she would like assistance unloading groceries into her car, but he would probably not bother asking a twenty-year-old who has purchased a box of corn flakes and a magazine the same question. 41 Ayres/Gertner, supra (fn. 20), at 91 (“Penalty defaults [are defaults that] are designed to give at least one party to the contract an incentive to contract around the default rule and therefore to choose affirmatively the contract provision they prefer.”). 42 On other occasions, a penalty default rule would penalize both parties for concealing information that makes the determination of their dispute easier for courts; in this way, Ayres and Gertner explain the then–U.C.C. § 2–201 zero-quantity provision, under which, if the parties have not agreed on the quantity, the courts would not fill in this gap and would not enforce the contract. Id., at 95–100, fn. 43.

Porat/Strahilevitz

13

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:59 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

An example penalty default rule used by Ayres and Gertner was the foreseeability requirement of Hadley v. Baxendale.43 Under this requirement, the aggrieved party is only entitled to compensation for foreseeable losses. Ayres and Gertner explain that without the foreseeability limitation on liability, an aggrieved party with unforeseeable losses would hide this information from the other party.44 The foreseeability limitation penalizes an aggrieved party that hides the information by barring recovery for his unforeseeable losses in case of a breach.45 In particular, if the aggrieved party is not the cheapest cost avoider or the cheapest insurer of his unforeseeable losses, he would disclose the potential losses to the other party. This disclosure renders the losses foreseeable, and the other party would take them into account in deciding whether to enter into the contract, how much to invest in precautions, and whether to perform or breach.46 28 Commentators criticize Ayres and Gertner’s penalty default rules theory from several angles. They argue that a penalty default rule would not necessarily force a contracting party to reveal private information because such a move might directly contradict its bargaining strategy47 or because the party might benefit from being pooled together with other parties, which could allow it to externalize costs to them.48 Eric Posner prominently argues that there are no penalty default rules in contract law, nor should there be any.49 This is because both majoritarian default rules and penalty default rules force contracting parties with private information, which prefer to opt out of the default rule, to reveal their private information to the other party, who would offer them a different contract in exchange.50 Opting out is costly, so a majoritarian default rule would function better than a penalty default rule, since it encourages fewer parties to opt out. It is possible that the minority’s total costs of opting out would exceed the majority’s total costs of doing so, but this is an unlikely scenario.51 29 We might better understand a penalty default rule as a species of minoritarian default rule, as Ayres and Gertner in fact acknowledge in an essay they published a decade after 27

43

Hadley v. Baxendale, (1854) 156 Eng. Rep. 145; 9 Ex. 341. Ayres/Gertner, supra (fn. 20), at 101–04. 45 Id. (arguing that the decision in Hadley is an example of a penalty default rule). 46 See Craswell, Contract Remedies, Renegotiation, and the Theory of Efficient Breach, 61 S. Cal. L. Rev. 629 (1988) (describing the various stages where the promisor takes decisions and incentives matter); Ulen, The Efficiency of Specific Performance: Toward a Unified Theory of Contract Remedies, 83 Mich. L. Rev. 341 (1984) (describing different remedies for breach of contract and examining when parties may take them into account). 47 Johnston, Strategic Bargaining and the Economic Theory of Contract Default Rules, 100 Yale L.J. 615, 617 (1990) (arguing that the Hadley default penalty rule will not incentivize promisees to reveal private information since revealing the value the promisee ascribes to the contract with the promisor would allow the promisor to raise the contract price substantially). 48 Adler, The Questionable Ascent of Hadley v. Baxendale, 51 Stan. L. Rev. 1547, 1551 (1999) (arguing that parties with private information would not reveal their types when they benefit from the cross subsidization entailed by pooling them with other parties). 49 Posner, There Are No Penalty Default Rules in Contract Law, 33 Fla. St. U. L. Rev. 563, 586–87 (2006). 50 Id., at 569–73. 51 Id., at 573 (arguing that examples of penalty default rules are either not default rules at all or can be explained by the majoritarian default rule theory); see also Bebchuk/Shavell, Reconsidering Contractual Liability and the Incentive to Reveal Information, 51 Stan. L. Rev. 1615, 1619–26 (1999) (assessing the Hadley default rule of limited liability and arguing that the rule entails some costs in different situations and should be adopted only where the parties would have most likely wanted it in advance, which makes it a majoritarian default rule); Johnston, supra (fn. 47), at 622–23 (arguing that the Hadley rule might be preferable to the parties ex ante and thus not a penalty rule). 44

14

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:59 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

they first proposed the penalty default rule idea.52 We believe that at least as personalized default rules are concerned, there could be minoritarian default rules, as we explain more fully below. But we also suspect that the rise of Big Data (described in Part III) will make penalty default rules less important, since firms are gaining access to a treasure trove of information about individual consumers. b) Minoritarian default rules as facilitators of personalized default rules. Minoritarian default rules could facilitate personalized majoritarian default rules. If sellers and courts have full information about buyers, default rules aimed at forcing buyers to reveal private information will be meaningless. Sellers and courts, however, often do not have full information about buyers’ preferences, characteristics, and traits, and personally tailoring default rules for them seems impractical. A default rule would encourage buyers to reveal their preferences, characteristics, and traits to sellers, either for a specific transaction or for many future transactions, by penalizing those buyers who could cheaply convey such information yet fail to do so. Consider again Example 2 (Specific performance or damages). Suppose sellers and courts cannot distinguish accurately between Steven and Sarah, and therefore it is impossible to tailor personalized default rules. Nevertheless, a default rule of damages could still change the outcome. If Sarah is aware of the damages default rule, she will reveal her preferences for specific performance to the seller or, alternatively, reveal her characteristics and traits to him, from which he would be able to deduce that unless they agree otherwise, her remedy will be specific performance. Thus, the damages default rule will penalize Sarah if she does not convey information to the seller about her preferences or characteristics. Could specific performance function in the same way? Under a specific performance default rule, Steven would arguably reveal neither his preference for damages nor his characteristics and traits because he is no worse off with specific performance than with a damages remedy. Although he is indifferent about the remedy, he may be better off with specific performance, since it would improve his bargaining position vis-à-vis the seller, for whom specific performance is typically more burdensome.53 But this analysis is incomplete. If the seller is able to structure the contract to reward buyers who are entitled to the less burdensome remedy, then both damages and specific performance could function effectively to force buyers to reveal their preferences, characteristics, and traits. Specifically, while a damages default rule would penalize Sarah ex post if she does not reveal her preferences or characteristics, specific performance would penalize Steven ex ante (higher price or less favorable contractual terms) if he does not reveal his preferences or characteristics. The choice between damages and specific performance should therefore hinge on the empirical question of who bears the lower costs in revealing their preferences or characteristics: Steven or Sarah? If the answer is Steven, specific performance should be the more efficient default rule, and if it is Sarah’s, damages should be the most efficient default rule. If there are more “Stevens” than “Sarahs” among buyers but it is much less costly for the “Stevens” than for the 52 Ayres/Gertner, supra (fn. 20), at 1600–02, 1606 (explaining that the penalty default rule is one type of minoritarian default rule, which is efficient when it is less costly for the majority to opt out than it is for the minority to do so). 53 See Schwartz, supra (fn. 35), at 274 (arguing that if damages are fully compensatory, adding the option of specific performance creates an opportunity for the promisee to exploit the promisor by threatening to compel performance when costs of performance are higher than the damages); Craswell, supra (fn. 46), at 636–38 (noting that specific performance, like overcompensatory remedies, has the potential to discourage efficient breaches).

Porat/Strahilevitz

15

30

31

32

33

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:32:59 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

“Sarahs” to reveal their preferences or characteristics, specific performance could be the efficient (minoritarian) default. 34 Under our personalized default rules theory, parties do not directly negotiate the terms of the contract but instead reveal information about their characteristics and traits, which in turn affect the contents of the default rules applied to them. This information could often be private and even confidential: not every sensitive, neurotic buyer would like to reveal these attributes to a seller. In other words, for some types of characteristics and traits, the default rules could be stickier than for others, and the people possessing the former characteristics and traits could be the minority. In the same way, some types of parties may have significant cognitive limitations or biases that would make it especially burdensome to reveal private information about their preferences, and these parties could be the minority. In such a situation, a minoritarian default rule could similarly work better than a majoritarian one.

4. Third-party effects The third approach to determining the content of default rules requires that the rules maximize social welfare generally, not just the welfare of the contracting parties. Contract law often takes negative effects on third parties as a central consideration in enforcing contracts. For example, an entire chapter of the Restatement (Second) of Contracts is dedicated to “Unenforceability on Grounds of Public Policy.”54 This chapter, however, is not about default rules but instead about mandatory, immutable rules: naturally, the parties are not allowed to opt out of these rules. Contract law doctrines, however, only rarely take positive effects on third parties into account,55 and externalizing benefit default rules are rare.56 36 In some instances, the personalization of default rules may produce benefits to third parties, or positive externalities, and the desire to promote such externalities may convince society to embrace personalization. For example, many jurisdictions confront the dilemma of how to encourage people to donate their organs after death to save other people’s lives. A possible solution is to have a default rule that is expected to be quite sticky: most people would not opt out, whatever the default rule is.57 Assuming the social goal is to find an optimum between fulfilling people’s wishes and third parties’ benefits (if these benefits were the only issue, a mandatory rule of donation would be the optimal solution58), tailoring personalized default rules to different groups in society could be an optimal solution. Thus, if there are groups in society – say, adherents of Shintoism – who are expected to object to organ donations and would opt out of any default rule that allows them,59 a no-donation default rule is the desirable one for them, since it would save the transaction costs of opting out. But if there are other groups in society that might weakly oppose donation but would not be willing to incur the costs of opting out, applying a default rule that is not majoritarian but balances possible donors’ weak preferences against the possible recipients’ strong preferences could better achieve 35

Restatement (Second) of Contracts § 178–199 (1981). See, e.g., id. § 207 (“In choosing among the reasonable meanings of a promise… a meaning that serves the public interest is generally preferred.”). But see Zamir, The Inverted Hierarchy of Contract Interpretation and Supplementation, 97 Colum. L. Rev. 1710, 1723–24 (1997) (“Despite [section 207’s] broad formulation, it is assumed that this rule only applies to contracts ‘which affect a public interest.’”). 56 See Ayres/Gertner, supra (fn. 20), at 1598–99 (discussing default rules which create positive externalities). 57 See Sunstein, supra (fn. 8), at 12–13. 58 See Strahilevitz, The Right to Destroy, 114 Yale L.J. 781, 807–08 (2005). 59 See Steinbuch, Kidneys, Cash, and Kashrut: A Legal, Economic, and Religious Analysis of Selling Kidneys, 45 Hous. L. Rev. 1529, 1566 n.268 (2009). 54 55

16

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

the social goal. Indeed, rules could be personalized to benefit the interest groups that in the absence of personalization would most forcefully oppose a rule that benefits third parties. This personalization would dampen political opposition. The discussion above highlights the applicability of our proposal for personalized 37 default rules to any concept of default rules that attributes at least some significance to the preferences of the individuals making the choices. Take our inheritance example60: suppose that under a given legal system the decedent’s preferences are not the only criterion in allocating assets, but other social goals are also considered important. As long as individuals are free to make wills, their preferences are legally significant. The law could then tailor personalized default rules that take into account people’s preferences, which vary among individuals, together with social goals, which are common to everyone. Thus, individuals with strong preferences would often make wills, and their preferences would govern the allocation of their estates. By contrast, individuals whose preferences are weaker would not make wills, and their estates would be allocated according to personalized default rules that are tailored to those people’s preferences, together with the social goals that the law considers important.

III. The feasibility of personalized default rules Part II showed how majoritarian and penalty default rules might be personalized. The 38 discussion so far implicitly has contemplated two different sorts of personalized default rules: crude and granular default rules. Crude personalized default rule takes a particular, observable characteristic, and 39 sorts individuals into different legal defaults based on whether they possess that characteristic. For example, if the state observes that men and women have systematically different preferences for how their estates should be divvied up among heirs, then the law might create one set of intestacy rules for men and another for women.61 Gender is easily observable, so the costs of determining which set of intestacy rules applies will be low. Indeed, inheritance law already employs gender-sensitive mandatory rules, as in the case of a few state dower statutes that are designed to protect the economic interests of widows, whom legislatures perceive to be more economically vulnerable than widowers following the death of a spouse.62 The preference-insensitive nature of these rules and their potential to reinforce stereotypes help make their persistence controversial. When a characteristic like age or gender becomes the basis for a waivable default 40 rule, concerns about limiting testamentary freedom disappear and antistereotyping considerations may be ameliorated. Thus, personalized default rules are already employed in some contexts. For example, some employers have implemented automatic enrollment for 401(k) retirement accounts, with a contribution level that is personalized based on the worker’s age.63 Furthermore, there are proposals to provide 60

See supra fns 1–5 and accompanying text. Assume for present purposes that such classifications are legally permissible, although this assumption may be unreasonable. We discuss the issue further infra in Sections IV.6.–7. 62 See, e.g., In re Estate of Miltenberger, 737 N.W.2d 513 (Mich. Ct. App. 2007) (rejecting equal protection challenges to Michigan’s statute providing that widows – but not widowers – are entitled to at least one-third of a deceased spouse’s land holdings). But cf Stokes v. Stokes, 613 S.W.2d 372, 376 (Ark. 1981) (invalidating a gender-sensitive dower statute on equal protection grounds). See generally Muller, Haven’t Women Obtained Equality? An Analysis of the Constitutionality of Dower in Michigan, 87 U. Det. Mercy L. Rev. 533 (2010) (surveying gender-sensitive dower statutes in various states). 63 We thank Lauren Willis for this example. 61

Porat/Strahilevitz

17

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

workers with investment vehicles that are tailored to the number of years they expect to continue working.64 Similarly, new cars can sense the weight of a passenger in the front seat and disable the air bag to reduce the risk of injury to a lighter passenger if the air bag is deployed in a crash.65 We can refer to approaches that use a single variable like gender, age, or weight as the sole basis for tailoring as “crude personalized default rules.” 41 Greater personalization is possible. Suppose that politically conservative and politically liberal women have different preferences with respect to the division of their estates. Suppose further that politically conservative women from cities and rural areas systematically differ in the way they prefer to divide their estates. In theory, there are a multitude of possible personalized default rules. Nevertheless, regularities exist, and the task of using these regularities to establish sufficiently large groups of like-minded individuals who can be assigned the same set of default rules implicates a tradeoff between precision and complexity.66 We will refer to precise default rules that employ many characteristics about individuals – including their past behaviors in similar circumstances – to predict the contractual or testamentary terms they would have opted for as “granular personalized default rules.” 42 The feasibility of employing crude personalized default rules is a straightforward matter. We need only show that a particular characteristic accurately predicts future behavior. That said, we will show why using crude personalized default rules is often less desirable than employing granular personalized default rules. In this Part, we therefore will focus on the feasibility of the granular defaults.

1. Big Data and Big Five An apparent hurdle in creating personalized default rules is the issue of conveniently identifying relevant default rules both ex ante by the parties and ex post by the courts. Suppose a legal dispute has arisen concerning ambiguity in a contract. Once the nature and the stakes of the dispute are clear to both parties, each will have an incentive to argue that she is the type of person who ought to be entitled to the personalized default rule that would cause the court to rule in her favor. In Example 2, both Steven and Sarah will argue that they are the types of people entitled to specific performance if this remedy creates an entitlement that strengthens their bargaining position relative to John’s.67 Is there a reliable way to prevent these problems of proof? We believe that in the era of Big Data, the answer to this question is yes. 44 Big Data is commonly defined as the process whereby computers sift through enormous quantities of data to identify patterns that can predict individuals’ future behavior.68 It depends on the combination of gigantic databases (typically cataloging consumer behavior) with predictive analytics. Firms spent $28 billion on Big Data in 2012, a number that was estimated to have grown to $34 billion in 2013.69 To put this 43

64

See Butler, American Paternalism and the One Fund Solution, 9 Wyo. L. Rev. 485, 521 (2009). Smith et al., Smart Defaults: From Hidden Persuaders to Adaptive Helpers, 18 (INSEAD Working Paper Series, Paper No. 2009/03/ISIC, 2009) available at https://flora.insead.edu/fichiersti_wp/inseadwp2009/2009–03.pdf. This proposal for “smart defaults” is similar to the “personalized defaults” approaches discussed in the legal literature, although their discussion of smart defaults is quite abbreviated. 66 See infra Section IV.4. 67 See supra Section II.2.b) and text accompanying (fn. 53). 68 Strahilevitz, Toward a Positive Theory of Privacy Law, 126 Harv. L. Rev. 2010, 2021 (2013). 69 Kolakowski, Big Data Spending Will Hit $28 Billion in 2012: Gartner, Slashdot (17 October 2012), http://slashdot.org/topic/bi/big-data-spending-will-hit-28-billion-in-2012-gartner/. 65

18

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

$28 billion number in perspective, it is an amount equal to the annual Gross Domestic Product of Jordan or Latvia70 yet with greater growth potential. What did all that money purchase? It is hard to know for sure since many uses of Big 45 Data are being kept as proprietary trade secrets. But in the past year or two, the news media has reported on a dizzying array of industries applying the tools of Big Data. Facebook’s new “graph search” feature seeks to employ the company’s Big Database to better predict which search results will be most useful to individuals who type in search queries.71 Big Data is a big industry in higher education.72 Big Data is a big business in medicine.73 It is all the rage in insurance.74 Researchers have shown how they can predict an individual’s race by analyzing online behavior75 and can make accurate predictions about an individual’s ideology by monitoring her television viewing habits.76 And the campaign to reelect President Obama was lauded (and criticized) for its sophisticated use of Big Data techniques to identify and energize the president’s partisans.77 These technologies have been employed to help businesses find customers who are profitable, patients who need special care, voters who are persuadable, and insureds who present good risks.78 Even brick and mortar outfits with familiar business models are using data-driven 46 strategies to personalize service in a way that will appeal to their customers. For example, restaurants are increasingly assembling dossiers on customers to enable them to remember whether particular patrons prefer black or white napkins and red or white wine.79 This information can then be shared with partner restaurants via Opentable. com’s reservation database.80 With the benefit of this data, savvy restaurants can provide a first-time diner with the same sort of personalized service that regulars from the neighborhood have long come to expect. Law is perhaps the primary major industry in which the effects of Big Data have not 47 been widely documented, although that is beginning to change, according to a recent 70 GDP – Countries – List, Trading Economics, http://www.tradingeconomics.com/country-list/gdp (last visited 13 February 2014). 71 See Sengupta/Cain Miller, Search Option from Facebook Is Privacy Test, New York Times, 19 January 2013, at A1, available at http://www.nytimes.com/2013/01/19/technology/with-graph-searchfacebook-bets-on-more-sharing.html. 72 See Perry, Please Be eAdvised, New York Times, 22 July 2012, at ED24, available at http://www. nytimes.com/2012/07/22/education/edlife/colleges-awakening-to-the-opportunities-of-data-mining.html. 73 See, e.g., Harris, Better Medicine, Brought to You by Big Data, GigaOM (15 July 2012, 6:00 AM), http://gigaom.com/cloud/better-medicine-brought-to-you-by-big-data/?utm_source=social&utm_medium=twitter&utm_campaign=gigaom. 74 See Sullivan, Credit Ratings Aid Marketers in Targeting Ads, Media Post News (17 August 2012, 4:29 PM), http://www.mediapost.com/publications/article/181075/credit-ratings-aid-marketers-in-targetingads.html#axzz2F2rHWmiH. 75 Croll, Big Data Is Our Generation’s Civil Rights Issue, and We Don’t Know It, O’Reilly Radar (2 August 2012), http://radar.oreilly.com/2012/08/big-data-is-our-generations-civil-rights-issue-and-wedont-know-it.html. 76 Carter, Republicans Like Golf, Democrats Prefer Cartoons, TV Research Suggests, New York Times Media Decoder (11 October 2012, 7:42 PM), http://mediadecoder.blogs.nytimes.com/2012/10/11/republicans-like-golf-democrats-prefer-cartoons-tv-research-suggests/?smid=tw-nytimes. 77 See Parsons/Hennessey, Obama’s Math Geeks Raised the Odds, L.A. Times, 14 November 2012, at A8, available at http://articles.latimes.com/2012/nov/13/nation/la-na-obama-analytics-20121113. 78 See, e.g., Singer, Shoppers, Meet Your Scorekeeper, New York Times, 19 August 2012, at BU1, available at http://www.nytimes.com/2012/08/19/business/electronic-scores-rank-consumers-by-potential-value.html (profiling eBureau, a technology company that uses data mining to determine which individuals are likely to be profitable customers for firms). 79 Craig, Getting to Know You, New York Times, 5 September 2012, at D1, available at http://www. nytimes.com/2012/09/05/dining/what-restaurants-know-about-you.html?pagewanted=all&_r=0. 80 Id.

Porat/Strahilevitz

19

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

article by Professor Katz.81 Katz identifies many applications of Big Data to the legal profession, suggesting its utility in predicting legal costs at the outset of a case, predicting outcomes in litigation, helping firms hire the right attorneys, and managing the discovery process.82 Our proposal suggests a different way in which the legal system can leverage the benefits of Big Data. Under certain circumstances, we want the courts (and advocates in the courtroom) to embrace the science of Big Data as a means of deciding what terms ought to be imported into an ambiguous contract or will. Furthermore, we propose that parties often would be able to use Big Data to predict beforehand what default rules will be applied to their contracts. We later explain in more detail how that would work, but before we do so, a few more words on Big Data and the “Big Five” personality characteristics are in order. 48 Journalists writing about Big Data have spilled much more ink discussing its proliferation than what makes it effective. At bottom, we believe a major reason why Big Data enables firms and government entities to predict future behavior is that patterns of purchases, mouse clicks, credit payments, and social network ties reveal fundamental aspects of individuals’ personalities and values.83 49 Psychologists understand human behavior largely in terms of the Big Five personality characteristics: extraversion, euroticism, agreeableness, conscientiousness, and openness.84 An enormous psychological literature has identified ways in which particular personality traits are more pronounced among people who engage in particular sorts of behaviors.85 For example, people who score highly on extraversion are more likely to disclose information about themselves on social networks,86 and people who score highly on conscientiousness are more likely to be politically conservative.87 Other research suggests that American college students score noticeably higher on personality tests measuring agreeableness than do their Western European counterparts.88 The Big Five is not the only valid framework for assessing personality. Various psychologists have categorized different dimensions of personality, such as authoritarian personality traits,89 which we will address in turn.90 81 Katz, Quantitative Legal Prediction – or – How I Learned to Stop Worrying and Start Preparing for the Data-Driven Future of the Legal Services Industry, 62 Emory L.J. 909 (2013). 82 Id. 83 Strahilevitz, supra (fn. 68), at 2022–24. 84 Barrick/Mount, The Big Five Personality Dimensions and Job Performance: A Meta-Analysis, 44 Personnel Psychol. 1, 1–5 (1991). 85 The legal literature employing the Big Five analysis in a sophisticated way, by contrast, is relatively sparse. For examples of successful interdisciplinary work of this sort, see Green/Kugler, When Is It Wrong to Trade Stocks on the Basis of Non-Public Information? Public Views of the Morality of Insider Trading, 39 Fordham Urb. L.J. 445 (2011), and Stevenson/Caldwell, Personality in Juror DecisionMaking: Toward an Idiographic Approach in Research, 33 Law & Psychol. Rev. 93 (2009). Although it characterizes individuals in a way that diverges somewhat from the Big Five framework, the Cultural Cognition Project has done the most influential legal work applying research about personality heterogeneity to legal problems. See, e.g., Kahan et al., Cultural Cognition and Public Policy: The Case of Outpatient Commitment Laws, 34 Law & Hum. Behav. 118 (2010). 86 Chen/Marcus, Students’ Self-Presentation on Facebook: An Examination of Personality and SelfConstrual Factors, 28 Computers in Hum. Behav. 2091, 2097 (2012); Ryan/Xenos, Who Uses Facebook? An Investigation into the Relationship Between the Big Five, Shyness, Narcissism, Loneliness, and Facebook Usage, 27 Computers in Hum. Behav. 1658, 1662 (2011). 87 Carney et al., The Secret Lives of Liberals and Conservatives: Personality Profiles, Interaction Styles, and the Things They Leave Behind, 29 Pol. Psychol. 807, 824 (2008). 88 Schmitt et al., The Geographic Distribution of Big Five Personality Traits, 38 J. Cross-Cultural Psychol. 173, 185 tbl.2 (2007). 89 See, e.g., Duckitt, Authoritarianism and Group Identification: A New View of an Old Construct, 10 Pol. Psychol. 63 (1989). 90 See infra Section III.2.b).

20

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

By employing Big Data, firms have found a substitute for administering complex 50 personality tests to each potential customer to identify her quirks and predilections.91 Because these firms are using publicly available data and proprietary data that is bought and sold in the marketplace, they can dispense with obtaining the consent of the individuals whose behavior they are studying. Moreover, because they will be studying a consumer’s revealed preferences rather than her responses to surveys (which might be slanted in ways the consumer believes will benefit her), firms may justifiably view the results of these quasipersonality tests as particularly reliable metrics. To be sure, we are not suggesting that the Big Five research unlocks every behavioral 51 mystery – the extant data suggest otherwise.92 Rather, our more modest claim is that personality profiling identifies many powerful tendencies among individuals and groups. Personality’s causal role in determining human behavior might also help alleviate some of the “black-box” concerns about Big Data’s use. Without some theory of why a consumer’s decision to purchase a particular product at Time 1 helps explain that same consumer’s creditworthiness at Time 2, there is a greater risk that the correlation between Time 1 and Time 2 behaviors is spurious and will not be useful in prediction.93 Worse yet, an unexplained, black-box correlation might be driven by considerations that the legal system has chosen to render illegitimate.94 A fascinating article by Gokul Chittaranjan, Jan Blom, and Daniel Gatica-Perez shows 52 the promise and potential of using data mining to identify individuals’ personality profiles.95 These three scholars administered personality tests to scores of Swiss smartphone users and then monitored the users’ smartphone activity over the next seventeen months. They found many significant correlations between particular personality traits and observed smartphone behavior. If you have a person’s cell-phone data and you know what to look for, you know a lot about what makes her tick. Along the way the tests showed that as a practical matter, it is easy to analyze automatically smartphone usage data to predict the personalities of individual phone users. The scholars summarized some of their main findings as follows: “The results clearly show that several aggregated smart-phone usage features could be predictive of the Big-Five personality traits… It was found that extraverts, who are characterized by talkativeness and outgoing nature, were more likely to receive calls and also spend more time on them… Agreeableness among females was associated with an increase in the number of incoming calls. Agreeable males were found to communicate with more number of unique contacts through voice calls. On the other hand, conscientiousness was associated with higher usage of the Mail app that could be used in a professional context and with lower usage of the Youtube application, which is likely to be used for entertainment purposes. Conscientious users were also likely to contact lesser number of unique people through voice calls. This conforms with their characterization in the literature as responsible and organized individuals. Interestingly, emotional stability was linked to higher incoming SMS. And high openness was associated with increased usage of Video/Audio/Music apps in females 91

Strahilevitz, supra (fn. 68), at 2023. See, e.g., Stevenson/Caldwell, supra (fn. 85), at 110–11. 93 Cf Bernstein, The Hidden Costs of Terrorist Watch Lists, 61 Buff. L. Rev. 461, 473–74 (2013) (discussing related problems in the context of governmental use of Big Data to predict national security threats). 94 See Tene/Polonetsky, Judged by the Tin Man: Individual Rights in the Age of Big Data, 11 J. on Telecomm. & High Tech. L. 351, 366 (2013). 95 Chittaranjan et al., Mining Large-Scale Smartphone Data for Personality Studies, 17 Pers. & Ubiquitous Computing 433 (2013). 92

Porat/Strahilevitz

21

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

and also with the usage of nonstandard calling profiles such as Beep and Ascending in the entire population.”96 This is an extraordinarily rich set of findings, and it suggests that Verizon, AT&T, Apple, Samsung, and other major firms in the cell-phone industry possess a treasure chest of personal information about their customers. Yet legal scholars have ignored their research. A follow-up Big Data project, in which a team of researchers from the Massachusetts Institute of Technology and the University of Trento analyzed social network ties and personalities of cell-phone users, suggests that in many respects behavioral data from smartphones can better predict individuals’ personalities than personality surveys themselves.97 This research confirms that behavioral data can predict personality, and we already know from the psychology literature that personality can predict behavior. The iPhones, not the eyes, turn out to be the windows into the soul. Is it any wonder that Google has gotten into the business of making its own cell phone?98 54 Smartphones are not the only Big Data tool that researchers can use to unlock individuals’ personality scores. In the last few years, there has been an explosion of research using the same tools to discern Big Five personality traits from social media usage data.99 Researchers at Microsoft and Cambridge University produced a thorough working paper in which they develop a model that can predict 33 % of the variation in extraversion, 26 % of the variation in neuroticism, and 17 % of the variation in conscientiousness through automated analysis of individuals’ Facebook activity.100 Evidently, what is true of smartphones is true of social networking platforms. And as the world moves toward an Internet of Things,101 it is conceivable that our televisions, refrigerators, and cars might be used to reveal aspects of our personalities that will help marketers predict our future behavior. 55 To be sure, sometimes Big Data has predictive power because it teases out regularities that have little to do with personality. For example, Target Corporation’s data miners identified a pattern whereby their female customers who suddenly started purchasing multivitamins and lotion were buying cribs and newborn diapers six months later.102 53

96

Id., at 449. Staiano et al., Friends Don’t Lie: Inferring Personality Traits from Social Network Structure, U. Trento Department Info. Engineering & Computer Sci. (Sept. 2012), http://disi.unitn.it/~staiano/ pubs/SLAPSP_UBICOMP12.pdf (“[W]e believe that our results have provided compelling evidence that mobile phones-based behavioral data can be superior to survey ones for the purposes of personality classification.”). 98 Pogue, Look Ma! A No-Hands Smartphone!, New York Times, 7 August 2013, at B1, available at http://www.nytimes.com/2013/08/07/technology/personaltech/the-moto-x-from-google-iphones-latestchallenger.html?pagewanted=all. 99 See, e.g., Hughes et al., A Tale of Two Sites: Twitter vs. Facebook and the Personality Predictors of Social Media Usage, 28 Computers in Hum. Behav. 561 (2012); Qiu et al., You Are What You Tweet: Personality Expression and Perception on Twitter, 46 J. Res. Personality 710 (2012); Stoughton et al., Big Five Personality Traits Reflected in Job Applicants’ Social Media Postings, 16 Cyberpsychology, Behav. & Soc. Networking 800 (2013); Farnadi et al., Recognizing Personality Traits Using Facebook Status Updates, Proc. Seventh Int’l AAAI Conf. on Weblogs & Soc. Media 14 (2013); Quercia et al., Facebook and Privacy: The Balancing Act of Personality, Gender, and Relationship Currency, Proc. Sixth Int’l AAAI Conf. on Weblogs & Soc. Media 306 (2012). 100 Bachrach et al., Personality and Patterns of Facebook Usage, Proc. 3rd Ann. ACM Web Sci. Conf. 24 (2012). 101 See Peppet, Freedom of Contract in an Augmented Reality: The Case of Consumer Contracts, 59 UCLA L. Rev. 676, 699 (2012) (defining the Internet of Things as “a connected web of ‘smart’ objects capable of generating and transmitting data on themselves”). 102 Duhigg, Psst, You in Aisle 5, New York Times, 19 February 2012, (Magazine), at MM30, available at http://www.nytimes.com/2012/02/19/magazine/shopping-habits.html?_r=0. 97

22

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

Through analytics, Target realized that multivitamin and lotion purchases were an indicator that the woman was pregnant, which she might not otherwise reveal to Target. The company used this information to its advantage and focused its efforts on keeping a new mother’s business despite her life-changing event of pregnancy, which marketing research has shown to disrupt buying behavior. If Target could make new moms into loyal customers, there was a greater chance that it could keep them as customers in the following years.103

2. Big Data in the law Big Data can be used to predict future behavior because the process of studying an 56 individual’s purchases, online searches, borrowing activity, and social network composition reveals aspects of that individual’s personality and preferences. Of course, it is one thing for well-capitalized firms and political campaigns to employ analytics at a high level and another for lawyers or judges to duplicate the sophisticated processes. The institutional-competence concerns are legitimate, especially at the present time, when courts have developed no expertise in profiling or in Big Data generally. There are several possible avenues by which Big Data could help personalize legal default rules: (1) Firms could use what they know about their customers to provide them with personalized default terms and prices in contracts that are determined at the time a contract is entered into and which any customer could see before she executes the contract. (2) Federal governmental regulatory agencies like the Consumer Financial Protection Bureau could identify particular default contractual provisions that are well suited to particular types of consumers, and require firms to offer the terms to customers with those profiles. These terms would also be specified at the time the contract is entered into and any customer (or firm) could see them before contract execution. (3) In cases where contracts are ambiguous or silent, courts could determine ex post what terms the parties probably would have specified in light of similarly situated parties’ choices and preferences. The legally relevant issue would be the parties’ observable characteristics and traits at the time they entered into the contract (or, in the case of probate matters, at the decedent’s death or the moment when the decedent lost decisional capacity), as well as the parties’ past behavior. Courts would rely on expert testimony to discern the contents of these personalized rules. In most disputes involving incomplete contracts or ambiguous terms, the parties would settle in the shadow of the (personalized) law. That is, they would anticipate how a court would most likely resolve their dispute if it were fully litigated and settle around that expected outcome. Due to concerns about institutional competence, we believe that the first and second 57 approaches are the most feasible and appropriate, although we shall spend some time discussing the third approach as well. We have already shown how Big Data and personalizing default rules could change 58 the law of inheritance.104 Let us now consider other important applications. a) Consumer contracts. Consumer law is perhaps the most natural field in which to 59 apply the personalized default rules approach. As we have explained, firms have an 103 104

Id. See supra text accompanying fns 1–5.

Porat/Strahilevitz

23

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

increasingly enormous amount of data on consumers’ preferences and characteristics, and they can use this data to tailor different default rules for their contracts. The parties can use this same data to settle disputes in the shadow of the law and courts can use it for adjudicating unsettled disputes. Since consumers are generally aware of their characteristics and traits, they will find the personalized default rules more predictable than the impersonal default rules currently applied to their contracts. A consumer who does not know what default rules apply to her contract would be able to easily verify it before or after the transaction takes place, either through the merchant who could provide her with the content of those rules at the point of sale or through intermediaries who could generate such information at low cost. 60 Consider Example 1 (Place of delivery), which suggests that while a default rule for consumers who are not disabled could be “delivery at the seller’s place of business,” a “delivery at the buyer’s residence” might be a better default rule for disabled consumers. There is little need for data to employ a personalized default rule approach in this case, and we would not be surprised to see courts reaching the same result through interpretation techniques. 61 In some industries, a default rule of “delivery at the buyer’s residence” could be an efficient minoritarian default rule that would facilitate personalized default rules. Thus, a store that sells medical equipment might have a relatively high number – but still a minority – of consumers who are disabled. Some of the disabilities may be visually hidden, and the disabled consumers might prefer not to disclose their disabilities verbally, especially if other customers are nearby. A default rule of “delivery at the buyer’s place” would encourage nondisabled consumers to ask for delivery at the seller’s place, with a possible price discount. b) Organ donation. The United States faces a severe shortage of organs for transplantation. About eighteen Americans die every day because of a shortage of organs available for lifesaving transplants, and nearly 114,000 Americans were on organ or tissue transplantation waiting lists as of 2012.105 In the United States, organ donors must opt in to organ donation, whereas in other nations people must opt out of donation, a design choice that has a very powerful effect on the prevalence of organ donation.106 63 We think the United States default rule on organ donations is almost certainly indefensible on welfarist grounds. Yet the rule persists. Perhaps personalization presents a superior alternative to the status quo’s impersonal opt-in rule. Although the question has not been studied extensively, some of the available research demonstrates that certain personality traits correlate with organ donation. For example, one Iranian study found that highly agreeable individuals were more likely to agree to donate the organs of a brain-dead loved one.107 A Turkish study found that conscientiousness and willingness to become an organ donor were significantly correlated, and the authors developed a personality model that could explain nineteen percent of the respondents’ variation in their intentions to become an organ donor.108 By contrast, a larger-scale

62

105 Evangelista, Online Organs, San Francisco Chronicle, 2 May 2012, at D1, available at http://www. sfchronicle.com/business/article/Facebook-hopes-to-expand-organ-donation-awareness-3525917.php. 106 See Sunstein, supra (fn. 8), at 4; Hawley et al., Increasing Organ Donation via Changes in the Default Choice or Allocation Rule, 18–19 (Ga. State Univ., Working Paper No. 2012–15, 2012), available at http://excen.gsu.edu/workingpapers/GSU_EXCEN_WP_2012–15.pdf. 107 Kalantari-Khandan et al., Personality Characteristics and Mental Health of Organ Donor and NonDonor Families, 16 World Applied Sci. J. 1183, 1187 (2012). 108 Demir/Kumkale, Individual Differences in Willingness to Become an Organ Donor: A Decision Tree Approach to Reasoned Action, 55 Personality & Individual Differences 63, 65, 67 (2013).

24

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

Dutch study found no significant correlations between any of the Big Five characteristics and people’s motivation to donate their body to science.109 To address this disagreement in the literature, Matthew Kugler and Lior Strahilevitz 64 ran a pilot survey on Mechanical Turk, using approximately 340 research subjects. They found no statistically significant relationships between any of the Big Five characteristics and one’s status as an organ donor or support for organ donation generally.110 Interestingly, however, when Kugler and Strahilevitz measured personality using an alternative metric – the Authoritarianism scale popularized by John Duckitt, Boris Bizumic, Stephen Krauss, and Edna Heled111 – they found consistent, significant correlations between political-psychological orientations and status and attitudes toward organ donation. Those who scored highest on Authoritarianism-Traditionalist scales were significantly less likely to be organ donors than those who scored lowest on those scales.112 Self-described social conservatives were also significantly less likely than selfdescribed social liberals to be organ donors and to support organ donation more generally. By itself, social conservatism explained nine percent of the variation in individuals’ propensity to be organ donors. Nine percent is a rather meaningful number by social science standards, although it likely does not represent a difference of sufficient power to warrant different defaults on a question like organ donation. But if a multifactor approach was capable of explaining organ-donation preferences more accurately, then personalization of default rules based on Authoritarianism scores and social conservatism might be appropriate. The same correlations that proved powerful in predicting whether an individual was 65 or intended to become an organ donor also predicted whether an individual was likely to support making presumed organ donation the default legal rule. Here, too, people scoring high on Authoritarianism-Traditionalist scales and those who self-identified as social conservatives were significantly less likely to support a default rule of presumed consent. Unsurprisingly, organ donors and those intending to become donors were far more likely to support a presumed consent default than nondonors. In the Kugler and Strahilevitz sample, 36 % percent of donors and intended donors supported a presumed consent default but only 8 % of nondonors supported such a default rule. (Big Five scores did not correlate with attitudes about the content of organ donation default rules in their sample.) The American rule requiring organ donors to opt in is popular domestically, notwithstanding it apparent responsibility for killing thousands of Americans annually. We are not certain that a sufficiently predictive model of organ donation based on 66 personality, political orientation, and other aspects of individuals’ identities can be developed to enable accurate personalization of defaults. If such personalization is possible, then it might permit policymakers to move past the tired opt-in versus optout dichotomy. Suppose that the Kugler and Strahilevitz pilot study is widely replicated 109

Bolt et al., Personality and Motivation for Body Donation, 193 Annals of Anatomy 112, 114 (2011). Agreeableness correlated most strongly with organ donor status and organ donation support, although this correlation was not significant. Openness and neuroticism also had the anticipated relationships with organ donation status and attitudes, although they were also not significant. 111 Duckitt et al., A Tripartite Approach to Right-Wing Authoritarianism: The AuthoritarianismConservatism-Traditionalism Model, 31 Pol. Psychol. 685 (2010). 112 The Authoritarianism-Traditionalism scale “expresses the value and motivational goal of maintaining traditional lifestyles, norms, and morality, and resisting ‘modern’ liberal, secular, open, lifestyles, norms, and morality.” Id., at 691. To measure Authoritarianism-Traditionalism, the scholars asked survey respondents, for example, about their belief in “old fashioned values,” their beliefs in strict adherence to “God’s laws about abortion, pornography, and marriage,” and whether young peoples’ experimentation with drugs, alcohol, and sex threatens societal success. Id., at 711. 110

Porat/Strahilevitz

25

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

in the United States. Then the law might provide that only very socially liberal people who score low on Authoritarianism-Traditionalism scales will be presumed to be organ donors by default, and the rest will be deemed nondonors. Big Data sources might clearly identify the orientation of a motorcycle victim brought into an emergency room so that the default rule could be determined. This default would then become the starting point for an examination of whether a fatally injured patient expressed a view on organ donation that deviates from the default. Employing such a personalized default rule could engender increased donation rates relative to the impersonal opt-in rule of the status quo, and it could also reflect individual preferences better than an impersonal opt-out rule. Such an approach could represent a workable compromise in organ transplantation law. c) Medical malpractice. The personalized default approach could also work in the medical malpractice context. Suppose that a doctor has prescribed a drug that, when taken for a prolonged period of time, causes a very unfortunate side effect in a very small number of cases (say, one in every 500,000). The drug is most effective when taken for a long time, but it is still somewhat effective when taken for just a week or two. The doctor fails to warn the patient about this particular side effect, and the patient suffers the effect and then sues the doctor for malpractice, alleging a failure to obtain informed consent. A key focus of the legal inquiry will be causation: Would the patient have consented to undergo the treatment even if she had been warned about the side effect? As long as the doctor has no concrete information about the particular patient’s wishes or expectations regarding disclosure, presently the law treats this inquiry as an objective one: What would a reasonably prudent patient have done?113 68 Our approach contemplates a rule whereby a physician can tailor her disclosure of risks to particular patients – even though she has no concrete information about the particular patient’s wishes or expectations regarding disclosure. The physician will then be judged based on whether her disclosure was appropriate for a particular patient type (rather than for the hypothetical reasonably prudent patient). 69 Big Data firms like the Fair Isaac Corporation (“FICO”) have already gotten into the business of using data mining to predict patients’ future behavior, as evidenced by the firm’s having recently launched its FICO Adherence Scoring. FICO Adherence Scores use information from a patient’s credit report to predict the likelihood that a patient will regularly take his prescription medication and otherwise adhere to medical advice.114 Suppose a doctor consulted a patient’s FICO Adherence Score, and FICO predicted that there was only a 5 % chance that the patient would take the medication for long enough to render the side effect a possibility. The doctor does some quick math and determines that the risk that this particular patient would suffer the side effect is 1 in 10 million. Given that any warning may cause psychosomatic symptoms or raise the likelihood of cognitive errors by the patient, the doctor elects not to warn the patient.115 On our analysis, a default rule of nondisclosure would be appropriate for this particular patient. 70 At the same time, if the same doctor were treating a different patient, one for whom FICO predicted a 95 % chance that the patient would continue taking the medication

67

113

See, e.g., Canesi ex rel. Canesi v. Wilson, 730 A.2d 805, 812 (N.J. 1999). Simon, New Medical FICO Score Sparks Controversy, Questions, CreditCards.com (28 July 2011), http://www.creditcards.com/credit-card-news/fico-score-medication-adherence-1270.php. 115 See generally Siegal et al., Personalized Disclosure by Information-on-Demand: Attending to Patients’ Needs in the Informed Consent Process, 40 J.L. Med. & Ethics 359, 360 (2012) (discussing the issues doctors consider when attempting to determine how much information to provide to patients in attempting to gain informed consent). 114

26

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

for long enough to trigger a possible side effect, then the court’s ruling could well come out differently. The odds of the side effect occurring for this patient are approximately 1 in 526,000,116 and these odds, while remote, might still be sufficient to warrant disclosure to the patient. Personalizing the default rule permits the physician to practice personalized medicine to a much greater degree – a development that could substantially advance the efficiency of health-care delivery.117 Pushing the point further, we might imagine ways in which other forms of Big Data 71 could affect the informed consent calculus. One of the other functions of credit scoring is to assess an individual’s tolerance for risk. Risk is apparently correlated across a number of life activities, such that individuals who drive in a risky manner make risky personal financial decisions as well.118 Suppose that a plaintiff’s consumer behavior profile reveals that she is an extremely cautious person. In this case, the law might impose heightened disclosure requirements on the physician. If the patient’s profile reveals that she is a devil-may-care consumer, then giving short shrift to disclosures of low risks may be appropriate for the physician in a world where disclosure may be both time consuming and potentially harmful to the patient’s emotional well-being. Such an approach to adjudicating medical malpractice cases, where the patient’s profile at the time the medication was prescribed is part of the factual record before the court, may help steer adjudicators away from the dangers of hindsight bias. In these cases, the judge or jury knows that a bad outcome has occurred and is tempted to think that a reasonable patient would have wanted to know about the possibility of such an outcome, even though the ex ante risk of this outcome was extremely remote.119 The (hopefully rare) patient whom FICO or other providers of analytics misunder- 72 stand would have the chance to opt out. Under a new version of informed consent, the physician may tell a patient, “This is the sort of person our analytics contractor thinks you are. If we have misunderstood you, please tell us now, because it will affect the facts I disclose to you and the circumstances that will prompt me to ask for further consent or clarification.”120 We will say more about this sort of personalized disclosure in Part V. In other contexts, personalized informed consent default rules could further the 73 interests of third parties. Consider vaccination: children are vaccinated from diseases, but it is often in a particular child’s best interest, strictly speaking, not to take the vaccine – which has possible side effects – because the rest of the population is vaccinated, thereby reducing the chances the child will come into contact with the disease. To avoid such free riding, a mandatory law could force vaccination. A softer approach would set an impersonal default rule according to which doctors could say nothing about side effects unless asked; they would proceed with the vaccination unless told otherwise. An even better approach – in a world where, say, 80–90 % vaccination 116 With a 95 % chance that the patient would consume the drug for a long enough time to render the side effect a possibility, the risk of the side effect is 19 times higher than with a 5 % chance. We assume for purposes of simplicity that once the patient uses the drug for a sufficient amount of time, the side effect’s likelihood does not increase with the time of consumption. Relaxing this assumption would enhance the risk differentials between the two hypothetical patients. 117 See Marchant, Foreword: Law and the New Era of Personalized Medicine, 48 Jurimetrics J. 131 (2008) (introducing a symposium on the law of personalized medicine). 118 Morrison/Gupta, Health Shocks and Household Financial Fragility: Evidence from Automobile Crashes and Consumer Bankruptcy Filings, 3 (13 February 2013) (unpublished draft), available at http:// economics.uchicago.edu/workshops/Morrison%20Edward%20Health%20Shocks.pdf. 119 See Rachlinski, A Positive Psychological Theory of Judging in Hindsight, 65 U. Chi. L. Rev. 571, 615–16 (1998) (discussing medical malpractice litigation and hindsight bias). 120 For a further discussion of the benefits and perils of such discussion, see infra text accompanying fns 225–230.

Porat/Strahilevitz

27

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

suffices to create herd immunity, and 10 % of the population is likely to suffer side effects from vaccination – would be to personalize the disclosure default. For example, the “no information unless the patient asked” default rule would not apply to patients whose attributes correlate most closely with those of patients who have suffered side effects in Food and Drug Administration trials. 74 Resources like FICO Adherence Scores are not the only tool that could be used to personalize the law of informed consent and medical malpractice. Some research suggests that Big Five neuroticism/emotional stability scores can help predict hypochondria.121 In light of concerns about a hypochondriac’s psychosomatic response to being told about a given side effect, physicians might appropriately elect to inform an emotionally stable patient about a very unlikely side effect while offering information about the same side effect to a highly neurotic patient only if asked. In this sense, using Big Data tools to assess personality, which can in turn help a physician personalize disclosures to a patient, could have significant therapeutic value. d) Landlord–tenant law. We believe that personalized default rules are appropriate in adjudicating disputes in property law as well. Suppose a landlord and tenant are involved in litigation. The tenant lives alone and has rented a two-bedroom apartment for $600 a month in a neighborhood where the average similarly sized apartment rents for twice that amount. The written lease specifies the rent, the term, and various other factors, but it says nothing about the quality of the apartment. Now suppose that a few months after the tenant moves in, plaster begins falling from the ceiling in the second bedroom, making it an unsafe space for sleeping, although the tenant continues to use the bedroom for storing personal belongings. Has the condition in the second bedroom amounted to a breach of the lease, such that if the ceiling is not repaired upon request the tenant can move out and stop paying rent? In most American jurisdictions, the answer to this question is yes. The condition of the ceiling constitutes a breach of the implied warranty of habitability, which is read into every landlord-tenant contract.122 In some jurisdictions, however, the implied warranty of habitability functions as a default provision that the parties can waive via explicit contract terms.123 76 American law has largely stuck with a one-size-fits-all approach to the implied warranty of habitability, although the limited exceptions are important for our purposes. As a general matter, the implied warranty of habitability will be read into any residential lease. But some jurisdictions hold that there will be no such warranty when the tenant rents a single-family home (as opposed to a unit in a multiunit building),124 and other jurisdictions recognize nothing akin to an implied warranty of habitability when nonresidential properties are leased.125 This granularity of the rules is based on common law courts’ suppositions that particular variables governing property ought to affect the tenant’s substantive legal rights.126 77 Our approach to personalized default rules posits that the characteristics of the tenant (and landlord) may be relevant to determining the appropriate missing term to impose on the contract. This is particularly true when the landlord has access to information relating to the tenant’s past behavior, characteristics, and traits, or to other data

75

121 Ferguson, Personality as a Predictor of Hypochondriacal Concerns: Results from Two Longitudinal Studies, 56 J. Psychosomatic Res. 307, 311 (2004). 122 See, e.g., Hilder v. St. Peter, 478 A.2d 202, 207 (Vt. 1984). 123 See, e.g., P.H. Inv. v. Oliver, 818 P.2d 1018, 1021–22 (Utah 1991). 124 See, e.g., Zimmerman v. Moore, 441 N.E.2d 690, 696 (Ind. Ct. App. 1982). 125 See, e.g., K & S Enters. v. Kennedy Office Supply Co., 520 S.E.2d. 122, 126 (N.C. Ct. App. 1999). 126 Note the similarities to the U.C.C.’s treatment of unique and nonunique goods. See supra fns 35–36 and accompanying text.

28

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

indicating the suitable default rules for the tenant. Compared with tenants with similar incomes, is this tenant routinely willing to sacrifice quality in return for cost savings? If so, the court ought to view the lease as lacking an implied warranty of habitability. Is the tenant someone who stays in nice hotels far more frequently than most travelers with similar incomes, pays for weekly maid service, and otherwise indicates a propensity for paying for comfort and pleasing aesthetics? If so, the court ought to view the lease as containing an implied warranty of habitability. Does the tenant score high on personality metrics measuring neuroticism, such that the prospect of problems with the ceiling will keep her awake at night? Or is she a very emotionally stable person who may be annoyed but will not be made anxious by her substandard ceiling?127 We are not suggesting that these intuitive correlations among purchasing history, 78 personality, and expectations for an apartment are airtight. We are articulating falsifiable hypotheses that ought to be tested empirically. But since at least the mid-1990s, consumer profilers have been able to analyze a broad swath of personal information relating to transactions and to use algorithms to identify “value oriented” or “Rodeo Drive Chic” consumers for marketing purposes.128 e) Labor law. American labor law is not often thought of in terms of default 79 rules, but defaults are very important in this field. More precisely, the default provision under the National Labor Relations Act is that workers are not unionized. If a group of workers mounts a unionization drive and a majority of the workers (or, in some cases, a majority of a subset of the non-management workers) within a workplace vote to unionize, then a union will be certified, and it will be authorized to bargain collectively on behalf of all the workers as a whole.129 Union certification efforts can be cumbersome, expensive, and contentious. At the same time, it seems plausible that American law’s chosen default rule is an appropriate one on majoritarian grounds – most American workers are nonunionized and have been for quite some time.130 Psychological studies have shown that personality characteristics correlate strongly 80 with membership in a voluntary union. In particular, the Big Five traits of extraversion and neuroticism both predict union membership, and the interaction of these two traits predicts union membership very strongly.131 Big Five personality characteristics also predict which industries individuals are likely to be drawn to and which individuals are most likely to thrive and retain their jobs in particular industries. For example, nurses who report high levels of neuroticism are much more likely to experience emotional exhaustion and burnout, which may cause them to leave nursing, although nurses with high levels of extraversion are likely to avoid burnout.132 And while politicians score 127 A counterargument is that an implied warranty of habitability should apply to any tenant unless he explicitly waves it. We develop this point further infra in text accompanying fns 174–175. 128 See Dwyer v. Am. Express Co., 652 N.E.2d 1351, 1353 (Ill. App. Ct. 1995). 129 What We Do: Conduct Elections, Nat’l Lab. Rel. Board, http://www.nlrb.gov/what-we-do/conductelections (last visited 17 November 2013). 130 Union Members Summary, Bureau of Lab. Stat., (24 January 2013, 10:00 AM), http://www.bls.gov/ news.release/union2.nr0.htm. An important caveat is in order. We do not know about workers’ preferences regarding unionization in an environment where there are no transaction costs for forming a union. Moreover, some workers’ decisions not to be part of a union may result from coercion or collective action problems. See Hirsch, Communication Breakdown: Reviving the Role of Discourse in the Regulation of Employee Collective Action, 44 U.C. Davis L. Rev. 1091, 1097, 1126–27 (2011). 131 Parkes/Razavi, Personality and Attitudinal Variables as Predictors of Voluntary Union Membership, 37 Personality & Individual Differences 333, 342–43 (2004). 132 See Zellars et al., Burnout in Health Care: The Role of the Five Factors of Personality, 30 J. Applied Soc. Psychol. 1570, 1588–89 (2000).

Porat/Strahilevitz

29

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

very high on extraversion and openness, bureaucrats do not.133 Managers and sales representatives show high levels of extraversion,134 and the unemployed commonly evince high levels of neuroticism.135 81 This kind of data suggests a radical possibility, which is that certain workplaces or industries, especially those containing high numbers of very extraverted and neurotic individuals, might be deemed unionized by default.136 Given the underrepresentation of highly neurotic individuals in the workforce, the nonunionized default plausibly makes sense for most workplaces. 82 At this point, we want to identify this kind of workplace profiling to determine the default rule as a theoretical possibility rather than as something we are advocating. Correlation and causation are distinct, and the factors that drive union membership continue to be debated.137 For example, it is plausible that extraversion and neuroticism explain the success of unionization campaigns rather than workers’ underlying preference for union membership. It is even conceivable that correlation runs in the opposite direction and that participation in a union makes workers more extroverted and neurotic. We would need to get a fuller sense of these causal variables before offering prescriptions for labor law. That said, depending on the results of future research, a prounionization default rule could be appropriate in some contexts.

3. Big Data guinea pigs Countries with enormous populations ought to take advantage of economies of scale. In this case, that would mean forgoing the careful monitoring of all their citizens’ choices and perhaps sidestepping some of the problems from inefficient social norms in the process. We therefore propose that American law ask one million guinea-pig residents to make active choices about their preferences, which the law would then data mine to identify the ways in which the other 314 million individual Americans are similar to the 1 million guinea pigs.138 The law would provide modest compensation to the guinea pigs for the costs they incurred in the process. The guinea pigs’ active choices would then become the personalized default choices for the people most similar to them across a variety of observable metrics. These surveys could be conducted through a governmental agency, like the Census Bureau or Consumer Financial Protection Bureau, or through an industry consortium. 84 A great deal of contract law scholarship concerns the extent to which consumers are rushed or inattentive and pay little attention to contract terms as a result.139 Yet, if one 83

133 Ashton et al., Personality Traits of Municipal Politicians and Staff, 50 Can. Pub. Admin. 273, 285 (2007). 134 Barrick/Mount, supra (fn. 84), at 19. 135 Murali/Oyebode, Poverty, Social Inequality and Mental Health, 10 Advances in Psychiatric Treatment, 216 (2004). 136 We ignore the (realistic) problem of multiple unions competing to represent the same workforce. See generally Akkerman, Union Competition and Strikes: The Need for Analysis at the Sector Level, 61 Indus. & Lab. Rel. Rev. 445, 446–49 (2008) (reviewing comparative data on competition among unions); Hodges, Southern Solutions for Wisconsin Woes, 43 U. Tol. L. Rev. 633, 647 (2012) (discussing informal noncompetition arrangements among unions in Virginia). 137 An introductory analysis of these questions is offered in Riley, Determinants of Union Membership: A Review, 11 Labour 265 (1997). 138 See infra Section III.3. 139 See, e.g., Marotta-Wurgler, Will Increased Disclosure Help? Evaluating the Recommendations of the ALI’s “Principles of the Law of Software Contracts”, 78 U. Chi. L. Rev. 165, 182 (2011) (pointing out that almost no consumer making transactions through the internet read the contract before accepting it).

30

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

in every 314 people is a compensated contract-law guinea pig,140 then the law might reasonably devote substantial resources to making sure that these guinea pigs are very well informed and have adequate time to consider the contractual options and associated tradeoffs. The guinea pigs would spend time reading the fine print so others do not have to. Once an entity – presumably a governmental agency – has assembled a large dataset to track the choices of these guinea pigs, the entity can identify behavioral patterns and facilitate efforts by firms to give each consumer a set of default contractual terms that mimic those chosen by the guinea pigs with the personalities and attributes most similar to hers. We envision a “clustering” approach for identifying coherent groups of people to whom particular personalized default rules will apply.141 Only the choices made by the guinea pigs prior to the time the contract at issue was executed would matter. In a recent article, Ayres and Schwartz propose that firms be required to survey their 85 customers about whether particular terms in a contract are consistent with their expectations.142 Terms that surprised many consumers or that had surprising and very bad consequences for a smaller number of consumers would need to be set apart in a special box designed to prompt consumers to pay more attention to such terms.143 A consumer-voting mechanism could ensure that the most surprising or most disadvantageous terms appear most prominently in the special box.144 Ayres and Schwartz’s proposal somewhat resembles our approaches to defaults and disclosure.145 But whereas Ayres and Schwartz propose an impersonal approach to determining which contract terms are problematic, our approach is personalized. It recognizes that different terms will be problematic to different types of people, so the “boxes” or defaults that different sorts of people are shown should differ systematically. Our “sampling” strategy mirrors the sorts of extrapolations that demographers and 86 survey researchers routinely use in their work.146 And the private sector already uses such strategies for predictive purposes. For example, Netflix’s Cinematch algorithm for movie ratings (a) analyzes the one- to five-star ratings provided by its users after they have seen a movie; (b) matches each user’s ratings with the ratings of other users in the Netflix database; and (c) uses these similarity scores to predict how much users will like particular movies. Users can then employ these predictions in deciding which films to rent or stream.147 The more films a user rates, the better the algorithm can personalize the user’s movie recommendations and the recommendations of similar Netflix customers. Of course, rating each movie on Netflix entails an active choice. Many Netflix users 87 do not bother to evaluate movies they have seen, perhaps because it is time consum140 The guinea pigs’ attributes and decisions would be closely scrutinized so that other people would not need to be subjected to high decision costs and such exacting scrutiny. 141 Obviously, the devil is in the details with respect to how these clusters will be created and what happens to consumers whose profiles place them at the borderline between different clusters. For further discussion, see Agrawal et al., High Performance Big Data Clustering, in: Catlett et al. (eds), Cloud Computing and Big Data, 2013, 192. 142 Ayres/Schwartz, The No-Reading Problem in Consumer Contract Law, 66 Stan. L. Rev. 545 (2014). 143 Id., at 583. 144 Id., at 584. 145 See infra Part V. 146 The closest analogue for the guinea pigs proposal in existing legal scholarship would be the sort of legal experimentation proposed in Abramowicz et al., Randomizing Law, 159 U. Pa. L. Rev. 929 (2011) (proposing randomized trials to test the efficacy of laws and regulations). For another recent article advocating greater regulatory experimentalism, especially continuous learning and updating, see Sabel/ Simon, Minimalism and Experimentalism in the Administrative State, 100 Geo. L.J. 53 (2011). 147 See Thompson, If You Liked This, You’re Sure to Love That, New York Times, 23 November 2008, (Magazine), at MM74, available at http://www.nytimes.com/2008/11/23/magazine/23Netflix-t.html?pagewanted=all&_r=0.

Porat/Strahilevitz

31

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:03 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

ing.148 And many Netflix users similarly do not use the “taste preferences” features, which permit users to specify how often they watch movies that can be characterized as “absurd,” “bawdy,” “cerebral,” “dark,” etc.149 One of the potential benefits of personalized default rules in a world of Big Data is that much of the data used to generate similarity scores and personalized defaults will be generated automatically, without requiring the user to do anything. It is almost tantamount to Netflix monitoring how many times a viewer laughed during a comedy, cried during a tragedy, or gasped during a horror flick. 88 A more modest alternative to guinea pigs would be to generate information necessary for personalizing default rules by asking individuals about their general preferences, characteristics, and traits, as well as about their past behaviors, and using this information to tailor default rules for them. An agency might distribute questionnaires to consumers, explaining that the answers will be used for personalizing default rules in their interactions with merchants. We predict that many consumers will answer the questionnaires, which should not be too intrusive, with the understanding that their answers would facilitate their receiving deals better adapted to their true preferences. The gist of the approach is to use information culled from a survey to modify defaults that a consumer will encounter. This blanket approach to personalizing default rules seems far more efficient than selective modifications of contractual boilerplate on a transaction-by-transaction basis. We propose that individuals should be able to see the “profile” constructed for them and change this profile if it does not fit their true preferences.150 89 In any event, in modern, high-stakes transactions, it is becoming increasingly common for sellers to have information about the consumers they are dealing with, which enables them to decide on pricing and service quality, pinpoint potentially fraudulent transactions, and evaluate the effectiveness of their marketing strategies.151 As the information age proceeds, it will be reasonable to assume that sellers “know their customers” and either already are or can easily become familiar with the personalized default rules that correspond to particular customers. 90 Consumers are less likely to have this sort of information about individual firms’ propensities, although in the case of large national firms or local firms with extensive Yelp profiles, the information asymmetries may be less pronounced. Imposing on consumers a burden to “know their sellers” is less justifiable, particularly when they are dealing with small-scale sellers in non-repeat-play environments.152

IV. Possible Objections and Limitations 91

Part III articulated a rather bold vision of personalized default rules. In this Part, we want to confront some potential objections to our proposal while conceding that some of these objections warrant limiting the appropriate scope for personalized default rules. 148 See id. Another reason users might not rate movies is that they do not think their opinion of a particular film is any of Netflix’s business (or the business of anyone with whom Netflix might share data). Seen from this perspective, seamless, automatic sharing is more troubling than sharing via forced choices, as occurs with Netflix’s system. 149 Taste Preferences, Netflix, http://dvd.netflix.com/TastePreferences (last visited 25 January 2014). 150 Securing individuals’ active participation in constructing their profiles could also occur under our guinea pigs scheme: each individual would be able to identify her profile and request changes to it. 151 See, e.g., Duhigg, supra (fn. 102). 152 For further thoughts on this point, see infra Section IV.3.

32

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:03 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

1. Cross Subsidies An obvious objection to our proposal relates to the equities and inefficiencies of cross 92 subsidization. In our analysis, two consumers might buy the same product for the same price but receive a different set of contractual rights as part of the transaction. That might be unjust and inefficient. Consider the following example: 93 Example 3: Right to Return. Dana is conservative, very careful in her behavior in all fields of life. She is a cautious consumer: before she buys anything, she consults Consumer Reports and asks for her friends’ advice. In the past, she has never returned a product she bought, unless it was defective. Jim is a risk taker who is quite impulsive excitable. He makes decisions fast, without consulting anyone. In the past, he returned products he bought several times just because he realized he should not have bought them in the first place. Both Dana and Jim have separately bought a new flat-screen television at the same store. After a day of using the new television, they realized that this purchase was a mistake. They want to return the product and get their money back. Should they be treated in the same manner? Under current law, the answer is yes. Whether the default rule is a “right to return”153 94 or “no right to return,” it would apply equally to Dana and Jim. If, however, personalized default rules are permitted and feasible, Jim would probably enjoy the right to return but Dana would not because Jim needs it more. Since the buyer’s exercise of such a right is costly for the seller and since both Dana and Jim paid the same price, the result would be that careful Dana subsidizes hasty Jim – a result that is both unjust and inefficient. Upon closer scrutiny, a personalized default rule in Example 3 is more just and 95 efficient than an impersonalized default rule, especially if the impersonalized rule contains a right to return.154 With an impersonalized default rule, both Dana and Jim pay the same price and get the same default rule of a right to return. But since Jim uses this right more often than Dana, Dana subsidizes Jim through the contract price. With a personalized default rule, the cross subsidization either disappears or at least 96 diminishes. It disappears if both the default rule and price are personalized. In this scenario, if either Dana or Jim gets the right to return, he or she pays a price reflecting the expected cost of each of them exercising the right, resulting in no cross subsidization. But even if prices are not adapted to the personalized default rule, the cross subsidization would diminish. This would occur because the efficient personalized default rule for Jim is probably no right to return, and the efficient counterpart rule for Dana is probable a right to return. Although Jim ostensibly needs the right to return more than Dana because he exercise it more often, he uses the right more frequently precisely because Dana subsidizes him. With such personalized default rules, there would still be some cross subsidization – now Jim subsidizes Dana – but the level of 153 In New York, unless the retailer opts out by displaying a “return and refund policy,” the default rule provides a right to return for cash up to thirty days after the purchase. See N.Y. Gen. Bus. Law § 218-a (McKinney 2012); see also, Cal. Civ. Code § 1723 (West 2009) (applying the same default as in New York). 154 If the impersonal default rule is “no right to return,” there is no cross subsidization, although not every consumer gets her preferred default rule, a problem that our personalized default-rules regime aims to resolve.

Porat/Strahilevitz

33

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:03 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

such subsidization would be lower than it would under an impersonal default rule where both Dana and Jim have the right to return.

2. Strategic Behavior Crude personalized default rules tied to an individual’s immutable characteristics, such as sex or age, alleviate significant concerns about strategic behavior. Under our proposal for granular personalized default rules, however, the products and services that an individual buys, the keywords he uses in his searches, the company he keeps, and various other aspects of his behavior can influence the terms under which he will purchase goods and services. When an individual consumer changes his behavior, he is simultaneously changing the identities of the guinea pigs with whom he will be compared. In effect, the consumer trades default rules: he replaces the rules selected by the guinea pigs who used to behave like him for the rules selected by the guinea pigs who behave like the “new him.” Given this possibility, there is a danger that individual consumers will engage in strategic behavior to ensure they are compared to the guinea pigs who have selected the most generous default terms. 98 To take a salient example of this problem, a Canadian credit-card issuer determined during the last decade that consumers who purchase carbon monoxide detectors or felt pads to be placed at the bottom of chair and furniture legs are exceptionally low credit risks.155 Evidently, people concerned about the dangers of carbon monoxide or intent on preventing scratches on hardwood floors are extremely careful, conscientious individuals with low discount rates; they are precisely the sort of people likely to repay loans on time.156 Before it publicized this finding, the credit-card issuer could use its knowledge of felt pad and carbon monoxide–detector purchases to price risk. But as soon as the correlation became public, its value diminished substantially. After all, felt pads and carbon monoxide detectors are relatively inexpensive compared to home-mortgage loans. It would be in everyone’s interests to stock up on these household products a few months before seeking to purchase a house, even if they had no intention of putting these items to their ordinary use. In this way, the strategic purchase of felt pads and carbon monoxide detectors would function as a smoke screen.157 99 Although the problem of strategic behavior is always an issue, we do not think it is particularly troublesome in this context. First, a great deal of predictive analytics is and will remain proprietary. Guessing which products function as felt pads will not be easy, and people who discover how to game the system will have little incentive to publicly disclose their success stories. Second, even when it becomes clear that certain types of behavior will be associated with more beneficial default terms in some contexts, employing smoke screens is costly. If people regularly purchase products they do not need, become Facebook friends with people they do not like, or develop hobbies they do not enjoy in order to enhance the quality of their personalized default profiles, they often will be making themselves worse off. Changing one’s behavior is a costly signal; it is not cheap talk. Much of the time it will be easier just to specify a different contractual term when entering into a contract – or simply pay a higher price – rather than putting on an elaborate and costly performance to achieve the same result. Third, while maintaining a charade may be easy for a short period of time, it gets harder for the 97

155 Duhigg, What Does Your Credit-Card Company Know About You?, New York Times, 17 May 2009, (Magazine), at MM40, available at http://www.nytimes.com/2009/05/17/magazine/17credit-t.html?pagewanted=all. 156 Id. 157 Strahilevitz, Signaling Exhaustion and Perfect Exclusion, 10 J. on Telecomm. & High Tech. L. 321, 327 (2012).

34

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:03 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

consumer (and easier for the seller to detect) with every passing day. Thus, in Example 3 (Right to Return), if hasty Jim pretended to be careful and hence obtained a right to return, after he abuses this right several times, merchants would recognize his true character and treat him accordingly. Fourth, on many occasions, consumers will not really benefit from pretending to be what they are not: possessing a specific character could benefit a consumer in one context but harm him in another context, and in many instances, faking it could cause a consumer to be presented with various default rules that do not fit him personally. While we think strategic behavior is a manageable problem associated with persona- 100 lized default rules, expanding personalization beyond waivable defaults would magnify the problem. Personalized default rules could potentially become so ingrained that sellers essentially refuse to bargain around them. In other words, firms might be willing to offer consumers contracts with personalized terms but might view negotiating around the personalized terms as prohibitive because of the high transaction costs. Such a progression away from personalized default rules and toward unwaivable “personalized terms” strikes us as a sufficiently thorny topic to warrant an article of its own. But we suspect society that will not need to cross this bridge, at least in the immediate future, in part because the strategic behavior problem would be substantially magnified in a world where most terms were nonnegotiable.

3. Abuse by Merchants Another potential objection to our proposal – mostly relevant to consumer law – is 101 that merchants could abuse the availability of large amounts of data to the detriment of consumers. When one party to a contract knows a lot about the other party’s preferences but not vice versa, the party with more knowledge may enjoy a substantial advantage over the other party and extract a larger share of the contractual surplus. This risk materializes once we assume that the market is not fully competitive and that consumers are not fully informed. To illustrate this point, suppose a merchant knows that a specific consumer is highly risk averse. The merchant may then overcharge this consumer for decreasing her risks. More generally, data about consumers’ preferences could indicate their willingness to pay for a certain product or service, thereby facilitating price discrimination. We concede that this objection has force, although it is important to understand 102 the precise nature of this force and the countervailing considerations. Firms already gather enormous amounts of information about individuals and use this information to boost their bottom line. This trend has its own economic rationale, and our proposal would have at most a marginal effect on it. That said, when all consumers are offered the same terms, price discrimination is easy to identify. When all consumers are offered personalized terms, however, price discrimination becomes much harder to spot. That which is hard to detect is more difficult to oppose and therefore tougher to deter. Of course, price discrimination has ambiguous welfare effects. It tends to raise 103 output, enabling consumers who otherwise could not obtain a good or service to do so, as well as to maximize producer surplus. The availability of personalized default terms is also designed to provide consumers with deals that better reflect their personalities and thereby encourage more consumers to enter into transactions with firms.158 Indeed, we might conceptualize personalizing contractual terms as enabling consumers to reclaim some of the firms’ surplus. Determining the dynamic effects of 158

For a lengthier discussion, see Strahilevitz, supra (fn. 68), at 2027–29.

Porat/Strahilevitz

35

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:03 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

these simultaneous changes in consumer welfare is extremely tricky. In any event, it would be unwise to reject personalization solely on price-discrimination grounds.

4. Uncertainty 104

105

106

107

108

According to a fourth objection, adopting a personalized default-rules regime would increase uncertainty, thereby making the law less effective in guiding people’s behavior. Personal default rules may also increase the costs of adjudication.159 Impersonal default rules, by contrast, avoid these drawbacks. An approach to contract law that locks in the guinea pigs’ choices before the contract’s execution would ameliorate any uncertainty created by personalized default rules. Any subsequent shifts in the guinea pigs’ choices would be irrelevant to the meaning of a contract. This is the primary approach we advocate here. Mechanically, a consumer would be entitled to ask at the point of sale what personalized terms her consumer profile generates for her, and the firm would respond with a disclosure of the terms. Because the entire process would be automated, producing this information for consumers would not slow down transactions. While the consumer’s review of the default terms might delay the transaction, the same is true of existing, written contract terms. To better understand the uncertainty objection to our personalized default-rules regime, reconsider Example 3 (Right to Return). If there is a one-size-fits-all default rule – either a right to return or no right to return – contractual parties could clearly understand whether in a specific transaction they have such a right. Similarly, in Example 2 (Damages or Specific Performance), if the choice of remedy is not contingent on the buyers’ characteristics and traits, both Steven and Sarah could know in advance that in the event of a breach, they are entitled to specific performance (or damages), regardless of the inferences which could be derived from their particular traits. With personalized default rules, there is more uncertainty: in the two examples above, contracting parties would find it harder to contemplate their substantive rights and remedies.160 The choice between personalized default rules and impersonal default rules only loosely tracks the choice between rules and standards, which commentators have thoroughly analyzed.161 Most importantly, rules are more costly to create, but standards are more costly both for individuals to interpret when deciding how to behave and for adjudicators to apply in evaluating past behaviors.162 At first glance, an impersonal default rule ostensibly resembles a rule while a personalized default rule seems to resemble a standard. Thus, in Examples 2 and 3, an impersonal default rule (such as “damages” or “right to return,” respectively) is a rule while a personalized default rule is a standard. The rules-versus-standards dichotomy is not identical to the impersonal-versuspersonalized default-rules dichotomy. In particular, there could be impersonal default rules that are standards (e.g., a duty of good faith) and personalized default rules that 159

Ayres, supra (fn. 9), at 13 (discussing the complexity costs of tailoring default rules). Cf Geis, supra (fn. 10), at 1124–29 (discussing transaction costs, and other costs, of tailoring default rules). 161 See, e.g., Kaplow, Rules Versus Standards: An Economic Analysis, 42 Duke L.J. 557 (1992) (describing the tradeoff between the use of rules and standards in law generally); Kennedy, Form and Substance in Private Law Adjudication, 89 Harv. L. Rev. 1685 (1976) (introducing the distinction between rules and standards). 162 Ehrlich/Posner, An Economic Analysis of Legal Rulemaking, 3 J. Legal Stud. 257, 262–71 (1974) (comparing rules and standards on the complexity dimension); Kaplow, supra (fn. 161) (comparing rules with standards on various dimensions). 160

36

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:04 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

are rules (e.g., different intestacy rules for men and women).163 Therefore, the crucial questions with personalized default rules are how to balance uncertainty with accuracy, better reduce transaction costs, encourage desirable behaviors, and meet people’s reasonable expectations. Would a personalized default rule typically be more associated with uncertainty than an impersonal default rule? Not necessarily. A consumer living in a world with impersonal default rules would need to invest resources in learning the content of the default rule (or bear the risks of failing to do so). A consumer living in a world with personalized default rules would need to invest resources in learning the content of whichever default rule applies to him, and he may also need to research other plausibly applicable default rules along the way. Critically, the consumer already knows a great deal about his preferences and characteristics, which are the factors driving the choice among multiple personalized default rules. Assuming that Big Data does what it is supposed to do – identify patterns of behavior among similarly situated people – then the consumer will be able to intuit the law’s contents based on what he himself would want, which would be a good proxy for the choices of guinea pigs just like him. It is therefore conceivable that the average consumer can discern the contents of applicable personalized default rules at a lower cost than he can discern the contents of an impersonal default rule, and he may very well be able to do so without consulting a lawyer.164 A caveat is in order. In our model, the guinea pigs are given more time and resources to make decisions, and it is conceivable that this extra time will cause them to make decisions that differ from the snap judgments that they (and those like them) would have made. If this gap is large, the effect will be greater consumer uncertainty combined with greater consumer satisfaction with their default choices. At worst, this seems likely to be a wash. Over time, many consumers may stop worrying so much about uncertainty, in the same way that consumers quickly overcame their widespread initial reluctance to purchase products over the Internet using credit cards.165 For the consumers who remain mistrustful, our proposal for personalized disclosure in Part V offers a novel strategy for ameliorating the uncertainty problem. In contracts between two consumers, especially consumers involved in non-repeatplay interactions, the uncertainty will rise dramatically, which is why we are quite skeptical about using personalized default rules in those contexts. But in contracts between a consumer and a profit-maximizing firm or between consumers involved in repeat-play interactions, the cognitive load faced directly by consumers should be more manageable.166 Contracting firms may face information asymmetries regarding consumer preferences, but reducing these asymmetries is one of the Big Data industry’s chief objectives. Matters would become more complicated if courts entered the business of personalizing default rules. If judges are not skilled at identifying litigants’ characteristics and preferences, then adjudicators’ cognitive loads will rise as a result of the shift from impersonal to personalized default rules. And as these cognitive loads rise, 163 See Geis, supra (fn. 10), at 1116–19 (distinguishing among simple rules, complex rules, simple standards, and complex standards). 164 Geis comes tantalizingly close to making this important point but instead goes in a more familiar direction, using the heterogeneity of actors to whom rules are tailored to discuss the transaction costs associated with rejecting a default rules. See id., at 1122–23. 165 Teo, Attitudes Toward Online Shopping and the Internet, 21 Behav. & Info. Tech. 259, 265 (2002) (noting that in 2002 consumer concerns about the security of financial information used to make purchases over the internet remained a significant impediment to e-commerce). 166 See supra text accompanying fn. 150.

Porat/Strahilevitz

37

109

110

111

112

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:04 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

the risk of judicial error increases, which will engender uncertainty for the parties themselves, even if these parties have perfect information about their own preferences and characteristics. As this analysis shows, the heightened uncertainty created by personalized default rules is likely to emerge indirectly, as a “shadow of the law” effect.167 113 These substantial concerns about personalization in the litigation context aside, there are plenty of other cases where personalized default rules promote accuracy without increasing uncertainty. For example, in our hypothetical transaction between a firm and a consumer, the firm knows the terms to which a particular consumer is entitled at the time of the purchase. The law could require the preservation of these terms, even if the consumer does not ask to see her personalized default terms. If a subsequent dispute arises, the court would have as much access to the personalized terms as it would to the written contract, which means the uncertainty associated with interpretation would be no greater than normal. Nor is the uncertainty concern serious with regard to our inheritance law example or other cases where the default rule is tailored according to a salient and easily observable characteristic like sex or age. Where a personalized rule is tailored to a defined social group (e.g., a default of no organ donation among Shintos), we can expect group members to learn the contents of the crude personalized default rule without having to investigate it.168 This brings us to a closely related objection: case law fragmentation. We turn next to this topic.

5. Case law Fragmentation Impersonal default rules minimize the fragmentation of the case law resolving contractual ambiguity. This is a key advantage. Personalized default rules, by contrast, would engender greater fragmentation in the legal precedents. Such fragmentation is a real drawback of judicial determination of personalized default rules, and this drawback may convince readers that personalization should be limited to ex ante contexts and that only firms and specialized agencies – but not courts – should use personalized default rules. 115 Presently, if a court interprets ambiguous contractual language, its interpretation will have precedential value and help clarify the law in future disputes arising out of contractual ambiguity. The precedential effect is most powerful in any future dispute arising between the same parties concerning the same ambiguity. In such a case, the earlier precedent has preclusive effect. Even here, though, the court may construe the same contractual language to mean different things if it identifies pertinent differences in the context of the contract negotiation.169 But the interpretation of language will certainly play a significant role in mitigating subsequent judicial uncertainty about the language’s meaning in future disputes.170 Still, the precedent may help reduce uncertainty with respect to similarly situated parties and similar contractual ambiguities. 114

167 See generally Mnookin/Kornhauser, Bargaining in the Shadow of the Law: The Case of Divorce, 88 Yale L.J. 950 (1979) (discussing shadow of the law effects in the context of divorce). 168 See supra text accompanying fns 57–60. 169 See Forbo-Giubiasco S.A. v. Congoleum Corp., 516 F. Supp. 1210, 1214 (S.D.N.Y. 1981) (“The fact that Congoleum used the identical language in the Giubiasco Related Company Clause as it had used six years earlier in the Krommenie Related Company Clause would suggest that the two provisions should be interpreted in the same manner only if the same negotiating context for both contracts existed. However, Congoleum has presented uncontradicted evidence (…) that the understanding between Congoleum and Giubiasco in 1971 was not the same as the understanding between Congoleum and Krommenie in 1965.”). 170 See, e.g., United States ex rel. B’s Co. v. Cleveland Elec. Co. of S.C., 373 F.2d 585, 588 (4th Cir. 1967).

38

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:04 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

To be sure, lawyers and judges will be able to distinguish precedents that are closely on point if they are sufficiently motivated to do so, but the greater the similarities in the contractual language at issue, the more difficult it will be to distinguish the precedents on contextual or other grounds. With personalized default rules, it becomes considerably easier to distinguish a 116 precedent that a judge disfavors. Even if the contractual language at issue in an earlier case is identical to the language at issue in the case before the court, a party would appropriately argue that the litigant in the earlier case and the litigant in the current case have sufficiently different personalities, attributes, and profiles to warrant divergent interpretations of the ambiguity. No two human beings are identical in every respect, and therefore the court will have to confront the question of whether litigant heterogeneity merits a different result in the face of linguistic and contextual homogeneity. This fragmentation of precedent seems likely aggravate uncertainty about the law’s content. Where personalized default rules make this problem particularly pronounced, they should be regarded more skeptically. The question is ultimately one of tradeoffs, and it is not clear whether the costs of 117 indirect uncertainty and case law fragmentation exceed the benefits of giving a greater number of individuals default rules that more closely approximate their preferences than impersonal default rules (if one adopts the majoritarian default rule theory). To the extent that readers are concerned about excessive fragmentation, they might support a scaled-back version of our proposal, whereby personalized default rules could only be employed to deal with contractual silence but not with contractual ambiguity. Under this modified approach, identical contractual language would usually mean identical things to different people, but the absence of a contractual provision would have different implications for different parties. Courts have occasionally confronted this fragmentation issue before. In one promi- 118 nent decision, the Fifth Circuit held that interpreting identical contractual language to mean different things in different contexts was justified, despite protests about the extent to which such results would destabilize existing contracts.171 If such an approach to interpretation is occasionally permissible when courts are engaged in ex post, holistic analyses of contractual meaning, then it ought to be even more palatable if undertaken in a rigorous, data-driven, ex ante way, which is our aspiration in advocating personalized default rules.172 We therefore conclude that uncertainty and precedent fragmentation are important but not necessarily decisive considerations in determining the desirability of personalized default rules.

6. Statistics, Stereotyping, and Valuable Default Rules Another possible objection to our proposal is similar to the one raised against 119 profiling in law enforcement, or more generally, against using statistical data for determining rights and duties. Statistical data does not focus on the individual parties; instead, it purports to establish factual findings and allocate rights and duties by using generalizations about the group to which the individuals belong, e.g., their sex, age,

Hall v. Fed. Energy Regulatory Comm’n, 691 F.2d 1184, 1194–95, 1195 fn. 19 (5th Cir. 1982). We note in passing that our analysis of personalization has implications for the law of class-action suits. Class certification is only appropriate under the Federal Rules of Civil Procedure if the “claims or defenses of the representative parties are typical of the claims or defenses of the class.” Fed. R. Civ. P. 23 (a)(3). As the consumers’ differences become increasingly regularized and the law comes to depend more heavily on the characteristics of particular consumers, the proper scope of class-action litigation will be substantially diminished. 171 172

Porat/Strahilevitz

39

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:04 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

120

121

122

123

race, religion, or other group.173 Such a method may contradict many people’s moral intuitions. Furthermore, using statistical data creates stereotypes by ascribing to people attributes they may not have. It is beyond the scope of this Article to discuss the pros and cons of using statistical data for allocating rights and duties and for law enforcement. We note, however, that any default rule, impersonal or personalized, is statistical in nature because it assigns rights and duties to individuals according to the averaged preferences of an entire population or a subset of people. Personalized default rules are just a better proxy – based on more accurate statistics – for the preferences of the specific party. Therefore, the objection to our proposal is not that it uses statistical data as such – this kind of data should be used regardless of the type of default rule – but instead that it creates undesirable stereotypes. Take the intestacy example. Suppose that we use different default rules for men and women: when there is no will, most of a mother’s estate goes to the children while most of the father’s estate goes to the children’s spouses. Such default rules could create (or strengthen) a stereotype that mothers care more about their children than fathers. We consider this objection in the next section. A variation of the objection discussed in this section relates to a subset of default rules pertaining to values that are central to our life and are not “mere” preferences. (We hereinafter call this subset of default rules “a fundamental values default rule.”). Here, the argument against a personalized default rule is that because of strong societal interests, only explicit waiver of the fundamental values default rule should count; this rule cannot be waived by statistics indicating that a particular person would have opted for waiver if he had been given the choice. To illustrate this point, let us return to our implied warranty of habitability example174 and to the question of whether personalized default rules should be used to leave some tenants without such a warranty provided they have not opted out of the default rule. If the implied warranty of habitability is a fundamental values default rule, then it might be appropriate to apply it to everyone regardless of his characteristics, attributes, and traits, while still respecting explicit opt outs for autonomy reasons. Note that this attitude could be considered a compromise between the warranty of habitability as an immutable rule – whereby even an explicit opt out is invalid – and the warranty as a personalized rule.175 We acknowledge that there could be fundamental values default rules that render personalized rules undesirable because there is too much at stake for society for the contracting parties’ preferences to be decisive. Perhaps individual preferences contribute to a collective action problem or the risk of error is asymmetric, such that society is more willing to tolerate the problem of people paying for a warranty they do not want than the problem of people being denied a warranty that they (unpredictably) badly need. Determining what counts as a fundamental value default rule will, of course, prove contentious as well, and different readers may wish to confront these questions through various frameworks. Put another way, reliance on personalized default rules is most appropriate in those settings where the law is most comfortable deferring to the preferences of the people bound by an arrangement. But domains where the law is uncomfortable with letting individual choices govern – whether based on paternalistic 173

See Stein, Foundations of Evidence Law, 206–07 (2005). See supra Section III.2.d). 175 Cf Ayres, Regulating Opt-Out: An Economic Theory of Altering Rules, 121 Yale L.J. 2032, 2084–09 (2012) (describing sticky default rules as quasimandatory rules and arguing that sometimes the law makes opting out of certain default rules hard, but not impossible, in order to enhance paternalism or prevent externalities). 174

40

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:04 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

grounds, deontological grounds, or concerns about collective action or asymmetric risk – personalized default rules will work less well.

7. Subordination, adaptive preferences, and personalization Sunstein’s paper on default rules provides an arresting example of an American 124 default rule that may be simultaneously antimajoritarian and constitutionally compelled. Sunstein draws on fascinating work by Liz Emens,176 which notes that overwhelming majorities of American women change their surnames when they get married but trivial numbers of men do so.177 An obvious potential implication of these data is that a personalized default rule is appropriate. Changing one’s name is time consuming,178 but most women will adopt their husband’s name upon marriage, so the law could just presume that women adopt their husband’s names while providing an opt out for women who wish to retain their names or hyphenate their last names. Men’s default would be no name change, again with an option to override this default upon request. Sunstein contemplates the possibility of using a personalized majoritarian default for 125 women’s marital name changes but then rejects the idea, noting that “a default rule of this kind would be discriminatory, and it would almost certainly be found unconstitutional.”179 While Emens does not deem unconstitutional a waivable default rule presuming women wanted to change their names, she does argue that compulsory name changes for women would be unconstitutional,180 and she makes a persuasive feminist case that state rules increasing the likelihood that women will adopt their husbands’ surnames are normatively undesirable.181 We will explore the descriptive constitutional claim shortly, but let us first address the normative issue. We are sympathetic to Emens’s concerns about pressuring women to change their 126 names in light of the sexist history of name-changing conventions. We also share her concern that adaptive preferences may be causing women to change their names.182 These strike us as good reasons for the law to continue employing an impersonal default rule according to which marriage does not entail a surname change.183 Many women will continue to change their names, overcoming the stickiness of the law’s default term.184 But nearly everything associated with marriage entails undoing a default choice. The default choice is to remain single. Once one decides to get married, the default choice is not to serve food at the wedding, to forgo flowers, to wear pajamas during the ceremony (or no clothing at all!), and to send no thank-you notes after receiving gifts. In short, defaults are not really relevant in these high-stakes settings. The point is simply that if the state adopts a popular but inegalitarian default, the result may reinforce existing gender inequality, both because of the power of inertia185 and because of the expressive dimen176 Emens, Changing Name Changing: Framing Rules and the Future of Marital Names, 74 U. Chi. L. Rev. 761 (2007). 177 Id., at 785–86; Sunstein, supra (fn. 8), at 25. 178 Emens, supra (fn. 176), at 809. 179 Sunstein, supra (fn. 8), at 34. 180 See Emens, supra (fn. 176), at 774. 181 See id., at 770–77. One can conceptualize Emens’s claim as an argument that the status quo has appropriately adopted an antimajoritarian, social-welfare-maximizing, impersonal default rule, where gender equality plays a decisive role in the social-welfare calculus. 182 See id., at 775–76. 183 Recall that under the default-rule theory that seeks to maximize social welfare, societal values should be taken into account. See supra Section II.4. 184 Emens, supra (fn. 176), at 813. 185 Id., at 815 (“[A]t least one study of marital names offers anecdotal evidence of a few women saying that they didn’t change their names because they couldn’t be bothered with the administrative hassle.”).

Porat/Strahilevitz

41

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:05 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

127

128

129

130

sions of the law.186 We therefore agree with Emens and Sunstein that a crude personalized default, with gender as the only variable,187 is normatively unattractive. We think a more compelling case can be made for a granular personalized default rule. If one examines the name change data on which Emens relies, it is apparent that there are various demographic characteristics that substantially affect the probability that a spouse will adopt her husband’s name upon marriage. A study of female Harvard alumnae showed that 20 % of them kept their surnames, whereas a study of the overall population found that only 10 % of married women did so. A more recent study of New York Times wedding announcements found that 29 % of marrying women whose nuptials appeared in the paper of record were keeping their surnames.188 Women with advanced degrees, women who married or became mothers later in life, graduates of elite universities, daughters-in-law of academics, and women whose husbands have PhDs were more likely to retain their surnames.189 Interestingly, demographic variables affecting name changes interact in somewhat surprising ways. Education levels were highly predictive of whether Caucasian women would retain their surnames, but education had no effect on African American women’s choices about keeping their surnames. African American women generally retain their surnames at significantly higher rates than Caucasian women.190 In light of this substantial variation, how should one feel about a highly granular personalized default rule? Suppose it turned out that Caucasian women who regularly shop at Wal-Mart, frequently dine at Cracker Barrel, dropped out of college, and are marrying spouses with similar characteristics adopt their husband’s surnames 98 % of the time, but that Asian American women who have a master’s degree in education, subscribe to the Vegetarian Times and Mother Jones, and take yoga classes adopt their husband’s surnames only 7 % of the time. Would it be normatively undesirable for the state to adopt as a default rule the assumption that Caucasian women with these characteristics would see their surnames changed upon marriage but the Asian American women would not? Imagine if the data showed that 88 % of male, vegan, Prius drivers with PhDs in philosophy adopt their wives’ surnames upon marriage. Why not flip the default for these husbands to a name change unless they opted out? The red tape associated with a name change is nontrivial,191 and it may be that at some point the demographic markers of an individual’s preferences with respect to name changes are sufficiently strong that we need not worry so much about the law’s expressive effects. Crude personalized default rules that are dependent on mere stereotypes are undesirable, but granular personalized default rules based on hard data and sound science may be desirable. Indeed, if data miners can drill down and find a set of men whose names ought to be changed by default, then even the expressive dimensions of the law may be ambiguous. What is more, the law’s discomfort with relying exclusively on problematic classifications like race and gender may become less pronounced if these factors are mixed with a number of nonsuspect classifications to generate a default rule.192 186 See McAdams, A Focal Point Theory of Expressive Law, 86 Va. L. Rev. 1649 (2000); Sunstein, supra (fn. 8), at 20 (discussing the state’s implicit endorsement via default rules). 187 The crude personalized default would be that men keep their surnames upon marriage, and women adopt their husbands’ surnames upon marriage. 188 Emens, supra (fn. 176), at 786 & n.85. 189 Id., at 787–89. 190 Id., at 788. 191 Id., at 817–18. 192 See Grutter v. Bollinger, 539 U.S. 306, 336–41 (2003) (holding that a holistic higher education admissions process in which race is one of many factors considered survives strict scrutiny).

42

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:05 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

Even crude, gender-based personalized decision rules may be appropriate when the 131 dangers of reinforcing an inegalitarian gender norm are minor. Nguyen v. INS, a 2001 Supreme Court case, is one of the key precedents governing the law’s use of gender proxies.193 At issue in Nguyen was a government policy that imposed greater burdens on people seeking American citizenship who claimed to be the children of U.S. citizens born out of wedlock. The illegitimate children of U.S.-citizen fathers born out of wedlock could only become citizens if their fathers legally legitimated them, if their fathers declared their paternity under oath, or if a court order determined their paternity.194 For the mothers, by contrast, maternity was presumed.195 Given the law’s structure, the gender-discriminatory provision functioned as a default 132 rule. While women’s offspring were presumed to be citizens, men had to opt in to citizenship (through a declaration of paternity or legitimation) to receive the same rights for their offspring. The Supreme Court held that the gender classification was justified by two factors: first, the government’s interest in ensuring that the person claiming citizenship and the U.S.-citizen father are indeed biologically related and, second, the state interest in ensuring that the person claiming citizenship has a meaningful relationship with the U.S.-citizen parent and, by extension, with the United States.196 The majority rejected the idea that its decision was based on outmoded gender stereotypes: “There is nothing irrational or improper in the recognition that at the moment of birth – a critical event in the statutory scheme and in the whole tradition of citizenship law – the mother’s knowledge of the child and the fact of parenthood have been established in a way not guaranteed in the case of the unwed father. This is not a stereotype.”197 The Court proceeded to hold that placing additional burdens in the path of the 133 illegitimate children of U.S.-citizen fathers was substantially related to achieving important governmental objectives.198 The Court emphasized that “Congress has not erected inordinate and unnecessary hurdles to the conferral of citizenship on the children of citizen fathers in furthering its important objectives.”199 The burdens are comparable on an applicant for citizenship and a woman defaulted into a surname change. The key considerations for a court would be whether accepting a default rule for surnames that is consistent with most American women’s preferences is “marked by misconception and prejudice” or shows “disrespect for either class.”200 A court would also ask whether the preferences in question might be adaptive and whether they were shaped by a history of patriarchy. In light of Nguyen, it is not certain that implementing a crude personalized default 134 rule for surname changes upon marriage would be unconstitutional as a positive matter; the question is a close one. We continue to think that such a rule is undesirable for reasons that feminist legal scholars like Emens have articulated. Having said that, an advantage of granular personalized default rules, as opposed to crude gender-based distinctions, is that it may be easier to achieve doctrinal unity and popular consensus around such solutions – at least in a world where people do not care much about 193

533 U.S. 53 (2001). Nguyen, 533 U.S., at 62. 195 Id. 196 Id., at 63–67. 197 Id., at 68. 198 Id., at 70. 199 Id., at 70–71. 200 Id., at 73. 194

Porat/Strahilevitz

43

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:05 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

information privacy. A classic efficiency-versus-equity tradeoff thus arises. Crude personalized default rules, which conveniently mitigate the uncertainty problem associated with personalization, compound the constitutional problems associated with personalization. 135 We can generalize from Emens’s example of name changes to any legal regime that incorporates a protected classification like race or gender into a granular personalized default rule. It is reasonable to survey the history of a state’s race and gender discrimination and conclude that such classifications ought to rarely be part of the state’s efforts to generate default rules. Indeed, as Sunstein notes, a major variable in determining whether the use of personalized default rules is appropriate is the trustworthiness of the “choice architects” who will frame and determine the contents of these rules.201 But because gender and race can be reliable predictors of current preferences and future behavior, entirely excluding these variables from an algorithm leaves a great deal of predictive power on the table.202 Most people would probably prefer an algorithm that knows their race and gender and, as a result, more accurately predicts their preferences over a system that excludes their race and gender from consideration and consequently provides them with less accurate default rules. Finally, the fragmented nature of the precedents in the gender- and race-discrimination context is important to underscore. The Supreme Court seems to be animated by different concerns in race cases than in gender cases, and antidiscrimination law is consequently far from coherent. While Nguyen‘s defaults are the most relevant precedent, a comprehensive explanation of the constitutionality of personalized default rules would require a law review article unto itself.

8. Privacy Information privacy restrictions make it more difficult to generate personalized default rules.203 Without the ability to track individuals online, access comprehensive public and private databases, and use various other Big Data strategies, it will be quite difficult for firms and courts to generate personalized default rules. In the European Union, where regulators have generally taken a more aggressive approach to data privacy than their American counterparts,204 such restrictions could well thwart the development of personalized default rules. 137 The privacy literature has long recognized the tradeoffs that information privacy entails. Scholars have explored the tension between privacy and security,205 privacy and antidiscrimination,206 privacy and gender equality,207 and privacy and innovation.208 We 136

201

Sunstein, supra (fn. 8), at 8. See In re Estate of Miltenberger, 737 N.W.2d 513, 519 (Mich. Ct. App. 2007) (“It remains an unfortunate fact that there are still circumstances in which the surviving wife may be significantly disadvantaged, in a way that surviving husbands generally are not, in the absence of dower, and the Legislature may properly consider such circumstances through the enacted dower statute.”). 203 Id., at 23. 204 Schwartz, The E.U.–U.S. Privacy Collision: A Turn to Institutions and Procedures, 126 Harv. L. Rev. 1966, 1974–79 (2013). 205 See, e.g., Posner, Privacy, Surveillance, and Law, 75 U. Chi. L. Rev. 245 (2008); Solove, Data Mining and the Security-Liberty Debate, 75 U. Chi. L. Rev. 343 (2008). 206 Strahilevitz, Information and Exclusion, 127–56 (2011). 207 See, e.g., Siegel, “The Rule of Love”: Wife Beating as Prerogative and Privacy, 105 Yale L.J. 2117 (1996). 208 See, e.g., Bernstein, The Transparency Paradox: A Role for Privacy in Organizational Learning and Operational Control, 57 Admin. Sci. Q. 181, 202–04 (2012); Cohen, What Privacy Is For, 126 Harv. L. Rev. 1904, 1918–27 (2013). 202

44

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:05 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

can understand the privacy-personalization tradeoff in similar terms. One of the unanticipated consequences of aggressive data-privacy regulations will be a series of shifts toward impersonal default rules and away from personalized default rules, from granular personalized default rules to crude personalized default rules, and (as we shall see) from personalized disclosure to impersonalized disclosure. The industry attack on “Do Not Track” rules that would govern the collection of 138 information about consumers’ internet activities has been largely focused on the benefits of personalized advertisements to consumers as well as their obvious benefits to industry. Making consumers aware of the potential benefits from personalized defaults and personalized disclosure may, in the long run, prompt fewer consumers to try to thwart tracking. After all, most consumers bring strongly pragmatic perspectives to privacy tradeoffs, and they are increasingly willing to share information about themselves when the benefits from sharing are increased and the threats from sharing are diminished.209 There is obviously another potential wrinkle here. The primary debate over Do Not Track has been about the appropriate default rules. Industry groups are open to permitting individuals to opt out of tracking, but they want to require an affirmative step by consumers to reject a protracking default rule embedded in web browsers.210 Many marketing firms have said they will not honor Do Not Track requests sent by consumers using Microsoft’s Internet Explorer, which turns on Do Not Track by default.211 Paradoxically, one way around the current stalemate may be to use our lack of 139 privacy to further privacy interests. If a consumer’s existing profile reveals that she cares a great deal about her own information privacy, and if her behavior mirrors that of guinea pigs who chose to protect their own privacy online, then it should be straightforward to enable Do Not Track by default for that user. Similarly, if a consumer’s existing profile reveals minimal concern for privacy and has similar characteristics to those of the guinea pigs who decided to enable tracking online, then permitting tracking ought to be the appropriate default option. Using personalized defaults in this way is appealing in contexts like online privacy where defaults appear to be very sticky.212 Note that although enforcing a Do Not Track rule against firms is costly, it is straightforward to enforce an evidentiary rule that limits the admissibility of information from tracking to affect the personalized default rule that applies to a particular consumer. Familiar problems of adverse selection and unraveling will remain, with bad-credit types and high-privacy-concern types potentially becoming pooled,213 but this is not a problem unique to personalization. At the margins, the benefits of personalized default rules will 209 Westin, “Whatever Works”: The American Public’s Attitudes Toward Regulation and Self-Regulation on Consumer Privacy Issues, Nat’l Telecomm. & Info. Admin., http://www.ntia.doc.gov/page/ chapter-1-theory-markets-and-privacy (last visited 2 October 2012); see also Consumer Intelligence Series, PricewaterhouseCoopers, Consumer Privacy: What Are Consumers Willing to Share? 1–4 (2012), available at http://www.pwc.com/us/en/industry/entertainment-media/assets/pwc-consumer-privacy-andinformation-sharing.pdf. Approximately 16 % of the American population is privacy unconcerned, with respondents saying that no corporate uses of their personal information would violate their personal privacy boundaries. Id., at 8. 210 Singer, Do Not Track? Advertisers Say “Don’t Tread on Us”, New York Times, 14 October 2012, at BU3, available at http://www.nytimes.com/2012/10/14/technology/do-not-track-movement-is-drawingadvertisers-fire.html. 211 Id. 212 The default provisions contained in the Gramm-Leach-Bliley Act concerning the downstream sharing of consumers’ financial information are extraordinarily sticky, with only 1 in 200 consumers opting out of the prosharing statutory default. See Janger/Schwartz, The Gramm-Leach-Bliley Act, Information Privacy, and the Limits of Default Rules, 86 Minn. L. Rev. 1219, 1230 (2002). 213 See Peppet, Unraveling Privacy: The Personal Prospectus and the Threat of a Full-Disclosure Future, 105 Nw. U. L. Rev. 1153, 1177–82 (2011); Strahilevitz, supra (fn. 68), at 2021–22, 2030.

Porat/Strahilevitz

45

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:05 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

prompt more consumers to surrender private information, a development that is positive in efficiency terms but problematic to theorists who argue that privacy produces positive externalities.214

9. “But I Can Change!” and opting in 140

Before turning to a further extension of personalization, we hope to clarify one last point about our proposal for personalized default rules. Sunstein notes that the “best default rules or settings for a particular person in one year might be very different from those in the next year. The default rules could change on a daily or even hourly basis.”215 We are skeptical about the underlying assumptions of this objection. We think that most choices about default rules are partially driven by values and personality characteristics, which longitudinal research shows to be quite stable once a person reaches adulthood.216 Personality seems to have a strong genetic component and be heritable.217 That said, we recognize that people sometimes change in ways that might cause them to want wholesale revisions in their preferences. We therefore want to underscore that personalization is itself a default rule that can be waived. Suppose a consumer has a change of heart. She recognizes in the past that she has been risk seeking, inattentive, and price insensitive. A divorce, a bankruptcy, or a stint in rehab convinces her that she ought to turn over a new leaf. Under our proposal, she need not be stuck with the choices made by her former self. To escape the consequences of her consumer profile, she may specify that she rejects personalized defaults. She can specify that she instead wants to contract for the impersonal majoritarian default rule, or an impersonal minoritarian default rule, or randomized selection of default rules, or any other set of decision rules to which the counterparty might agree. Indeed, with the consent of the counterparty, a consumer might specify (through a contract) that the contract will be governed by the personalized default rules that would apply to a (presumably admirably rational) third party. “We hereby reject the Porat-Strahilevitz proposal for personalized default rules as a basis for interpreting this contract” would be a valid and enforceable contractual provision, as would, “We hereby agree that the promisee shall be entitled to the personalized default rules that would apply were this to be a contract between the promisor and Ralph Nader.” Alternatively, we are quite open to tweaking our proposal so that present-day impersonalized default terms are the default choice for everyone, and individuals who would like their terms to be personalized opt into doing so, either with a blanket opt in or particular opt ins for personalization in discrete settings.

214 See, e.g., Allen, Coercing Privacy, 40 Wm. & Mary L. Rev. 723 (1999); Schwartz, Property, Privacy, and Personal Data, 117 Harv. L. Rev. 2055 (2004). 215 Sunstein, supra (fn. 8), at 53. 216 A longitudinal study that tracked people’s personality over a nine-year period from ages thirty-three to forty-two found very high levels of consistency in each of the Big Five characteristics among both men and women. See Rantanen et al., Long-Term Stability in the Big Five Personality Traits in Adulthood, 48 Scandinavian J. Psychol. 511, 515 (2007) (showing an average correlation coefficient of .85 for men and .78 for women). Personality tends to be far less stable across longer periods of time or between young adulthood and late adulthood, although some studies show correlation coefficients in the .20 to .38 range for some of the Big Five across decades. See, e.g., Hampson/Goldberg, A First Large-Cohort Study of Personality-Trait Stability over the 40 Years Between Elementary School and Midlife, 91 J. Personality & Soc. Psychol. 763 (2006); Soldz/Vaillant, The Big Five Personality Traits and the Life Course: A 45-Year Longitudinal Study, 33 J. Res. Personality 208 (1999). 217 Jang et al., Heritability of the Big Five Personality Dimensions and Their Facets: A Twin Study, 64 J. Personality 577, 577 (1996).

46

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:06 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

V. Personalized disclosure The question of default rules has long vexed legal scholars and prompted an enormous academic literature. In the last few years, the topic of disclosures has become another hotbed of legal scholarship. In their noteworthy recent work, Omri Ben‐Shahar and Carl Schneider argue that disclosure to consumers rarely achieves what its advocates claim, in part because disclosures have a pronounced tendency to become longer and more complicated over time.218 Disclosure mandates accumulate in legislation and regulations, and as a result, the disclosures themselves get so lengthy and cumbersome that consumers stop reading them entirely. Our personalized-disclosure solution to the problems that Ben‐Shahar and Schneider identify should be obvious to readers by now, and it is surprising that it is an approach largely absent from the broader literature on disclosure. We have already shown how personalization might improve doctor-patient disclosures.219 In this part, we extend the idea to disclosure more broadly. Let us begin with the least controversial form of personalized disclosure: tailored communicative strategies. Advertisers have long understood that particular communicative methods might have varying effectiveness in reaching individuals with divergent personalities. This basic Madison Avenue assumption is buttressed by the available social science research.220 Now suppose that the government wishes to convey a particular message to the citizenry in a way that will maximize its impact. Suppose the government seeks to encourage new parents to put their infants to sleep on their backs to reduce the risk of Sudden Infant Death Syndrome (“SIDS”). Or suppose the state wants to convince as many people as possible to evacuate from an area in the path of a Category 4 hurricane, both to reduce the residents’ risk of death and to minimize the threat to first responders who might have to rescue those who refuse to evacuate. A government that knows nothing about its individual citizens’ personalities must broadcast a message designed to appeal to the majority. A government that knows its citizens’ personalities and can narrowcast to defined groups of individuals with common traits might tailor the state’s warnings in a way much more likely to persuade them.221 Just as it is sensible for the state to use different languages to communicate with non-Englishspeaking populations or to tailor to teens public service ads that are designed to discourage smoking, it is appropriate for the state to try to use psychological insights to tailor the packaging of information in a manner designed to prevent the unnecessary deaths of, say, babies or police officers. If tailoring the communicative strategy to a group of people with similar personality traits is uncontroversial, then altering the contents of the message should be acceptable as well, as long as all the information provided is truthful. Where consumers are purchasing items online, we propose a regime whereby their Big Data profiles help determine which disclosures they see. The advantages of such a regime are apparent. When online disclosures occur today, single males who live alone are shown warnings about the effects that 218

Ben‐Shahar/Schneider, The Failure of Mandated Disclosure, 159 U. Pa. L. Rev. 647, 684–90 (2011). See supra Section III.2.c). 220 See, e.g., Hirsh et al., Personalized Persuasion: Tailoring Persuasive Appeals to Recipients’ Personality Traits, 23 Psychol. Sci. 578 (2012). 221 Cf id., at 580 (“Tailored messages are considerably more effective than one-size-fits-all campaigns, and the effectiveness of tailoring increases with greater customization and adaptation to the unique features of the recipient.” (citations omitted)). 219

Porat/Strahilevitz

47

141

142

143

144

145

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:06 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

prescription medication may have on pregnant women. Childless seniors living in agerestricted communities are warned about how household goods may have small parts that toddlers can break off and swallowed. Devout Mormons are warned about the effects of mixing a particular pharmaceutical with alcohol. The proliferation of warnings targeted toward a small set of potential consumers greatly lengthens disclosures, heightening the risk that a consumer will fail to see the one or two warnings that are very pertinent to people just like him. Too much disclosure can be as bad as too little disclosure because both result in a consumer retaining too little pertinent information. We submit that the disclosure strategy can be rescued and rejuvenated by a personalization strategy that makes the disclosures each consumer sees shorter and more relevant. 146 As technology improves, we would anticipate this sort of personalization of disclosures occurring even in brick-and-mortar supermarkets, shopping centers, and hardware stores.222 Twentieth-century disclosure technology involved a printed label with finite space and constraints on how much manufacturers could shrink font sizes to cram more information into those spaces. Twenty-first century disclosure technology ought to take advantage of the fact that a significant proportion of consumers now shop with smartphones that can scan bar codes.223 Personalized disclosure applications would enable a consumer to scan a product at the point of sale and to see only the disclosures and warnings likely to be relevant to him. We believe the health and safety gains from such innovation could be substantial. Personalizing disclosures will reduce the time that’s wasted when people have to see irrelevant disclosures and reduce the frequency with which people fail to notice a key disclosure that is buried amid many irrelevant disclosures. 147 We believe this proposal for personalized disclosure is novel. Although we think our idea is intuitive, we are unaware of any academic literature discussing the prospects of using Big Data to personalize the disclosures that individuals receive. The new literature on “Smart Disclosure” proposes that in some contexts, like cellphone plans or energy usage, vendors would be required to provide consumers with data about their individual usage patterns, which consumers could then use to compare products and services on shopping web sites. Armed with this information, consumers could see what different firms charge customers whose expected use of a service mirrored theirs.224 Our approach has many of the same virtues, although it aims to be less cumbersome and therefore more useful to the consumer; it also tries to mitigate an information overload problem that Smart Disclosure does too little to solve. Another close cousin in the literature is a recent article by Gil Siegal, Richard Bonnie, and Paul Appelbaum that discusses personalized disclosure in medicine.225 Their version of “personalized disclosure” differs from ours, and we think it lacks some of the advantages of our approach. Their first approach to personalized 222 A thoughtful article exploring possible uses of these technologies in consumer contracting at the point of sale is Peppet, supra (fn. 101), at 676. 223 Kumar, Shoppers with Smartphones Put Retailers in Glass Boxes, St. Louis Post-Dispatch, 15 February 2011, at A1, available at http://www.stltoday.com/business/local/mobile-shoppers-shakingup-retail-industry/article_4fccd0af-7447–5f2f-8f81–45d52e56520b.html. 224 See, e.g., Thaler/Tucker, Smarter Information, Smarter Consumers, Harv. Bus. Rev., Jan.–Feb. 2013, at 3; Brody, Product Attribute Information and Product Use Information: Smart Disclosure and New Policy Implications for Consumers’ Protection, (25 September 2013) (unpublished manuscript), available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2142734. 225 Siegal et al., supra (fn. 115), at 359. Peppet’s article has one sentence flagging the possibility that the warnings a customer sees at the point of sale could depend on the preferences that the customer previously entered into his smart device. Peppet, supra (fn. 101), at 730.

48

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:06 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

disclosure asks patients at the outset whether they would like to receive: (a) very detailed and precise disclosure of side effects and medical risks, including information likely to interest only a small subset of patients; (b) moderately detailed and general disclosure of side effects and risks, where minor and insignificant risks are not disclosed to the patient; or (c) very basic disclosures, such as the reasons for the treatment and the likely period of time the patient will have to miss work.226 They view the patient’s choice about how much disclosure to receive as legally significant: “once a patient has stated his preferences and the procedure has taken place, he may no longer argue in court that the informed consent process was inadequate in that it failed to provide him with the information he needed.”227 Siegal and his coauthors also identify a second form of personalized disclosure, which 148 they seem to prefer. Under that approach, disclosure would occur via software that enabled the patient to click on hyperlinks to find out more about particular risks, side effects, or tradeoffs.228 The software would record a transcript of what the patient asked to see and did not ask to see, and this transcript would be admissible evidence in any subsequent litigation over informed consent.229 We think Siegal’s proposal is a step in the right direction, but as Big Data proliferates 149 and the technologies underlying FICO Adherence Scoring improve, we think there is a strong case to be made for preferring our version of personalized disclosure. Answering many questions about whether one wants to read a particular paragraph may increase the stress levels of patients, particularly those who know that by selecting the minimal disclosure option or failing to click on a particular hyperlink they will be waiving various legal rights. A regime that scrutinizes the choices that representative guinea pigs have made with the benefit of full information may be a more sensible way to proceed.230 Indeed, guinea pigs might work differently in the personalized-disclosure context. 150 We envision compensating guinea pigs to read various disclosures and to evaluate (both immediately and several weeks after the treatment at issue) how useful the disclosed information proved to be. Non-guinea-pig patients would then be paired with the choices made by the guinea pigs with personalities and attributes most similar to theirs. The key point is that different warnings will be helpful in different ways to different sorts of people. Personalized disclosure thus identifies the warnings that were useful to “people like you” or “people like those in your household” and only provides you with those warnings unless you opt for more complete disclosure. Parents whose children have peanut allergies will constantly see peanut-related warnings about products they are considering – including perhaps an “Are you sure?” message in the checkout line; parents whose own children do not have allergies but who may be bringing snacks for a kindergarten class will need to opt in to receive allergen information when circumstances require such additional precautions. We anticipate that these sorts of personalized disclosures will save consumers a great deal of time. More importantly, however, they will prompt more consumers to actually read carefully health and safety disclosures. 226

Siegal et al., supra (fn. 115), at 361–62. Id., at 362. 228 Id., at 363. 229 Id., at 364. 230 To the extent that a patient or consumer is informed of the right to opt out, it may cause some stress, especially during the transition to personalized disclosure. Decisions to trust guinea pigs will resemble leaps of faith that would decrease as time passes and as consumers become more comfortable with using guinea pig’ preferences as proxies. 227

Porat/Strahilevitz

49

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:06 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

We anticipate that such personalized disclosures are likely to take root in the area of consumer warnings and may spread to other domains as well. For example, a smartphone application that knows, based on Big Data and guinea pigs, that you are likely to be concerned about particular sorts of risks can also learn that you are concerned by particular contractual provisions. Most people may not care about the terms of clickwrap software agreements, but some users may be sensitive about particular rights, responsibilities, and waivers. Through automation, an application could do what a good lawyer already does – read the contract and advise the client about provisions that may be problematic in light of the client’s idiosyncrasies. Here, again, consumers could benefit from the close scrutiny that compensated guinea pigs would devote to reading all the pertinent contractual provisions. 152 Personalization may play a similar role in the context of governmental disclosures. For example, just as we might want the government to tailor its warnings about SIDS to audiences with different personalities,231 it may make sense for the government to target air quality warnings directly to asthmatics (and their parents) instead of broadcasting such warnings through mass-media outlets unlikely to pay them much heed. A city government that knows our daily commute patterns (because we have agreed to share them) can let us know about accidents along the route while staying silent about accidents on other highways in the metropolitan area. Under the status quo, consumers and voters can always “pull” such information out of the public sphere, but doing so can be difficult and costly. Personalized disclosure may often be the most efficient mechanism for pushing the right information to the right people, assuming the state can be trusted to put information about individual citizens to appropriate uses. 153 Finally, we think there is an important role to play for personalized disclosures in personalizing default rules. Of course, some consumers will respond better than others to the possibility that they are entering into a contract whose terms are dependent on choices made by others. Consumers whose profiles suggest that this level of uncertainty is likely to upset them might receive additional disclosures about anticipated directions of those changes and be given easy opportunities to reject such changes. Consumers whose profiles indicate an interest in saving money wherever they can – even if it means more onerous contractual terms – might receive regular notices about terms that could be modified if the customer wishes to realize a cost savings. Other consumers, who rarely elect to pay less in exchange for fewer contractual rights, would receive fewer notices of this kind. In short, there are many ways in which personalized disclosure could address some of the complexity problems that arise with personalized contracts. Personalized disclosures can help consumers determine what their existing profiles indicate about the meaning of a contract they are contemplating signing and how their profiles are influencing the contractual terms. Where similar guinea pigs were not unified over which terms were best, the consumer may be presented with active choices among several default terms or instructions on how the default might be altered. 154 While there are many objections to personalizing default rules, we think that the objections to personalized disclosure are fewer and less significant. As with default rules, an individual could always request disclosure of a greater amount of information than what personalization suggests, and we would want these choices to be honored. Given this possibility, it is hard to imagine that individuals would engage in strategic behavior to affect the disclosures to them, and a personalized disclosure regime can easily accommodate changes in individuals’ personalities and preferences. Concerns about 151

231

50

See supra text accompanying fn. 220.

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:06 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

cross subsidies do not arise with respect to personalized disclosure, nor do concerns about uncertainty or fragmentation. And it seems unlikely that the government would find constitutional objections to personalized disclosure – the state regularly makes judgments about which messages should be conveyed to which audiences, and it seems hard to believe that even race-based messaging, such as extra warnings to African Americans about the dangers of sickle cell anemia, is constitutionally problematic. The potential downsides of personalized disclosure, then, seem confined to misgivings about stereotyping and privacy. There may also be worries about whether courts are really willing to countenance the possibility that someone might not receive a warning about an extremely unlikely side effect of a drug based on their personality profile, and then, due to some fluke, this low-probability side effect manifests itself.232 In such circumstances, courts should not award compensation. Social insurance, rather than the tort system, is the best mechanism for compensating victims, given the inability of would-be defendants to fully reap the benefits of nondisclosure resulting from personalization. To summarize, we think that personalized disclosures may be the wave of the future, 155 too. They have the potential to minimize consumers’ information-overload problem and to prompt them to start paying attention once again to pertinent disclosed information. And they even have the potential to alter, for the better, the way that contracting is done.

VI. Conclusion The idea of personalized default rules has been “in the air” for several decades. 156 Although the origins of our inquiry can be found in Ayres’s essay, which was published twenty years ago, no one has developed a comprehensive account of personalized default rules. Sunstein took the idea an important step further and pointed out some of the main benefits and drawbacks of a personalized default rule regime compared with impersonal defaults and active choices. Our Article, then, has finally developed a comprehensive framework for understanding the theoretical and practical issues arising in the implementation of personalized default rules. Along the way, we have contributed several innovations. For example, we have shown 157 how providing a limited number of guinea pigs with resources to make rational decisions, and using these guinea pigs’ choices to generate the default rules that will be presented to the most similar members of the general public, makes personalization substantially more attractive. We have explained how majoritarian and minoritarian default rules might be made more effective through personalization. And we have broken down the category of personalized default rules into crude personalized defaults (which are more easily predictable for the parties, applied with more certainty by adjudicators, less precise, and more impervious to strategic behavior) and granular personalized defaults (which have the opposite costs and benefits). Perhaps most interestingly, we have shown that personalization may present an important way forward, not only for default rules but also for various disclosures to consumers and the citizenry. As we have demonstrated, the most powerful critiques that have been lodged against disclosure are largely products of disclosure’s impersonal nature. The disclosure strategy can be resuscitated via personalization. Why has it taken the literature so long to reach this juncture? We believe the answer 158 is that until recently, technological constraints would have rendered our approach 232 By contrast, we would support legal liability for negligent or reckless profiling that results in harm because an individual was incorrectly categorized.

Porat/Strahilevitz

51

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:07 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

wildly unrealistic. But the Big Data revolution fundamentally changed the equation, at least in the United States. Now more than ever, implementing a personalized default rule regime is attainable, and personalized disclosures are within reach, given minor improvements in the social science research and applicable technology. Big Data has its own economic momentum, but we have endeavored to show how its development might open up new possibilities for legal intervention. To that end, we call on legislatures and courts to respond to the challenge proposed in this Article by considering personalized default rules for consumer contracts, contracts between repeat players, inheritance law, medical malpractice, and landlord-tenant law. 159 Legislatures should consider tailoring personalized default rules, at least in those areas when it is quite obvious that the law’s goals could be better achieved with such rules and where implementing them is feasible and not too costly. Thus, in inheritance law, intestacy rules should be personalized in accordance with existing data, provided a bit more research is conducted about whether the preferences and characteristics of intestates differ from those of testators of the same gender. Courts hearing medical malpractice suits should allow doctors to argue that they adopted a disclosure practice that is consistent with the personal characteristics of their patients, as revealed by FICO Adherence Scores and other data-driven patient profiling technologies. Courts should also avoid applying constitutional rules developed before personalization could be contemplated in a manner that suffocates personalized rules in their infancy. Regulators should fund pilot projects to facilitate personalized disclosure, and legislators might create safe-harbor provisions to encourage manufacturers, retailers, and service providers to begin innovating with personalized disclosures in the private sector. 160 We realize that personalizing default rules and disclosure is costly. There is a tradeoff here, somewhat similar to the rules-versus-standards tradeoff, between certainty and accuracy: more personally detailed default rules could increase accuracy but at the same time create uncertainty for courts applying default rules to disputes and for private actors trying to anticipate what courts might do. Because the tradeoffs are significant, we advocate beginning with personalized default rules in the easiest cases, followed by incremental advances if the early results are promising. 161 Personalized default rules and personalized disclosure are just two important pieces of a more ambitious idea, which is personalized law in general. One could imagine a legal system where criminal law, constitutional law, tort law, and property law are personally tailored to people’s preferences and characteristics. Indeed, aspects of these bodies of law are already personalized to some degree. Consider insanity defenses or the Sentencing Guidelines in criminal law, litigant sensitivity in First Amendment law,233 the eggshell skull doctrine in tort law,234 and the focus on a landowner’s “distinct, investment-backed expectations” in takings doctrine. We might anticipate far more granular and data-driven personalization in each of these domains during the coming years. Envisioning such a legal system is beyond our present project, and the conversation becomes increasingly fraught as the idea is stretched from contract to criminal

233

See Samaha, Litigant Sensitivity in First Amendment Law, 98 Nw. U. L. Rev. 1291 (2004). See, e.g., Klimchuck, Causation, Thin Skulls, and Equality, 11 Can. J.L. & Jurisprudence 115, 124 (1998). 234

52

Porat/Strahilevitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/01_Porat Strahilevitz PP Complete - ls edits2.3d from 30.10.2020 11:33:07 3B2 9.1.580; Page size: 160.00mm  240.00mm

A. Personalizing Default Rules and Disclosure with Big Data

settings.235 Nevertheless, we believe the case for trying personalized default rules and personalized disclosure in various contexts is sufficiently compelling to warrant nearterm experimentation. 235 One research paper suggests that juvenile recidivists and nonrecidivists differ significantly in terms of Big Five agreeableness and neuroticisms, with repeat offenders scoring higher on neuroticism and lower on agreeableness. The paper also finds that offenders in general are less agreeable and less open to new experiences than nonoffenders. van Dam et al., PEN, Big Five, Juvenile Delinquency and Criminal Recidivism, 39 Personality & Individual Differences 7, 14 (2005). If these results are generalizable – the research is very sparse – they suggest that personality might be relevant in personalizing sentencing and parole decisions as well as in developing offender profiles. That said, we have normative misgivings about such an approach to criminal law in light of concerns about stigmatization, incorrect use, and manipulability. These concerns become particularly forceful in light of personality’s strong genetic component. See Jang et al., supra (fn. 217), at 577. But if determining personality through Big Data is relatively straightforward, if personality might predict aspects of criminality and recidivism, and if the government has access to a large store of data about individuals, then it is conceivable that governments around the world already employ (or will soon employ) such techniques.

Porat/Strahilevitz

53

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law* 1

The most fundamental feature of negligence law is the “reasonable person” standard. This feature bases negligence law on a strictly objective foundation: It requires people to behave in the prudent way that, as Holmes explained, the ordinary, typical member of their community observes. In this Article we argue that with the increasing availability of information about actors’ characteristics, negligence law should give up much of its objectivity by allowing courts to “subjectify” the standard of care – that is, to tailor it to the specific injurer’s tendency to create risks and his or her ability to reduce them. We discuss the effects of this personalization of the standard of care on injurers’ and victims’ incentives to take care, injurers’ activity levels, and the injurers’ ex ante investments in improving their skills. We also discuss justice considerations as well as the feasibility of personalization with the aid of “Big Data.”

I. Introduction The law takes no account of the infinite varieties of temperament, intellect, and education which make the internal character of a given act so different in different men.1 3 The most fundamental feature of negligence law is the “reasonable person” standard. This feature bases negligence law on a strictly objective foundation: It requires people to behave in the prudent way that, as Holmes explained, the ordinary, typical member of their community observes.2 The standard of care is uniform across the population, rarely varying with the skills and dangers of each actor. 4 This Article challenges the reasonable person paradigm. We argue that with the increasing availability of accurate information about actors’ characteristics, negligence law should give up much of its objectivity by allowing courts to “subjectify” the standard of care – that is, to tailor it to the specific actor’s tendency to create risks and her ability to reduce them. Rather than addressing each actor as a nondistinct member of a large pool and commanding her to meet the level of reasonable precautions that correspond to the average competence within the pool, a personalized negligence law would separate the actor from the pool and require her to meet her own customized standard of care. The reasonable person standard, traditionally derived from an aggregate relevant pool, would be replaced by the 2

* For very helpful comments we thank Hanoch Dagan, Bar Dor, Michael Frakes, Daniel Hemel, Saul Levmore, Jonathan Masur, Anthony Niblett, Julie Roin, Catherine Sharkey, Michael Trebilcock, Ernest Weinrib, David Weisbach, and workshop participants at the University of Chicago, the Hebrew University, the NYC Tort Group, Tel‐Aviv University, University of Toronto, UCLA, and the 2015 American Law & Economics Association meeting. We also thank Daniel Kopilov, Peter Salib, and Dana Zuk for excellent research assistance. We are grateful for financial support from the Coase‐Sandor Institute at the University of Chicago Law School and the Cegla Center for Interdisciplinary Research in the Law at Tel‐Aviv University. This chapter was originally published in 91 NYU Law Review 627 (2016). 1 Holmes, The Common Law, Little, Brown, & Co. 1881, 108. 2 Dobbs, The Law of Torts, § 117, at 277 (1st edn. 2000) (“The duty owed by all people generally – the standard of care – is the duty to exercise the care that would be exercised by a reasonable and prudent person under the same or similar circumstances (…).” (footnote omitted)); id. § 122, at 290 (“A reasonable person will act in the light of (…) knowledge shared by the community generally (…).”).

54

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

“reasonable you” standard – a personalized command that is based on information about this actor’s specific characteristics. The idea that standards of care ought to be personalized to the particular circumstances of the particular defendant may strike our readers as old news. Surely, a doctor is required to perform a treatment at a more advanced level of care than a layperson, and a physically disabled person may be allowed to satisfy a lower level of precaution. An actor who has special knowledge or experience may be required to do more than one who has not. Despite this intuition, tailored standards of care are the exception in tort law, not the rule. From its early days, negligence law has wrestled with the personalization problem. When a cognitively limited defendant who caused a fire asked the court to acknowledge his incompetence and apply a more forgiving standard of care, the court – in a landmark decision – refused, explaining that it would be impossible for negligence liability to be “co-extensive with the judgment of each individual, which would be as variable as the length of the foot of each individual.”3 The court, instead, chose “to adhere to the rule which requires in all cases a regard to caution such as a man of ordinary prudence would observe.”4 Holmes later explained that this approach was justified by the “impossibility of nicely measuring a man’s powers and limitations.”5 Yet, over time, negligence law has created subcategories of actors, lowering or raising the standard of care within each category to reflect special skills. For example, children or the physically disabled may be held to lower standards of care (although their license to engage in the activity in the first place may be more stringent).6 And, conversely, medical professionals are held to higher standards than nonprofessionals.7 Personalized negligence law – the reasonable you standard – takes this already familiar (but sparingly applied) approach of partitioning injurers into relevant classes, and expands it to its conceptual limit. Whereas the heightened standard of care for doctors creates a specific pool (all doctors in the relevant practice, or the specific advanced specialty), it still relies on average competence within a defined pool to determine what is reasonable. Personalized negligence law creates a pool of one for each defendant. What is reasonable for this defendant would be determined, not by reference to the average traits of some larger reference group to which this defendant belongs, but only according to the information available about this defendant. Consider, for example, a typical problem addressed by negligence law: What is a reasonable driving speed in treacherous road conditions? Imagine that a sixty-five-yearold driver, cruising at thirty-five miles per hour, injures a child who jumped into the street chasing a ball. Under prevailing negligence law, the court assumes that the driver is no different than any other driver in the population and would accordingly set the standard of care according to the capabilities, the reaction time, and the tendency to inflict harm that the court expects the average driver to have. If at a speed of thirty-five miles per hour, the average brake time for drivers is thought to be short enough even in relation to the risk of children playing in a residential neighborhood, the sixty-five-yearold driver would not be regarded as negligent. Under personalized negligence law, the capabilities of the average driver are not relevant. First, it might be true that the average sixty-five-year-old driver has inferior 3

Vaughan v. Menlove (1837) 132 Eng. Rep. 490, 493; 3 Bing. 468, 475. Id. 5 Holmes, supra (fn. 1), at 108. 6 See infra Section II.1 (describing lower standard of care for children and physically disabled). 7 See infra Section II.2 (describing heightened standard of care sometimes imposed on those with elevated capacity). 4

Ben-Shahar/Porat

55

5

6

7

8

9

10

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

driving capabilities and slower response times than the average driver. If so, a speed of thirty-five miles per hour might not be negligent for younger drivers, but would be negligent for the elderly.8 Merely partitioning the population of drivers and deriving the applicable standard of care from a smaller subset of the population would be a first, albeit crude, step in personalization. 11 But personalization would not stop there. Not all sixty-five-year-old drivers are alike. The courts might have additional information about the specific defendant, which would allow for further refinement of the standard of care. Some of that information might relate to his past driving experience. This would allow the court to make a statistical inference about the defendant’s risk “type” and adjust his standard accordingly. Such personalization based on past experience is similar to the “experience rating” methodology that insurers use in inferring idiosyncratic risk, which then impacts the pricing of automobile insurance policies.9 It is also similar to the approach used in criminal law – treating past offenders differently than first timers.10 12 More interestingly, some of the additional information deployed in constructing the reasonable you standard might relate to the defendant’s other characteristics, beyond his past driving record. It is information reflecting on his driving capabilities, other risky activity he takes, and his skills in reducing these risks. This information would allow the court to make more reliable inferences about the risk that this defendant creates, the risk he should have created, and the precautions he should have taken given his characteristics. With the aid of more advanced information tools – including what has come to be known as “Big Data”11 – courts might know that the defendant is very risk averse (or risk preferred), that he engages in frequent activities that make his instincts and reactions faster (or slower) than those of the average driver, or that in other parts of his life he is generally a very careful (or careless) person. A clumsy, impulsive, or prone-to-lapses person may need to be confronted with a more demanding standard of care. Again, similar to the “feature rating” methodology that insurers use to rate policyholders, courts could use statistical correlations in assessing the risk posed by the defendant. Taking into account every known relevant factor would assist the court in setting the more efficient reasonable you standard, where the level of care is tailored to a specific individual’s risk creation and capacity to mitigate that risk. 8 It might be that older age brings more experience and responsibility, which could pull to the other direction. This is typically the case with very young drivers versus older drivers, but not when the ages are, say, forty-five and sixty-five. See, e.g., Charlton et al., Older Driver Distraction: A Naturalistic Study of Behaviour at Intersections, 58 Accident Analysis & Prevention 271, 277 (2013) (“[O]lder drivers selfregulate by limiting their engagement in secondary activities when the driving task is more challenging compared with less demanding situations.”); Horberry et al., Driver Distraction: The Effects of Concurrent In-Vehicle Tasks, Road Environment Complexity and Age on Driving Performance, 38 Accident Analysis & Prevention 185, 189–90 (2006) (studying effects of distractions upon driving performance and finding that drivers over age of sixty tend to drive more slowly and cautiously while distracted in order to compensate for slower reaction times). 9 See Norberg, The Credibility Approach to Experience Rating, 1979 Scandinavian Actuarial J. 181, 181–82 (“[Driver] premiums are adjusted annually according to bonus rules, which are to the effect that drivers with a favourable claims record are allowed a premium deduction (bonus), whilst those with an unfavourable one will experience a premium increase (malus).”). 10 Dana, Rethinking the Puzzle of Escalating Penalties for Repeat Offenders, 110 Yale L.J. 733, 735 (2001) (“The legal system punishes repeat offenders more severely than nonrepeat offenders. Second-time offenders receive more severe punishment than first-time offenders; repeat offenders with many previous offenses receive more severe punishment than repeat offenders with a few previous offenses.”). 11 See Porat/Strahilevitz, supra Part 1.A (discussing concept of “Big Data,” its legal applications, and its possible role in the personalization of default rules).

56

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

This Article examines the case for personalized negligence law along two channels of inquiry. The first channel is normative: Does personalization advance the goals of negligence law – efficient deterrence and just compensation? In exploring these questions, one of the major contributions of this Article is the distinction between skill-based and risk-based personalization. The Article demonstrates the effects of personalization along those dimensions in various ways. The first dimension – skill-based personalization – addresses each actor’s subjective ability to take precautions. This dimension measures how effectively an actor can take precautions to reduce the risk of injury. Thus, if there is a costly technological device that drivers can use to reduce risks of driving, but which requires high technical skills to use effectively, it might be cost justified to require only the technically skillful drivers to use it. The second dimension – risk-based personalization – addresses each actor’s riskiness. This dimension measures the different risks that actors create at any given investment in care. Risk-based personalization would place a greater precaution burden on actors who, at any given level of care, create higher risks. For example, drivers with poor eyesight, or who are easily distractible, create higher risks at any given speed, and should be required to take more precautions and drive slower than those with better eyesight or instincts. In reality, actors’ characteristics may combine both idiosyncratic skill and riskiness. A driver may be both highly skilled and low risk. She may have good eyesight and good instincts (that is, low risk), but also high proficiency in utilizing accident prevention technologies. In such cases, personalization should take into account both aspects, which might counteract. Conversely, a low-skill driver would not be required to install a costly skill-intensive precaution technology. But if she is also a high-risk type, she would be required to take high alternative precautions. This distinction between skillfulness and riskiness is fundamental to our analysis. To be sure, under some abstractions the two dimensions may be regarded as equivalent. Both high risk and low skill increase the injurer-specific expected risk. It might be thought, then, that unifying the two dimensions into a single analysis of personalized expected risk would economize the analysis without sacrificing insight. We show throughout the Article that this reductive form is too impoverished to capture important insights of legal design. Skill-based personalization raises a host of different concerns and implementation problems than risk-based personalization. In an important set of cases, we find that risk-based personalization is easier to defend, both on efficiency and justice grounds. The effects of personalization on efficient precaution taking are just the tip of the iceberg. The Article identifies a wealth of effects that personalized standards of care would have on injurers’ precaution, activity levels, and ex ante incentives to invest in reducing their harmfulness. Relative to a regime of uniform standards, personalization leads to more efficient precaution and has the potential to alleviate the excessive-activity distortion inherent in negligence rules. Currently, negligence law incentivizes actors to become more skilled at harm reduction but not to reduce their riskiness when possible. This latter effect can be tackled, we show, if personalization is designed correctly. Personalized standards also affect victims in predictable ways. It might be intuitively thought that by facing personalized standards of care by injurers (say, drivers each obeying a different, personalized speed limit), victims endure a more uncertain and volatile environment, thereby diminishing their ability to take efficient contributory care. Not so. While drivers’ speed – and other precautions – may vary more under personalized standards, the risks that they pose to victims may in fact be less variable and more easily mitigated. Ben-Shahar/Porat

57

13

14

15

16

17

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

The Article then turns to provide some benchmark observations about how to evaluate personalization from a justice perspective. Under some conceptions of corrective justice, personalization may be regarded as problematic. Its primary prescription – to adjust the injurer’s obligation based on the cost of care – infringes the notion of equality between the injurer and victim, because it allows the injurer to unilaterally draw the line between his and the victim’s rights. We disagree with this evaluation. Why should a particularly skilled injurer not owe a heightened duty of care to a victim, and be required to correct this victim’s harm when breaching the duty? Moreover, personalization is not merely about different burdens of care, but also about different risks which different injurers create. We argue that raising the liability standard for people who create above-average risk and lowering the standard for people who create belowaverage risk is required by any plausible corrective justice account. Lastly, we comment on the distributive justice aspects of personalization. True, it treats similarly situated injurers differently, and it might expose some victims to higher risks of physical injury. But, quite intuitively, personalization has the potential of promoting equality among injurers along important aspects, while at the same time increasing victims’ safety. 19 After presenting the normative inquiry, the Article turns to a positive channel. It asks whether personalization can be implemented. What are the information obstacles to personalization and how can they be addressed? What sources of information might be harnessed to the personalization enterprise? Not surprisingly, we envision a process that relies on advances in information technology, from in-depth screenings of individuals to statistical analysis of large data. If Big Data is reliably predictive in high-stakes industries like financial services, insurance, and increasingly in medicine, why not utilize this predictive power in law? 20 It is not enough, however, to show that more data and better screening could be deployed by courts in adjudication. The challenge for a successful negligence regime is to show that actors would be able to anticipate the more refined burdens and adjust their behavior. Otherwise, if the greater ex post accuracy does not translate to ex ante behavior, it might merely impose excessive information costs.12 Recognizing this dilemma, we make the counterintuitive argument that personalization could make it easier, not harder, for actors to predict the standard of care applicable to them. People often know better what is reasonable for them to do, given their idiosyncratic characteristics. It is harder to know what the average skills and risks are. Thus, in the driving example, the prevailing reasonable person standard asks the driver to meet a standard of care tailored to the impersonal reasonable driver. But he is not this driver, and he would need much more information than mere self-introspection to figure it out. 21 This Article fits within a literature that examined the optimal tailoring of legal rules.13 The idea of personalizing default rules, for example, has been studied in various contexts by several authors,14 and further expanded recently by Ariel Porat and Lior 18

12 See Kaplow, A Model of the Optimal Complexity of Legal Rules, 11 J.L. Econ. & Org. 150, 151 (1995) (arguing that low information costs for enforcement authority improve complex rules’ efficiency); cf Kaplow, General Characteristics of Rules, in: Bouckaert/De Geest (eds.), Encyclopedia of Law and Economics, Volume 5, 2000, 502, 504 [hereinafter Kaplow, General Characteristics] (discussing connection between increased specificity in the law and greater information costs to both government enforcement and private actors seeking to comply). 13 At the most general level, Kaplow’s work on the optimal precision of legal rules lays a foundation for the inquiry into tailoring any legal command. Kaplow, General Characteristics, supra (fn. 12), at 502–07 (discussing possible problems caused by rule precision and analyzing its negative and positive effects). 14 See Ayres/Gertner, Filling Gaps in Incomplete Contracts: An Economic Theory of Default Rules, 99 Yale L.J. 87, 89–95, 97–98, 115–18 (1989) (differentiating between tailored, untailored, and penalty default rules in contract law and providing theory for when courts should fill contractual gaps using each

58

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

Strahilevitz.15 In the torts literature, early law and economics writers recognized that tailored duties could improve efficiency.16 Judge Richard Posner and Steven Shavell have separately explained that the reason why the standard of care is not adapted to the specific injurer is the saving of administrative costs.17 Shavell further showed that if courts are constrained to apply a uniform standard of care for all injurers, they should minimize the costs of some injurers taking too much, and others taking too little, care.18 These writers, as well as Warren Schwartz in an excellent article, have recognized that personalized standards of care might have problematic effects on the level of activity.19 They recognized that uniform standards could drive out the activities of very high-risk injurers.20 Finally, both Shavell and Schwartz recognized that the incentives to make ex ante investments (such as being sober while driving or acquiring information about risks) would be affected by a personalized standards regime.21 None of these authors, however, have examined the distinction between injurers who vary by skill and injurers who vary by riskiness.22 The Article proceeds as follows: Part II introduces the concept of personalizing the 22 standard of care and outlines some of its appearances in prevailing tort law. Part III develops the claim that personalizing the standard of care is generally more efficient than having a “one-size-fits-all” standard of care. Part IV looks at personalizing the standard of care from a justice perspective, showing that while corrective justice notions might be consistent with personalization in only some cases, distributive justice method); Geis, An Experiment in the Optimal Precision of Contract Default Rules, 80 Tul. L. Rev. 1109, 1114–15, 1129–59 (2006) (offering models of tailored and untailored default rules under particular sets of assumptions to analyze welfare implications of trading off precision against complexity); Sunstein, Deciding by Default, 162 U. Pa. L. Rev. 1, 7–10, 56–57 (2013) (differentiating between impersonal default rules, active choosing, and personalized default rules, and concluding that the choice between regimes is dependent on costs of decisions and errors, and therefore varies between target groups). 15 Porat/Strahilevitz, supra Part 1.A (suggesting use of Big Data to personalize disclosures, thereby increasing their relevance and effectiveness). 16 See Landes/Posner, The Economic Structure of Tort Law, 124 (1987) (arguing that uniform standard creates two effects of misallocation: injurers with low marginal costs of taking care would have no incentive to take care beyond the reasonable person standard, even though it would be socially desirable for them to do so, and injurers with slightly higher than average marginal costs of care would nevertheless adhere to the uniform standard so as to avoid bearing all liability). 17 Posner, Economic Analysis of Law, 218 (8th edn. 2011) (arguing that reasonable person standard adhered to by courts is justified by administrative costs courts would bear in attempting to measure actual individual costs of each party); see Shavell, Economic Analysis of Accident Law, 89 (1987) (suggesting that classification of injurers would allow courts to set optimal level of care for each class type and therefore “it is socially desirable for courts to acquire information about an injurer’s type if the cost of doing so is sufficiently low”). 18 See Shavell, supra (fn. 17), at 86–88 (showing that if courts cannot determine an injurer’s type, they would choose single care level that is optimal for average type of injurer). 19 E.g., Schwartz, Objective and Subjective Standards of Negligence: Defining the Reasonable Person to Induce Optimal Care and Optimal Populations of Injurers and Victims, 78 Geo. L.J. 241, 247–50 (1989) (“[A] single standard is preferable to a rule of subjective negligence because, unlike a rule of subjective negligence, it creates self-enforcing incentives for optimal behavior in deciding whether to engage in the activity.”). 20 E.g., Shavell, supra (fn. 17), at 91 (“[I]t may be socially beneficial for courts not to reduce due care for types of injurers for whom the socially optimal level of care would be low, because such types of injurers may thereby be induced not to engage in the activity.”). 21 Id. at 92 (“[I]f due care equals the socially optimal level, then injurers will be led to choose both the socially optimal level of prior precautions and the socially optimal level of care.”); Schwartz, supra (fn. 19), at 254–57, 259 (arguing that subjective negligence standard generally results in underinvestment of precaution). 22 But see Korsmo, Lost in Translation: Law, Economics, and Subjective Standards of Care in Negligence Law, 118 Penn St. L. Rev. 285, 292–95 (2013) (making this distinction, but not exploring full set of incentive effects due to two types of personalization).

Ben-Shahar/Porat

59

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

considerations support personalization in many more cases. Part V explains how personalization could be broadly implemented in negligence law with the aid of Big Data, among other things. It also makes the claim that even if full-fledged personalization could not be implemented (yet), two alternative would be (i) to apply a gradual personalization regime, according to which courts would have to choose whether to impose a high, medium or low level of care at any specific case brought before them; and (ii) to utilize presumptions designed to elicit private information out of litigants. Lastly, the Conclusion summarizes our proposal for personalization of the standard of care, pointing out several options for personalization, and offers a few extensions to other fields of the law.

II. Personalized negligence under existing law Current law does not personalize standards of care. It adheres, instead, to a regime of uniform, nonpersonalized, standards. The Third Restatement states: “A person acts negligently if the person does not exercise reasonable care under all the circumstances.”23 Reasonable care requires balancing the “foreseeable likelihood that the person’s conduct will result in harm, the foreseeable severity of any harm that may ensue, and the burden of precautions to eliminate or reduce the risk of harm.”24 The Third Restatement clarifies that its balancing approach is identical to the reasonably careful person approach “[b]ecause a ‘reasonably careful person’ (…) is one who acts with reasonable care (…).”25 The “reasonably careful person” standard is explicitly objective and, therefore, nonpersonal.26 The law does not generally ask whether a given person took as much care as she personally ought to have taken, given the risk she creates and the risk reduction skills she has. Rather, it insists that individuals be judged according to the standard of an external reasonable actor, representing some aggregate community measure.27 24 Objective standards do not mean one-size-fits-all. The present objective regime permits some partition of the reference group against which an actor’s behavior is judged. While the partition does not go so far as to personalize negligence law, courts have been willing to adjust standards of care to account for several special human characteristics that are thought to have a strong correlation with riskiness of actors and with the effectiveness of their precautions. These characteristics, discussed below, include inherently diminished physical and cognitive capacity; enhanced special skills, intelligence, or knowledge; and doctors and medical institutions with either enhanced or diminished resources. 23

Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 3 (Am. Law Inst. 2010). Id. This wording indicates that the Restatement endorsed the Hand Formula for determining negligence. See United States v. Carroll Towing Co., 159 F.2d 169, 173 (2d Cir. 1947) (determining liability on whether burden of adequate precautions is smaller than multiplication of damages caused by their probability). 25 Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 3 cmt. a (Am. Law Inst. 2010). 26 While the Restatement determines “reasonable care” by considering the objective primary factors set out in § 3, considerations of more personal characteristics such as age and knowledge are permitted for particular categories of cases. See Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 3 cmt. d (Am. Law Inst. 2010) (setting out certain categories of torts where more personalized factors should be considered). 27 See, e.g., Rappaport v. Nichols, 156 A.2d 1, 8 (N.J. 1959) (“[T]he standard of care is the conduct of the reasonable person of ordinary prudence under the circumstances.”). 23 24

60

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

1. Diminished capacity Tort law treats several groups of people with diminished capacity differently, applying 25 a separate standard of care for each. These groups include the physically disabled, children, and the mentally disabled. a) Physically disabled. Actors with physical disabilities generally face a standard of 26 care in accordance with their condition: “The conduct of an actor with a physical disability is negligent only if the conduct does not conform to that of a reasonably careful person with the same disability.”28 For example, a blind or deaf person is only required to take the contributory precautions reasonable in light of her limitation.29 This adjustment of the standard of care is often downward. For example, required precautions for the blind or deaf do not include looking or listening for a train at a railroad crossing.30 This is consistent with what later in the Article we call skill-based personalization: People whose skill in taking precautions is lower (or whose private cost of taking precautions is higher) should optimally take less care.31 But, the adjustment of standards may also go the opposite way, raising the burden of precautions. A paralyzed driver whose physical disability diminishes his control of the car might be required to take additional precautionary measures that an able-bodied driver would not be required to take, such as installing special steering mechanisms or special brakes.32 This is consistent with what later in the Article we call the risk-based personalization: people whose conduct creates higher risk should take more care. b) Children. Children face standards of care distinct from, and generally lower than, 27 those of adults: “A child’s conduct is negligent if it does not conform to that of a reasonably careful person of the same age, intelligence, and experience (…).”33 This, again, is consistent with skill-based personalization: “A child may be so young as to be manifestly incapable of exercising any of those qualities of attention, intelligence and judgment which are necessary to enable him to perceive a risk and to realize its unreasonable character.”34 This adjustment is more finely personalized: If the child has Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 11 (Am. Law Inst. 2010). E.g., Muse v. Page, 4 A.2d 329, 331 (Conn. 1939) (“[R]easonable care in the case of one with such defective vision as the plaintiff had, is such care as an ordinarily prudent person with a like infirmity would exercise under the same or similar circumstances.”); Fink v. City of New York, 132 N.Y.S. 2d 172, 173 (Sup. Ct. 1954) (ruling that deaf mute hit by fire truck sounding its alarm is free from contributory negligence, having exercised necessary due care allowed by his affliction). 30 See, e.g., Iron Mountain R.R. Co. v. Dies, 41 S.W. 860, 862 (Tenn. 1897) (“These obligations to stop and look and listen [before going over the tracks of a railroad] must receive a reasonable construction and interpretation (…). [A party] cannot be required to listen if he is deaf (…).”). 31 A related justification is that adjustment of the standard of care affords people with physical disabilities some security in living their daily lives. See Dobbs, supra (fn. 2), § 119, at 282 (noting that in some cases the rule is “especially protective of persons with disabilities or physical limitations”); see also Dorfman, Negligence and Accommodation: On Taking Others as They Really Are, 12–13 (27 December 2014) (unpublished manuscript), http://works.bepress.com/avihay_dorfman/18/ (noting that cases in which physical disadvantage warranted watered-down standard of care were cases of contributory or comparative negligence, whereas cases concerning conduct of tortfeasors did not make allowance for her physical disability). 32 While the law does not require sighted individuals to use seeing eye dogs or canes to navigate public walkways, a blind person who fails to do so and is injured can be considered contributorily negligent. See, e.g., Poyner v. Loftus, 694 A.2d 69, 71–72 (D.C. 1997). 33 Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 10(a) (Am. Law Inst. 2010); see also, e.g., Hoyt v. Rosenberg, 182 P.2d 234, 236 (Cal. Dist. Ct. App. 1947) (“While a minor, like an adult, is required to exercise ordinary care he is only required to exercise that degree or amount of care that is ordinarily exercised by one of like age, experience and development.”). 34 Lutteman v. Martin, 135 A.2d 600, 602–03 (Conn. C.P. 1957). 28 29

Ben-Shahar/Porat

61

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

different intelligence and experience than children of comparable age, the standard would be further adjusted. It could shift upwards: “[A] child who has not yet attained his majority may be as capable as an adult.”35 And it can shift downwards if a child of a given age is demonstrably less capable than his or her peers – perhaps because of immaturity or other developmental delays.36 c) Mentally disabled. In general, tort law makes no allowance for mental disability or insanity: “An actor’s mental or emotional disability is not considered in determining whether conduct is negligent, unless the actor is a child.”37 But in one specific area, standards of care may be adjusted downwards for mentally disabled individuals. This is in determining whether a mentally disabled plaintiff was contributorily negligent.38 Lowering the standard of contributory care for mentally disabled victims shifts greater liability and cost of precaution to their negligent injurers, and relieves these victims of some of the losses they would have otherwise had to bear. Adjusting the standard of care of the mentally disabled victim – but not the injurer – is a manifestation of the idea (also embedded in the egg-shell skull principle)39 that the defendant “takes the victim as she finds him.”40 29 It is something of a mystery why tort law treats the mentally disabled differently from the physically disabled and children. One possible justification is evidentiary: it is relatively easy to determine physical disability and the age of a child but relatively difficult to verify the specific effects of mental illness.41 This justification ignores the fact

28

35 Id. at 603. For example, children are often held to a higher standard of care, similar to that of adults, when performing what are considered “adult activities” such as driving an automobile or operating a snowmobile. See, e.g., Dellwo v. Pearson, 107 N.W.2d 859, 863 (Minn. 1961) (“While minors are entitled to be judged by standards commensurate with age, experience, and wisdom when engaged in activities appropriate to their age, experience, and wisdom, (…) in the operation of an automobile, airplane, or powerboat, a minor is (…) held to the same standard of care as an adult.”). Bernstein’s view is that by partaking in such activity, a child “assume[s] the combination of selected risks, pleasures, and accountability that characterizes autonomous adult life” and therefore “must accept (…) the rigors of adult-level reasonable care.” Bernstein, The Communities that Make Standards of Care Possible, 77 Chi.-Kent L. Rev. 735, 759 (2002). 36 See Soledad v. Lara, 762 S.W.2d 212, 214 (Tex. App. 1988) (“[T]he fact that a child is mentally retarded, or that he is unusually bright for his years is to be taken into account.”). 37 Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 11(c) (Am. Law Inst. 2010); see also Johnson v. Lambotte, 363 P.2d 165, 166 (Colo. 1961) (quoting 44 C.J.S. Insane Persons § 122 (1936)) (“The general rule is that an insane person may be liable for his torts the same as a sane person, except perhaps those in which malice and, therefore, intention are necessary ingredients.”); Burch v. Am. Family Mut. Ins., 543 N.W.2d 277, 280 (Wis. 1996) (“[A] tortfeasor’s mental capacity cannot be invoked to bar civil liability for negligence.”). 38 See, e.g., Birkner v. Salt Lake Cty., 771 P.2d 1053, 1060 (Utah 1989) (noting that “[i]n contrast to the use of an objective standard in cases of primary negligence, the majority of courts have adopted a more compassionate stance regarding the contributory negligence of the mentally impaired,” specifically that “[t] hose who are insane are incapable of contributory negligence, whereas lesser degrees of mental impairment should be considered by the jury in determining whether the plaintiff was contributorily negligent”); Snider v. Callahan, 250 F. Supp. 1022, 1023 (W.D. Mo. 1966) (“[W]ith respect to contributory negligence, in Missouri and in many other states a subjective standard is applied to children and persons suffering from a mental deficiency.”). See generally Flynn, Contributory Negligence of Incompetents, 3 Washburn L.J. 215 (1964) (discussing case law examples of contributory negligence by mentally ill tortfeasors). 39 See, e.g., Vosburg v. Putney, 50 N.W. 403, 404 (Wis. 1891) (“The rule of damages in actions for torts (…) [is] that the wrongdoer is liable for all injuries resulting directly from the wrongful act, whether they could or could not have been foreseen by him.”). 40 See Dorfman, supra (fn. 31), at 33–34 (justifying asymmetrical measurement of reasonable care across defendant/plaintiff divide by notion that tortfeasors should take potential victims as they find them). 41 See Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 11 cmt. e (Am. Law Inst. 2010) (explaining that limited or moderate mental disorders, as opposed to psychoses, are

62

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

that the law takes a highly granular approach to determining standards of care for children, including their mental development (but perhaps children are less likely to bluff cognitive impairment in legal proceedings). It also runs against the increased accuracy of psychiatric diagnosis42 that might improve the abilities to assess the skills and riskiness of a defendant alleged to be mentally disabled. Another justification for the reluctance to personalize standards of care for the mentally disabled is that doing so would hurt the incentives of their caretakers.43 But this justification, if it has any force, should apply even more strongly to children than it does to mentally disabled adults. Children are more likely to be under the direct supervision of a caretaker who could take supervisory care. And yet children’s standards are adjusted downwards, effectively exempting their caregivers from the onus of step-in care, while the same forgiving standards are denied for the mentally disabled.

2. Elevated capacity In an apparent asymmetry, tort law principles allow courts to take into account 30 elevated capacity more broadly than diminished capacity. First, elevated capacity is relevant generally as a category and is not limited to a closed list of cases. Thus, “[i]f an actor has skills or knowledge that exceed those possessed by most others, these skills or knowledge are circumstances to be taken into account in determining whether the actor has behaved as a reasonably careful person.”44 Second, elevated capacity is relevant not only when it is inherent, but also when it is deliberately acquired.45 However, in practice this principle of elevated capacity is applied inconsistently. For example, courts have been willing to account for certain kinds of special skill – like medical training46 – while ignoring others – like professional skill as a driver.47 Defendant’s special skills are most often taken into account in cases where the 31 defendant’s profession is relevant to the injury. For example, doctors are held to a standard of care for their patients that is considerably higher than the reasonable person standard.48 disregarded partly due to “the problems of administrability that would be encountered in attempting to identify them and assess their significance”); cf Seidelson, Reasonable Expectations and Subjective Standards in Negligence Law: The Minor, the Mentally Impaired, and the Mentally Incompetent, 50 Geo. Wash. L. Rev. 17, 29 (1981) (differentiating between the reasonable expectations of a plaintiff facing a minor to those of one facing a mentally disabled defendant, and arguing that “to give the [defendant] the benefit of the less demanding standard, when the [plaintiff] has no knowledge, actual or constructive, of the first actor’s impairment, would impose on the [plaintiff]”). 42 See Korrell, The Liability of Mentally Disabled Tort Defendants, 19 L. & Psychol. Rev. 1, 35 (1995) (“The task [of distinguishing between legitimate and spurious defenses] may be simpler now than when this rationale was first offered, given the increased accuracy of psychiatric diagnosis under the new Diagnostic and Statistical Manual (DSM‐III‐R) and the modern recognition of physiological indicators of mental disorders.” (footnote omitted)). 43 Id. at 29–30 (presenting and refuting caretaker incentive justification). 44 Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 12 (Am. Law Inst. 2010). 45 See, e.g., Everett v. Bucky Warren, Inc., 380 N.E.2d 653, 659 (Mass. 1978) (holding hockey coach to higher standard of care due to his substantial experience and knowledge). 46 See Martinez v. Cal. Highway Patrol, No. F056592, 2010 WL 625838, at *7 (Cal. Ct. App. Feb. 24, 2010) (finding material issue of fact as to whether highway patrol officer was negligent for having carelessly extracted accident victim from car, taking into account “that [the officer] had received medical training and recertification as an [Emergency Medical Responder] at the CHP, and that his training would have included teaching patient assessment related to C‐spine precautions”). 47 See, e.g., Capital Raceway Promotions, Inc. v. Smith, 322 A.2d 238, 246–47 (Md. Ct. Spec. App. 1974) (affirming trial court’s instruction not to hold professional race car driver to higher standard of care). 48 See Palandjian v. Foster, 842 N.E.2d 916, 920 (Mass. 2006) (“[A] specialist should be held to the standard of care and skill of the average member of the profession (…) [practicing] the specialty, taking into account the advances in the profession.” (quoting Brune v. Belinkoff, 235 N.E.2d 793, 798 (Mass. 1968) (first alteration in original))).

Ben-Shahar/Porat

63

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

The same is true (although this varies across jurisdictions),49 for example, of architects, physical therapists, and engineers.50 Even construction workers have been held to a standard of care that reflects their familiarity with heavy machinery.51 32 It is unclear to what extent tailored standards based upon professional experiences are personalized within the profession. For example, in medical malpractice, the law requires care commensurate with the “average qualified practitioner.”52 But the level of care to which a doctor is held – the “average” against which the doctor is evaluated – varies quite a bit by specialty. Specialists are held to a higher standard when treating an illness that falls within their purview.53 Some courts have gone even further, holding that, whatever the medical standard of care is, each individual doctor is required to make decisions to the best of her own judgment, when the doctor has superior knowledge, skill, or intelligence in reducing risks inherent to a common practice.54 33 Considerations of special skill, knowledge, and intelligence are, for the most part, a one-way street. While all courts are willing, in a variety of circumstances, to raise standards of care above the level of the “reasonable person” for individuals with enhanced capacity, they refuse to decrease standards of care for individuals with lower-than-average skill, knowledge, or intelligence55 (with the exceptions, as we saw, of children, physically disabled, and sometimes mentally disabled victims).56 Part of our goal in Parts III and IV below is to offer a possible rationale for this asymmetric personalization regime.

3. Resource-based personalization Precautions are costly, and individuals face different resource constraints that vary the level of care they can optimally satisfy. While in general negligence law resources do not matter in setting the standards of care,57 in medical malpractice law they do. 35 We saw that doctors are generally required to provide care that is at least as good as the average qualified medical practitioner, perhaps adjusted upwards to account for 34

49 For example, in Cervelli v. Graves, 661 P.2d 1032, 1037–39 (Wyo. 1983), the court refused to hold a professional truck driver with over ten years of truck driving experience to a higher standard of care. See also State v. Robbins, 246 P.3d 864, 867 (Wyo. 2011) (reaffirming ruling in Cervelli). 50 See, e.g., Simon v. Drake Constr. Co., 621 N.E.2d 837, 839 (Ohio Ct. App. 1993) (architects); Rehab. Care Sys. of Am. v. Davis, 73 S.W.3d 233, 234 (Tex. 2002) (physical therapists); Affiliated FM Ins. v. LTK Consulting Servs., 243 P.3d 521, 529 (Wash. 2010) (en banc) (engineers). 51 See Hill v. Sparks, 546 S.W.2d 473, 476 (Mo. Ct. App. 1976) (ruling that earth moving machine operator was negligent for failing to warn decedent “[d]espite his knowledge and experience”). 52 Palandjian, 842 N.E.2d at 920–21. 53 See supra (fn. 48). 54 Toth v. Cmty. Hosp. at Glen Cove, 239 N.E.2d 368, 372–73 (N.Y. 1968) (“[A] physician should use his best judgment and whatever superior knowledge, skill and intelligence he has. Thus, a specialist may be held liable where a general practitioner may not.” (citation omitted)). 55 See, e.g., Stevens v. Fleming, 777 P.2d 1196, 1199 (Idaho 1989) (plurality opinion) (“Individual inexperience is not a legitimate reason for a lower standard of conduct.”); Summerill v. Shipley, 890 P.2d 1042, 1046 fn. 7 (Utah Ct. App. 1995) (“[The defendant]’s inexperience or lack of knowledge cannot excuse his actions if the jury finds that the reasonable person would have acted differently in his place.”); see also Restatement (Third) of Torts: Liability for Physical and Emotional Harm § 12 cmt. b (Am. Law Inst. 2010) (“The fact that a person is below average in judgment, knowledge, or skills is generally ignored in considering whether the person is negligent (…).”). 56 Supra Section II.1. 57 Arlen, Should Defendants’ Wealth Matter?, 21 J. Legal Stud. 413, 428 (1992). Arlen opposes this notion, arguing that under the assumption that individuals are risk averse, optimal deterrence requires account of wealth differences. Id. at 419–27. For a discussion regarding the possible usage of information concerning wealth in the design of negligence standards, see infra Section V.2.d).

64

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

personal expertise. But what is the reference group from which the average qualification is derived? One dimension of reference is geographical. Traditionally, medical malpractice law has taken as the relevant reference the practices of other doctors from the same locality as the doctor under scrutiny.58 More recently, reference groups have been broadened to include similar localities, either across the state or across the country59 (an expansion designed to prevent groups of small-town doctors from shielding themselves from liability by collectively refusing to update methods of care to conform to modern practices).60 Such regional variations in the standard of care are certainly a partial response to 36 perceived variances in levels of physician skill or knowledge.61 But, they are also explicit responses to variations in medical resources. As one court explained, “[i]n applying this standard it is permissible to consider the medical resources available to the physician (…). [S]ome allowance is thus made for the type of community in which the physician carries on his practice.”62 Resource-based adjustments in standards of care apply to hospitals, as well.63 Hospitals 37 serving smaller communities may not be asked to maintain the same medical equipment as their larger neighbors, even if such absence means lower care. Interestingly, such considerations can be relevant even where the hospital’s alleged negligence is not in the provision of medical treatment. If, for example, the hospital applies only limited security and supervision, enabling a patient to escape the hospital and later suffer due to lack of proper treatment, the hospital’s resources are deemed relevant.64 In one such case, the court held that “[t]he protection of patients is not a medical function of a hospital; rather, it is a service provided by a hospital to its patients, and the ability of a small rural hospital to provide such a service is limited by its location and resources.”65

4. Personalization through insurance? Insurance is another mechanism by which personalized standards of care may 38 emerge. When potential injurers purchase liability insurance, the premiums they pay are tailored to the risks they create. Any information that insurers have about factors affecting these risks – including injurers’ skills and riskiness – would be reflected in the premium. For example, auto liability insurance would offer discounts to drivers whose known characteristics are correlated with low risk (such as high grades in college) or with high skill (driving experience). If insurers were able to observe the precautions injurers take and adjust premiums 39 accordingly, would insurance policies incentivize injurers to adopt personalized levels of 58 See Dobbs, supra (fn. 2), § 244, at 635, fn. 1 (discussing rule’s origin in Small v. Howard, 128 Mass. 131 (1880)). 59 See, e.g., Bahr v. Harper-Grace Hosps., 528 N.W.2d 170, 172 (Mich. 1995) (“[T]he standard of care for general practitioners is that of the local community or similar communities, and is nationwide for a specialist.” (footnote omitted)). 60 See, e.g., Pederson v. Dumouchel, 431 P.2d 973, 977 (Wash. 1967) (“The fact that several careless practitioners might settle in the same place cannot affect the standard of diligence and skill which local patients have a right to expect. Negligence cannot be excused on the ground that others in the same locality practice the same kind of negligence.”). 61 See, e.g., Geraty v. Kaufman, 162 A. 33, 36 (Conn. 1932) (“[W]e recognize that a country surgeon should not be expected to exercise the high degree of skill possessed by eminent surgeons living in large cities and making a specialty of surgical operations.”). 62 Brune v. Belinkoff, 235 N.E.2d 793, 798 (Mass. 1968). 63 See Johnson v. Wills Mem’l Hosp. & Nursing Home, 343 S.E.2d 700, 702 (Ga. Ct. App. 1986) (applying locality rule in wrongful death action against small rural hospital with limited resources). 64 See id. at 701–02. 65 Id. at 702.

Ben-Shahar/Porat

65

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

care, reflecting their idiosyncratic traits? Could insurance therefore substitute for the tort system by privately regulating a personalized standard regime?66 40 The idea that insurance can promote personalized care is intuitive, and it seems consistent with ways in which insurance personalizes policy premiums according to risk. Insurance is the business of data sorting, and it is indeed at the forefront of personalized underwriting. For example, insurers install “telematics” devices in cars and collect data about driving patterns, thereby fine tuning the premiums that drivers pay.67 But perhaps surprisingly, insurance is not going to be effective in personalizing the safety measures and care levels that policyholders adopt. 41 Insurers set premiums according to what they predict to be the expected liability of the policyholder – this is the payout that they would have to make under the liability coverage. Since the expected liability depends on the standards set by tort law, the premiums insurers charge would merely reflect the uniform standards prevailing under the legal system. Even if the insured-injurer is high skill or high risk, and should ideally take a high level of care, the insurer has no reason to create incentives to take more than an average level of care. Requiring policyholders to do more than is required by law would not lead to lower premiums, and thus policyholders would reject such policies. 42 What about policyholders with low skills or with low risks? Would insurers incentivize them to take the personalized and optimal level of care, which is less than the legally required uniform level? No. Here, too, insurers would not go any further than inducing the care levels that injurers would take absent insurance, under a uniform standard of care. Under such a regime, as we will explain below, injurers have the incentive to meet the uniform standard of care and bear no liability. Inducing people to take lower than the required level of care would only expose insurers to liability coverage.68

5. Summary 43

Our brief survey demonstrates the existence of some personalization in negligence law. This is only crude personalization, partitioning the population of injurers into subgroups that, as a general approximation, have different skills or a different degree of riskiness.69 Sometimes it is more finely done, as in the case of children, where the courts are willing to look at their individual developmental stage. We also saw that personalization is, in professional cases, unidirectional: Only higher but not lower skills, knowledge, and experience are taken into account in setting the standards of care. Hence, while tort law seems open to the idea of personalization of standards of care, its progress thus far in that direction has been, at best, partial and inconsistent.

66 See generally Ben‐Shahar/Logue, Outsourcing Regulation: How Insurance Reduces Moral Hazard, 111 Mich. L. Rev. 197, 220 (2012) (examining ways in which insurance regulates standards of safety). 67 See Litman, Victoria Transp. Policy Inst., Pay-As-You-Drive Insurance: Recommendations for Implementation, 7 (2011), http://www.vtpi.org/payd_rec.pdf (describing use of GPS transponders installed in cars to price insurance policies based on location and time driven); Woodyard, Drivers May Lower Insurance Premiums by Getting Monitored, USA Today (14 March 2011, 10:23 PM), http://www. usatoday.com/money/autos/2011–03–14-Progressive-electronic-check-system.htm (discussing such a program implemented by Progressive Insurance). 68 Later, we explain that under certain assumptions low-skill or low-risk injurers will self-personalize under a uniform standard of care, i.e., take a low level of care and pay damages. See infra Section III.1.c). Insurance coverage would not change injurers’ incentive to self-personalize. 69 For a theoretical analysis of the role of subgroups in negligence law and their relation to objective and subjective standards, see Bernstein, supra (fn. 35).

66

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

III. The efficiency of personalized standards Part II presented the law of negligence as a system of uniform standards. Some 44 pockets of personalization are recognized, but they are the exception, not the rule. We now turn to the core of the Article – the normative comparison between uniform and personalized standards. In this Part, our yardstick is efficiency. Later, in Part IV, we analyze personalization from both corrective and distributive justice perspectives. Our analysis in this Part compares uniform and personalized standards along several dimensions: the efficiency of the levels of care and levels of activity of injurers, the efficiency of victims’ care, and the effect on injurers’ ex ante investments in reducing their harmfulness. Along each of these dimensions, we examine the two types of personalization – skill-based and risk-based – and demonstrate their centrality to any analysis of personalization. Throughout this Part, we present our claims through analysis of a simple numerical 45 example. Most of the insights arising from this example are general. When necessary, we expand the analytical framework beyond the simple setting.

1. Levels of care Assume that injurers can each take precautions that reduce the probability of 46 accident, but not its magnitude. Suppose, for simplicity, that these individuals interact with potential victims and may cause a harm of $100 to a victim. The effectiveness of precautions for a “representative” injurer is shown in Table 1. Table 1 Level of Care

Cost of Care

Probability of Harm

Expected Social Cost

Low

$6

22 %

$28

Medium

$16

10 %

$26*

High

$26

2%

$28

* Lowest social cost.

Looking at the “Expected Social Cost” column in Table 1, we see that the lowest social 47 cost is realized when the injurer takes “medium” care. Without more information on the specific competence of each potential injurer, the optimal uniform standard of care should be “medium,” imposing an average cost of $16 on all potential injurers. The expected social cost would be $26. But now suppose that injurers are heterogeneous and that the numbers in Table 1 are 48 merely averages. Assume that the court has reliable information about idiosyncratic traits of the injurer-defendant, and that this information fits one of two categories. The first category is information on the “skill” that the injurer has in reducing risks – how costly it is for the injurer to meet each level of care in Table 1. The second category is information about the riskiness of each injurer – the likelihood at any precaution level that the harm would happen. Let’s examine what the best use of such information is. a) Skill-based personalization. The simplest way to capture the idea that injurers 49 have different risk-reduction skills is to vary the cost they have to incur in order to reach each of the three discrete levels of care – low, medium, and high. More skilled injurers can achieve the same reduction in risk as unskilled injurers by spending less on Ben-Shahar/Porat

67

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

care.70 For example, some drivers are more competent in operating sophisticated technical equipment and therefore can more effectively reduce risks with such equipment; some doctors are more experienced than other doctors and therefore can more quickly and cheaply diagnose certain patients. It is possible, of course, that the same injurer would be skillful with one type of precaution and unskillful with another type of precaution. That should affect her choices not only as to the level of care with a certain precaution, but also as to the specific type of precaution she takes. Personalization, in other words, is both qualitative as well as quantitative. In our analysis below, for simplicity, we focus only on the quantitative aspect: For any particular type of precaution, how should the level be set across different injurers? 50 We can assume that there is a spectrum of skill, ranging between the highest- and lowest-skilled injurers. Relative to the representative injurer depicted in Table 1, the highest-skilled injurer can spend 50 % less at each level of care to obtain the same risk reduction, whereas the lowest-skilled injurer must spend 50 % more at each level of care. Table 2 summarizes the precaution choices for these two extreme types of injurers (which we label “skilled” and “unskilled”). Table 2 Level of Care

Cost of Care Skilled

Unskilled

Low

$3

$9

Medium

$8

High

$13

Probability of Harm

Expected Social Cost Skilled

Unskilled

22 %

$25

$31*

$24

10 %

$18

$34

$39

2%

$15*

$41

*Lowest social costs.

51

Notice that the average of skilled and unskilled injurers yields exactly the representative injurer depicted in Table 1. If the standard is set uniformly for all injurers irrespective of their skill – what we call a uniform standard regime – the most efficient level would be “medium,” and the social cost would be $26. But society can do better. If the standard is set in a personalized manner, it would vary across injurer types. Looking at the “Expected Social Cost” dual columns in Table 2, we see the lowest cost is obtained when the skilled injurer takes “high” care and the unskilled injurer takes “low” care.71 Instead of requiring all injurers to take “medium” care, as prescribed under the uniform standard regime, the law can differentiate the standard of care according to the skill of the injurers and reduce the expected social costs. If, for example, there are equal numbers of skilled and unskilled injurers, the expected social cost will be $23 (the average of $15 and $31, the lowest attainable social costs for skilled and unskilled injurers, respectively) – lower than under a uniform standard regime ($26).72 70 Schwartz illustrates this point by presenting a graph which compares the marginal cost curve of taking care for a blind person alongside a similar graph for a sighted person. The ensuing conclusion is that as the former bears higher costs for each level of care, it is efficient for him to take less care. See Schwartz, supra (fn. 19), at 243. For a different graphical illustration of this argument, see Korsmo, supra (fn. 22), at 309–10. 71 It is assumed, for now, that under the uniform standard regime the low-skilled injurer abides by the medium standard of care. This assumption will be revisited, and the resulting discussion refined, below. See infra Section III.1.c). 72 More generally, if injurers’ skill varies along a continuum, anywhere between the +50 % and −50 % range (all relative to the representative injurer depicted in Table 1) there is a threshold of care cost above which an injurer’s standard of care should be scaled down to “low,” and another threshold of care cost

68

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

It is possible that the standards injurers face could be further refined, applying more 52 than two high/low adjustments. Following the same logic, this would generate even greater precaution efficiency. The practical implementation burdens of such continuous personalization will be discussed in Part V. The observation that skill-based personalization is more efficient than a uniform 53 standard is wholly intuitive. It pays to impose higher burdens on the more competent actors to take advantage of their greater productivity. Thus, the driver who is more competent in operating sophisticated technical equipment should probably use it, while the less competent driver perhaps should not.73 Similarly, the experienced doctor who can diagnose a patient in minutes but who failed to do so, should be considered negligent, while a less experienced doctor, who needs much more time to diagnose a patient and failed to do so, perhaps should not be considered negligent (assuming, in both cases, that the doctor has a small amount of time to invest in each patient because of a sudden overload of work). But a less intuitive aspect is the effect of personalized standards on the overall costs 54 imposed on differently skilled injurers. Personalized standards, although imposing more differentiated levels of care, impose less differentiated costs of care on the various types of injurers. Under uniform standards, the skilled and unskilled injurers have to take the same level of care (“medium”), but they bear differentiated costs of $8 and $24, respectively, to satisfy it. Under personalized standards, they have to take different levels of care. The skilled injurer has to take “high” care but can do so relatively cheaply and incurs a cost of $13. The unskilled injurer has to take “low” care but in a relatively expensive manner and incurs a cost of $9. This illustrates a general point: Skill-based personalization counteracts people’s unequal skills, offsetting the high cost of compliance with scaled-down standards.74 b) Risk-based personalization. Assume now that injurer types vary according to a 55 different attribute: the riskiness of their conduct. For the same level of care, “safe” injurers create lower risk than “risky” injurers. For example, some drivers create higher risks on the road, even when driving at the same speed, because they have poor instincts or inferior driving abilities relative to other drivers; some doctors create higher risks in performing medical procedures, even when they use the same tools and procedures, because they are less experienced and knowledgeable than other doctors. Note that experience and knowledge in some occasions, and for some tools and procedures, might affect skillfulness, as we demonstrated in the previous section, but in other occasions, and for some other tools and procedures, might affect riskiness. Thus, when injurers below which an injurer’s standard of care should be scaled up to “high.” To determine the thresholds, we look for multiples of the cost of care, a and b, such that low care and high care become more efficient than medium care: (1 + a)6 + 0.22 x 100 < (1 + a)16 + 0.10 x 100; (1 + b)26 + 0.02 x 100 < (1 + b)16 + 0.10 x 100. This yields a > 20 % and b < –20 %. When the skill level of the injurer is sufficiently low that the cost of taking each level of care rises by more than 20 % or more relative to the average injurer, the standard of care should be adjusted downwards; and when the skill is sufficiently high that the cost of taking each level of care falls by more than 20 % or more relative to the average injurer, the standard of care should be adjusted upwards. 73 Korsmo criticizes the concept that unskilled injurers should take less care, ergo act in a less prudent fashion. This theoretically sound notion, he argues, may lead to absurd results. See Korsmo, supra (fn. 22), at 316–17 (“The assumptions of the Standard Model actually suggest that unskilled drivers should be allowed to drive faster than skilled drivers. They suggest that unskilled drivers should be allowed to engage in more distractions than the skilled (…). Something is evidently amiss with the Standard Model, when translated into actual legal prescriptions.”). 74 For further discussion of this last point, see infra Section IV.2.a).

Ben-Shahar/Porat

69

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:13 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

lack experience or knowledge with respect to a certain precaution, they should often be required to meet lower standards of care with respect to that specific precaution. That lack of experience or knowledge, however – namely, the low skill in taking certain precautions – makes these injurers more risky and requires them to take other alternative precautions, as we explain below.75 56 Again, we assume that injurers’ riskiness is distributed randomly, anywhere on a continuum between safe and risky. Specifically, relative to the representative injurer, safe injurers impose a risk that is 50 % lower, whereas risky injurers impose a risk 50 % higher.76 Table 3 summarizes the care choices for the safest and for the most risky injurers. Table 3 Care

Cost of Care

Probability of Harm

Expected Social Cost

Safe

Risky

Safe

Risky

Low

$6

11 %

33 %

$17*

$39

Medium

$16

5%

15 %

$21

$31

High

$26

1%

3%

$27

$29*

* Lowest social costs.

57

Notice again that the average of safe and risky injurers yields exactly the representative injurer depicted in Table 1. But the optimal personalized standards are different than the uniform standard. The lowest social cost is obtained when safe injurers take “low” care and risky injurers take “high” care.77 Relative to the most efficient uniform standard (“medium”), social costs are reduced. If, for example, injurers are either safe or risky with equal likelihood, the expected social cost under a personalized standards regime will be $23 (the average of $17 and $29) – lower than under a uniform standards regime ($26). 75

See infra text accompanying fns 77–79. It should be noted that variation according to risk of harm could be captured also as variation according to cost of care. If care is defined as the cost to achieve a given reduction in the probability of accident, then the two attributes – skill and riskiness – would be synonymous. Thus, presenting the case of personalization according to risk of harm does not add a new theoretical insight, but merely replicates the effect described in the case of personalization according to cost of care. It is present here, nevertheless, in order to set the stage for the legal applications. The mathematical similarities between the two forms of variations have been noted in previous writings on the topic. See Shavell, supra (fn. 17), at 73 (“[R] eference will be made, for simplicity, only to differences in parties’ cost of taking care, although what will be said will plainly bear equally on differences in the effectiveness of their exercise of care.”); Korsmo, supra (fn. 22), at 292 (“From a purely mathematical perspective, the distinction between the two scenarios is, indeed, seemingly inconsequential.”). Korsmo nevertheless devotes a significant portion of his article to an analysis of the differences between the two variations, and suggests a method for determining which one should be applied. See id. at 319–37. 77 More generally, if injurers’ riskiness varies anywhere between the +50 % and −50 % range, there is a threshold of riskiness above which the injurer’s standard of care should be scaled up to “high,” and another threshold of riskiness below which the injurer’s standard of care should be scaled down to “low.” To determine the thresholds, we look for multiples of the probability of harm, s and t, such that low care and high care become more efficient than medium care: 6 + (1 + s)0.22 x 100 < 16 + (1 + s)0.10 x 100; 26 + (1 + t)0.02 x 100 < 16 + (1 + t)0.10 x 100. This yields s < –0.167 and t > 0.25. When the probability of harm at every level of care is scaled down by more than 16.7 % or more relative to the representative injurer, the standard of care should be adjusted downwards; and when the probability of harm is scaled up by more than at least 25 %, the standard of care should be adjusted upwards. 76

70

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:13 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

Notice, also, that in terms of the distribution of burdens, we get an opposite effect to 58 the one we saw under skill-based personalization. Risk-based personalized standards impose more differentiated costs on the different types of injurers than uniform standards. Under uniform standards, the safe and risky types bear the same cost of $16 to meet the “medium” standard of care. Under personalized standards, they have to bear costs of $6 and $26, respectively.78 The main result arising from Table 2 – that injurers who create lower risks should 59 take lower care – is intuitive.79 It pays to impose higher burdens on the more risky actors since any additional burden would produce more risk reduction for the high-risk actor than for the low-risk actor. Thus, the high-risk driver with the poor instincts should take more care than the driver with the sharper instincts. Similarly, the high-risk doctor with less experience and knowledge should take more care than the more experienced and knowledgeable doctor. Our conclusion, that the less experienced and knowledgeable doctor or driver should 60 take more care, does not contradict the previous claim that such doctor, being less skillful, should take less care. There are two ways to reconcile these conflicting findings. The first is by recognizing that care and precautions have multiple dimensions, and thus the conflicting conclusions apply to different dimensions of care. A doctor should take a low level of the type of precautions that she is unskillful in deploying. That, in turn, makes her relatively riskier and justifies imposing upon her a higher level of care with respect to other precautions. The second way to reconcile the conflicting findings is to view them as “all else equal” effects, each pulling in a different direction. The overall net effect would then be the combination of the two forces. If, for example, only one type of precaution is available, a driver who is both riskier and low skill may in the end be required to take either a higher or lower level of care, depending on which effect dominates. It is therefore possible that despite the low skill in applying this singledimensional precaution, the high-risk driver may be required after all to take a higher level of care. c) Self-personalization. The reason uniform standards are not as efficient as perso- 61 nalized standards is the incentive they provide injurers to abide even by inefficient standards of care. Injurers have this incentive because of what is known as the “discontinuity” feature of negligence law: that the failure to meet the standard – even a small margin of departure – would give rise to full liability for the entire harm suffered by the victim.80 Thus, even when injurers recognize the standard to be inefficiently tailored to their skill or riskiness, as the uniform standard would often be, they nevertheless abide by it and incur inefficient precaution costs, to avoid the even greater lump sum liability. There is, however, an important caveat to this “discontinuity” feature. If failure to 62 meet the standard of care results only in incremental liability – only for the additional harm due to the gap between actual care and the legal standard of care – the incentive to abide by an inefficient standard of care is attenuated. An injurer might prefer to 78

For further discussion of this last point, see infra Section IV.2.a). Korsmo illustrates this point by presenting a graph showing the accident costs for each level of care for both the skilled and unskilled injurers. As the former’s costs are lower, they intersect with the ascending precaution costs at an earlier stage, leading to the conclusion that skilled injurers should take less care. See Korsmo, supra (fn. 22), at 323–24. 80 This discontinuity and its behavioral consequences were originally explained by Cooter. Cooter, Economic Analysis of Punitive Damages, 56 S. Cal. L. Rev. 79, 80–89 (1982). Cooter later explained that this discontinuity is due to incomplete information available to the courts or the probabilistic nature of the causal connection. Cooter, Punitive Damages for Deterrence: When and How Much?, 40 Ala. L. Rev. 1143, 1155 (1989). 79

Ben-Shahar/Porat

71

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:13 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

disregard the standard and assume such incremental liability. Such an injurer would take efficient care and pay a little extra in liability.81 Accordingly, the distortion arising from uniform standards is not as large as our analysis above stated, and the benefit of shifting to personalized standards is correspondingly smaller.82 Nevertheless, and despite the self-correcting mechanism of injurers ignoring inefficient uniform standards, we now show that such self-correction will not always occur and therefore personalized standards continue to have a systematic efficiency advantage. 63 Consider, first, skill-based variation among injurers. We saw that skill-based personalized standards would require unskilled injurers to take “low” care and the highskilled injurers to take “high” care, and we argued that these are improvements relative to the “medium” care that all injurers take under the uniform standards regime. But would injurers indeed abide by the “medium” care standard under a uniform standards regime? 64 Not necessarily. To be sure, skilled injurers would. For them, the “medium” care standard is a boon. It is cheaper than the more efficient “high” care personalized standard. The skilled injurers would be delighted to qualify for a liability safe harbor by investing less than efficiently. But unskilled injurers would have a different incentive. They would choose to disregard the inefficiently burdensome “medium” standard, even if this means that they would be found liable. Returning to Table 2, the unskilled injurers, taking the efficient “low” level of care at a cost of $9 would create some exposure to liability. But not for all harms: They would be liable only for harms that are due to the difference between taking “low” and “medium” care. Since the shift from “medium” to “low” raises the expected harm from $10 to $22, the expected liability of the unskilled injurers who ignore the “medium” standard and take “low” care is only $12 (the difference between $22 and $10) and not $22 (as we previously assumed). For them, taking “low” care at a cost of $9 and incurring the expected liability of $12, for a total cost of $21, is cheaper than incurring no liability by satisfying the “medium” care standard at a cost of $24. 65 Here, the advantage of personalized standards is diminished because some unskilled injurers would be self-driven to take the efficient care level, even under a uniform standard. This is a general observation: Any time the idiosyncratic cost of care to injurers is high enough to justify a lower personalized standard, this injurer would also have the incentive to ignore the uniform standard and take a lower level of care. In other words, the unskilled injurer would always self-personalize.83 The advantage of personalized standards is then solely due to their effect on the upper side of the 81 Grady and Kahan have demonstrated that the discontinuity of liability, as well as the risk of burdening the negligent injurer with liability for more than the harm he caused, completely disappears when causation rules are properly applied so that the injurer is liable only for those harms that would not have been created had he behaved reasonably. Grady, A New Positive Economic Theory of Negligence, 92 Yale L.J. 799, 812–13 (1983); Kahan, Causation and Incentives to Take Care Under the Negligence Rule, 18 J. Legal Stud. 427, 428–29 (1989). 82 Landes and Posner were the first to note that under a uniform standard of care rule, injurers with very high costs of taking care would not comply with the uniform standard but choose instead the standard of care which is efficient for them. Landes/Posner, supra (fn. 16), at 125. Schwartz furthers this notion by dividing the group of high-cost injurers who choose not to comply with the uniform standard into two subgroups: one which chooses to engage in the activity and one which refrains from doing so, as the benefits it derives are exceeded by costs of care and costs of harm. Schwartz, supra (fn. 19), at 249–50. This result is viewed by Schwartz as an advantage for the uniform standard over the subjective one, as it creates “self-enforcing incentives for optimal behavior in deciding whether to engage in the activity.” Id. 83 More generally, in the numerical example of Table 2, unskilled injurers have an incentive to selfpersonalize if: (1 + a)6 + (0.22 – 0.10) x 100 < (1 + a)16. Thus, anytime unskilled injurers have a cost that is more than 20 % higher than the representative injurer, they would self-personalize and take “low” care. This, recall, is also the cost threshold that justifies a reduction of the standard of care from “medium” to

72

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:13 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

distribution of injurers – the skilled injurers – who otherwise are happy to satisfy a uniform standard, what is for them inefficiently low care. The same one-sided self-personalization occurs in the case of risk-based variations 66 across injurers. Once we relax our assumption that injurers abide by the standard of care, and instead assume that injurers do what is less costly for them and that they bear incremental liability (no “discontinuity”), then, again, some injurers will ignore the inefficient uniform standard – they will self-personalize and behave efficiently. Specifically, safe injurers will ignore what is for them an inefficiently high uniform standard. Taking the personally efficient “low” care, at a cost of $6, and bearing the expected liability of $6 (the difference between the actual expected harm of $11 and the expected harm of $5 that would have resulted had he abided by the required “medium” standard of care) is less costly for the safe injurer than incurring no liability by satisfying the “medium” care standard at a cost of $16.84 Here, too, the advantage of personalized standards arises only from their effect on risky injurers. These risky types would be content to meet the “medium” level of care required under the uniform standards regime, rather than the costlier “high” level of care under a personalized regime. Personalization corrects this distortion. d) Summary. We examined the efficiency of shifting from uniform to personalized 67 standards in environments in which injurers vary across two harm-relevant dimensions: skill in taking precautions and underlying propensity to impose risks. There are other dimensions along which standards can be differentiated (for example, the magnitude of harm), but the discussion above already demonstrates several basic insights that apply to all cases. First, differentiating the standards can improve incentives for care. If information is available about the different risks and prevention costs, and if injurers can anticipate the differentiated standards they face, personalized standards are more efficient than average standards. Second, the examples above draw out some basic principles in the design of 68 personalized standards. Should injurers who impose a higher expected harm face stiffer standards? Upon first reflection, one might intuitively conjecture that such harmful injurers should always be confronted with higher standards of care. The analysis shows, however, that this intuition is only partially valid. We saw that risky injurers – who impose higher probabilities of accidents at each level of care relative to safe injurers – should indeed face higher standards of care. But we also saw an effect in the opposite direction: Unskilled injurers – who impose a greater risk because they are less effective in taking care and can only achieve accident prevention at higher cost – should face lower, not higher, personalized standards relative to the skilled injurers. Skilled injurers are less harmful but should nevertheless face higher standards of care due to their relative effectiveness. We also saw that personalized standards impose a different cost of compliance on 69 different types of injurers. Here, too, it might be conjectured that the distribution of burdens would exhibit more variance under a personalized standards regime. But, again, this is not always so. When injurers vary in their costs of care (skilled versus unskilled), personalized standards can reduce, rather than increase, the disparity in the burdens of compliance. “low.” We can conclude that the incentive to self-personalize for the unskilled occurs if and only if it is efficient. 84 In a similar fashion, Korsmo argues that under the variation in which injurers differ by riskiness, it is the less risky injurer who would find it too costly to adhere to the reasonable person standard and would therefore abide by her lower subjective standard, thereby creating a “pocket” of strict liability. See Korsmo, supra (fn. 22), at 327–29.

Ben-Shahar/Porat

73

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:13 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law 70

Finally, we examined the incentives of injurers to self-personalize – a type of selfselection that might occur even under uniform standards, and might lead injurers to take differentiated levels of care notwithstanding the crude, uniform legal standard. This would happen only if we assume continuity – rather than discontinuity – of liability, namely, if we assume that failure to take the required level of care exposes the injurer only to the incremental losses caused by this failure, and not others. In this environment of self-personalization, personalized standards continue to be more efficient, but solely due to their effect in increasing the burden of care upon some injurers.

2. Levels of activity 71

A standard result in the economic analysis of negligence law is the activity-level distortion. When injurers conform to the standard of care they bear only some of the full social cost of their activity (they bear the cost of care but not the residual expected harm) and therefore engage in excessive levels of activity.85 In this section we ask how the activity-level distortion would be affected by personalized standards of care. We make two distinct observations. First, personalizing the standard of care according to skill (but not risks) could further distort, rather than improve, injurers’ activity levels. Second, we identify a novel regime that combines both personalized and uniform standards, which improves both care and activity levels.

a) Improving or distorting levels of activity. We saw in Section III.1 above that the standard of care would generally be higher the more skilled and risky the injurers are, exceeding the average uniform standard for the upper half of the population of injurers. How would that affect their levels of activity? 73 Raising the standard of care for some injurers reduces their activity-level distortion, while lowering it for others exacerbates this distortion. In the example in Table 2, raising the standard for skilled injurers from “medium” to “high” raises their cost of care from $8 to $13. At “medium” care, the negative externality from their activity was $10 (the expected harm which they do not have to bear). At “high” care, the negative externality is only $2. Since it is this negative externality that drives the activity-level distortion, shrinking it from an absolute magnitude of $10 to $2 reduces the distortion. 74 While standards are raised for skilled injurers, they are lowered for the unskilled types, from “medium” to “low” care. Here, the activity-level distortion is aggravated. At “medium” care the negative externality from the activity of the unskilled injurers was $10. At “low” care, it was $22. As the externality rises from an absolute magnitude of $10 to $22, the distortion grows. 75 Thus, under skill-based personalization, the unskilled are led to engage in more undesirable activity, while the skilled are led to engage in less. Commentators have long noticed one side of this result – the increasingly inefficient activity levels by the unskilled – and invoked it as a primary argument against personalization.86 If people who cannot take effective care were only required to meet their low personalized standard, others would be imperiled by the greater risk they impose. The neighbors of the unskilled injurer, says Holmes, “accordingly require him, at his proper peril, to

72

85

Shavell, supra (fn. 17), at 23–24. See, e.g., Landes/Posner, supra (fn. 16), at 126 (arguing that when a uniform standard of care, as opposed to an individualized standard of care, is applied to a child, who is unable to attain a high level of care, his parents are incentivized to prevent him from driving); Schwartz, supra (fn. 19), at 246 (“A rule that only requires the injurer to take what is for her optimal care while engaging in the activity cannot achieve the optimal result. Under such a rule, some injurers who should not engage in the activity will nevertheless do so.”). 86

74

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

come up to their standard, and the courts which they establish decline to take his personal equation into account.”87 But what they have not noticed is the other side of the coin: the increasingly efficient activity levels by the skilled injurer. Still, in normal circumstances, the added inefficient risks posed by the unskilled 76 due to personalization more than offsets the improved activity levels of skilled injurers.88 The reason is subtle. The activity distortion is due to the expected harm that a standard-abiding injurer does not bear. The more care an injurer takes, the lower this expected harm, but the marginal reduction has a diminishing property. When the unskilled injurer shifts from the uniform standard to the personalized low care, the increase in expected harm is greater in absolute value than the decrease that occurs when the skilled injurer shifts from the uniform to the personalized high care. As a result, the overall distortion of activity under skill-based personalized standards increases. But this is not the case under risk-based personalization. Here, safe injurers take 77 lower care and thus engage in more inefficient activity (in our example in Table 3, they now create uncompensated expected harm to victims of $11, up from $5 under uniform standards). But risky injurers take higher care and engage in less inefficient activity ($3 of uncompensated external harm, down from $15). Because, all else equal, risky injurers create larger harms, the effect of curbing their activity level more than offsets the increase in activity by the safe injurers. As a result, the overall distortion of activity under risk-based personalized standards decreases. Here, the desirable effect of personalization on activity levels adds to their effect on care levels to bolster the efficiency of the regime. b) Activity levels with self-personalization. We concluded that personalized stan- 78 dards have a mixed effect on activity levels. Relative to uniform standards, they produce two effects. On the upside, personalized standards reduce the distortion in activity levels for injurers who now face higher standards (skilled or risky injurers). On the downside, they worsen the distortion for injurers who now face lower standards (unskilled or safe injurers). We now argue that the downside is actually smaller than what the analysis above suggested. Return to the environment in which injurers may self-personalize. We showed in 79 Section III.1.c) that under a uniform standard regime, in which liability is only for harm caused by untaken care (“continuous” liability), it would be rational for unskilled or safe injurers to ignore the uniform standard (“medium”) and take instead the personally efficient care (“low”). In this setting, personalized standards have a smaller distorting effect on activity levels. To see this, consider the case of risk-based personalization. Under uniform standards, the safe injurer’s activity level would depend on whether he self-personalizes. If he doesn’t – if he abides by the uniform standard – he takes “medium” care, he is not found liable, and thus incurs a private cost of $16, and imposes an uncompensated expected harm of $5. If, instead, the injurer does selfpersonalize and takes “low” care, he is found liable for the incremental harms that would have been prevented had he taken “medium” care, and thus incurs a private cost of $12 ($6 cost of “low” care plus $6 expected liability), and imposes an uncompensated expected harm of $5. This illustrates a general pattern: Self-personalization does not affect the size of the uncompensated harm the injurer inflicts on victims ($5 either way), but it does reduce the private cost of activity to the injurer. As a result, with self87

Holmes, supra (fn. 1), at 108. This was the case in our numerical example: The increased externality for the unskilled injurer was $12, and the reduced externality for the skilled injurer was only $8. 88

Ben-Shahar/Porat

75

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

personalization, safe injurers (those whose private benefit is between $12 and $16) engage in a higher activity level and inflict the externality.89 80 Thus, once we allow for the possibility of self-personalization under uniform standards, activity level is higher than we initially calculated. This should be obvious – the only reason the injurer self-personalizes is to reduce the private cost of activity. This means that moving to a regime of personalized standards imposes a smaller increase in the activity level of safe and unskilled injurers than we otherwise calculated. The activity-level downside of personalization is thus smaller. c) A hybrid regime. Personalized standards have a downside: They bring about an increase in the activity level of some injurers. Indeed, it is this concern that led commentators to conclude that a uniform standards regime is superior.90 We now argue that this concern should not trump categorically the case for personalized standards. A personalized standards regime can be designed to apply only when it does not distort activity levels. We show that an unambiguous improvement in both care and activity can be obtained if personalization is done selectively. 82 Consider a “hybrid” regime in which each type of injurer faces a standard that is the greater of the pure personalized standard and the one-size-fits-all uniform standard. Personalization, in other words, can only operate to increase, but not to decrease, the standard of care. In the case of skill-based personalization (the example in Table 2), skilled injurers face the “high” standard (the higher among {high, medium} – the optimal personalized and the optimal uniform standards), whereas unskilled injurers face the “medium” standard (the higher among {low, medium}). Under this hybrid regime, skilled injurers would take more efficient care and activity levels than they would under a pure uniform standards regime; and unskilled injurers would take the same care and activity levels as they would under a uniform standard regime. This regime is generally better than uniform standards, due to the improvement in care and activity by the skilled type. The same logic applies to the case of risk-based standardization: Under the hybrid regime, the risky injurer will be required to meet a high standard, while the safe injurer will be required to adhere to a medium standard.

81

3. Victim care In this section we examine how personalization of injurers’ standards of care affects the efficiency of victim precaution. We assume, for the purpose of this discussion, that victims are homogeneous. To be sure, victims vary in many ways as well, which could also justify personalization of standards of contributory care. The question in this section, however, is different. Does the case for personalization of injurers’ standards depend on its effect on victims’ behavior? 84 If a victim can adjust her own level of care to the personalized standard and the idiosyncratic conduct of each injurer, the case for personalizing injurers’ standards of care would only be bolstered. Such injurer-only personalization would improve not only injurers’ behavior, but also victims’. For example, if pedestrians can adjust their precautions to the different dangers that different drivers facing different standards of care impose, a law that sets personalized standards for drivers would induce pedestrians to vary their precautions efficiently. Facing skilled injurers who take more care, victims would adjust their care downwards and save some unnecessary precautions. 83

89 The same analysis shows that with self-personalization, more unskillful injurers engage in the activity and inflict the externality than without self-personalization. 90 Shavell, supra (fn. 17), at 91.

76

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

But what if victims have to set their level of care without observing the personalized 85 standards and behavior by injurers? What if, when a car is approaching, the pedestrian cannot observe the skill and the standard of care of the specific driver? Victims may observe the distribution of injurers, and take a uniform level of contributory care that best responds to this distribution of risk. In this setting, personalized injurer standards pose a challenge. Rather than facing injurers who all take uniform care, victims now interact with injurers who take varying levels of care. It might be thought, then, that in the personalized standards environment, victims would have a more difficult optimization problem to solve – how to respond to a volatile care environment. For example, what care should a pedestrian crossing a busy intersection take, now that cars travel at different personalized speeds? Given this difficulty, the concern is that victims might “play it safe” and take high uniform care. If so, victims’ care would be more costly. The analysis below shows that these concerns are not generally valid. a) Skill-based personalization. We saw in Table 2 that under a uniform standard 86 and in the absence of self-personalization, all injurers would take a medium level of care (costing $8 and $24 to the skilled and unskilled, respectively), and the residual risk of harm facing the victim would be 10 %. We also saw that under a personalized standards regime, the skilled injurer would be asked to take high care (costing $13), leaving a residual risk of 2 %; and the unskilled injurer would be asked to take low care (costing $9), leaving a residual risk of 22 %. Under a uniform standard regime, then, the victim faces the same risk regardless of 87 the injurer’s type – here, a 10 % probability of accident. Under a personalized standard regime for injurers, the victim faces a variance of risks – here, either 2 % or 22 %. In which setting will the victim’s care be more effective? Assuming that injurer and victim care are strategic substitutes (more care by one party 88 makes it optimal to take less care by the other), as the injurer’s skill increases, the victim’s optimal care decreases. The skilled injurer leaves a residual risk of only 2 %, so there is less value to additional precaution by the victim than when the injurer is unskilled and the residual risk is 22 %. But if the victim’s care cannot be tailored (either because the victim cannot know which injurer he faces or because precautions are “lumpy” and cannot be varied across injurers), it is possible that the overall contribution of the victim to accident prevention would be diminished. Relative to the case of uniform standards, where all injurers impose on the victim a 10 % risk of uncompensated accidents, the victim will now seek the most efficient response to an environment that contains injurers of both high risks (22 % chance of harm) and low risks (2 %). In this environment, it may well pay off for the victim, especially the risk-averse type, to take more care relative to the uniform injurer standard case. Such added care is a waste vis-à-vis the high-skill injurers, but it is all the more valuable vis-à-vis the low-skill injurers who expose the victim to a high risk of uncompensated accident. For example, if different cars on the road drive at different personalized speeds and pose differential risks, a potential victim’s care would necessarily be too high against slow-driving cars and too low against fast-driving cars. It might be optimal for victims in such an erratic environment to react primarily to the subset of low-skill injurers (who create high uncompensated risk), notwithstanding the redundancy of such effort in relation to the high-skill injurers.91 91 This observation does not change if injurers self-personalize under a uniform standard regime. We showed that under uniform standards, low-skill injurers self-personalize and take the socially optimal low level of care (because saving in precaution costs outweighs the cost of liability they expect to incur by failing to meet the uniform standard). See supra Section III.1.c). Accordingly, low-skill injurers behave the same under either a personalized or a uniform standards regime, and impose the same high risk of accident upon victims. With self-personalization, low-skill injurers end up incurring some liability, which

Ben-Shahar/Porat

77

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law 89

Note, however, that even if victims’ care is less efficient under a personalized standards regime, the overall effect on bilateral care under a personalized standards regime cannot be less efficient. We saw that personalized standards unambiguously improve the efficiency of injurers’ care. Under a personalized standards regime that takes victim care into account in designing injurers’ standards, it’s always possible to achieve the outcome of uniform standards, by unifying the different types of injurers’ standard of care. Thus, if the costs of personalized standards on victims’ care are higher than the benefits of personalized standards on injurers’ care, efficiency would require that all injurers stick to a uniform standard. After all, the presence of victims and their care is a crucial factor in efficiently personalizing the standards of care, as well as in whether to personalize them. It is also possible that given the tradeoff of personalization and victims’ care, personalization would be done partially. Thus, if personalization, without taking into account victims’ care, requires that given their skillfulness some drivers will impose a risk of 30 % and the others a risk of 50 %, given victims’ care, personalization might end up with the former drivers being allowed to impose a risk of 35 %, and the others a risk of 40 %.

b) Risk-based personalization. Victims’ care would be unambiguously more efficient under a regime that personalizes injurers’ standards of care according to the risk they pose. We saw that under a uniform standard all injurers would take a medium level of care, but would impose different risks: 5 % versus 15 % residual probability of harm by the safe and risky types, respectively. We also saw that under a personalized standards regime, the safe injurer would be asked to take low care, leaving a residual risk of 11 %; and the risky injurer would be asked to take high care, leaving a residual risk of 3 %. 91 Here, the effect of personalized injurer standards over victims’ care is unambiguous and desirable. Despite the fact that different injurers are asked to take different personalized care levels, victims overall face a less disperse distribution of risk. Under uniform injurer standards, victims faced actors who cause either 5 % or 15 % probability of harm, whereas under personalized injurer standards the probabilities of harm are both lower and less dispersed (11 % and 3 %). Since the efficiency of victim care depends on the residual probability of harm, personalized standards allow victims to confront a less erratic distribution of such probabilities. As injurers are induced to behave in a way that compensates for their different risks, victims take more efficient care.

90

4. Ex ante investment in improving private characteristics 92

Personalized standards reflect injurers’ observable idiosyncratic properties – individual traits that affect their ability to reduce the risk of accidents. How do these traits form? The analysis so far assumed that people vary exogenously, and that the law merely observes – but does not influence – the development of personal traits. In this section we relax this assumption. We assume instead that traits are determined by investments that people make: Drivers could improve their skills, for example, by taking driving classes and training;92 doctors could also improve their skills, for example, by reduces the incentive of victims to take care. Nevertheless, with or without self-personalization, a uniform standard regime imposes on victims a uniform risk of uncompensated accident (in our example, 10 % risk of $100 in harm). 92 See, e.g., Dorn/Barker, The Effects of Driver Training on Simulated Driving Performance, 37 Accident Analysis & Prevention 63, 68 (2005) (“It would appear that professional driver training affects simulated driving performance with trained drivers demonstrating a potentially safer driving style than untrained drivers.”); Isler et al., Effects of Higher-Order Driving Skill Training on Young, Inexperienced Drivers’ On-Road Driving Performance, 43 Accident Analysis & Prevention 1818, 1820–25 (2011) (showing that

78

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

reading more professional materials and participating in more conferences;93 employers can acquire more sophisticated tools and train their personnel to reach higher skill or pose lower risk to outsiders. We ask whether such investments would be affected by personalized negligence law. 93 Specifically, we address a powerful objection to personalized standards – that they undermine injurers’ incentives to improve. If injurers anticipate that such investments would in turn raise the precaution burdens imposed upon them, their incentives to make the investments would weaken.94 They might even be incentivized to take action to diminish, rather than improve, their harm-reduction traits. For example, if personalization indeed hurts incentives to acquire skill, a firm might prefer employing less skillful employees for risky activities (like driving), thus reducing their expenditures on care. a) Skill-based personalization. High skill warrants a high standard of care. We saw 94 in Table 2 that under a personalized standard regime, the unskilled injurer would face a “low” standard (at a cost of $9) whereas the skilled injurer would face a “high” standard (at a cost of $13). Imagine that each injurer begins as unskilled, but that prior to interaction with the victim the injurer could become skilled by spending a lump sum cost of k. At what levels of k would it be socially desirable to spend it? At what levels would it be in the private interest of the injurer to make this investment? aa) Personalized Standards. (i) Social optimum. If the injurer remains unskilled, 95 the optimal level of personalized care would be “low” and the resulting social cost of his activity would be $31. If, instead, the injurer becomes skilled, the optimal level of care would be “high” and the social cost would be $15. Thus, the social gain from investment in skill is $31 – $15 = $16. It is socially desirable to make the investment in skill if k < $16. (ii) Private incentives. The unskilled injurer faces a cost of care of $9, whereas the 96 skilled injurer faces a cost of care of $13. Here, investing in becoming skilled is privately undesirable: Not only does the injurer enjoy none of the savings such investment yields socially, but he is saddled with a higher cost of compliance. This is a general problem. The injurer’s investment in skill improvement reduces his cost of taking care – a social benefit that is also a private benefit. But it also leads to an upward adjustment of the level of care – another social benefit but one that creates a private loss. This suggests that not enough investment in human capital would be made, and that – as conjectured above – skill-based personalized standards undermine ex ante investment. Relatedly, firms might prefer employing less skillful employees.95 bb) Uniform standards with full compliance. (i) Social optimum. Since the stan- 97 dard of care does not change for those who become skilled, and assuming that injurers comply with the optimal uniform standard (“medium”), the social value of the investyoung, inexperienced drivers who receive training aimed at improving skills such as situational awareness and hazard anticipation perform significantly better at driving-related simulations). 93 See, e.g., Davis et al., Impact of Formal Continuing Medical Education: Do Conferences, Workshops, Rounds, and Other Traditional Continuing Education Activities Change Physician Behavior or Health Care Outcomes?, 282 J. Am. Med. Ass’n 867, 867 (1999) (analyzing previous studies concerning CME (continuing medical education) and concluding that there is some evidence that interactive, as opposed to didactic, CME sessions can effect changes in professional practice and, on occasion, in health care outcomes). 94 Schwartz, supra (fn. 19), at 254–57 (arguing that while a personalized standard would yield optimal investments in the ability to take care if courts were to measure said expenditure and take it into account, in a more feasible scenario where ex ante investments in skill are disregarded, injurers would not have a high enough incentive to do so and would underinvest in the ability to avoid harm to victims). 95 See supra text accompanying fn. 94.

Ben-Shahar/Porat

79

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:14 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

ment is the reduced cost of compliance with the uniform standard, from $24 to $8. The investment should be made if k < $16. (ii) Private incentive. Under a uniform standard with full compliance, the injurer who invests in skill recoups the entire social saving, reducing her cost of compliance from $24 to $8. Here, investment would be optimal. Since there is no accompanying increase in the standard, there is no divergence between the private and social incentive to invest.96 99 This analysis demonstrates a robust observation. Skill-based personalization destroys the incentives to invest in improved skills. Under a uniform standards regime, investment is optimal because the investing injurer captures the entire social surplus from the improved skill. In contrast, under a skill-based personalized standards regime, the injurer does not enjoy the full social surplus from the investment in reducing the cost of care, and may even be worse off. 100 To be sure, the problem of distorted ex ante investment under a personalized standards regime can be resolved if courts could monitor such investment. If a court has enough information to set standards that reflect, not existing skills, but optimally acquired skills, injurers would be prompted to make the optimal investment. If, for example, a doctor could not invoke his low skill in defense against malpractice and would instead be held to the optimally acquired skill, personalization would clearly outperform a uniform standard. But the information burden is high: It is not enough for the court to set a standard based on optimal hypothetical skill across the entire population. For this to be a personalized standards regime, the optimally invested skill would then have to vary by the idiosyncratic investment traits of each injurer.97 98

101

b) Risk-based personalization. High risk warrants a high standard of care. We saw in Table 3 that under a personalized standard regime, the risky injurer would face a “high” standard (at a cost of $26), whereas the safe injurer would face a “low” standard (at a cost of $6). Imagine that in the absence of ex ante investment, the injurer would be the risky type, and that it would take an investment of k to become a safe type. At what levels of k would it be socially desirable to spend k? What if the investment of k were privately undertaken?

102

aa) Personalized standards. (i) Social optimum. If the injurer remains risky, the lowest social cost of his activity when he takes “high” care is $29. If, instead, the injurer invests in becoming safe, the lowest social cost, when he takes “low” care, is $17. Thus, the social gain from the investment is $29 – $17 = $12. It is worth making the investment in safety if k < $12.

103

(ii) Private incentive. The risky injurer faces the cost of “high” care of $26, whereas the safe injurer faces the cost of “low” care of $6. The injurer would make the investment if k < $20. Here, investing in safety is privately desirable. In fact, the private 96 If injurers self-personalize, it is socially desirable to make an investment under uniform standards if k < 13, since the unskilled injurer takes “low” care and imposes a social cost of $31, whereas the skilled injurer takes “medium” care and imposes a social cost of $18. The private incentive is the same – make the investment in skill if k < 13. Without the investment, the unskilled injurer self-personalizes to “low” care and faces a cost of care of $9 and liability of $12, for a total private cost of $21, whereas the skilled injurer takes “medium” care and incurs a cost of $8. Here, too, investment is generally optimal. While the level of care does increase with improved skill, the injurer enjoys the entire social benefit – a lower cost of care and the net reduction in expected harm. 97 As to the concern that firms would prefer hiring low-skilled employees for risky activities, see supra text accompanying fn. 94, this concern would be avoided if the standard of care would be personalized according to the reasonable hiring policy of the firm, which is very hard to define given the many variables involved.

80

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

value of the investment is greater than the social value, suggesting that too much investment would be made. The example exposes one side of a general problem of inefficient investment, 104 although the direction of the distortion may go either way (too much or too little investment). The private value of lowering one’s riskiness is the reduced cost of complying with the lower personalized standard. But the social value contains an additional component beyond the reduction in the cost of care: the change in expected harm. This change in the expected harm is due to two factors. First, the expected harm goes down because, all else equal, the safe injurer poses a lower probability of accident. Second, the expected harm goes up because the safe injurer takes lower care. In the example, the second effect was stronger than the first, and so the expected harm caused by the safe injurer increased relative to that of the risky injurer (from $3 to $11). This is why the private incentive to invest was too high. But in other situations, the first effect could be stronger than the second, in which case there is an additional social benefit to the investment that is not captured by the injurer, and the incentive to invest in safety under a personalized standards regime would be too small. bb) Uniform standards with full compliance. (i) Social optimum. Since the stan- 105 dard of care does not change for those who invest in becoming safer, and assuming that all injurers comply with the optimal uniform standard (“medium”), the social value of investment is the ensuing reduction in the probability of harm at the uniform level of care, from 15 % to 5 % (and the expected harm from $15 to $5). It is worth making the investment if k < $10. (ii) Private incentive. Under a uniform standard with full compliance, the injurer 106 who invests in becoming safer receives no benefit, as she would have to continue and comply with the same standard at the same cost. Accordingly, investments that are socially desirable are not made. Injurers may gain other benefits from becoming less risky, which are not captured in the example, such as the reduction of self-risks. Still, the point remains: Uniform standards generate too little investment.98 In comparison to personalized standards, the investment under a uniform standards 107 regime would be unambiguously lower. The reason, as we just saw, is that personalization allows the investing injurer to capture some benefits of his investment.99 98 The same is true for uniform standards with self-personalization. Here, a safe injurer ignores the uniform standard, takes “low” care, and imposes a social cost of $17, whereas a risky injurer meets the uniform standard and imposes a social cost of $31. Thus, it is socially desirable for the injurer to make the investment to become safe if k < $14. The private incentive is small. The injurer who invests in becoming safe enjoys a reduction of private cost from $16 (the cost of meeting the “medium” standard of care) to $12 (the cost of meeting the “low” standard of care, $6, plus the expected incremental liability of $6). Thus, the injurer would make the investment if k < $4. Another way to explain this distortion – too little incentive to invest ex ante in reducing one’s riskiness – is by recognizing some positive social value of the investment, which the injurer cannot expropriate. If the injurer does not invest in becoming safe, he takes “medium” care and pays zero damages. If, instead, he does invest, he self-personalizes, takes “low” care, and pays some damage. The investment creates a benefit for the victim, in the form of some expected damages. Since the injurer does not internalize this benefit, his investment is too low. Consequently, injurers have deficient incentives to become safe. 99 At the same time, we cannot rule out the nagging possibility that the overinvestment problem of personalized standards might be worse than the underinvestment problem of uniform standards. In the example, if k < $12, the investment is efficient but would only be made under a personalized standards regime. If $12 < k < $20 the investment is inefficient but would still be made under a personalized standards regime. In this case, the overall cost of accidents, inclusive of the cost of the ex ante investment, would be higher and less efficient under a personalized standards regime. And, finally, if k > $20, the investment would not be made under either regime, injurers would remain risky, and there will be no difference between the two regimes.

Ben-Shahar/Porat

81

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

5. Summary We compared two types of personalization – skill-based and risk-based – along four dimensions: care, activity, victim’s contributory care, and ex ante investment. The main themes that emerged are the following: 1. Personalization improves injurers’ level of care. 2. The main distortions that personalization may cause are three: excessive activity levels, costlier victim care, and weak incentives to invest ex ante. 3. The activity-level distortion applies to the unskilled and safe injurers, and it is due to the lower personalized standards they face. It does not apply to skilled and risky injurers; for them, personalization reduces the activity-level distortion that is otherwise ingrained in a uniform standard negligence regime. Accordingly, a onesided application of the personalization regime – increasing the standards of care for the skilled and risky injurers but providing no reduction for the unskilled and safe – would unambiguously improve injurers’ activity level. 4. The victim care distortion applies only to skill-based personalization and is due to the increased variance in risks that victims face. Risk-based personalization, by contrast, reduces this variance and may improve the efficiency of victim care. Therefore, a personalization regime based solely on risks would unambiguously decrease victim costs of care. 5. The ex ante investment distortion applies mainly to skill-based personalization; with risk-based personalization, injurers typically have more efficient incentives to invest in decreasing their riskiness than under a uniform standard regime. Here, too, a personalization regime based solely on risks would unambiguously improve injurers’ incentives to invest in reducing their harmfulness. 6. The gap in incentives between a personalized standard regime and a uniform standard regime narrows once self-personalization under a uniform standard regime takes place. Specifically, the gap narrows with respect to the incentives of the unskilled and safe (who self-personalize under a uniform standard), but not with respect to the skilled and risky (who do not self-personalize). 109 Table 4 summarizes our main conclusions (naturally, not all nuances are captured by the table). Each of the four columns is a different regime, distinguished by the type of personalization (skill versus risk) and the direction of standard adjustment (upward versus downward). A “+” sign means that the specific personalized standards regime is more efficient along that aspect than uniform standards. 108

Table 4: Efficiency effects of personalization Skilled: up

Unskilled: down

Risky: up

Safe: down

1. Level of Care

+

+*

+

+*

2. Activity Level

+

–

+

+

3. Victim Care

–

–

+

–

4. Ex Ante Investments

–

–

+

+

* Under the assumption of self-personalization, personalization has no effect compared to a uniform standard.

82

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

IV. Justice considerations The analysis in Part III identified the incentive effects of a personalized negligence 110 regime and evaluated them along the total welfare metric. While the overall effect is ambiguous, we nevertheless identified several clear advantages to personalized standards, suggesting that in a large set of circumstances it is a superior regime. The Article does not end here, however, because two additional aspects need to be considered. One is a feasibility aspect: Do courts have the institutional capacity to implement personalized standards? This will be taken up in Part V below. The other is a normative aspect that often plays a central role in evaluating tort law doctrines – whether the rule is just. We offer in this Part a preliminary analysis of the justice considerations that might be relevant to the personalization of negligence law.

1. Corrective justice Corrective justice imposes primary duties on actors to refrain from injuring others, 111 and to repair injuries that were caused by the violation of the primary duties.100 It mandates that the negligent injurer should compensate the victim for her losses if, by his wrongdoing, he infringed on his duty not to harm the victim (or to create unreasonable risk of harm), and thus violated the equality between the parties.101 Compensation is aimed at rectifying the injustice done by the wrongdoer to the victim.102 Under a prominent corrective justice account, what constitutes an unreasonable risk 112 created by the injurer toward the victim has to be determined without regard to the burden of reducing the risks on the injurer.103 Being negligent is not merely failing to take cost-justified care (as it is in economic analysis of negligence, under the Hand formula). Rather, and regardless of the cost, the injurer’s duty has to comport with a reasonable conception of liberty and security for the victim.104 In this light, a party may be held negligent even if the cost of untaken care is too high, under a cost-benefit analysis. 100 Theories of the Common Law of Torts, Stan. Encyclopedia Phil., (22. September 2003) http://plato. stanford.edu/entries/tort-theories. 101 See Aristotle, The Nicomachean Ethics, 77–78 (David Ross trans., Oxford Univ. Press 2009). 102 See Coleman, Risks and Wrongs, 367–69 (1992) (justifying liability for negligence under corrective justice framework); Weinrib, The Idea of Private Law, 145–70 (1995) (discussing negligence law under corrective justice theory). In recent years, Goldberg and Zipursky have argued that tort law’s goal is to allow a remedy for victims of wrongdoing, rather than restoring them to the position they would have been in but for the wrongdoing. See Goldberg, The Constitutional Status of Tort Law: Due Process and the Right to a Law for the Redress of Wrongs, 115 Yale L.J. 524 (2005) (advocating for recognition of right that empowers individuals to seek redress against persons who have wronged them); Goldberg/ Zipursky, Unrealized Torts, 88 Va. L. Rev. 1625, 1643 (2002) (arguing that some avenue of recourse other than private violence must be made available to victim); Zipursky, Civil Recourse, Not Corrective Justice, 91 Geo. L.J. 695 (2003) (arguing that tort system requires injurer to cover injury of specific party and entitles that party to recover for specific injury from defendant); Zipursky, Rights, Wrongs, and Recourse in the Law of Torts, 51 Vand. L. Rev. 1, 82–90 (1998) (arguing wronged plaintiff is only entitled to use civil legal system to exact compensation from defendant). 103 See Weinrib, supra (fn. 102), at 147–52 (contrasting American approach, which compares risk and cost of precautions in order to determine what constitutes reasonable care, and English approach, which ignores cost of precautions altogether in formulating proper standard of care). 104 See Coleman, Legal Theory and Practice, 83 Geo. L.J. 2579, 2603–04 (1995) (arguing that objective standard of care comports with reasonable conception of liberty and security); Marshall, On the Idea of Understanding Weinrib: Weinrib and Keating on Bipolarity, Duty, and the Nature of Negligence, 19 S. Cal. Interdisc. L.J. 385, 398 (2010) (describing how objective standard of reasonable care reflects equal

Ben-Shahar/Porat

83

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

If one accepts the irrelevance-of-cost premise, then skill-based personalization would seem to conflict with corrective justice. An injurer’s skill, as we defined it above, is primarily a measure of his burden in reducing risks – the very factor this conception of corrective justice rejects. An actor with relatively low skill should not be permitted to satisfy a more lenient standard, and conversely, an actor with above-average skills should not face a higher standard. 114 We do not accept this irrelevance-of-cost premise. As argued by one of us previously, even under a corrective justice account, negligence and unreasonable risks could not be meaningfully defined without considering the burden of care imposed upon the injurer.105 If a technological shock made it ten times cheaper to administer some care measure, doesn’t the victim’s interest in security entitle her to expect an increase in the amount of care used to protect her? In fact, it is hard to see why the corrective justice account would oppose a personalized increase in the standards of care. Even if injurers should not be allowed to argue that because of their low skills the “average” burden of care is too heavy for them and should be reduced, victims should be allowed to argue that because of the injurer’s high skills the “average” burden of care is too lenient and has to be increased.106 115 Finally, while the case for skill-based personalization might conflict with some conceptions of corrective justice, the case for risk-based personalization would only be bolstered by this normative framework. The focus on the duty not to expose victims to unreasonable risk means that injurers whose conduct imposes relatively high risk should do more to reduce it than injurers whose same conduct imposes a lower risk. Otherwise, if both are held to the same standard, they would expose victims to different levels of risk. Indeed, we saw in Part III that risk-based personalization reduces the variance of risks created by injurers.107 Personalized standards therefore reduce to what should be considered an anomaly under the corrective justice account – that some victims are presented with greater uncompensated harms than others. 113

2. Distributive justice 116

Personalization has distributive consequences. First, by treating different injurers differently, it raises questions of distributive justice across injurers. Indeed, we saw that personalization may increase or decrease the variance in costs of care borne by injurers. Such unequal allocation of the burden of care among similarly situated injurers might be considered unjust, violating the requirement to treat like cases alike.108 But is it? Are status of parties). But see Rachlinski, Misunderstanding Ability, Misallocating Responsibility, 68 Brook. L. Rev. 1055, 1057 (2003) (arguing that subjective standard of care comports with corrective justice rationales and that “[b]y comparing the conduct of ordinary people to that of an idealized superhero, the law allocates fault where none exists and labels reasonable conduct as unreasonable”). 105 See Porat, Questioning the Idea of Correlativity in Weinrib’s Theory of Corrective Justice, 2 Theoretical Inquiries L. 161, 167 (2001) (arguing that one cannot characterize risk as reasonable or unreasonable without considering burden of care). 106 This counterargument can be derived from the justification for ignoring the injurer’s burden of care. As Weinrib put it, the injurer should not be allowed to unilaterally draw the line between his and the victim’s rights. Weinrib, supra (fn. 102), at 152. This justification does not necessarily imply that the injurer with high skills should not do more than the injurer with average skill to protect the victim’s rights. 107 See supra Section III.3. Indeed, it might even happen – as we have demonstrated – that personalization could make the risks created by the more risky injurer lower than those created by the less risky one. 108 See Keren-Paz, Torts, Egalitarianism and Distributive Justice, 5–7 (2007) (explaining that distributive justice theory is based on formulation of proportion between participants regarding their possession of criteria for distribution and arguing that it is seemingly unjust to impose different standard of care on

84

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

injurers similarly situated if they have different skills or create different risks? We address this question in Subsection a) below. Second, personalization of standards of care changes the allocation of accident costs 117 between injurers and victims, trading victim harm for injurer care. Low skill injurers, for example, are asked to take less care even though this might result in higher harm. While justified under cost benefit analysis, does this result conform to principles of distributive justice? Can precautions and harm be treated at equal footing? Should the goal of preventing harms be treated with priority over saving in precautions?109 These questions are the topic of Subsection b) below. a) Among injurers. Personalization replaces a uniform one-size-fits-all standard with 118 a scheme that has higher variance. Engaged in the same conduct, different injurers are asked to meet different standards. But the distributive impact of this greater variance of standards depends on how it affects the distribution of burdens. Consider, first, skill-based personalization, which requires more skilled injurers to 119 take more care. The skilled injurers have to meet more burdensome standards, but at the same time they are able to meet any standard at a lower private cost. (This, recall, is the very definition of injurer skill: more impact for any $1 of care). Under a uniform standard regime, both the skilled and unskilled injurers are required to take the same level of care, requiring the unskilled injurers to spend more than the skilled injurers. Raising the standard for the skilled injurers and lowering it for the unskilled injurers counteracts this unequal cost-burden and contributes to a more equal allocation of burdens. This argument may seem to hold greater merit when skills are distributed exogen- 120 ously and are uncontrolled by injurers, as in the case of inherited physical and cognitive abilities that are determined by nature. There is some unfairness in the distribution of endowments and it is offset by graduated duties. But what if skills are acquired by injurers through deliberate investment in human capital and precaution aids, as examined in Section III.4 above? Should high-skill injurers be denied the cost saving they worked hard to acquire? Should noninvesting low-skill injurers be rewarded with a lower standard and lower burden? A possible defense of personalization even along the dimension of deliberately acquired skills would focus on overall progressivity of social policy. Often, individual skills – even if acquired by deliberate investment – are also correlated with other privileges and advantages in society at large. If skilled people are on average better off, if they are more likely to tap into socially funded opportunities, if social institutions allow them disproportionate access to the opportunities to invest in skill and to benefit from their acquisition – then an offsetting burden to meet heightened standards would not violate an overall scheme of distributive justice and may well improve it.110 two similarly situated injurers). For a different argument stating that a subjective standard of care can promote distributive goals, see Logue/Avraham, Redistributing Optimally: Of Tax Rules, Legal Rules, and Insurance, 56 Tax L. Rev. 157, 238 (2003). 109 See Keating, Pricelessness and Life: An Essay for Guido Calabresi, 64 Md. L. Rev. 159, 178–80 (2005) [hereinafter Keating, Pricelessness] (stating that legal rules cannot trade severe injuries for trivial precautions borne by others); see also Esper/Keating, Putting “Duty” in Its Place: A Reply to Professors Goldberg and Zipursky, 41 Loy. L.A. L. Rev. 1225, 1248–49 (2008) (arguing that “[s]acrificing an urgent interest – the interest in avoiding premature death or devastating injury – for the sake of trivial gains to others cannot be justified”); Keating, Reasonableness and Rationality in Negligence Theory, 48 Stan. L. Rev. 311, 355 (1996) (stating that “the harm threatened by accidental injury and death is generally disproportionate to the harm threatened by increased precaution costs”). 110 See Seidelson, supra (fn. 41), at 44–45 (explaining why subjective standard of care is also justified according to distributive justice principles).

Ben-Shahar/Porat

85

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law 121

The picture is exactly the opposite when we evaluate the fairness of risk-based personalization. Recall that with such personalization there is more – not less – variance in costs of care than under a uniform standard regime. Is this variance justified by distributive justice considerations? Is it justified that the risky injurers would be required to spend more in reducing risks? As with variance in skills, to answer this question it is important to know the reason for the variance in riskiness. If risks are the manifestation of natural characteristics, then greater variance in costs of care due to personalization would not be supported by distributive justice. All the more so if uncontrollable riskiness is correlated with lower overall wealth or wellbeing.111 If, instead, injurers are able to reduce their risks by investing money, time, and efforts, rewarding such investors with lower care burdens is appropriate. In this case, the greater variance in the costs of care achieved through risk-based personalization would be justified. But again, the picture might flip when the distribution of advantages and burdens is viewed more broadly. As with acquired skills, it is possible that those who were able to reduce their riskiness have also managed to systematically recoup more advantages and benefits across various social activities, and are better off overall. Granting them yet another advantage – lower standards of care – would violate distributive justice.

b) Victims versus injurers. So far we have discussed distributive effects among injurers. Now we turn to victims: Are the effects of personalization on victims justified by conceptions of distributive justice? Is victims’ safety compromised and placed at an inferior normative status relative to injurers’ attributes?112 123 As we have seen, personalization raises the standard of care for skilled and risky injurers. At the same time it decreases the standard of care for unskilled and safe injurers. Victims of some injurers are therefore safer, whereas victims of other injurers are less safe. Still, as long as victims are equally likely to face all types of injurers, the greater efficiency of personalized standards suggests that the overall shifting of losses to victims – namely, only those losses that injurers are unfit to prevent – conforms to the distributive goals of tort law.113 124 We cannot, however, make the stronger claim – that under personalization victims face overall fewer risks and uncompensated losses. Skilled and risky injurers take more care and reduce risks, but unskilled and safe injurers take less care and increase risks, relative to uniform standards. This ambiguity remains even if under uniform standards injurers self-personalize. Recall that under uniform standards, unskilled and safe injurers may choose to ignore the standard, take lower (and more efficient) care, and bear negligence liability. Compared to a personalized standards regime, under a uniform standard with self-personalization victims incur two conflicting effects. On the downside, they receive less care from the skilled or risky injurers. On the upside, they receive full compensation from the unskilled and safe injurers who self-personalize and are found to be negligent. It is impossible to determine unambiguously which effect dominates. 125 As a caveat, it is possible to apply a partial personalization regime that would also make victims unambiguously better off, and thus not conflict with a victim-oriented fairness baseline. Clearly, victims would be better off if personalization is applied

122

111 Also, risky people may be injured more often, and pay higher insurance due to insurers’ reliance on experience rating in determining premiums. Norberg, supra (fn. 9). 112 See Keating, Pricelessness, supra (fn. 109), at 179–80 (arguing that victim’s severe injuries should not be tradable for injurer’s abilities to take precautions). 113 See Logue/Avraham, supra (fn. 108), at 237–38 (arguing that subjective standard of care may conform to distributive justice principles).

86

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

asymmetrically, raising the standard of care only for the skilled and risky, and preserving the uniform average standard for the rest. They would enjoy higher safety due to higher care taken by some injurers, without the downside of lower safety (or lower compensations) otherwise.

V. Broadening personalization We now turn to more a pragmatic question: Is it realistic to expect courts to implement personalized standards, and for people to correctly anticipate these burdens? Our discussion so far showed that personalization – if done correctly – can provide efficiency and fairness gains which current law does not realize. But does it create informational burdens too heavy for the legal system to bear? Can courts do what has become common practice in many industries and utilize more fine data to set personalized standards of care? If so, how far should personalization go? In this Part, we argue that any personal information that is feasible for courts to reliably collect and for individual actors to reliably foresee should be factored into personalized standards. This includes information about individual characteristics, including physical, genetic, cognitive, and emotional, as well as information about individual resources and past experience. The information could be collected through standard “low-intensity” methods such as past records, observable traits, tests, and screens. The information could also be collected through “high-intensity” methods such as Big Data and machine-learning prediction methods. While the feasibility of some of these methods may still be limited by technological and legal constraints, our goal is to demonstrate the enormous potential that nonpersonalized negligence law is threatening to leave untapped. As the amount of relevant information may be large and growing, the implementation of personalized standards is limited by several constraints. First, courts may be only partially able to translate personalized data into individual standards, lacking the actuarial expertise to make the fine-tuned, continuous adjustments. This problem can be solved, we argue, by a regime of qualitative step-adjustments in the standards – similar to the sentencing guidelines approach in criminal law. Second, personalized standards can have the desirable deterrent effect only to the extent that actors can anticipate them. Calibrating the standards too finely along a continuous range could create uncertainty among actors, which itself distorts care choices.114 We argue, perhaps counterintuitively, that it often would be easier for injurers to anticipate personalized standards than uniform ones, because they know more about their own characteristics than about the general distribution of characteristics in society. Some of the evidentiary proposals presented in this Part may strike readers as a fantasy. They create a different model of information acquisition by courts than the traditional rules of evidence and civil procedure. We nevertheless present these ideas as 114 See Calfee/Craswell, Some Effects of Uncertainty on Compliance with Legal Standards, 70 Va. L. Rev. 965, 966 (1984) (analyzing inefficient effects of uncertain legal standards). Calfee and Craswell have further developed their analysis in later articles. See Craswell/Calfee, Deterrence and Uncertain Legal Standards, 2 J.L. Econ. & Org. 279, 280 (1986) (formalizing and extending analysis of authors’ prior work on inefficient effects of uncertain legal standards while focusing on three factors – the shape of the uncertainty in the law, the costs and benefits created by the activity, and the penalty structure for violating the law – that determine which incentives are likely to dominate in an uncertain legal system); see also Ben-Shahar, Should Products Liability Be Based on Hindsight?, 14 J.L. Econ. & Org. 325, 325 (1998) (exploring ramifications of determining product liability in hindsight and noting uncertainty effects that it creates upon manufacturers).

Ben-Shahar/Porat

87

126

127

128

129

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

a benchmark for discussion. Our argument, in a nutshell, is that if procedural and ethical rigidities can be overcome, the law could make advances similar to ones made in areas like medicine, insurance, marketing, or education. There is a large potential for improving the deterrent effect of negligence law, without sacrificing (and perhaps also promoting) important notions of corrective and distributive justice.

1. Procedures for implementing personalized standards The first question any personalization regime has to address is the degree of granularity. A more granular regime distinguishes individuals more finely and adjusts the standards based on more factors, sensitive to more kinds of information. At the extreme, the most granular regime requires courts to tailor the standard of care for each injurer along a continuum, shifting it up or down in response to every bit of individualized information (the “continuum mode” of personalization). 131 Choosing the optimal granularity of a personalized standards regime is a problem of information costs. First, it might be costly for courts to collect the information necessary to tailor different standards of care for each and every injurer. It is cheaper and easier for courts to avoid the information-rich inquiry of personalized standards and implement a one-size-fits-all uniform standard. Even if personal information is collected and presented at trial, there are limits to courts’ abilities to process the available data and translate it accurately into adjustments of the standard of care (and even more so when jurors are involved). This requires actuarial expertise that courts often lack. 132 Second, like courts, injurers facing personalized standards need to take into account personal traits when trying to anticipate and perform their duties of care and understanding how courts would require them to behave. Is it realistic to expect injurers to make such informed assessments? Can they adapt their behaviors to the standard of care they are required to meet under the continuous mode? Are uniform standards easier to anticipate? 133 It might seem, intuitively, that the information problems faced by courts and by injurers in a personalized standards regime are similar. Since, by definition, personalized standards rely on more richly tuned and finely partitioned information, they inflict on all participants in the regime, including courts (ex post) and injurers (ex ante), a more daunting informational task. But upon further reflection we claim that this conjecture is false. In fact, it is easier for injurers to anticipate what is reasonable for them, given their personal characteristics, than to extrapolate what is reasonable for the average person in society. We know our riskiness and skill better than we know the societal distributions of these traits, and we can act intuitively upon this self-knowledge. True, people may learn or infer the uniform societal standards without having to know the exact distributions over the entire society by observing past cases and by following societal norms. But in a regime that relies on ex post standards (which, unlike ex ante rules, do not articulate bright-line commands) such learning of what the uniform standard requires is slow and imperfect. Personalized standards, by contrast, require no learning, as much as they harness information injurers already have about themselves.115 130

115 This is not always the case. New drivers, for example, might be unaware of their skills, and learning, by observing the uniform standards applied to all, might sometimes be a better option than adapting their behaviors to their perceived subjective skills. There is also the risk that actors, under a personalized standard regime, would be “over-optimistic” about their abilities, and that might distort their incentives, encouraging them to create too high risks. That problem, however might distort incentives also under a uniform standard regime. Furthermore, risk aversion might counteract overoptimism under a personalized standard regime.

88

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

The problem for injurers is that while they may have a good sense of what 134 individually optimal behavior is, given their idiosyncratic traits, they still need to anticipate the personalized standards that an imperfect court would impose on them. Even if courts set personalized standards that are unbiased, their tailoring would have some degree of inaccuracy (random errors). Having to anticipate such imperfect tailoring of personalized standards, injurers’ informational burden would be compounded. To overcome some of the information costs that courts face, and to help injurers 135 predict how standards would be personalized, we propose two procedural refinements to a personalized standards regime: (1) gradual personalization and (2) presumptions. a) Gradual personalization. A continuous mode of personalization – under which 136 every bit of personalized information can shift the standard incrementally – would likely involve excessive information costs. It would be too costly for courts to implement case by case; and too costly for injurers to anticipate the patterns of courts’ judgments. To reduce these information problems, personalized standards may be set along a 137 scheme of discrete qualitative steps – what we call “gradual personalization.” According to this scheme, courts would have to choose among a limited number of standards – for example, a three-step scheme of high, medium, and low – and pigeon-hole injurers into these groups. Gradual personalization is similar to a sentencing guidelines scheme that provides qualitative step-like adjustments to judgments based on case-specific characteristics, but stops short of the continuous mode. For example, while drivers’ skills and riskiness may vary along a continuum, justifying driving their cars at a different speed under similar circumstances of the road and traffic, the gradual personalization scheme would require them to drive at low, medium or high speeds. Thus, at similar situations, one driver would be expected to drive no more than fifteen miles per hour, another driver up to twenty miles per hour, while a third driver would be allowed to reach twenty-five miles per hour. Gradual personalization would certainly be an improvement compared to the current 138 uniform standards rule. It is more practical and easy to implement than a pure personalization rule. And the degree of granularity (the number of steps) would depend on the variance of personal attributes and the costs for courts of making finer personal determinations, and for injurers of anticipating these partitions. b) Presumptions116. The information cost burden on courts can be reduced by a 139 scheme that elicits private information from injurer defendants (and does not rely on utilization of Big Data). Consider, first, a simple presumption that injurers are average skilled.117 Unless information is presented to rebut this presumption, courts would set the uniform midlevel standard of care. The injurer-defendant may present evidence to rebut the presumption.118 Assume that courts have no upfront information, but can tell if the information provided by the litigants is credible. Low-skill injurers would have the incentive to produce such evidence to the court and reveal their true type, so as to enjoy a more forgiving standard. The problem, of course, is that high-skill injurers would not. Some evidence regarding the high skill of a particular defendant may be obtained by the plaintiff through discovery and presented to court; and sometimes courts would be able to infer, from defendant’s 116

We are grateful to Michael Trebilcock for suggesting that this procedure be examined. Similar analysis would apply to evidence regarding the riskiness of the individual injurer. 118 The victim-plaintiff may present evidence too, but it would largely be based on information obtained through discovery from the defendant. 117

Ben-Shahar/Porat

89

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

silence, that she is the high-skill type. But such information would unravel only partially, and the separation of types would not be perfect. 140 To improve the incentives to provide courts with the defendants’ private information, imagine instead a more strategic design of the presumption. Consider a presumption of high skill: If there is no additional information, the standard of care is set at the very high level, reflecting the optimal care for the most skilled injurer. This presumption, too, can be rebutted by evidence the injurer produces. Now, however, all injurers along the spectrum of skill (other than those at the very top) would have the incentive to reveal themselves – to demonstrate that the standard is unrealistically demanding in their particular circumstances. With more complete unraveling of private information, courts end up with as much information as injurers privately possess. 141 Using such a presumption is not cost free. Some defendants would face the dilemma of whether to reveal medical, psychological, or other sensitive information. Revealing information relating to various kinds of low skill might have adverse consequences elsewhere, for example in the workplace. Still, it is quite common for tort plaintiffs to reveal sensitive information, for example pertaining to medical conditions. In the same way that a plaintiff should consider whether to reveal private information for the purpose of expanding tort liability and recovery, defendants would elect whether to reveal private information to qualify for a lower standard of care.

2. Which personal information? Another aspect of accuracy, apart from granularity as discussed above, addresses the types of information a personalized standards regime incorporates. It seeks to distinguish people according to their tendencies to create risks and their capabilities to prevent them. But which information should be drawn upon? Which individual characteristics should be the basis for personalized negligence law? 143 Our discussion in this section is intended to begin charting the informational potential that personalized standards could unleash. This includes physical and genetic information about people as well as personality information including cognitive skills and emotional aspects. It could be learnt either from general data and statistics relating to the injurer’s attributes such as age, gender, education, and profession, or from personal information collected directly through medical, physical, or psychological tests and from past behavior that resembles the behavior in question. It could also be inferred from past behavior that is different from the behavior in question but could reveal capabilities which are relevant to the assessment of the behavior in question. 144 The information relevant to setting personalized standards can be collected through traditional methods such as public records or examination scores, but it could also be collected from large digital databases – Big Data. The term Big Data refers to databases with enormous quantities of information.119 Data mining – the process of discovering human behavior patterns in these large-scale databases – allows predictions of future behavior across many dimensions. Big Data analysis can predict various risks, personal characteristics, preferences, and many other aspects relevant in the determination of optimal legal standards. 142

119 See Colonna, A Taxonomy and Classification of Data Mining, 16 SMU Sci. & Tech. L. Rev. 309, 329–31 (2013) (defining Big Data and explaining how data mining works); Porat/Strahilevitz, supra Part 1.A (discussing usage of Big Data and explaining how it can be used for personalizing default rules).

90

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

a) Physical characteristics. Different people impose different risks on others based 145 on their physical characteristics. For example, a short driver might create higher risks than tall drivers toward both other drivers and pedestrians, because she might only have a narrow vision of the road;120 drivers with impaired vision are likely to impose higher risks on others;121 and the same is true with respect to drivers who have hearing difficulties.122 Higher risks might justify demanding more precautions from the actor creating the risks. Research conducted in the field of system design of planes and automobiles indicates 146 a relationship between certain human traits and the capabilities to perform a certain task. For example, research conducted by J.E. Korteling showed – quite unsurprisingly – that older drivers (sixty-one to seventy-three years old) and drivers with brain injury history have significantly longer reaction time than younger drivers (twenty-one to forty-three years old).123 Age is also a significant factor in predicting drivers’ ability to avoid lane crossing124 and their braking response time.125 In our example in the Introduction, a sixty-five-year-old man probably imposes higher risk on others than the average driver, and his reaction time is probably longer than that of the average driver. The higher risk would optimally require him to drive slower, while the longer reaction time might justify relaxing the “reaction time standard.” b) Cognitive and emotional characteristics. Risk creation is also linked to mental 147 and cognitive capabilities and traits. For example, a driver with high spatial awareness can better avoid dangerous situations and should face an elevated standard that would prompt her to utilize more of her skill.126 Human traits such as impulsivity, risk taking, and sensation seeking increase the likelihood of a person to engage in dangerous activities, thereby imposing risks on others.127 Therefore, a sensation-seeking doctor 120 See Dobbs, supra (fn. 2), § 119, at 283 fn. 10 (citing Mahan v. State, 191 A. 575, 580 (Md. 1937) (holding that driver whose short stature imposed limitations on her vision is expected to exercise “greater watchfulness” to avoid injuring others)). A few car manufacturers, being aware of short drivers’ visibility problem, offer some models for shorter people. See Kronenberg, 5 Best Cars for Short Drivers, TheStreet (20 September 2013), www.yahoo.com/autos/s/5-best-cars-for-short-drivers-213753032.html; Rogers, Better Cars for Short and Tall Drivers, Wall Street J. (9 October 2013, 7:10 PM), www.wsj.com/articles/ SB10001424052702304626104579123411103492676. 121 See Ball et al., Visual Attention Problems as a Predictor of Vehicle Crashes in Older Drivers, 34 Investigative Ophthalmology & Visual Sci. 3110, 3118 (1993) (showing that older drivers with severe sensitivity loss in both eyes have twice the number of crashes than older drivers with normal visual field sensitivity). 122 See Hickson et al., Hearing Impairment Affects Older People’s Ability to Drive in the Presence of Distracters, 58 J. Am. Geriatrics Soc’y 1097, 1101–12 (2010) (showing that people with moderate to severe hearing impairment had significantly poorer driving performance in presence of distracters than those with normal or mild hearing impairment). 123 See Korteling, Perception-Response Speed and Driving Capabilities of Brain-Damaged and Older Drivers, 32 Hum. Factors 95 (1990) (describing experiments regarding reaction-time tasks and driving tasks that were conducted to identify variables that may be sensitive to the effects of aging). 124 See Szlyk et al., Relative Effects of Age and Compromised Vision on Driving Performance, 37 Hum. Factors 430, 430–36 (1995) (describing experiment held in order to determine effects of age and compromised vision on driving skills). 125 See id. at 435 (showing that older groups had poorer driving-related skills than younger groups on simulator missions). 126 See Wochinger/Boehm-Davis, U.S. Dep’t of Transp., The Effects of Age, Spatial Ability, and Navigational Information on Navigational Performance, (1995) (showing that navigational ability, which is linked to car accidents’ involvement, declines with age due to decrements in spatial ability and perceptual speed). 127 See Zuckerman/Kuhlman, Personality and Risk-Taking: Common Biosocial Factors, 68 J. Personality 999, 1000 (2000) (explaining that some personality traits, such as sensation seeking, are relevant to risk-taking inclination).

Ben-Shahar/Porat

91

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

might be more prone to appraise risks as lower than a low-sensation-seeking doctor. We might want to require the former doctor – or his employer – to take extra precautions before making crucial decisions involving risk estimation. 148 A conscientious person tends to be more organized and prefer planned rather than spontaneous behavior.128 This tendency has a clear implication for the way different people perform their tasks and the precautions they could take to reduce risks. It might be reasonable to have different demands and expectations from actors who tend to be planners (and may be more responsible, organized, and reliable)129 and from actors who are spontaneous. Those demands and expectations might change across activities. Sometimes we might demand that actors who are less organized take more precautions to decrease risks (in the case of doctors, for example), while sometimes the less organized and more spontaneous actors might be the ones more capable to react to unexpected circumstances (say, unexpected risks in the road) and that might affect the standard of care most suitable for them. 149 Reaction time while performing a dangerous task depends, we saw, on physical aspects, but also on psychological factors such as fatigue, aging, history of brain damage, and use of drugs.130 A surgeon who suffers from sleep deprivation is likely to impose a higher risk to patients than other surgeons. Also, when the time of performing a task increases – such as when the operation on a patient becomes longer – the surgeon’s fatigue increases, resulting in significant increase in reaction time and in risk to patients.131 This might justify an increase in the standard of care from doctors who suffer from sleep deprivation, for example by requiring them to take longer breaks in extended shifts, or if this is impossible, requiring them to take more precautions as the operation progresses. And, conversely, when taking a break or taking other precautions is impractical, it might be justified to relax – rather than elevate – the standard of care. As urgency rises and care becomes costlier, the optimal level should correspondingly adjust. 150 Big Data can be a reliable source for learning about injurers’ cognitive skills and intelligence, sometimes more so than direct exams, because it is not as manipulable (people may underperform on exams if high scores raise their burden of care). For example, according to some studies, intelligence and cognitive abilities can be predicted to a high degree of accuracy based on records of users’ “likes” on Facebook. One study found that strong predictors of high intelligence included “likes” to the Facebook pages for “Thunderstorms,” “The Colbert Report,” “Science,” and “Curly Fries,” and that low intelligence correlated with “likes” to the pages for “Sephora,” “I Love Being A Mom,” “Harley Davidson,” and “Lady Antebellum.”132 Similarly, a person’s level of education can be inferred by analyzing search terms and web pages accessed by her,133 although in 128 See Quercia et al., Our Twitter Profiles, Our Selves: Predicting Personality with Twitter, in 2011 IEEE International Conference on Privacy, Security, Risk and Trust and IEEE International Conference on Social Computing 180, 180 (2011) (analyzing relationship between personality and different types of Twitter users). 129 See Golbeck et al., Predicting Personality with Social Media, in The 29th Annual CHI Conference on Human Factors in Computing Systems 253, 254 (2011) (presenting method of predicting human personality through information available in Facebook profiles). 130 Gawron, Human Performance Measures Handbook, 41 (2000). 131 See Harris et al., Performance, Workload, and Fatigue Changes Associated with Automation, 5 Int’l J. Aviation Psychol. 169, 176–85 (1995) (discussing influence of workload and fatigue in multitask environment). 132 Kosinski et al., Private Traits and Attributes Are Predictable from Digital Records of Human Behavior, 110 Proc. Nat’l Acad. Sci. U.S. 5802, 5804 (2013). 133 See Murray/Durrell, Inferring Demographic Attributes of Anonymous Internet Users, in: Masand/ Spiliopoulou (eds.), Web Usage Analysis and User Profiling, (2000), 7, 14–18 (showing that demographic

92

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

most cases, for personalizing the standard of care, the level of education can be more easily learned from direct resources. Or, there is some evidence that users with different personalities prefer different website categories. For example, people with a tendency to be well-organized prefer websites such as kodak.com, education.com, exct.net, ecnext. com, and ecollege.com.134 The tendency of a person to be well-organized could be a consideration in setting a personalized standard of care for him.135 Big Data analysis can also help courts identify risk-taking inclination, which could be essential for setting a personalized standard of care. Thus, one study has found that tendency towards risky driving is correlated with risky financial behaviors.136 Knowing how people invest might tell us also how they drive. Of course, not every behavioral study published in a social science journal should 151 budge the standard of care. Many findings are preliminary and perhaps questionable. The point we stress is the power of statistical analysis of Big Data to pick up factors that, if confidently identified, can tell us a significant amount about people’s riskiness and their skill in accident prevention. Another type of information relevant to the determination of standards of care is 152 behavioral genetics information. It connects mental and cognitive abilities and hormonal and neurological influences.137 Emerging developments in brain imaging technology enable better understanding of human behavior. One such development is MRI testing, and its complement fMRI (functional magnetic resonance imaging), which allows examination of the way the brain works during the performance of particular tasks.138 The MRI tests measure changes in blood oxygenation levels in order to identify which regions of the brain work during a specific task.139 Although MRI images require substantial interpretation, they are considered valuable in demonstrating cognitive processes and have in fact been proposed as a tool in tort cases, and some uses of this technique in criminal cases has already begun.140 MRI tests, and neuroscience more generally, could inform the court as to how to define the standard of care in a more concrete and nuanced manner than currently done.141 Specifically, some research has shown a correlation between impulsivity, emotional 153 reactions, and violent behaviors on the one hand, and specific activity in several areas in facts such as gender, age, income, marital status, and level of education can be inferred through usage information analysis). 134 Kosinski et al., Personality and Website Choice, 22 June 2012 (unpublished manuscript), http:// research.microsoft.com/en-us/um/people/pkohli/papers/kskbg_acmwebsci_2012.pdf. 135 As we argued earlier in this section, it might be reasonable to demand that less organized actors take more precautions in order to decrease the risks they create. 136 See Morrison et al., Health Shocks and Household Financial Fragility: Evidence from Automobile Crashes and Consumer Bankruptcy Filings, 3 (Coase-Sandor Inst. for Law & Econ., Working Paper No. 655 2013), https://www.researchgate.net/publication/251315742_Health_Shocks_and_Household_Financial_Fragility_Evidence_from_Automobile_Crashes_and_Consumer_Bankruptcy (explaining that persistent financial distress may encourage risky behavior). 137 See Plomin/Caspi, Behavioral Genetics and Personality, in: Pervin/John (eds.), Handbook of Personality, 2nd edn. 1999, 251, 261–67, (providing overview of research on contribution of genes to personality). 138 See Eggen/Laury, Toward a Neuroscience Model of Tort Law: How Functional Neuroimaging Will Transform Tort Doctrine, 13 Colum. Sci. & Tech. L. Rev. 235, 240–41 (2012) (explaining fMRI technique). 139 Jones et al., Brain Imaging for Legal Thinkers: A Guide for the Perplexed, 2009 Stan. Tech. L. Rev. 81, 84–85. 140 See Eggen/Laury, supra (fn. 138), at 249–52 (describing various cases, mainly criminal, in which fMRI has been used in courts). 141 See Smith Churchland, Moral Decision-Making and the Brain, in: Illes (ed.), Neuroethics, (2006), 3, 10–14 (arguing that fMRI can be used in order to identify neurobiological differences between voluntary and involuntary action).

Ben-Shahar/Porat

93

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

the brain on the other.142 It has shown a significant neurological basis of aggressive and violent behaviors.143 Once a defendant undergoes MRI or fMRI tests, the findings can be used by courts for personalizing the standard of care. Thus, if those tests point to the defendant’s impulsiveness and aggressiveness, courts might make the proper adjustment in the standard of care. c) Past behaviors. We distinguish between similar past behaviors and different past behaviors. Similar past behaviors can often be a good proxy for the defendant’s abilities and tendencies regarding risk creation and precaution taking. Thus, a driver’s record of traffic violations could be used to personalize her standard of care.144 Information about a doctor’s past malpractice behavior might also be used by the court in personalizing the standard of care.145 On many occasions, this kind of information is available through official records.146 155 More problematic is the usage of information about different past behaviors of the defendant and learning from these about her capabilities as a potential wrongdoer. As we have explained, in the era of Big Data it is no longer difficult to collect information about the defendant’s past behavior as a consumer, driver, employee, patient, student, and in many other capacities. As we have demonstrated, this past behavior might be associated with specific capabilities and traits which are relevant to the process of personalizing the standard of care. 156 Using past behavior as a predictor of risk and as a factor in determining the optimal precaution is a hallmark of insurance actuarialism – a practice known as experience rating. Every driver is familiar with the increase in insurance premium after an accident. This technique – personalizing the premium charged to each policyholder based on past behavior – is founded on the same tailored-treatment logic as personalized standards of care. In the insurance context, the use of Big Data and high-intensity information models is their bread and butter. Auto insurers, for example, invite policyholders to install data recording devices in their cars, which transmit information to insurers about driving habits, risk taking, and the competence of the driver – information that is then factored into the personalized pricing of the auto insurance policy.147 While courts cannot base judgments on similarly installed recorders of conduct, they can tap into any available resource of personal information to observe past behavior and adjust the standard accordingly.

154

142 See Raine/Yang, Neural Foundations to Moral Reasoning and Antisocial Behavior, 1 Soc. Cognitive & Affective Neuroscience 203, 205 (2006) (summarizing brain imaging studies conducted on antisocial, violent, and psychopathic groups). 143 See Bufkin/Luttrell, Neuroimaging Studies of Aggressive and Violent Behavior: Current Findings and Implications for Criminology and Criminal Justice, 6 Trauma Violence & Abuse 176, 179–84 (2005) (discussing neuroimaging studies that demonstrated abnormal brain activity in aggressive and violent offenders). 144 See Colonna, supra (fn. 119), at 360 (showing that data mining can be used to track past traffic violations). 145 See Public Use Data File, Nat’l Prac. Data Bank, www.npdb.hrsa.gov/resources/publicData.jsp (last updated Feb. 2016) (containing selected variables from medical sources concerning physicians, dentists, and other licensed health care practitioners). 146 See Colonna, supra (fn. 119), at 358 (discussing huge amount of data that law enforcement agencies acquire through Big Data records); Porat/Strahilevitz, supra (fn. 11), at (explaining how companies use publicly available data as well as proprietary data to gauge customer preferences). 147 See Tuttle, Big Data Is My Copilot: Auto Insurers Push Devices that Track Driving Habits, Time (6 August 2013), http://business.time.com/2013/08/06/big-data-is-my-copilot-auto-insurers-push-devices-that-track-driving-habits/ (reporting on new Big Data devices that help insurers profile their customers’ driving habits).

94

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

d) Resources. Another source of information, often readily available, is about 157 people’s resources. It is sometimes argued that the wealth of the injurer should be factored into the design of negligence standards, perhaps because high-resource injurers can more easily afford greater expenditures on care.148 Inasmuch as such wealth-based standards are aimed at improving wealth distribution in society, income taxes and fiscal policies are thought to be superior tools, in the sense that they achieve redistribution more efficiently and comprehensively.149 It might be thought, for example, that a small rural hospital should not be held to the same standards of medical care as a large city hospital, because the smaller facility cannot afford and should not be asked to make the same level of expenditure in advanced medical equipment. The small hospital may well face a lower standard of medical care, but not because of “affordability” or wealth. Rather, because it treats smaller populations, the value of investment in some devices is lower and insufficient to justify the costs.

VI. Conclusion This Article examined the justifications for a new type of negligence law – 158 abandoning the objective reasonable person standard and adopting instead a personalized subjective standard of care. It identified several important efficiency advantages to the selective adoption of personalized standards, and argued that tort law’s other possible objectives, including corrective and distributive justice, would also be served. Our analysis reveals that personalization could be made in two dimensions: the skill 159 dimension and the risk dimension. Indeed, the efficiency considerations (level of care, activity level, victim care, and ex ante investment) as well as the justice considerations (corrective and distributive justice) often vary depending on whether personalization is made according to the skill or according to the risk dimension, and also whether it is done to increase or decrease the standard of care relative to the uniform standard. Table 5 summarizes all the considerations, according to the skill-risk and increasedecrease (up-down) dimensions. Table 5: Effects of personalization Skilled: up

Unskilled: down

Risky: up

Safe: down

1. Level of Care

+

+*

+

+*

2. Activity Level

+

–

+

–

3. Victim Care

–

–

+

+

4. Ex Ante Investments

–

–

+

+

B. Corrective Justice

+**

–

+

+

A. Efficiency

148 See, e.g., Arlen, supra (fn. 57), at 421–23 (arguing that care expenditures impose lesser burden on the rich and thus can be raised more than on the poor). 149 See Kaplow/Shavell, Why the Legal System Is Less Efficient than the Income Tax in Redistributing Income, 23 J. Legal Stud. 667, 667‐69 (1994) (arguing that tax policies are more efficient than legal regulation in achieving distributive goals).

Ben-Shahar/Porat

95

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law Skilled: up

Unskilled: down

Risky: up

Safe: down

1. Among Injurers

+***

+***

–***

–***

2. Injurers vs. Victims

+

–

+

–

C. Distributive Justice

* Under the assumption of self-personalization, personalization has no effect compared to a uniform standard. ** Corrective justice might require the skilled injurer to do more than average, but would not allow the unskilled injurer to do less than average. *** This conclusion might change if skills/safeness are deliberately acquired.

As the table indicates, the most favorable case for personalization is increasing the standard of care for risky injurers. It might raise some distributive justice concerns, especially if riskiness is exogenously determined. Increasing the standard of care for skilled injurers also has many more pros than cons, although it does raise concerns about potential victims’ level of care, and might also create inefficient incentives to invest in improving one’s skills. As we have explained, however, in the process of personalization, victim’s care should be considered, and this might limit the extent of personalization.150 And the ex ante investment problem could be attenuated if personalization takes into account the optimal investment that the injurer should make in improving his skills.151 161 Personalization requires enormous amounts of information and much expertise in applying it, and we argued that advances in information technology could put the legal system on the path to such information-rich procedures. Even if the legal system lags behind other institutions in Big Data advances, we argued that in the short run using a gradual mode of personalization – by applying several discrete steps within the standard of care – is relatively easy to implement. 162 Like any other use of Big Data, privacy concerns might slow down the personalization of the standard of care. We believe they should not. One such concern is that the usage of Big Data in courts would encourage further collection of sensitive information, which may be used to infringe people’s privacy. But data about human characteristics has yielded enormous returns and will continue to be collected and used for commercial purposes, so there is no reason to assume that the further use of it by courts for a public purpose would have any significant effect on its already-occurring collection.152 Another concern is that Big Data and fMRI tests would infringe on the privacy of the particular injurers sued in court since it exposes personal characteristics. There are ways to protect people from embarrassing revelations and restrict their use to trials without abandoning the entire project. And it would be in the interest of many injurers to voluntarily subject themselves to such screening, if they expect the findings to reduce their personalized standards of care. Such voluntary submission to screening would thus occur along the familiar unraveling dynamic,153 because those injurers who refused to cooperate in 160

150

See supra text accompanying fn. 91. See supra text accompanying fn. 96. 152 See Porat/Strahilevitz, supra Part 1.A (noting tradeoff between privacy and personalization and contending that following industry’s lead in making consumers aware of benefits of personalization may engender greater willingness to share). 153 See Baird et al., Game Theory and the Law, 2, 89–90 (1994) (defining game theory concept of “unraveling” as “situations in which the ability of people to draw inferences from silence leads to the revelation of information”). 151

96

Ben-Shahar/Porat

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/02_Ben-Shahar-Porat-auth24.3d from 30.10.2020 11:34:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

B. Personalizing Negligence Law

tailoring the standard of care would be suspected by courts of being injurers for whom a high standard of care is appropriate. This paper studies personalization of negligence law, but there is no reason to stop 163 there. One of us previously suggested personalization of other areas of law – disclosures and default rules.154 We can also think of personalized regulatory standards, personalized penalties, and a host of applications of the idea of personalized standards beyond the realm of tort law. Personalization is the trajectory of many other social and private institutions, like insurance, medicine, education, employment, product design, and advertising. In all these areas, personalization has yielded substantial progress, even if some of its risks have to be monitored and regulated. In the same way that personalized medicine can save lives and avoid inefficient uniform treatments, personalized safety standards can reduce the social costs of accidents. How long will negligence law resist this enormous value of information? 154 See Porat/Strahilevitz, supra Part 1.A (proposing use of Big Data to personalize default terms in contracts and wills and to target disclosures based on relevance to recipient).

Ben-Shahar/Porat

97

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:38 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards* 1

Scholars have examined the lawmakers’ choice between rules and standards for decades. This chapter, however, explores the possibility of a new form of law that renders that choice unnecessary. Advances in technology (such as big data and artificial intelligence) will give rise to this new form – the microdirective – which will provide the benefits of both rules and standards without the costs of either. Lawmakers will be able to use predictive and communication technologies to enact complex legislative goals that are translated by machines into a vast catalog of simple commands for all possible scenarios. When an individual citizen faces a legal choice, the machine will select from the catalog and communicate to that individual the precise context-specific command (the microdirective) necessary for compliance. In this way, law will be able to adapt to a wide array of situations and direct precise citizen behavior without further legislative or judicial action. A microdirective, like a rule, provides a clear instruction to a citizen on how to comply with the law. But, like a standard, a microdirective is tailored to and adapts to each and every context. While predictive technologies such as big data have already introduced a trend toward personalized default rules, in this chapter we suggest that this is only a small part of a larger trend toward context-specific laws that can adapt to any situation. As that trend continues, the fundamental cost trade-off between rules and standards will disappear, changing the way society structures and thinks about law.

I. Introduction 2

Imagine a world where lawmakers enact a catalog of precisely tailored laws, specifying the exact behavior that is permitted in every situation. The lawmakers have enough information to anticipate virtually all contingencies, such that laws are perfectly calibrated to their purpose – they are neither over- nor underinclusive. Now imagine that when a citizen in this world faces a legal decision, she is clearly informed of exactly how to comply with every relevant law before she acts. This citizen does not have to weigh the reasonableness of her actions, nor does she have to search for the content of a law. She just obeys a simple directive. The laws at work in this world are not traditional rules and standards. Instead, they take a new form that captures the benefits of both rules and standards without incurring the costs. This new form – we call it the microdirective – is the future of law. * We thank Benjamin Alarie, Omri Ben-Shahar, Stavros Gadinis, Saul Levmore, Ariel Porat, and participants at the Granular Norms conference at Villa Vigoni, Italy, the Junior Business Law Conference, the STILE conference in Rome, the Annual Meeting of the Canadian Law and Economics Association, the Law, Economics and Business Workshop at Berkeley Law, the Law & Economics workshop at the University of Michigan Law School, the Society of Institutional and Organizational Economics meetings in Paris, France, and workshops at The University of Chicago Law School, the Faculty of Law, University of Toronto, and Washington University St. Louis for helpful conversations and comments. We also thank Michael Alcan, Elizabeth Kiernan, and Erica Yang for excellent research assistance. The Richard Weil Faculty Research Fund, the Paul H. Leffmann Fund, and the SSHRC provided generous support. In the interests of disclosure, one of the authors (Niblett) is a co-founder of a start-up bringing machine learning to the law (Blue J Legal). This chapter is a shortened version of a longer piece published in 92 Indiana Law Journal 1401 (2017).

98

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:38 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

When lawmakers enact laws today, they must choose between using rules and using standards to achieve a desired goal.1 This choice requires a trade-off between certainty and calibration. Rules provide certainty through clear ex ante statements of the content of the law.2 But rules are costly to design because lawmakers must, at the outset, identify and analyze all the various scenarios to which rules might apply. Rules can also be imprecise and error prone. Because they are defined ahead of time, they can be poorly calibrated3 to the events as they actually occur. Standards, on the other hand, are adjudicated after the fact. As a result, lawmakers avoid high up-front design costs. Moreover, when applied after the fact, standards can be precisely tailored or calibrated to a specific context as it actually arose. But they also generate ex ante uncertainty because regulated actors do not know up front whether their behavior will be deemed by the adjudicator to comply with the standard.4 We suggest that technological advances in predictive and communication technologies will render this trade-off between rules and standards unnecessary. A new form of law, the microdirective, will emerge to provide all of the benefits of both rules and standards without the costs of either. These microdirectives will provide ex ante behavioral prescriptions finely tailored to every possible scenario. The first technology to consider is predictive technology. Innovations in big data and artificial intelligence will make it increasingly easy to predict the outcomes that certain behavior will produce. Lawmakers will ultimately have the ability to cheaply gather information and use predictive algorithms and big data to update the law instantly based on all relevant factors. In effect, this lowers the cost of designing precise, finely calibrated laws. The second technology to consider is communication technology. Ubiquitous and instantaneous communication capabilities will reduce the uncertainty of law. From the vast catalog of rules generated by predictive technology, communication technology will be able to identify the rules applicable to an actual situation and inform the regulated actor exactly how to comply with the law. It will be able to translate all the information into a single behavioral directive that individuals can easily follow. To see how the mechanism might work, consider the regulation of traffic speed. In a world of rules and standards, a legislature hoping to optimize safety and travel time could enact a rule (a sixty miles-per-hour speed limit) or a standard (“drive reasonably”). With microdirectives, however, the law looks quite different. The legislature 1 See, e.g., Kaplow, Rules Versus Standards: An Economic Analysis, 42 Duke L.J. 557, 561 fn. 6 (1992); McGinnis/Wasick, Law’s Algorithm, 66 Fla. L. Rev., 991, 1027 (2014). 2 The literature on this distinction is vast. See Farnsworth, The Legal Analyst: A Toolkit For Thinking About The Law, 163–71 (2007); Raz, Practical Reason and Norms (1990); Sunstein, Problems with Rules, 83 Cal. L. Rev. 953, 961–62 (1995); see also Ehrlich/Posner, An Economic Analysis of Legal Rulemaking, 3 J. Legal Stud. 257 (1974); Sullivan, The Supreme Court, 1991 Term, Foreward: The Justices of Rules and Standards, 106 Harv. L. Rev. 22 (1992). See generally Kaplow, supra (fn. 1); Schauer, The Tyranny of Choice and the Rulification of Standards, 14 J. Contemp. Legal Issues, 803, 803 fn. 1 (2005). 3 We use the term calibration to denote the fit of a law to its legislative purpose. For example, a fiftyfive miles-per-hour speed limit may be poorly calibrated because it is too low when the weather is perfect and the roads are clear and too high when the weather is bad and the roads are crowded. Another term could be “inclusiveness.” The fifty-five miles-per-hour speed limit is both under- and overinclusive because it prohibits some desirable behavior (driving sixty miles per hour on a sunny day) and allows some undesirable behavior (driving fifty miles per hour on a rainy day). See Diver, The Optimal Precision of Administrative Rules, 93 Yale L. J. 65, 73–74 (1983); Kaplow, supra (fn. 1), at 565; McGinnis/Wasick, supra (fn. 1), at 1030–31; Schauer, The Convergence of Rules and Standards, 2003 N.Z. L. Rev., 303, 305–09; Schauer, supra (fn. 2), at 803–04; Sunstein, supra (fn. 2), at 992. 4 Kaplow, supra (fn. 1), at 569, 575 fn. 42, 587–88; Sunstein, supra (fn. 2, at 974–77; see also Craswell/ Calfee, Deterrence and Uncertain Legal Standards, 2 J.L. Econ. & Org. 279 (1986); Kennedy, Form and Substance in Private Law Adjudication, 89 Harv. L. Rev. 1689–701 (1976).

Casey/Niblett

99

3

4

5

6

7

8

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:38 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

9

10

11

12

13

merely states its goal. Machines then design the law as a vast catalog of context-specific rules to optimize that goal. From this catalog, a specific microdirective is selected and communicated to a particular driver (perhaps on a dashboard display) as a precise speed for the specific conditions she faces. For example, a microdirective might provide a speed limit of 51.2 miles per hour for a particular driver with twelve years of experience on a rainy Tuesday at 3:27 p.m. The legislation remains constant, but the microdirective updates as quickly as conditions change. In this chapter, we explore whether this example could become the model for law more broadly. Our long-run prediction is that microdirectives will become the dominant form of law, culminating in the death of rules and standards. But even if that full evolution does not happen, microdirectives are certain to become a viable alternative for many laws. This short-run phenomenon is of great importance, as even a limited spread of microdirectives has the potential to change the way laws are structured and thought about generally. This advent of microdirectives may take various paths. In the simplest story, the legislature uses the new technology and communicates the command to the citizen. We use this example to illustrate the concept. More realistically, however, the technology will often be implemented at the administrative level by regulators and enforcement agencies. Lawmakers may still enact standards, but administrative agents will convert them to microdirectives. A third possibility is that private citizens will generate the microdirectives. Citizens using private predictive technology may inform themselves of the most reasonable action in any particular situation. As that private technology gets better, two things will happen. First, failure to use the technology will become a per se violation of a legal standard. And, second, the technology will be able to predict judicial outcomes. Both effects will result in citizens using private technology to derive a simple microdirective for how to comply with the law. For all of these paths, the result is that laws that look like standards to the legislatures will appear as simple and easy-to-follow directives to the regulated individual. This form of law is neither a standard nor a rule. It provides the certainty of a rule and the calibration of a standard, with none of the decision costs associated with either. Moreover, the law, in application, morphs from a standard (for the legislature) to a set of complex rules (within the machine process) to a simple command (for the citizen). We describe the rise of microdirectives as the death of rules and standards. One might alternatively frame the coming change simply as the death of standards. After all, micro-directives are ex ante rules that govern behavior. The driver in our example is told exactly how to behave ex ante. In that framing, technology has reduced the cost of precise ex ante rule making. Rules will no longer be over- and underinclusive. As a result, the rationale for using standards goes away. That is consistent with the conventional law-and-economics definition of a rule as having ex ante content (relative to the regulated actor). But the lawmakers are not enacting rules. The lawmakers need not spend the time to prescribe precise rules. They can enact broad standards and let the machines do the rest. Indeed, from the perspective of the lawmakers, it is the death of rules. The framing is less important than the recognition that microdirectives will change the foundational nature of law. Our analysis is positive rather than normative. One might think of perfect calibration of laws to legislative goals as problematic in a system with multiple branches and checks and balances. Indeed, our analysis implies a reduced role for judges and perhaps the need for institutional reforms to preserve important aspects of our current system. Others may view micro-directives as a threat to privacy and autonomy. The easier it is 100

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

for the government to learn information about the behavior of an individual and use technology to predict outcomes, the more the government can micromanage to achieve desired social results. Finally, some may have concerns about ethics and moral health in a world where many important decisions are automated.5 We do not take a side on these normative questions. The primary contribution of this chapter is to explore the most far-reaching effects of 14 technology on the general structure of law. This contribution builds on and connects with two strands in the law-and-technology literature. The first strand looks at the effects that predictive technology has on the legal services industry.6 The second strand looks at the nature of personalized default rules.7 We suggest, however, that these strands understate the momentous effect that the 15 coming technological revolution will have on law.8 By connecting the growing literature on technology and the law to the literature on rules and standards, we show that the same technology that will bring us automated compliance lawyers and personalized default rules will also bring us the microdirective. And that change in the form of law will have broader consequences than retail personalization of law. Indeed, microdirectives have the potential to bring wholesale institutional changes to our entire system of laws and the way we choose to regulate behavior. In the next part, we spell out how technology will affect the administration of law and 16 the structure of legal content. We outline two types of technology that will lead to a dramatic reduction in the cost of calibrating and communicating ex ante legal directives, thereby eliminating the need to choose between rules and standards. The analysis is presented in four sections. First, we briefly review the distinction between rules and standards and outline the cost choices presented by the dichotomy. Second, we set out our core theory that technology will fundamentally change those cost choices. Third, we provide two examples to demonstrate how predictive and communication technologies will pave the way for microdirectives that capture the benefits of both rules and standards. Fourth, we discuss how the emergence of microdirectives can take place through different branches of lawmaking or can be driven by private actors with access to predictive technology.

5 See, e.g., Shiffrin, Inducing Moral Deliberation: On the Occasional Virtues of Fog¸ 123 Harv. L. Rev. 1214 (2010). 6 Susskind, The End of Lawyers? Rethinking the Nature of Legal Services (2010); Susskind, Tomorrow’s Lawyers: An Introduction To Your Future (2013); Henderson, A Blueprint for Change, 40 Pepp. L. Rev. 461 (2013); Katz, Quantitative Legal Prediction – Or – How I Learned To Stop Worrying and Start Preparing for the Data-Driven Future of the Legal Services Industry, 62 Emory L.J. 909, 914–15 (2013); Kobayashi/Ribstein, Law’s Information Revolution, 53 Ariz. L. Rev. 1169 (2011); Henderson, From Big Law to Lean Law, 38 Int’l Rev. L. & Econ. 5 (2013); Ribstein, The Death of Big Law, 2010 Wisc. L. Rev. 749; Sheppard, Incomplete Innovation and the Premature Disruption of Legal Services, 2015 Mich. St. L. Rev. 1797. 7 See, e.g., Porat/Strahilevitz, supra Part 1.A; Ben-Shahar/Porat, supra Part 1.B; Geis, An Experiment in the Optimal Precision of Contract Default Rules, 80 Tul. L. Rev., 1109 (2006); Sunstein, Deciding by Default, 162 U. Pa. L. Rev. 1 (2013). 8 The closest work to ours is that of McGinnis/Wasick, supra (fn. 1). Though they reach strikingly different conclusions, McGinnis and Wasick begin in the same place as we do, asking how technological advances that reduce information costs will affect the balance of rules and standards. Focusing primarily on legal search technology and the ability to predict judicial outcomes, they predict a world where standards and dynamic rules are favored over simple rules. Building on this analysis, we add in the effects of communication technology and machine learning to show that standards and rules (simple and dynamic) will no longer be viable forms of law.

Casey/Niblett

101

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

II. The emergence of microdirectives and the decline of rules and standards 1. Background: Rules and standards 17

18

19

20

21

Rules are precise and ex ante in nature. Rules indicate to an individual whether certain behavior will violate or comply with the law. When a rule is enacted, effort must be undertaken by lawmakers to give full and precise content to the law before the individuals act. Standards, on the other hand, are imprecise when they are enacted. The exact content of the law comes after an individual acts, as judges and other adjudicators determine whether the individual’s specific behavior in a particular context violates the standard. Generally, lawmakers incur both error costs and decision costs when enacting a law. Error costs arise when a law is over- or underinclusive; the law allows behavior that should be prohibited, or prohibits behavior that should be allowed. Errors can be reduced as lawmakers exert greater effort to get the law right. But this requires information and deliberation. Reducing error costs imposes decision costs on the lawmakers. Additionally, regulated individuals face a cost in figuring out whether their behavior complies with the law. When the application of the law to a particular situation cannot be easily predicted, the individual incurs cost of legal uncertainty. Error, decision, and uncertainty costs arise in different ways for rules and standards. The classic models in the rules-versus-standards literature conclude that, for several reasons, standards tend to perform better when the behavior of the regulated actors is infrequent and heterogeneous.9 First, when behavior of regulated actors is infrequent, standards generate lower decision costs because the content of the law only needs to be decided in the infrequent event that the relevant context actually arises. Rules, on the other hand, require ex ante decisions about all future possible scenarios. Where behavior is infrequent and heterogeneous, lawmakers must make many more decisions if they want to write rules that are as precise in application as a standard that is adjudicated ex post would be. Rules do, however, impose lower decision costs when behavior is frequent and homogeneous. Economies of scale kick in and a law need only be enacted once rather than litigated over and over again.10 Second, error costs for standards are lower when behavior is infrequent and heterogeneous because the adjudicator determining the content of the law ex post has more information than the ex ante lawmaker. The adjudicator has additional context not available to the ex ante lawmaker and has the benefit of hindsight in identifying which factors are relevant.

9 See, e.g., Kaplow/Shavell, Economic Analysis of Law, in: Auerbach/Feldstein (eds), Handbook of Public Economics, Volume 3, 2000, 1665, 1744–45. 10 Strict application of the doctrine of precedent also introduces economies of scale for standards, but it does so in a way that turns the standard into a rule. See, e.g. Holmes, Jr., Justice of the Supreme Judicial Court of Massachusetts, The Path of the Law, Address at the Dedication of the New Hall of the Boston University School of Law (Jan. 8, 1897), in 10 Harv. L. Rev. 457 (1987); Niblett, Case-by-Case Adjudication and the Path of the Law, 42 J. Legal Stud. 303 (2013).

102

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

On the other hand, adjudicator competency and bias complicate this simple model of error costs.11 Ex post adjudication may suffer from hindsight bias12 and from biases based on the personal characteristics of particular individuals.13 Such biases can manifest themselves in arbitrariness, political favoritism, covert influence, inconsistency, and discretionary justice14 even when judges believe they are being unbiased.15 Ex ante lawmakers law makers and regulators may, of course, also be biased.16 But the biases exhibited in ex post adjudication are particularly costly. Hindsight bias is more pervasive and difficult to minimize for ex post adjudication. Additional biases based on personal characteristics of an individual are also more likely for ex post adjudication and may be particularly pernicious and harmful to social objectives. The presence of biased adjudicators, thus, alters the error-cost trade-offs between rules and standards and weakens any claims that standards have lower error costs. A third cost comparison is also relevant when assessing the relative merits of rules and standards: the uncertainty cost imposed on the regulated actor in understanding whether her behavior complies with the law. Uncertainty about the content of a law is greater with standards than with simple rules. When regulated by a simple rule, an individual will more likely know whether her behavior is allowed or prohibited. When regulated by a standard, on the other hand, the individual does not know how any particular judge with wide discretion will apply the standard to the facts. She may not know what behavior a judge will consider reasonable. The choice between using a rule or a standard to achieve a particular policy objective is therefore a question of weighing and trading off these costs. We predict that advances in technology will fundamentally change that trade-off.

22

23

24

25

2. Technology will facilitate the emergence of microdirectives as a new form of law Two types of technology will lead to the death of rules and standards and the rise of 26 microdirectives: predictive technology and communication technology. The first will facilitate lawmakers’ efforts to craft precise ex ante context-specific rules that provide the nuance and specificity traditionally associated with standards. The second will allow for the translation of those nuanced and specific laws into simple directives that are communicated to the regulated actors in a timely manner. a) Predictive Technology. Predictive technology, driven by ever increasing computa- 27 tional capacity, will allow lawmakers to sculpt more perfect ex ante laws.17 Computation 11 See, e.g., Raz, supra (fn. 2); Schauer, Playing by the Rules: A Philosophical Examination of Rule-Based Decision-making in Law and in Life (1991); see also Kaplow, supra (fn. 1), at 609. 12 See generally Kahneman, Thinking, Fast and Slow (2013); Jolls/Sunstein/Thaler, A Behavioral Approach to Law and Economics, 50 Stan. L. Rev. 1471, 1523–27 (1998); Rachlinski, A Positive Psychological Theory of Judging in Hindsight, 65 U. Chi. L. Rev. 571 (1998). 13 See, e.g., Rachlinski/Johnson/Wistrich/Guthrie, Does Unconscious Racial Bias Affect Trial Judges?, 84 Notre Dame L. Rev. 1195 (2009); see also Jolls/Sunstein, The Law of Implicit Bias, 94 Cal. L. Rev. 969 (2006). 14 See, e.g., /Miles/Sunstein, The New Legal Realism, 75 U. Chi. L. Rev. 831 (2008); Niblett, Tracking Inconsistent Judicial Behavior, 34 Int’l Rev. L. & Econ. 9 (2013). 15 See, generally, Kahneman, supra (fn. 12); Jolls/Sunstein, supra (fn. 13), at 970–71; Rachlinski et al., supra (fn. 13), at 1201–04. 16 See, e.g., Bainbridge, Mandatory Disclosure: A Behavioral Analysis, 68 U. Cin. L. Rev. 1023 (2000); Choi/Pritchard, Behavioral Economics and the SEC, 56 Stan. L. Rev. 1 (2003). 17 In a different context, Professor Michael Abramowicz identified the power of predictive decision making to “take (…) advantage of the best of both the world of standards and the world of rules.” Abramowicz, Predictive Decisionmaking, 92 Va. L. Rev. 69, 74 (2006).

Casey/Niblett

103

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

power is growing at exponential rates. The consistent trend of the last fifty years suggests that that power will, by the end of this century, be more than one trillion times greater than what it is today.18 With even a fraction of that processing power, tomorrow’s computers will be able to gather and analyze more facts than any human lawmaker or judge. Lawmakers will be able to direct a machine to analyze a massive amount of data instantly to predict which rules can precisely achieve a policy objective. 28 Relying on the machines to observe and analyze more relevant facts, lawmakers will make better predictions about the impact of a law and will face reduced error costs. Lawmakers will no longer have to think up rules to enact laws. Judges will no longer have to examine citizens’ decisions on a case-by-case basis in order to apply laws. And the laws will be highly calibrated to policy objectives with no chance of judges introducing bias or incompetence. Of course, the calibration need not be perfect, it only needs to be better than the calibration associated with the alternatives of legislated rules and adjudicated standards. 29 As a practical matter, the result will be a new hybrid form of law that is both rule and standard. The lawmaker can set a broad objective, which might look like a standard. But the predictive technology will take the standard and engineer a vast catalog of contextspecific rules for every scenario. But that is only the first half of the story.19 b) Communication Technology. In the second half, the communication technology will simplify that context-specific catalog of rules into clear microdirectives for the regulated individuals. Without that simplification, the catalog of rules would be too complex and pose significant compliance challenges. It would be impossible for people to learn, remember, and process all of the requirements contained in the catalog. But advances in communication technology will produce microdirectives that reduce or eliminate those compliance costs and prevent uncertainty costs that might otherwise arise. 31 The mechanism for translation is straightforward. Communication technology will gather and transmit information about the scenario in which the individual finds herself, identify the applicable rule from the vast catalog, and then translate that into a simple directive that is communicated back to the individual when she needs it. In this way, microdirectives will turn hundreds or thousands of context-specific, machinegenerated rules into simple directives that are easy to understand and follow. The law controlling a particular scenario may take into account hundreds or thousands of factors, but the individual will receive a simple command like a red or green light. When the output from the predictive technology is translated into a microdirective, citizens will be able to act as if they are taking into account more relevant factors than are humanly possible.

30

32

c) Summary. To summarize, these technologies will combine to do the following. First, they will take a standard-like policy objective, analyze its application in all possible 18 This estimate is based on a trend known as Moore’s Law. See generally Moore, Cramming More Components onto Integrated Circuits, Electronics, 19 April 1965; Lundstrom, Moore’s Law Forever?, 299 Science 210, 210 (2003). 19 The discussion of predictive technology here and throughout this chapter assumes a consequentialist approach to law. For a consequentialist, the content of the law is driven by a prediction of the outcome of behavior. For nonconsequentialist theories, the use of the technology is slightly different. But the trend toward microdirectives will likely be the same. For example, imagine that a lawmaker wants to prohibit certain behavior she deems immoral regardless of the consequences of that behavior. She does not want to list out all permutations of immorality, so a rule will not work. Instead, she can start with a standard – immoral activity is prohibited – and then identify samples of immoral behavior to feed into a machine. The machine can then use analytic and pattern recognition technology to determine whether other new scenarios would be deemed immoral by the lawmaker.

104

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

contexts, and create a vast catalog of legal rules – each of which is tailored to best achieve the objective in a specific scenario. Second, when a regulated actor is in any actual scenario, the technologies will search the vast catalog and identify the specific rules that are applicable. Third, they will translate those rules into a simple microdirective on how the regulated actor can comply with the law. Fourth, they will communicate that microdirective to the regulated actor in a timely and efficient manner.

3. Examples To demonstrate the point, we present two stylized examples.

33

a) Predictive technology in medical diagnosis. In this subsection, we provide an example that demonstrates how improved predictive technology – technology that allows lawmakers to better predict the outcomes of actions – will foster microdirectives. Suppose you are a legislator. You are charged with determining when doctors should be liable for performing a risky surgery on a patient. How can you best regulate doctors’ behavior? How can you best draft a statute that will help doctors understand when their behavior complies with or violates the law? How many of the specific details should you include in the statute? How many of these details can be postponed until we have more information about how doctors behave in each case? One option is to provide doctors with a clear and simple bright-line rule that dictates the circumstances under which surgery should or should not be conducted. This simple rule provides great certainty to the doctors and is easily enforced; either a doctor complied with the rule or she didn’t. A simple, precise ex ante rule would be your preferred method if similar patients frequently present with the same symptoms.20 Under these circumstances, a rule would be preferred because the content of the law can be established just once, and there are enormous benefits from economies of scale. But a doctor’s decision to operate on a patient frequently turns on many different factors. A “one size fits all” rule here would likely not be optimal. Any simple bright-line rule you enact will likely be overinclusive and underinclusive compared to an optimal decision rule. There will be some patients who receive surgery who do not need it (type I errors); there will also be other patients who do not receive surgery who do need it (type II errors). To overcome these errors, you may try to write a more complex rule. To formulate this rule, you may try to think up many different scenarios, where you imagine different types of patients presenting with various symptoms. A complex rule is preferred if the cost of thinking and writing the rules is very low and the cost of doctors understanding and being able to comply with such a complex rule is also low. But it is often very costly for legislators to think up and write down all contingencies. Further, the more complex the rule you write, the more difficult it becomes for a doctor to follow.21 Rather than implement a rule, another option you have is to enact a standard and evaluate the conduct of a doctor after the decision to operate (or not operate) has been made. That is, the decision to hold a doctor liable would be made once all the circumstances of the particular case are known.22 For example, the legal standard might 20

See Kaplow, supra (fn. 1), at 573–77. Kaplow, A Model of the Optimal Complexity of Legal Rules, 11 J.L. Econ. & Org. 150, 151 (1995); see also Diver, supra (fn. 3), at 73–74. 22 As Henry Hart and Albert Sacks note: “The wise draftsman (…) asks himself, how many of the details of this settlement ought to be postponed to another day, when the decisions can be more wisely and efficiently and perhaps more readily made?” Hart/Sacks, The Legal Process: Basic Problems in the Making and Application of Law, 157 (1958). 21

Casey/Niblett

105

34

35

36

37

38

39

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:39 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

40

41

42

43

44

45

stipulate that all doctors must take “reasonable care” in determining whether to operate on patients. This provides doctors with greater flexibility to decide whether or not the patient needs surgery. But it also provides an ex post adjudicator with the flexibility and discretion to determine what is meant by “reasonable.” If a patient suffers harm as a result of a doctor’s decision, then a judge can look at all the facts as they actually occurred and make an informed decision as to whether the doctor took reasonable care. A standard would be better than a rule if patients and symptoms are heterogeneous and the likelihood of two patients with the same background and symptoms is very low. There are, of course, costs associated with implementing and enforcing a standard. First, the cost of deciding each case is not zero. There are decision costs of learning the best course of action the doctor should have taken in the circumstances. Second, a judge may apply the standard incorrectly, either due to error or to bias. Third – and importantly – unlike a clear rule, a vague standard creates a great deal of uncertainty for the doctor. A doctor may not know how a judge will decide any given case; further, different judges may decide inconsistently. If doctors are risk averse, a vague law can chill socially desirable behavior,23 and the uncertainty may generate considerable expense in the form of compliance costs. But in our hypothetical situation, let’s suppose that a standard is optimal. Let’s assume that the question of surgery rarely arises and that patients are highly diverse, both in terms of health backgrounds and in terms of the symptoms they present. Formulating detailed rules that cover all those situations and being able to communicate these complex rules to doctors would be difficult, and a simple rule would create high error costs. Case-by-case adjudication is not costless but it is preferred in our example because the infrequent cost of determining the content of the law ex post is lower than the costs of trying to specify the law up front in all potential situations, many of which will never arise. Now let’s examine how technology will eliminate this trade-off between rules and standards. Suppose that you learn of the existence of a diagnostic machine that is designed to predict when surgery is required. The machine takes into account relevant facts about the patient24 – her history, the symptoms, and other relevant information – to provide a best guess as to whether the patient requires surgery. You, the legislator, have access to this machine. How does this predictive machine affect your decision to enact a rule or a standard? The answer turns on two factors. First, how good is the machine at accurately predicting outcomes? If the predictive technology is very powerful and the machine is able to provide precise and accurate information, then this points in favor of using the machine to create a rule, rather than relying on a judge to adjudicate a standard. Second, can this information be easily communicated to a doctor? That is, can lawmakers provide the doctor with timely notice of what behavior will comply with or violate the law? Consider two scenarios: Scenario 1: A terrible predictor In scenario 1, the machine is very poor at predicting when a patient requires surgery. The machine essentially randomizes patients for surgery. The machine generates both type I and type II errors. One might think of the technology as a simple coin toss: heads for surgery, tails for no surgery. 23

See Craswell/Calfee, supra (fn. 4), at 298–99. In practice, the machine would actually take into account relevant information about the doctor as well, such as his track record with surgeries of the relevant type. 24

106

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

Scenario 2: A perfect predictor In scenario 2, the machine can predict with 100 % accuracy whether a patient requires surgery or not. The machine instantly examines the patient’s history and symptoms, analyzes millions of prior cases, and reads all articles in medical journals. It then makes a perfect prediction. It is better than any human at determining whether it is optimal to have surgery. There are no type I errors: patients who do not need surgery are not designated for surgery. There are no type II errors: patients who need surgery are designated for surgery. Under scenario 1, the technology should have no effect on your decision as a regulator to implement a rule or a standard. You should implement a standard and determine liability on a case-by-case basis, learning more about doctors’ behavior over time. Under scenario 2, however, the optimal form of the law will be different. The machine’s predictions provide the exact content of the law. The machine provides microdirectives for each and every scenario. The over- and underinclusivity associated with simple rules have disappeared. There are no errors (type I or type II) in this scenario. And the costs incurred in thinking up and formulating such a complex rule have already been incurred in the development of this machine. The justification for relying on ex post adjudication of standards – reducing the error costs of rules – is gone. Further, we have an added benefit of eliminating uncertainty for the doctors. If they follow the directive of the machine, they know they will not be held liable. The emergence of microdirectives and the death of rules and standards as we know them do not rely on perfect predictive technology. Rather, as the predictive technology gets better and better, we move away from the world of scenario 1 and towards the world of scenario 2. There will come a point where the technology is good enough that the costs of using a microdirective are sufficiently low so that there is no longer any need to use traditional rules or standards. A caveat is necessary. This tipping point can only be realized if the rules generated by the machine can be easily communicated to doctors. That is, the legislator has to be able to provide the doctor with a quick and simple answer to the question of whether the patient requires surgery. Doctors would find it difficult to follow complex, computer-derived rules. Regulated actors have neither the desire nor the time to thumb through thousands of pages of legislation and understand complex algorithms. Rather, lawmakers need some form of technology to allow a doctor to easily input all the relevant facts about a patient and receive an instant output that dictates whether or not the patient requires surgery. One might imagine a web-based program or mobile app, where the doctor can quickly and easily enter all relevant facts, submit the information, and instantly receive a binding ex ante opinion. Such technology is emerging and will be able to transform the complex rules generated by machine prediction into a simple directive that the doctor can follow. The costs to the doctor in understanding the complex rule will be dramatically reduced as this technology improves. Even though the rule will be highly complex and based on a sophisticated algorithm, from the perspective of the doctor, the rule will be simple: operate or do not operate.25 We explore communication technology further in our second example. 25 It may seem odd at first that lawmakers are in the business of diagnostic technology. But this is no different from what judges do in medical litigation. Judges hear expert testimony and decide ex post whether certain behavior was reasonable. In our example, lawmakers just use expert technology to do that ex ante. It is true that the role of the doctor has changed – diagnostic judgment is less important – but

Casey/Niblett

107

46

47

48

49

50

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law 51

52

53

54

55

56

57

b) Communication technology in traffic laws. In this subsection, we highlight the way improved communication technology will facilitate microdirectives. Machines can almost instantaneously gather information, process it, and produce a useable output that directs how individuals should behave. Traffic lights provide an example of this type of technology. They communicate the content of a law to drivers at little cost and with great effect. This notice technology – combined with technology for predicting traffic patterns and driver behavior – creates an environment where lawmakers are able to replace vague standards and simplistic rules with crisp and increasingly complex microdirectives. Electric traffic lights communicate to drivers precisely when they are required to stop and when they may proceed. Traffic lights appear to generate very simple rules: if the light is red, you must stop; if the light is green, you may go. But these rules are simple only from the perspective of the driver. From the perspective of the lawmakers, the underlying rules are complex. The simplest underlying rule may dictate that cars must stop during regular, alternating time intervals. In more complex examples, the time intervals can vary by intersection, direction of traffic, or time of day. If promulgated without traffic lights, these rules would be far too complex. Drivers would have to consult tables that matched intersections, times, and directions with prescribed intervals of stopping. They would also have to consult precise clocks to determine when the intervals start and end. The traffic light translates complexities into a simple command. From the driver’s point of view, the lights provide a directive that is easily understood. And the lawmaker’s cost of giving notice is low. Electric traffic lights take advantage of significant economies of scale that enable lawmakers to make complex rules, translate them into simple directives, and deliver notice of the required behavior to many drivers. Moreover, while the command of the traffic light remains simple, the substance of the underlying rules is becoming more complex. Predictive analysis facilitates this process. Stopping at a red light when an intersection is deserted is wasteful and costly. That rule is overinclusive. It would be better if the directive to the driver could change depending on the circumstances (as it would with a standard). To address this, traffic lights in some jurisdictions already contain sensors that detect and predictively analyze traffic flow and adjust the timing of red and green lights accordingly. Some traffic lights contain detectors allowing emergency service vehicles to “preempt” the signal and expedite their journey. In the near future, these systems will take into account more variables, such as the number of cars, speed of travel, or type of intersection. They might even take into account personal characteristics of a vehicle’s driver or passengers.26 In the not-so-distant future, a traffic-light system may know that a passenger in a regular vehicle requires medical attention and give the rushing driver a series of green lights all the way to the hospital. The progress of traffic lights shows how lawmakers can define optimal policy outcomes (e.g., travel times and accident rates) and machines can generate a catalog of rules and exceptions to achieve those outcomes. And yet – even while the lawmakers enact a standard and the machines generate an increasingly complex catalog of rules underpinning the operation of traffic lights – from the perspective of the driver, the law will remain constant and straightforward: a simple stop-go directive. that is the inevitable result of advances in diagnostic technology. Our point is simply that in the hands of lawmakers the technology also changes the role of law. When the technology is only available to the private actors – the doctors in this example – then the evolution of rules into standards takes a slightly different path. 26 Cf Porat/Strahilevitz, supra (fn. 7).

108

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

This phenomenon is not limited to traffic law. The forces at work here are 58 ubiquitous. The invention and mass adoption of Internet technology has facilitated instantaneous and cheap communication between individuals across all domains. It also, importantly, allows for immediate communication between lawmakers and individuals.

4. The different channels leading to the death of rules and standards We have, until now, spoken generally of lawmaking by a legislature. That is by no 59 means the only avenue. Microdirectives can emerge through two other channels: (1) nonlegislative (regulatory or judicial) law making; and (2) private use of technology by regulated actors.27 We discuss each in turn. a) The production of microdirectives by nonlegislative lawmakers. Legislatures are 60 not the only lawmakers with access to technology. In many cases, the lawmaking power is entrusted to a regulator or enforcement agent. In other cases, judges make law. Those entities can also use technology to create and communicate micro-directives to regulated actors. aa) Regulatory microdirectives. It is likely to be more politically feasible for regulators to develop microdirectives than legislators. The legislative path to enacting a computer algorithm is complicated. Pork barrel and horse-trading amendments to an algorithm do not make for successful programming. On the other hand, a regulator tasked with enforcing some legislated standard might easily adopt an algorithm-driven system of microdirectives.28 The pressures on a budget-constrained regulatory body will push the agency toward adopting technology. Likewise, trends towards cost-benefit analysis and requirements that regulations be shown to be cost justified29 are likely to accelerate agency adoption. Predictive technology facilitates such cost-benefit analysis, reduces uncertainty costs to the regulated actors, and cuts down on ex post adjudication costs. Congress could enact a standard and direct that these standards be administered by an algorithm-based system of microdirectives overseen by regulators or the regulators could themselves decide to implement the standard in that manner. Advance tax rulings provide an example of an area for regulators to use microdirectives.30 As it currently stands, taxpayers may seek clarification of vague standards in the law by asking the tax authority to examine their tax arrangements and determine whether they comply with the code.31 A taxpayer may ask the tax authority to give a ruling on a matter that takes into account a number of factors such as: Am I a resident of the United States for tax purposes? Or, are my workers independent contractors or are they employees?32 These advance tax rulings bind the tax authority to the tax arrangements set out in the ruling, but only for the one specific taxpayer.33 Essentially the taxpayer is asking the 27 Here, we discuss different channels that technology will affect the law. In other work, we have discussed the incremental nature of these changes. See Casey/Niblett, Self-Driving Laws, 66 U. Toronto L. J. 429. 28 McGinnis/Wasick, supra (fn. 1), at 1042. 29 See generally, Sunstein, The Cost-Benefit State: The Future of Regulatory Protection, 2002. 30 See generally Romano, Advance Tax Ruling and Principles of Law, 2002; Givati, Resolving Legal Uncertainty: The Unfulfilled Promise of Advance Tax Rulings, 29 Va. Tax. Rev. 137, 144–47 (2009). 31 26 C.F.R. § 601.201 (2002); Givati, supra (fn. 30), at 149–52. 32 Romano, supra (fn. 30), at 80. 33 In the United States, these rulings (“private letter rulings”) are “binding on the IRS if the taxpayer fully and accurately described the proposed transaction in the request and carries out the transaction as

Casey/Niblett

109

61

62

63

64

65

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

tax authority to turn an ex post standard into a specific rule that applies solely to her circumstances. These advance rulings have a variety of benefits. Most prominently, they provide greater legal certainty to the taxpayer.34 They eliminate the uncertainty costs of the standard.35 But such rulings can be costly to generate.36 The tax authority is essentially engaged in personalized rule making. It is incurring high ex ante decision costs by enacting a rule that applies to just one taxpayer.37 66 Now imagine the tax authority could create a system where a taxpayer simply turns to a machine to answer her tax questions. She could, for example, turn to an agency website or a mobile app. She could ask the machine whether her tax arrangements will expose her to liability and the machine could quickly read the entire tax code, all relevant cases, all associated regulations, and all relevant advisory opinions. The machine could immediately provide an answer to the taxpayer’s question.38 67 The tax authority, thus, could use this artificially intelligent machine to provide advance tax rulings. Depending on the underlying objective of the legislature, the tax authority could use the machine to identify optimal rules that allow it to generate more revenue with greater efficiency and fewer distortions on market behavior. It could use this technology very broadly to choose very specific rules that are highly calibrated to legislative objectives without introducing compliance costs that would otherwise be associated with such complexity. 68 If regulators adopt these technologies, the answers provided by the tax authority would essentially become the red or green lights of tax law. Even though the underlying tax laws would be very complex, the directives provided to an individual would be simple. Any enforcement agent could adopt technology of this kind. As predictive technology makes it easier to automate such regulatory advance rulings and ensure their accuracy, they will become a common mechanism for the adoption of machinegenerated microdirectives.39 bb) Judicial microdirectives. Aside from the legislator and the regulator, there is, of course, a third potential rule maker: the judge. But, as they currently function, judges do not quite fit into this model of law making. To be sure, judges could use artificial intelligence and big data to apply standards or complex rules.40 But judges are not – at least in a formal sense – regularly in the business of providing ex ante notice of the outcomes of hypothetical scenarios. 70 For better or worse, advisory opinions are frowned upon by the American judicial system. Judges might use the predictive technology to refine the law ex post. But without

69

described.” Understanding IRS Guidance – A Brief Primer, IRS (last updated 6 July 2016), https://www. irs.gov/uac/Understanding-IRS-Guidance-A-Brief-Primer; see also 26 C.F.R. § 601.201(a)(1)–(2), (l); Givati, supra (fn. 30), at 149–50. 34 Romano, supra (fn. 30), at 77–78. 35 Givati, supra (fn. 30), at 147. 36 Romano, supra (fn. 30), at 277–80. 37 See Givati, supra (fn. 30), at 149. 38 For more on this process, see Alarie, The Path of the Law: Toward Legal Singularity, 66 U. Toronto L. J. 443 (2016); Alarie/Niblett/Yoon, Using Machine Learning to Predict Outcomes in Tax Law, 58 Can. Bus. L. J. 231 (2016). 39 There will be some areas in law where the provision of advance directives is problematic. Tax provides a salient example. For some things, the lawmaker and the individual have aligned incentives. The individual wants to comply with the lawmaker’s policy objectives and certainty makes compliance more likely. But for other things, the individual wants formal compliance with law but would prefer to avoid the policy objective. In other words, the individual is looking for a loophole. If the law provides a clear rule and the regulated individual would prefer to circumvent that rule, then certainty provides a road map for avoidance. See Weisbach, Formalism in the Tax Law, 66 U. Chi. L. Rev. 860 (1999). 40 Cf Porat/Strahilevitz, supra (fn. 7).

110

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

notice to the regulated actors, those specific rulings impose some of the same costs as standards. For example, if judges announce that all negligence cases will be decided using a computer algorithm, a regulated actor without access to the algorithm would still be faced with nothing more than a standard that imposes uncertainty about how the judges will apply that standard. It would make little difference to the individual that the actual judge happens to be a computer. Things change if the regulated actors have access to the algorithm that judges will use. 71 In that world, the regulated actors can predict the outcome with precision. If judges commit to using a certain technology that is available to the public, that would be equivalent to providing advance rulings.41 This would essentially shift the judge’s role to that of ex ante regulators. While not implausible, we think the avenues of legislative and regulatory rulemaking will be more pervasive. There is another way that judges could be involved in the promulgation of micro- 72 directives. Just as legislatures could set a broad policy objective and delegate the rule making to an agency, so too could the courts. In deciding cases, courts can announce a standard that blesses any rule that results from a process aimed at the correct policy objective and that takes into account the relevant factors. The agency could then create an algorithm that does exactly that. This “second-order regulation” by the court would send a message to the agencies on how to design the algorithm to ensure compliance. Here again it would be the agencies and enforcers who have the ultimate responsibility for implementing the machine algorithm to promulgate microdirectives. b) An alternative path: Private use of technology by regulated actors. Predictive 73 technology will be available to private actors. And, in some cases, private actors may have more advanced proprietary technology than legislatures, regulators, or courts. Private use of predictive technology will lead to the emergence of microdirectives. There are two main ways this can occur. The first path is through the interplay of reasonableness, industry standards, and 74 technology. In our medical example above, imagine that the machine that predicts medical outcomes is available not to lawmakers but only directly to doctors. As the technology becomes more accurate we can expect more and more doctors to use it. At some point, it is likely that courts will begin to deem it per se unreasonable to not use such advanced technology. Imagine an orthopedic practice today that did not use an xray machine42 or a colorectal specialist who refused to perform colonoscopies in diagnosing colon cancer.43 As technology becomes more accurate and widespread, the likelihood that courts will base a reasonableness standard on the use of that technology increases. The proliferation of these technologies across industries will cause behavior that complies with standards to function exactly as if it were complying with a microdirective promulgated by the predictive technology. The second path is a softer version of our main thesis. This path does not require 75 lawmakers to use technology. Individuals can use predictive technology to provide predictions on how judges will apply a standard. In this way, technology improves on 41 In a slightly different but related context, one commentator has suggested that judges could bind themselves to textualist interpretations of statutes by using computers to derive the meaning of text. Cooper, Judges in Jeopardy!: Could IBM’s Watson Beat Courts at Their Own Game?, 121 Yale L.J. Online 87 (2011), http://yalelawjournal.org/forum/judges-in-jeopardy-could-ibms-watson-beat-courts-at-theirown-game. 42 Doctors have frequently been held negligent for failing to order x-rays. See, e.g., Rudick v. Prineville Mem’l Hosp., 319 F.2d 764 (9th Cir. 1963); Webb v. Lungstrum, 575 P.2d 22 (Kan. 1978); Betenbaugh v. Princeton Hosp., 235 A.2d 889 (N.J. 1967). 43 Doctors who failed to order a colonoscopy have been held negligent. See, e.g., Morse v. Davis, 965 N. E.2d 148 (Ind. Ct. App. 2012).

Casey/Niblett

111

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:40 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 1. The Concept of Personalized Law

the role of lawyers as compliance advisors.44 When lawyers provide compliance advice, they are, in part, predicting how ex post adjudicators will apply a standard.45 As computers can gather and analyze more and more prior cases, they will outperform lawyers at this task. On first blush, this advance would appear to reduce the compliance cost of standards. But it does so in a way that effectively turns the standard into a microdirective, as it reduces the costs of legal uncertainty because it tells the individual exactly how to behave. As Oliver Wendell Holmes noted, a prediction of a judicial outcome is the law.46 76 Advances in big data and artificial intelligence will spawn intelligent machines that can predict legal outcomes with great accuracy.47 In our traffic example, imagine that traffic is regulated only with yield signs that impose a reasonableness standard. But in this world, consumer technology has advanced to a stage where it can predict when a court will deem yielding to be required under the standard. This private technology provides a mechanism for informing the driver when she must stop under the law. The technology gathers the relevant facts, applies the standard to those facts as a judge would, and provides predictive analysis. 77 Even though we have standards and private technology, the resulting behavior looks as if we had public traffic lights with underlying complex rules. And compliance is as simple for the driver as it would be with a microdirective. The driver simply gets a message saying stop. She does not have to even take mental note of the underlying facts. As technology makes ex post adjudication more predictable, citizens treat a prediction as a rule. They receive directives ex ante and have little uncertainty about how the law requires them to behave. 78 This may lead lawmakers to simply enact those predictions as law. It is possible, though, that lawmakers may deem fully predictable ex post adjudication to be the satisfactory equivalent of a microdirectives and not take the final step to formalize the microdirectives into law. But from the individual’s perspective the transformation will be already complete. Drivers will know to stop when the technology in their car gives a signal – the equivalent of a red light.

III. Conclusion As machines become increasingly intelligent, and continue to outperform human judgment, the influence of artificial intelligence will spread far and wide. The technologies we have discussed are already being used by doctors to detect cancers, by consumers to optimize their search for products, and by financial advisors to provide advice. 80 The legal system will not be immune from this trend. We have suggested throughout this chapter that this technological revolution will dramatically alter the foundational structure of law as we know it. Predictive technology will generate greater ex ante 79

44 See generally DeStefano, Taking the Business Out of Work Product, 79 Fordham L. Rev. 1869 (2011); Parker, Lawyer Deregulation via Business Deregulation: Compliance Professionalism and Legal Professionalism, 6 Int’l J. Legal Prof. 175 (1999); see also Millman/Rubenfeld, Compliance Officer: Dream Career?, Wall St. J. (15 January 2014, 8:13 PM), http://www.wsj.com/articles/ SB10001424052702303330204579250722114538750. 45 See Kaplow/Shavell, Private Versus Socially Optimal Provision of Ex Ante Legal Advice, 8 J.L., Econ. & Org. 306 (1992). See also LoPucki/Weyrauch, A Theory of Legal Strategy, 49 Duke L.J. 1405 (2000). 46 Holmes, supra (fn. 17), at 461 (“The prophecies of what the courts will do in fact (…) are what I mean by the law.”). See also Abramowicz, supra (fn. 17). 47 See Katz, supra (fn. 11); see also Porat/Strahilevitz, supra (fn. 7), at .

112

Casey/Niblett

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:41 3B2 9.1.580; Page size: 160.00mm  240.00mm

C. The Death of Rules and Standards

information that can be used by lawmakers to write highly specific, complex laws. And individuals will receive notice of these complex laws in a simple form thanks to technological advances in communication. This will be the death of rules and standards and the rise of microdirectives. These developments will have profound implications for the role of judges, legisla- 81 tors, regulators, lawyers, and individuals in the legal system. But beyond that, we will have to change the way we think and talk about law. Technological changes that vastly improve ex ante information will also breathe new life into old law-and-economics models that began with an assumption that lawmakers and citizens have full information. Friction in these models caused by imperfect and asymmetric information has provided a fertile source of material for critics, both inside and outside the field of law and economics. But these models will be given renewed importance. Similarly, the public choice literature will have an increased emphasis on how legislators choose objectives, rather than how they implement laws, while academic interest in subjects such as judicial behavior will dissipate. All of this is to say that legal institutions of all types will change radically. We are 82 witnessing an information revolution. And, like other technological revolutions, it will precede a legal revolution. The industrial revolution, for example, saw human labor replaced by machine labor and the cost of transportation fell markedly with inventions such as the steam engine. It greatly reduced transaction costs and had widespread impact on all spheres of law including contract law, property law, employment law, criminal law, and tort law. The information revolution has already resulted in dramatic changes in the world of 83 commerce. For example, companies such as YouTube, Uber, and Airbnb have disrupted and uprooted heavily regulated and stable industries. The coming technological revolution will lead to similar disruption of the legal services industry, but the effect on law will be much deeper and far wider. It will affect the very structure of legal commands and the way we, as a society, choose to govern the behavior of citizens.

Casey/Niblett

113

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/03_Casey and Niblett, DORS – Final (January 2019).3d from 30.10.2020 11:34:41 3B2 9.1.580; Page size: 160.00mm  240.00mm

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:08 3B2 9.1.580; Page size: 160.00mm  240.00mm

PART 2 CRITIQUE AND THEORETICAL PERSPECTIVES D. The Law between Generality and Particularity. Chances and Limits of Personalized Law* I. Introduction Over the recent years, a legal debate has emerged around “granularization of legal 1 norms” or – as it is sometimes put differently – “personalized law”.1 This debate is inspired by two parallel technical developments, both related to the emergence of highcapacity algorithms: (i) The significant increase in data collection and data availability, generally discussed under the term “Big Data” and (ii) the omnipresence of modern communication systems which distribute information to each single person and thus link this person to the rest of society, which can be illustrated by the term “Big Link”.2 While the protagonists of granularization do not fully agree with regard to how those tools could change and reshape the legal landscape, the general idea can be summed up as follows: In a first step, the huge set of data available about each individual could be analyzed with the help of algorithms in order to discover, through proxy-based correlations, behavioral patterns and preferences, and to take those individual character traits into account for the elaboration of personalized legal commands. In a second step, the Big Link, i.e. elaborated communication devices, would make it possible to communicate these commands to the respective individual or to those who have to take an adjudicatory decision (judges or agency officers).3 As a consequence, legal norms would tend to become granular, the law as such would turn to be greatly (or even infinitely) personalized. From that perspective, the relationship between law and the new technical 2 phenomenon of Big Data and Big Link would be remarkable: The basic concepts of the law tend to remain stable for quite a long time even if they are facing technical inventions. In fact, conventional legal principles are usually quite resilient in handling, coordinating and regulating any kinds of changes in society. In general, this capacity also applies to algorithms, Big Data and Big Link.4 But as far as personalization of legal commands is the issue, the current debate is also about how a new * The following article has been developed from the presentation given by Hans Christoph Grigoleit on 3 March 2017 at the Villa Vigoni, Menaggio, Italy. We have greatly enriched the contents and arguments and selectively added references. 1 Most commonly, the term “personalization” is used (e.g., Ben-Shahar/Porat, supra Part 1.B; Porat/ Strahilevitz, supra Part 1.A. In order to emphasize that the described phenomenon only concerns a move towards particularity, or hyper-tailoring of legal commands, and is not necessarily linked to the humanistic, dignitarian ideal of individuality or personality, others prefer to use the notion “granularization” (e.g. Busch, Granular Society – Granular Law?, ). We will mostly stick to the common use of “personalization”, because this notion is content-richer insofar as it describes in addition to the granularization phenomenon its dimension: the person. We emphasize, however, that personalization in that sense is purely descriptive, and does not convey any dimension of dignitarian self-fulfillment. 2 With regard to the emergence of those two technologies cf Casey/Niblett, supra Part 1.C. 3 For a more detailed model cf Porat/Strahilevitz, supra Part 1.A. 4 Cf the first distinction under II.2.a).

Grigoleit/Bender

115

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

technological invention could influence the very structure of the law and its doctrines. From a historical perspective, few technical inventions had the potential to challenge conventional legal structure in that way – one might think of the development of writing thousands of years ago or, a couple of hundreds of years ago, the emergence of print media. This extraordinary perspective makes it worthwhile to rethink law in between generality and particularity and to analyze limits as well as potentials of personalization of the law. 3 While these prospects seem huge, we would nevertheless suggest to approach the topic with some humility: At this juncture, it is not at all clearly foreseeable what the exact technical implications will be that may be used to overturn the structure of the law. Furthermore, as of today, the potential of personalization may have been sketched out in rough lines, but has not been described in depth and specificity. Against this background and with some anticipation of the following considerations, it is somewhat premature and “greatly exaggerated”5 to invoke “the death of rules and standards”6 – as (widespread) death invocations usually are if they relate to fundamental categories of the law. 4 Taking on the great potential of personalization and yet following the path of humility, we will, in the first part of our work, venture to structure the personalization discourse by introducing some fundamental dichotomies which may lead to useful notional distinctions (II.). Indeed, our limited knowledge about what will be technically possible and the unprecedented nature of many problems creates a discourse in which notions often remain foggy. Based on the findings of the notional part, the article will then concentrate on what we would like to call the “evolutionary perspective”. Here, we will explore the potential of personalization with regard to specific issues, elements and structures of “the law, as we know it”, such as torts, default rules, and disclosure (III.). The evolutionary perspective will suggest that the principle of equality and the model function of the law provide important insights into where the law should halt in its oscillation between generality and particularity. As the discourse on personalization reaches beyond the traditional elements of the law, we will then, in the final part of this paper, turn to what we would like to call the “revolutionary perspective”: the potential of changes in the very structure of the law. (IV.).

II. Distinctions and notional specifications 5

In our view, the discourse on personalization of the law – while being enormously important – is in a very early and therefore immature stage. The statements made do not always meet the degree of precision that is required to fully appreciate them as legal arguments or doctrines. Therefore, we would like to suggest some distinctions (or: dichotomic notions) which might be quite helpful to structure the issues and arguments in the personalization discourse. When technology meets law, distinctions can be related either to the technology innovations (1.) or to the law (2.).

1. Technology-related specifications 6

Approaching the discourse around personalization, a first helpful question may be whether the intended personalization depends on new technologies – such as Big Data 5 Mark Twain, letter from 31 May 1897 (“The report of my death was an exaggeration.”), becoming “the report of my death has been grossly exaggerated” in Paine, Mark Twain Biography. The Personal and Literary Life of Samuel Langhorne Clemens, 1910, ch. 197. 6 Casey/Niblett, supra Part 1.C.

116

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

and powerful algorithms – or whether this personalization could already be implemented in conventional ways (a). If the latter is the case, the issues of personalization can be discussed without problems of feasibility or data protection specifically related to the use of new technologies. If the use of new technologies is necessary, then a second question arises: Are we discussing problems related to the rule-elaboration, which would require the collection of a huge amount of data (Big Data problems), or are we discussing problems with regard to the rule-communication-process, which would require the single individual to be steadily connected to a communication network (Big Link problems) (b). a) Technological vs. conventional. It is first worthwhile to distinguish tendencies of 7 personalization that are linked to new technological innovations (such as algorithmbased data analysis)7 from arising tendencies that are unrelated to the technological development. One may note that “non-technological” personalization is one of the most obvious international trends in the development of private law, prominently reflected in the emergence of specific fields of law, such as, for instance, commercial law, labor law, and consumer protection law. New forms of contracts developed for very specific situations with a specific set of default rules, such as – from the German point of view – the Consumer Construction Contract (Verbraucherbauvertrag), a newly introduced sub-form of the general service contract which is now codified in §§ 650i et seq. in the German Civil Code (BGB). Furthermore, there is a discussion on the relevance of individual character traits with respect to the standard of reasonable care – a classic issue of negligence law in torts as well as in criminal law and an issue that is much older than algorithm-centered approaches.8 Even from a historical perspective, one may turn to the General Land Law for the Prussian States from 1794, with its more than 19,000 provisions all but general, which can be qualified as a significant, yet basically failed move towards particularity. In the context of these “conventional” or “historical” tendencies, it may be useful to 8 explore how the particular effect could be defined that powerful algorithms, Big Data and Big Link may contribute to the development of the law. In general, it is said that new technologies help to overcome transaction and compliance costs linked to an expensive information gathering and also error costs with regard to rule-production (ex ante) and adjudication (ex post).9 In fact, transaction and error costs or just the limited capacity of timely reacting may have prevented the legislator from enacting a set of special default rules for each single sales contract of a particular washing machine. However, the “conventional legislator” does in certain areas move towards persona- 9 lization while, on the whole, he is not exhausting even the simple potentials of personalization. In German contract law, for instance, a particular law of “service contracts for buildings” has been codified in §§ 650a et seq. of the German Civil Code (BGB), whereas the legislator did not deem necessary to provide for a specific subset of sales law for buildings. What were the reasons for this differentiation? Supposedly, the legislator is reacting on different needs of the business practice. Whereas this practice may deal perfectly well with general norms in one area, it became apparent that in other areas there might be a real or perceived need of personalization. From the legislator’s perspective, the issue is whether the underlying purpose-analysis can be accounted for by an optimized general rule, or only by a more personalized, more complex provi7

Which is not the only form of technical personalization, cf Busch, infra Part 4.N. With regard to this debate in the German and US context cf infra (fn. 28 et seq.). 9 Cf, e.g., Porat/Strahilevitz, supra Part 1.A. 8

Grigoleit/Bender

117

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

sion.10 It is not obvious yet in which way new technologies would change the needs for particularity or the sufficiency of generality in specific fields, depending on the needs of business. 10 The dichotomy between technology-driven personalization on the one hand and conventional personalization on the other may also prove to be noteworthy to attach specific problems of personalization to their source. For instance, there are technologydependent concerns of feasibility and constitutional issues. In particular, they might reflect the delegation of power issue which is linked to the use of algorithms, away from the political and legal arena, and towards experts such as computer scientists, or even non-human mechanisms which we – as humans – only partly understand.11 Or they might rather deal with privacy and data protection law. But then, there are also problems of personalization that are not rooted in the technical design of personalized law. This is important to note, because those “non-technological” concerns would not lose their strength even if all the technology-related concerns might be designed away. In that regard, we may consider issues of discrimination and equality before the law, or the loss of a public forum in which general rules can be discussed productively given their particular scope of applicability. 11

b) Rule-elaboration (Big Data) vs. rule-communication (Big Link). Once an issue of personalization can be identified as technology-related, it is useful to further specify its source within the field of technology in order to sharpen the practical, ethical, or constitutional issues of personalization. One further distinction is that between the phenomenon of algorithm-based rule-elaboration on the one hand and the algorithmbased rule-communication on the other. Both aspects of personalization are closely linked and intertwined, since personalized law cannot necessarily be perceived as a whole on a human scale, may change rapidly according to changes of the circumstances and therefore needs special tools of promulgation and communication. But, generally speaking, it is useful to be aware of whether an issue basically relates to the practical implication of rule-communication or to the technological material background of rule elaboration such as the use of Big Data. For instance, issues relating to separation of powers, equality or data protection, are primarily related to the level of rule-elaboration. Issues such as whether the individual can be required to be available for personalized commands at any time, or whether this conflicts with the personality- and privacyrelated right to be left alone (or to leave one’s iPhone at home), and practical concerns with regard to rural areas, are primarily linked to the level of rule-communication (Big Link).

2. Law-related specifications 12

In the same way as one should reflect whether it is really new technology that is at stake and which part of it is in question, one should also try to achieve more precision concerning the question of whether it is really the law that is concerned by the 10 With regard to the notion of complexity cf Kaplow, Rules versus Standards: An Economic Analysis, (1992) 42 Duke LJ 557, 565 and 586 et seq.; Kaplow, A Model of the Optimal Complexity of Legal Rules, (1995) 11 J L Econ & Org 150, 150; cf also with regard to the connection between tailoring and complexity Geis, AALS Section on Contracts Symposium Empirical Scholarship: What should we study and how should we study it?: An Experiment in the Optimal Precision of Contract Default Rules, (2006) 80 TulLRev 1109, 1115 et seq. 11 On that point Waldman, Power, Process, and Automated Decision-Making, (2019) 88 Fordham LR 613, 617 et seq. (opacity-problem), 627 (engineer-problem); Breiman, Statistical Modeling: The Two Cultures, (2001) 16 StatSci 199, 206 ff; Selbst/Barocas, The Intuitive Appeal of Explainable Machines, (2018) 87 Fordham LR 1085, 1092 et seq.

118

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

personalization project (a)). Indeed, an alternative to developments in the law is that private individuals use new technologies to personalize their behavior and relations. The law – in its conventional generality – would then have to deal with this phenomenon by either encouraging or deterring the conduct in question. Arguments, for example with regard to data protection, might change their weight significantly depending on whether it is a private entity that pursues personalized distinctions based on data sets, or if the state acts similarly. The same is true for considerations of behavioral economics around taste-shaping and legal commands. If one has concluded that it is the law itself that is the object of personalization, this object should be analyzed with more accuracy: Is it legal rules or standards that are at stake (b)? And furthermore: should personalization be pursued by an “evolutionary” approach, retaining and re-interpreting the elements of the law or should the very structure of the law be overturned by “revolutionary” developments (c)? a) Factual-private vs. legal-public. One first distinction refers to who is the moving 13 power of personalization: Is it a private actor or the government (in the widest sense)? From the viewpoint of the law, any tendency of personalization realized by private actors belongs to the sphere of facts. Government-based personalization by means of the law is – on the other hand – of normative quality. It is obvious that the law must deal with the omnipresent algorithmic use of Big Data by private actors (regulation of personalization).12 This exercise, however, does not necessarily require significant changes within the law. In other words: regulation of personalization is possible without personalization of regulation. With already existing legal tools, the law can react in two distinct ways to private use 14 of algorithms: First, the law can limit this use or otherwise seek to undertake counterregulation (prohibition-dimension). Second, the law can prescribe the use of algorithmdriven instruments by private actors (obligation-dimension).13 In contrast, an example of personalization of the law would be the tailoring of default rules according to the personal preferences of the parties – as it is suggested by some legal scholars.14 The prohibition-dimension of regulation of personalization may be illustrated by 15 legal demands to racially debias algorithms used by private companies in order to personalize advertisement or prices or the credit-ranking system. An example for regulation of personalization in the obligation-dimension would be to require a doctor to use Big Data and algorithms in order to personalize the diagnosis of an illness. Obviously, the law demands optimal exploitation of the factual possibilities of science – just as a doctor is also required to use other tools technology in the best interest of her patient. In this obligation-dimension, one may also refer to a (general) obligation to disclose personalized information. The degree to which those cases of personalization are imputable to the government or will fall – beyond the mere general obligation to personalize – within the discretion of the private actor, depends on the margin of discretion the government grants them. This margin gets close to zero in case the government would prescribe the use of certified algorithms. 12 With regard to this perspective, which could be called “regulation of personalization” instead of “personalized regulation”, cf, e.g., the contribution of Hacker/Petkova, Reining in the Big Promise of Big Data: Transparency, Inequality, and New Regulatory Frontiers, (2017) 15 Nw J Tech & IP 1, 7 et seq. For a comprehensive view of both phenomena cf also Busch, infra Part 4.N. 13 In that context, e.g., Busch, The future of pre-contractual information duties: from behavioural insights to big data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Elgar 2016; Busch, Implementing Personalized Law: Personalized Disclosures in Consumer Law and Data Privacy Law, (2019) 86 UChiLRev 309. 14 Cf infra (fn. 37).

Grigoleit/Bender

119

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 16

Even though there may be overlaps, especially from a policy-perspective, there is a fundamental difference between the government reacting on private use of Big Data and using Big Data to improve the law. In the first instance, the principle of party autonomy applies. There is no reason to generally deviate from liberal principles with regard to Big Data use by private persons. These principles render legal restrictions of private behavior and agreements exceptional and they subject restrictions to a – widely recognized – burden of argumentation, ethically as well as economically.15 Consequently, this burden of argumentation has to be met (only) by any restriction of Big Data use. In this, the legislator will apply basically the same considerations as with regard to other interferences with parties’ conduct.

b) Rules vs. standards. Once it is determined that the object of regulation is a legal rule (personalization of regulation), one can further examine this object and specify it with regard to the distinction between rules and standards. Rules and standards are mainly distinguishable according to the degree of ex ante and ex post-precision: Whereas rules contain (ideally) precise commands designed ex ante,16 standards defer the definite rule-making (within the scope of the standard) to a moment after the facts occurred to which the standard is to be applied.17 From an ex ante-perspective, rules are therefore more precise.18 This distinction also has consequences in a separation of power context: Whereas a rule strengthens the law-maker, e.g. the parliament, a standard delegates power to the law-applier, e.g. the judges. 18 The distinction between rules and standards might raise the level of precision as to what exactly is the legal format of personalization: Is it a rule that is drafted ex ante or is personalization only a manner of ex post-application of a standard? The latter procedure is often called personalization of standards, but in reality, it is about the personalization of ex post created rules. Indeed, the personalization of a standard as such is possible only to a very limited extend: One might think about personalizing the scope of application of a standard (extrinsic standard tailoring) and say, e.g. that a good-faith-in-performance-standard is replaced by a within-30-day-delivery-rule in a commercial context or vice versa. But distinguishing different shades of reasonableness (intrinsic standard tailoring), i.e. introducing sub-standards, is an intricate task. Our intellect has already some difficulties to distinguish “utmost care” from “normal care” or “diligentia quam in suis”. It is definitely not possible to make a difference between a hundred different shades of care. If this kind of intrinsic standard-tailoring is to be pursued, the only way to achieve it is to get the standard closer and closer to a rule. 19 Therefore, personalization in the context of standards practically always relates to ex ante or ex post created rules: What authors may often mean when they speak about personalization of the reasonableness-standard is to replace the standard by ex ante created rules which take into account a significant amount of individual characteristics

17

15 In more detail Grigoleit, Mandatory Law: Fundamental Regulatory Principles, in: Basedow/Hopt/ Zimmermann (eds.), Max Planck Encyclopedia of European Private Law, Volume II, Oxford University Press 2012, 1126, 1128 et seq. 16 We do not qualify a certain generality in the scope of application as an element of the rule-definition. Therefore, what would be defined as an “administrative act” in the Continental legal tradition, (cf, e.g., Maurer/Waldhoff, Allgemeines Verwaltungsrecht, 19th edn., C.H. Beck 2017, § 9 mn. 15 et seq.) can also be addressed as a rule, because it contains an ex ante-command. 17 Cf Kaplow, supra (fn. 10), at 557, 559 et seq. (arguing that “the only distinction between rules and standards is the extent to which efforts to give content to the law before or after individuals act”); this definition is broadly received, cf, e.g., Geis, supra (fn. 10), at 1109, 1117. 18 Presenting this as an alternative definition (in this direction Ayres, Preliminary Thoughts on Optimal Tailoring of Contractual Rules, (1993) 3 S Cal Interdis LJ 1, 8 et seq.) is misleading, as both definitions describe the same phenomenon and complement each other.

120

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:09 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

and which are therefore highly complex and personalized. Alternatively, “personalization of standards” may refer to retaining the standard and to postulating that the standard should be applied by taking into account numerous data and therefore by creating ex post rules of the same complexity as the ex ante created rules would be.19 Thinking about the impossibility of standard tailoring as such and about the abovementioned implications of rules and standards with regard to the distribution of power within a legal system, might give new input for the personalization project and the issues regarding separation of power arising from it. c) Evolution (changes within the system) vs. revolution (changes of the system). 20 Finally, the goal of legal personalization can either be pursued within the legal system or it can be taken to overturn the very structure of the system. We define the former perspective as “evolutionary” and we will, under this heading, deal with concepts that tie in with conventional elements of the law, such as the reasonable standard of care in torts,20 default rules in contracts or regulatory areas,21 or disclosure rules.22 Beyond such evolutionary perspectives, some authors suggest systemic changes of the existing overall legal structure. They predict an era of governance by “micro-directives” and the “death of rules and standards”, in which the judiciary as well as the legislative branch cease to exercise their traditional functions.23 These suggestions can truly be qualified as “revolutionary”. This distinction is essential because the reservations to be raised against personaliza- 21 tion depend on the benchmark of personalization and on its degree: For instance, as we will elaborate later, there is good reason to be skeptical about default rule personalization, but there are likewise valid rationales to embrace disclosure personalization. And it is yet another thing to make predictions about the overall development of a legal system and develop a utopian (or dystopian) image of what our society will look like. Indeed, even though one might favor certain elements of particularity, the “death of rules and standards”24 still seems somewhat exaggerated.25 While we generally tend to a skeptical approach, the dichotomy between the notions “evolutionary” and “revolutionary” seems to provide a useful structure for the further content of this paper.

III. Evolutionary perspectives On the basis of these distinctions, we will now turn to and illustrate what we have 22 called evolutionary perspectives. According to our general research focus, we will concentrate on specific doctrines of private law which we find meaningful and illustrative in the context of personalization. We will start with the standard of reasonable care in torts (1.) and then move on to the law of contracts where we will analyze default rules and disclosure duties (2.). As a matter of course, our considerations might 19

Cf the examples given by Ben-Shahar/Porat, supra Part 1.B. Ben-Shahar/Porat, supra Part 1.B. Porat/Strahilevitz, supra Part 1.A; Sunstein, Deciding by Default, (2013) 162 UPaLRev 1; Barr/ Mullainathan/Shafir, The Case of Behaviorally Informed Regulation, in: Moss/Cisternino (eds.), New Perspectives on Regulation, The Tobin Project 2009, 25, 41 et seq. (especially 43). 22 Porat/Strahilevitz, supra Part 1.A; Busch, The future of pre-contractual information duties: from behavioural insights to big data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, (Elgar 2016); Busch, supra (fn. 13), at 309. 23 Cf e.g. Casey/Niblett, supra Part 1.C; Casey/Niblett, Framework for the New Personalization of Law, (2019) 86 UChiLRev 333. 24 Cf Casey/Niblett, supra Part 1.C. 25 Cf supra (fn. 5). 20 21

Grigoleit/Bender

121

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

be transferable to other areas, as negligence, mandatory rules, default rules, and disclosure duties are phenomena with universal relevance.

1. Torts: The reasonable standard of care 23

In order to analyze the potential of personalization with regard to the reasonable person standard (bonus pater familias, bon père de famille, besonnener Durchschnittsmensch), it is useful to build on the notional specifications introduced in the previous chapter. In doing so, we will first point out that personalizing the reasonable standard of care is not about a personalization of the standard as such (a), and then we will suggest to distinguish personalization with regard to ex post-rule-making in application of the standard (b) from ex ante-rule-making replacing the standard (c).

24

a) Limits of personalization of standards. Let us turn back to the distinction between rules and standards introduced above. When it is said that the “reasonable person” standard should be replaced by a “reasonable you” standard,26 those authors do not suggest to personalize the standard as such and to introduce different levels of reasonableness. What they actually aim at is either to replace the standard by ex ante created, personalized rules, or to personalize the ex post created rules that specify the standard. The choice between both approaches tends to depend on who is considered to be the better decision-maker: the judge or the legislator. Indeed, standards imply a general delegation of rule-making to a later moment and to a situation in which more information about the matter in question is available, i.e. to judges. Given the special expertise of judges, some arguments might lead in the direction of personalizing ex post created rules. It is therefore sensible to concentrate on the personalization of ex postcreated rules in the application of the general reasonableness-standard.

b) Ex post-rule-personalization. The question of whether individual character traits should be taken into account in the application of a standard, i.e. make the standard more complex,27 is not new. The general line – with a lot of dispute in the details – could be described as follows: 26 It is common ground that the standard of care is responsive to “above-average” capacities or qualifications by which the individual in question could have avoided the damage done.28 This is in accordance with the principle that everybody has to apply the degree of care that is reasonable under the particular circumstances: It would be unreasonable not to use (special) capacities and tools to prevent an accident. In that sense, the law already applies a high degree of particularity. Doctors, for example, are held liable if they do not exercise what can be expected of a doctor.29 Consequently, the reasonable person is held to be a reasonable doctor, driver and engineer specified

25

26

Ben-Shahar/Porat, supra Part 1.B. With regard to complexity and tailoring cf Kaplow, supra (fn. 10), at 557, 590 et seq. (at II.C.); Geis, supra (fn. 10), at 1109, 1115 et seq. 28 With regard to this tendency for the US context Ben-Shahar/Porat, supra Part 1.B; with regard to German tort law cf RGZ 68, 422, 423; BGH NJW 1995, 1150, 1151; Wagner, in: Säcker/Rixecker/Oetker/ Limperg (eds.), Münchener Kommentar. BGB, Volume 6, 7th edn., C.H. Beck 2017, § 823 mn. 38, 43; Larenz, Lehrbuch des Schuldrechts, Allgemeiner Teil, Volume I, 14th edn., C.H. Beck 1987, § 20 III, 285 et seq., and with regard to German criminal law cf Roxin, Strafrecht. Allgemeiner Teil. Grundlagen. Der Aufbau der Verbrechenslehre, Volume I, 4th edn., C.H. Beck 2006, § 24 A mn. 53 et seq., mn. 57 et seq., mn. 61 et seq.; Duttge, in: Joecks/Miebach (eds.), Münchener Kommentar. BGB, Volume 1, 3rd edn., C.H. Beck 2017, § 15 mn. 95 et seq. (higher standard of care if tortfeasor or criminal has special skills or knowledge). 29 Cf e.g. Martinez v Cal Highway Patrol [2010] Cal CA, Fifth Appellate District, [2010] Cal. App. Unpub. Lexis 1317. 27

122

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

empirically through how other doctors, drivers and engineers would act. Algorithms and Big Data can help in the adjudication process (ex post-rule-making) to clarify capacities or character traits even further. But in that sense, the use of algorithms is not clearly distinct from a conventional expert simulation of the potentially tortious scene. One may think, for instance, of an accident, where the relevant standard of care may be conventionally determined based on models created by an expert and based on the visible destructions on the car. In this context, algorithms may help to perfect the existing law, but they would not change its overall structure. With regard to “below average”-capacities, there is a certain (while not clear-cut) 27 tendency not to individualize, i.e. to maintain a general, normative standard, which transforms negligence law in quasi-strict-liability-law.30 This general standard may largely be qualified as a “short-cut”: By applying the general standard, the judge does not need to establish further details of the case, because she can resolve it based on general considerations. But what actually justifies liability is to go back to previous reproachable actions of the defendant. Let us consider, for instance, a blind person who rides a bicycle and causes an accident. In this case, the law may conclude that the individual did not orderly monitor the traffic and therefore caused an accident. This would amount to applying the standard of ordinary care without considering the particularities of the case. Alternatively, the law may conclude that not properly monitoring the traffic is nothing for which one can reproach a blind person. Obviously, though, the blind individual cannot be fully exonerated. She will be held responsible at least for riding a bicycle in the first place without being able to sufficiently control the resulting risks. This approach appears to be more sensible, since the specification of the standard of care implicates the definition of behaviorguiding rules and the law cannot expect someone to do something impossible (impossibilium nulla obligatio est31). In that way, the ex post created rule also matches with the hypothetical ex ante-definition. The legal system is inevitably called upon to determine the level of risk it tolerates. 28 This determination is essential and apart from the task of tort law to merely specify the required risk-avoiding conduct by giving a balanced definition of behavior-guiding rules. Of course, it is also conceivable that the legal system determines to permit a higher level of risk, e.g. for the disabled or for children, in order to foster their participation in society.32 Whether those slight differences of risk levels are really apt for adjudication and whether the law of torts should be open for this kind of distributive 30 For the American context cf Ben-Shahar/Porat, supra Part 1.B; in the US case law, some groups receive “special treatment”. In case of minors, for instance, age is, generally speaking, factored in (Charbonneau v MacRury, Supreme Court of New Hampshire, 1931, 84 N.H. 501, 153 A. 457, 73 A.L.R. 1266), but not if the negligence of the minor is at stake in the context of the doctrine of respondeat superior (Hill Transp. Co v Everett [1944], US Circuit CA, First Circuit, 1944. 145 F.2d 746) or if the minor engages in adult activity (Dellwo v Pearson [1961] Minn SC, 259 Minn. 452, 107 N.W.2d 859). In case of a physically disabled person, it was allowed to personalize “down” (Memorial Hospital of South Bend, Inc v Scott, Supreme Court of Indiana, 1973, 261 Ind. 27, 300 N.E.2d 50), but for a mentally disabled person, to personalize “down” is generally not permitted (Vaughan v Menlove [1837] Common Pleas, 132 ER 490, 3 Bing NC 468 (Eng.)). For the German context cf RGZ 68, 422, 423; BGH NJW 1995, 1150, 1151; Wagner, in: Säcker/Rixecker/Oetker/Limperg (eds.), Münchener Kommentar. BGB, Volume 6, 7th edn., C.H. Beck 2017, § 823 mn. 38, 43 (objective standard but modified insofar as the reference point is a person of the same social background (Verkehrskreis), taking into account e.g. age or – contrary to the US system – mental illness). 31 General principle in law, already found in Roman Law, cf Digest D 50, 17, 185. 32 With regard to this objective in the context of required supervision of children by their parents to avoid responsibility according to § 832 BGB, cf BGH NJW 1984, 2574, 2575; 1980, 1044, 1045; Spindler, in: Bamberger/Roth/Hau/Poseck (eds.), Beck’scher Online Kommentar. BGB, 51st edn., C.H. Beck 2019, § 832 mn. 20.

Grigoleit/Bender

123

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

element is at least questionable. Indeed, other mechanisms such as state subsidized insurance might be a better solution which help to collectivize the cost of integration. Behavior-guidance and liability are then detached. 29 In conclusion, the potential of personalization of ex post created rules is largely independent from the technological development. Rather, technology intervenes in such cases as a tool to better understand the underlying facts of the case. 30

31

32

33

34

c) Ex ante-rule-personalization. In cases of what we would like to refer to as ex anterule-personalization, the law may – flexibly and in an individualized manner – prescribe ex ante what the courts would otherwise find ex post by specifying a standard. This kind of individualized ex ante-commands, replacing the standard, is also partly exercised already by rather coarse technological means. For instance, on highways, driving limitations sometimes vary already now according to traffic, time of the day or to rainy and sunny weather. A further move towards particularity seems possible through extensive use of new technologies. An example for such a development that has been suggested in legal literature would be the communication and application of individualized speed limits according to age, health, and other individual features.33 Such micro-commands could be thought of in many other situations. With regard to the current system of traffic law, such individualized speed limits would have two (in principle) positive effects. Firstly, it would allow above average drivers to go faster than the current one-size-fits-all speed limits and thereby increase freedom without significantly higher risk for other people’s safety. Secondly, it would make visible ex ante the limitations of below average drivers. Ideally, however, the driving behavior of below average drivers would not change, since the status quo prescribes already that below average drivers are not allowed to go up to the speed limit if driving thereby became unsafe. Indeed, speed limits are always qualified as maxima, while more restrictive commands may be derived from the general standard of care ex post, which for the German context is provided for under § 3 I 2 of the road traffic regulations (StVO). The increase in safety would therefore not consist in requiring below average drivers to drive slower (this is already now required via the general standard of care), but to clearly communicate the limitation to the driver through a simple rule without relying on her own judgment of her capacities. It is still quite uncertain whether it will be feasible to develop algorithms, which reliably indicate personalized speed limits and lead to noticeable distinctions. Even if one presumes the technical feasibility of defining and communicating such microcommands, three34 reservations should be considered. They concern the consequences of personalized ex ante rules in the social sphere and are not meant to categorically object to the personalization of ex ante created rules but rather to add to the critical reflections in a potential implementation process. First, on a practical level, it has to be explored how the individualized guidance affects the overall flow of persons and goods. The degree of specificity in speed-limitations might not only depend on individual capacities, but also on the general flow of traffic, especially with regard to one-lane-roads. 33

E.g., Ben-Shahar/Porat, supra Part 1.B. Sometimes, an additional argument is brought up: individualization would affect the overall standing of an individual in society and it might be harmful for the self-perception of one individual if that person visibly receives significantly lower speed-limits as her co-driver of the same group of age. We would suggest, however, that this argument merely addresses self-perception and thus might not be strong enough to limit above average drivers in their freedom of going faster or even to risk other people’s lives by not communicating clear limits to below average drivers. 34

124

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

Second, on an educational level, the move towards particularity will probably affect 35 the ability to learn from the behavior of others and the sanctions they receive, as other people’s behavior cannot serve as a role-model to the same extent as it can now. The taste-shaping power of the law is therefore at least reduced. On that token, it is also unclear how the law will be taught in a world of particularity. Third, on a constitutional and ethical level, at least some potential proxies for 36 capacity, such as age, race and gender, may not be eligible to be taken into account by algorithms. Therefore, the algorithm-based system of personalized commands might have to deviate from its one premise to mirror capacity as accurately as possible. Whether it is still worthwhile to go down the road of personalization if some character traits are to be excluded from consideration, is at least questionable. But even outside the domain of strict scrutiny, equality-issues arise that will be explored below when we examine general constitutional limits of personalization in the context of revolutionary approaches.

2. Contracts In contract law, personalization has mainly been discussed with regard to default 37 rules and to disclosure duties35 – even though personalization of mandatory contract law in general has also entered the discussion recently.36 While there are considerable reservations with respect to personalization of default rules (a), new technologies may bear a higher potential with regard to personalizing information which has to be disclosed (b). a) Default Rules. Default rules are presented by some authors as particularly well 38 suited for personalization.37 A widely held understanding of default rules underpins this assumption: Default rules are perceived as a state service that reduces transaction costs by (empirically) reflecting the parties’ preferences, so that they do not have to contract around.38 Algorithm-based use of Big Data could therefore help to determine proxies for our behavioral patterns and preferences which then could serve as a basis for a more granular default rule system. We will critically reflect this theoretical assumption and argue that default rules are not exclusively empirical translations of one’s preferences, but always partly normative, i.e. value-based statements of the legislator (aa), and that (at least) therefore,39 constitutional principles, especially the principle of equality, limit the move towards particularity significantly (bb).40 35

Cf, e.g., Porat/Strahilevitz, supra Part 1.A; Sunstein, supra (fn. 21). Ben-Shahar/Porat, supra Part 1.B. 37 In particular Porat/Strahilevitz, supra Part 1.A; Sunstein, supra (fn. 21), at 1 et seq. 38 In the context of personalization cf Porat/Strahilevitz, supra Part 1.A; generally in this direction cf Shavell, Damage measures for breach of contract, (1980) 11 Bell J Econ 466, 466 ff; Schwartz, Proposals for Products Liability Reform: A Theoretical Synthesis, (1988) 97 Yale LJ 353, 361; Cziupka, Dispositives Vertragsrecht. Funktionsweise und Qualitätsmerkmale gesetzlicher Regelungsmuster, Mohr Siebeck 2010, 291 et seq., 339 et seq. (“staatliche Serviceleistung”); Maultzsch, Die Grenzen des Erfüllungsanspruchs aus dogmatischer und ökonomischer Sicht, 2007, 207 AcP 530, 547 (“Dienstleistung des Staates”). 39 Constitutional principles should – at least in the German context according to article 1 III of the German Fundamental Law (GG) – apply, even if default rules were to be perceived as pure state service and justified through parties’ preferences (contra Schmidt-Kessel, Europäisches Vertragsrecht, in: Riesenhuber (ed.), Europäische Methodenlehre: Handbuch für Ausbildung und Praxis, 3rd edn., De Gruyter 2015, 373, 385 (mn. 26 et seq.). However, as soon as one recognizes a normative element in each default rule, the argument in favor of the “special nature” of default rules collapses. 40 More in detail with regard to theoretical and constitutional limits to personalization of default rules cf Bender, Grenzen der Personalisierung des dispositiven Rechts, in GJZ Tagungsband 2019 (forthcoming). 36

Grigoleit/Bender

125

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

aa) Theoretical objections to the purely empirical approach to default rules. While it is true that default rules must be in line with the parties’ typical preferences,41 this assumption does not exhaust their theoretical background. In many jurisdictions, in particular under German law, default rules are also attributed a normative element, in the sense that they do not only mirror the parties’ preferences, but also define a fair balance of the parties’ interests.42 This notion is clearly expressed by § 307 II No. 1 of the German Civil Code (BGB). Under this provision, “standard terms” are generally qualified as unfair and invalid if they are contrary to the basic principles laid down in default rules unless this contradiction can be justified by legitimate interests.43 40 Even from an economic perspective one may note that mirroring preferences is not a goal in itself, but only a tool to reach a higher goal: to maximize the joint surplus of contracts and therefore overall wealth. By designing default rules according to this goal, the economic perspective is basically oriented at idealized preferences of homines oeconomici44 with regard to surplus maximizing contract terms. Those idealized preferences might and very probably will, but not necessarily have to correspond to real preferences. By idealizing preferences, it is also possible to largely deal with the problem of opposed preferences in the contractual context: Parties are presumed to favor joint surplus maximizing default rules even though one party would prefer a default rule that is less surplus maximizing, but instead has beneficial distributive effects for her.45 In addition, it is noteworthy that even the perspective of idealized preferences may be abandoned, if the overall goal of surplus maximization demands that a so-called penalty or information-forcing default rule would have more beneficial effects.46 Indeed, penalty default rules incentivize the parties to an opt-out by deviating from what they would have wanted so that they disclose information. 41 Insights of behavioral economics reinforce the (potential) detachment from real preferences to an even further extent. The phenomenon of “bounded willpower” leads to understand preferences as endogenous47: They have no “natural pre-existence” in relation to the background to be determined as a matter of law. Preferences are therefore necessarily influenced by the law, they are sometimes inexistent and they steadily change. Constructing a comprehensive default rule system based on them, without additional normative or efficiency assumptions, appears to be hardly feasible. In that sense, the purely empirical approach has a circular tendency. Moreover, diverse

39

41 See on this relation with more detail Grigoleit, Anforderungen des Privatrechts an die Rechtstheorie, in: Jestaedt/Lepsius (eds.), Rechtswissenschaftstheorie, 2008, 51, 59. 42 Canaris, Die Bedeutung der iustitia distributiva im deutschen Vertragsrecht: Sitzungsberichte der Bayerischen Akademie der Wissenschaften, Volume 7, 1997, 54; see also BGH NJW 1964, 1123 (“Gerechtigkeitsgehalt”); critically presenting Cziupka, supra (fn. 38), at 90 et seq. 43 Cf, e.g., BGH NJW 2003, 888, 889 et seq.; Wurmnest, in: Säcker/Rixecker/Oetker/Limperg (eds.), Münchener Kommentar. BGB, Volume 6, 8th edn., C.H. Beck 2019, § 307 mn. 38; Schmidt, in: Bamberger/ Roth/Hau/Poseck (eds.), Beck’scher Online Kommentar. BGB, 51st edn., C.H. Beck 2019, § 307 mn. 54 et seq. 44 Emblematically, Schwartz/Scott, Contract Theory and the Limits of Contract Law, (2003) 113 Yale LJ 541, fn. 35, assimilate preferences with economically rational behavior. 45 In case of purely distributive terms, additional normative criteria are needed. On that point BenShahar, A Bargaining Power Theory of Default Rules, (2009) 109 Colum LRev 396 (suggesting that default rules should mimic the bargaining power of the parties). 46 Ayres/Gertner, Filling gaps in incomplete contracts: An economic theory of default rules (1989) 99 Yale LJ 87, 95; Sunstein, supra (fn. 21), at 35 et seq.; Porat/Strahilevitz, supra Part 1.A. 47 Thaler, Doing Economics Without Homo Economicus, in: Medema/Samuels (eds.), Foundations of Research in Economics: How Do Economists Do Economics?, Elgar 1996, 227, 230 et seq.; Hacker, Verhaltensökonomik und Normativität, Mohr Siebeck 2017, 75 et seq.; Korobkin, Status Quo Bias and Contract Default Rules (1998) 83 Cornell LRev 608, 611 and 625 et seq. (against the “preference exogeneity assumption”).

126

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:10 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

status quo-biases and similar psychological phenomena48 imply a certain “stickiness” of default rules.49 This stickiness may not only be implemented as a tool for taste-shaping in steady interaction with endogenous preferences. It also implicates that the default rule order constitutes an initial distribution of rights and duties which influences the bargaining process and which also has effects on wealth distribution between the parties.50 Default rules in that sense may be understood as “alienable property rights” in the terminology of Calabresi and Melamed51: If one presumes that designing default rules is an initial distributive decision, then one should also require a justification beyond the empirical reflection of preferences.52 These considerations suggest, in our view, that the creation of default rules always 42 contains normative elements beyond mere preference-imitation. The requirement of a justification based on efficiency and fairness greatly confines the potential of personalization. bb) Constitutional limits: equality and freedom. The need for a normative justifica- 43 tion of default rules opens the floor for a constitutional review of personalized, preference-based default rules. We will concentrate on aspects related to equality (i.) and freedom (ii.). (i) Equality. Liberal constitutions contain, among other fundamental rights, a 44 principle of equality (e.g. article 3 in the Fundamental Law (GG) in Germany or the Equal Protection Clause of the US Constitution, contained in the 14th Amendment for the states and incorporated into the 5th Amendment for the federal level). Default rule-personalization leads to potentially discriminatory classifications: Either two people are treated differently because of their different preferences (inter-preferenceclassification). For example, the risk-averse person A receives a warranty and therefore has to pay a higher price, but person B does benefit from the default warranty and pays a lower price accordingly. Or the discrimination may stem from giving effect to one’s person’s preferences which then touches the interests of third persons in a potentially discriminatory way (intra-preference-classification). To give one example that exceeds the limits of contract law but is yet quite telling: a person with 48 Sunstein, supra (fn. 21), at 11 et seq.; Tversky/Kahneman, Judgment under Uncertainty: Heuristics and Biases, (1974) 185 Science 1124; Korobkin/Ulen, Law and Behavioral Science: Removing the Rationality Assumption from Law and Economics, (2000) 88 Cal LRev 1051; Thaler, Towards a Positive Theory of Consumer Choice, (1980) 1 J Econ Behavior & Org 39; Korobkin, Status Quo Bias and Contract Default Rules, (1998) 83 Cornell LRev 608; Kahneman/Knetsch/Thaler, Anomalies: The Endowment Effect, Loss Aversion and Status Quo Bias, (1991) 5 J Econ Perspectives 193; Kelman, A Guide to Critical Legal Studies (HUP 1987) 114 et seq.; further Unberath/Cziupka, Dispositives Recht welchen Inhalts? Antworten der ökonomischen Analyse des Rechts, (2009) 209 AcP 37, 72 et seq.; Eidenmüller, Effizienz als Rechtsprinzip. Möglichkeiten und Grenzen der ökonomischen Analyse des Rechts, 4th edn., Mohr Siebeck 2015, 131 et seq. 49 On this point inter alia Ben-Shahar/Pollow, Default Rules in Private and Public Law: Default Rules in Economic Relationships: On the Stickiness of Default Rules, (2006) 33 Fla St U L Rev 651; Sunstein, supra (fn. 21), 11 et seq. 50 Herdegen, in: Maunz/Dürig (eds.), GG-Kommentar, Volume I, 87th edn., C.H. Beck 2019, Art. 1 III mn. 68 (dispositive rules “verschieben die Kräftebalance im Vorfeld privatautonomer Regelung.”); Kähler, Begriff und Rechtfertigung abdingbaren Rechts, Mohr Siebeck 2012, 48 et seq. (“Abbedingungslasten”); Schwab, A Coasean Experiment on Contract Presumptions, (1988) 17 JLS 237, 265 (“contract presumptions may frequently operate more like property entitlements”); different on a model-basis Ayres/Gertner, Strategic Contractual Inefficiency and the Optimal Choice of Legal Rules, (1992) 101 Yale LJ 729, 732 and 737ff (“irrelevance conjecture”). 51 Calabresi/Melamed, Property Rules, Liability Rules, and Inalienability: One View of the Cathedral, (1972) 85 Harv LRev 1089. 52 Regarding behavioral tools in general cf Hacker, supra (fn. 47), at 205 et seq., who also invokes the necessity of a deontological justification.

Grigoleit/Bender

127

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

old-fashioned views on “the family” might receive the algorithm-based intestate rule that his illegitimate child will not be her heir. 45 These examples lead to the question how default rule classifications can be justified. The different treatment of individuals is particularly problematic when it is – according to empirical preferences – based on objectionable criteria that trigger some kind of elevated scrutiny. In the context of the US constitution, the classification might have to live up to the standard of strict scrutiny, which requires a compelling state interest and is mainly triggered by race,53 or heightened scrutiny, which requires an (exceedingly) important governmental objective and is triggered by gender or illegitimacy.54 In the German context, similar criteria are named in article 3 III of the Fundamental Law (GG), in case of which the classification must meet the highest justificatory standards.55 Especially giving effect to (discriminatory) preferences is not a justification that could meet those high argumentative burdens and justify discriminatory intra-preferenceclassifications. Indeed, the US Supreme Court convincingly held in a case involving race that “[p]rivate bias might be outside the reach of the law, but the law cannot, directly or indirectly give them effect.”56 Whether preference-implementation can be seen as a compelling state interest or an (exceedingly) important government objective that could justify the use of race or gender as proxies for preferences in the context of interpreference-classifications is at least doubtful. 46 Even with regard to the “regular” rational basis test in the context of US constitutional law or the “control of arbitrariness” of article 3 I of the German Fundamental Law (GG), it is questionable whether empirical preferences (alone) can be taken as (absolute) justifications. An additional, normative, and policy-based justification should be required to bridge the gap between “is” and “ought to”.57 Certainly, this teleological foundation points, as a general principle, to the parties’ preferences, but only because and insofar as the preference-implementation generally serves the higher policy-goal like wealth maximization.58 Not the preferences alone, but the preferences in connection with this policy-goal satisfy the rationality-requirement of the law.59 47 These reservations have been made based on the simplistic assumption that algorithms can perfectly mirror peoples’ preferences. Of course, justificatory problems are exacerbated if one takes into account the possibility of approximation errors60: Algorithms approximate preferences by analyzing available data and especially this data might not perfectly reflect what a person actually and currently wants. One may think, for instance, of a situation in which a person recently experienced a significant decrease in her degree of risk aversion and this development has not yet been reflected by the data on which the algorithm determines the degree of risk aversion. Since the only justification of the individualized rule is the presumed 53

Cf, e.g., Korematsu v United States, 323 US 214 (1944); McLaughlin v Florida, 379 US 184 (1964). Cf Craig v Boren, 429 US 190 (1976); US v Virginia, 518 US 515 (1996). 55 BVerfG NVwZ 1999, 756; NJW 2004, 1095 (1096); cf BVerfG, 28 January 1992 (judgment), Az. 1 BvR 1025/82, juris mn. 55, BVerfGE 85, 191; BVerfG, 7 October 1957 (order), Az. 1 BvL 1/57; juris mn. 54, BVerfGE 7, 155; cf Kischel, in: Epping/Hillgruber (eds.), Beck’scher Online-Kommentar. GG, 41st edn., C.H. Beck 2019, Art. 3 mn. 214; Langenfeld, in: Maunz/Dürig (eds.), GG-Kommentar, Volume 1, 87th edn., C.H. Beck 2019, Art. 3 III mn. 72. 56 Cf Palmore v. Sidoti, 466 US 429 (1984). 57 Hume, A Treatise of Human Nature, 1739, Reprinted from the Original Edition in three volumes and edited, with an analytical index, by Selby-Bigge, OUP 1896, 244 et seq. (Book III, Part I, Section I). 58 Cf already under III.2.a.bb. 59 With regard to the rationality requirement (“Rationalitätsgebot”) cf Grigoleit, Dogmatik – Methodik – Teleologik, in: Auer/Grigoleit/Hager et al. (eds.), Privatrechtsdogmatik im 21. Jahrhundert: Festschrift für Claus-Wilhelm Canaris zum 80. Geburtstag, De Gruyter 2017, 239 et seq. 60 With regard to the possibility of error cf Waldman, supra (fn. 11), at 613, 617 et seq. 54

128

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

individual preference, the actual absence of this preferences makes the rule arbitrary. In contrast, in a system of generalized legal rules, the justification of default rules transcends individual preferences. Fairness, efficiency, or wealth maximization give content to this supra-individual justification. In sum, only in case of parties designing the content of a contract, i.e. only in case the parties actively opt out in compliance with the requirements of altering rules, their autonomy can justify the content of the rule, because there is no risk of approximation errors. (ii) Freedom. Personalizing default rules according to people’s preferences could – at 48 first glance – promote individual autonomy and freedom, because personalized law possibly corresponds to specific preferences to a larger extent than general law. However, paradoxically, one may argue with some thrust that personalization will in fact not foster, but rather threaten private autonomy and individual freedom. Indeed, personalized default rules will become particularly sticky: Personalized law 49 tends to change according to the individual characteristics of the parties and to the circumstances of the case, which in turn will likely lead to a lack of “individual transparency”61 – in addition to transparency-problems linked to the algorithmic ruleelaboration.62 Private actors that are confronted with personalized, and therefore constantly changing default rules have three hypothetical options: They may firstly consult their default rules – potentially communicated to them through modern communication technology – in every single contractual encounter. They can secondly preventively opt out in every single case. Or they can thirdly blindly trust the personalized default rules created for them. The first option, though, i.e. acquiring knowledge of the personalized default rules, is rather theoretical: Parties will not be able to acquire knowledge of a comprehensive set of default rules for every single contractual encounter. This information overload would not only increase transaction costs compared to a general default rule system, but it also reaches the limits of feasibility. The second option, the preventive optout, is feasible, but it would also increase transaction costs compared to a general default rule system, because one function of (elaborated) default rules precisely is to lower the contracting parties’ load to write extensive contracts.63 Since the first two options would increase transaction costs compared to general default rules, economic rationality would push the contracting parties towards the third option. However, this option makes default rules quasi-mandatory. Individual autonomy may therefore decrease in favor of the “smarter algorithm”. In that context, one should note that it has become not clear yet how the interaction 50 between personalized default rules on the one hand and the setting of the price should look like, if market principles are to be maintained.64 As any default rule of contract law 61 On that requirement cf Hacker, Nudge 2.0: The Future of Behavioural Analysis of Law in Europe and Beyond. A Review of ‘Nudge and the Law. A European Perspective, edited by Alemanno/Sibony, (2016) 2 EurRevPrivL, 297, 308 et seq.; Di Porto/Rangone, Behavioural Sciences in Practice: Lessons for EU Rulemakers, in: Alemanno/Sibony (eds.), Nudge and the Law. A European Perspective, Hart Publishing 2015, 29, 49; van Aaken, Judge the Nudge: In Search of the Legal Limits of Paternalistic Nudging in the EU, in: Alemanno/Sibony (eds.), Nudge and the Law. A European Perspective (Hart Publishing 2015), 83, 94. 62 Similarly for every use of algorithms Waldman, supra (fn. 11), at 613, 614, 618 et seq.; Breiman, supra (fn. 11), at 199, 206 et seq.; Selbst/Barocas, supra (fn. 11), at 1085, 1092 et seq. 63 Grigoleit, Mandatory Law: Fundamental Regulatory Principles, in: Basedow/Hopt/Zimmermann (eds.), Max Planck Encyclopedia of European Private Law, Volume II, Oxford University Press 2012, 1126, 1127. 64 Whereas Waldman, supra (fn. 11), at 613, 614, 624 et seq., links the use of algorithms to a neoliberal tendency, we would – also or even to the contrary – predict a dimension of increased government intervention and market restriction.

Grigoleit/Bender

129

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

is likely to be price sensitive in some way, the personalized choice of a default rule would consequently correspond to a personalized price for each purchaser that has to be agreed to by the seller. As the price is typically the most essential factor, would it not be desirable that the buyer gets a choice of different “rule/price-packages”? Or does the algorithm know of her preference in advance? But if the algorithm is so smart, is there still a need for contractual decisions, contracts or of contract law at all? This point further illustrates the freedom-limiting dimension of particularly sticky personalized default rules. 51 Again, this tendency is even more problematic if the potential of approximation error is taken into consideration: If just the preference and no broader policy reason served as justification of a default rule, then the lack of this (wrongly determined) preference would lead to a lack of justification altogether, because no additional fairness or efficiency considerations would be available to justify that rule. We already developed this point in the context of the principle of equality and the requirement to justify classifications. Similar considerations apply with regard to the question of whether freedom-limitations due to the quasi-mandatory character of default rules can be justified. b) Disclosure rules. Information duties, which are growing in number in several contemporaneous legal systems, are generally based upon the notion that one party would benefit from information that the other party can (effortlessly) provide.65 Under European consumer protection law, for instance, this general principle is specified by many information requirements reflecting the legislators’ assumption that under the legal conditions, the transfer of information is appropriate. However, as the conditions are drawn quite generally, e.g. relating to all b2c relationships, the individual need cannot be met accurately. Therefore, the legislators tend to adopt a wide range of information requirements in order to fulfill all potential needs of information, which leads to an information overload and information mismatch.66 This suggests that personalizing disclosure might be appropriate and significantly less problematic than personalization of default rules.67 53 Algorithm-based analysis of Big Data may help to determine which information is actually necessary to be disclosed.68 For example, a person’s past behavior might show that she is particularly interested in whether her food contains gluten. In that case, gluten-components in food might be the object of personalized information disclosure. The specific functioning of this mechanism of personalized disclosure still requires some further research. Only some general tendencies can be described here, which are based on the above-mentioned dichotomy between private (factual) and public (normative): A system of personalized disclosure is very likely to work in a way that requires private actors to personalize the information that has to be disclosed. Those private actors could probably use the algorithmic know-how and the data sets of their customers which they gained in the process of personalizing advertisement in order to personalize disclosure as well. In that scenario, the law as such, that is the norm

52

65 For general considerations concerning information duties see, Eidenmüller/Faust/Grigoleit/Jansen/ Wagner/Zimmermann, Towards a revision of the consumer acquis, (2011) 48 (4) Common Market Law Review 1077, 1112–1116; Faust/Grigoleit, Informationspflichten: Grundlegende Weichenstellungen, in: Eidenmüller/Faust/Grigoleit/Jansen/Wagner/Zimmermann (eds.), Revision des Verbraucher-acquis, Mohr Siebeck 2011, 193–197. 66 Ben-Shahar/Schneider, The Failure of Mandated Disclosure, (2011) 159 UPaLRev 647, 684 et seq.; Porat/Strahilevitz, supra Part 1.A; Busch, supra (fn. 13), at 309, 315. 67 Likewise Porat/Strahilevitz, supra Part 1.A. 68 Busch, The future of pre-contractual information duties: from behavioral insights to big data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Elgar 2016, 221 et seq.; Busch, supra (fn. 13), at 309; Porat/Strahilevitz, supra Part 1.A; Busch, infra Part 4.N.

130

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

requiring private actors to personalize information, remains general. The personalization of disclosure therefore concerns the private dimension and presents itself as “regulation of personalization” in its “obligation-dimension”, not as “personalization of regulation”. Things would be different in case the state would provide every specific piece of information that should be provided in a personalized way. Then the personalization would concern the law itself (“personalization of regulation”). Further research should also think about whether uniquely personalized information should be provided or whether personalized information should only be a first step (tier onedisclosure), which is completed by a set of retrievable general information (tier twodisclosure), giving interested individuals the option to deepen their knowledge with regard to a specific issue. Those active choice-models69 have the potential to alleviate some objections also in other areas, e.g. default rules.

IV. Revolutionary perspectives We will now turn to what we call revolutionary perspectives – i.e. to concepts of 54 changes which involve personalization and relate to the very structure of the legal system as a whole. In this regard, our bottom line refers to the elementary distinction between changes in the factual environment of the law and changes of the law itself. The potentials of personalization are “revolutionary” in that the factual environment is changing fundamentally and private actors will have to personalize their behavior (1.). However, we do not expect that Big Data and Big Link will revolutionize the overall structure of the law (2.).

1. The digital revolution as a factual phenomenon The potential of personalization in the factual environment of the law has already 55 emerged above when we outlined “evolutionary” developments with regard to specific fields of the law (a). With a little more specificity, we would like to highlight the aspect of “dealing with probabilities” as a field of major developments (b). a) Generalizing the field-specific approach. As we have seen during the analysis of 56 specific elements of the law, algorithms and Big Data will mainly have a place with regard to the factual level and thereby indirectly influence the normative judgment. Indeed, new technological inventions will increase the knowledge of private actors and thereby enable them to make distinctions where they formerly could not see any differences. The doctor, for instance, will be required to take into account technological inventions with the potential of a more personalized treatment. By the same token, retail sellers might be required to algorithmically analyze the data of customers not only to personalize advertisement, but also to personalize disclosure. The law, however, may still retain its conventional general standards. It is the level of complexity in applying the law that increases because more factual information will be considered in their application. In other words: The technologically driven shift towards particularity will fundamentally change the factual environment, but not (necessarily) the legal structure. “Revolutionary” novelty will not come along with significantly different legal norms or “micro-directives”, but it will concern the ability of private actors to personalize. It will be the task of the law to guide this factual or private personalization in a beneficial way. b) Especially: dealing with probabilities. The use of new technologies in factual 57 settings will be particularly promising in areas in which a risk-evaluation, based on 69

Cf Sunstein, supra (fn. 21), at 38 et seq.

Grigoleit/Bender

131

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

probabilities, is required. Indeed, algorithmic analysis of data will enable those who are confronted with a prognostic decision to personalize their decision according to the particular information available in a special case. This general move towards particularity can already be observed. In California, predictive algorithms are used to codetermine whether to grant bail or not70 and automatic tax control programs evaluate the risk of tax evasion also in Germany.71 By the same token, insurance companies, too, already use algorithms to evaluate the prospects of minor cases.72 The fact that dealing with probabilities is particularly prone for personalization, has a specific reason: In case of prognostic decisions, the statistical statement that shows a certain tendency based on large numbers is more important than the correctness of every individual case. In such a context, private actors are therefore willing to accept the de-humanization of the decision-making process and to defer to the statistical statement of algorithms.

2. Limitations to changes of the very structure of the legal system 58

If personalization by algorithms is reflected as a source of structural changes within the overall legal system, some reservations should be taken into account, which may be sketched out around the above mentioned dichotomic distinction between technology and law: In a first step, we will present a psychological objection which has to do with the use of algorithms, i.e. the technical side of the personalization project (a).73 In a second step, we will turn to constitutional problems linked to the personalization of legal commands, i.e. the legal side of the personalization project (b). One may note that, even though the following considerations are aiming at what we call the revolutionary perspective, they convey general risks of personalization, which might – to a lesser degree – also be considered when dealing with personalization in a fieldspecific setting. In that sense, this part, which is dedicated to limits of revolutionary approaches, also illustrates some general problematic tendencies and risks related to personalization.

a) Legitimacy-related concerns with regard to algorithmization. Algorithmization of the structure of the law may give rise to a legitimacy- or acceptability-issue in an empirical sense of the term.74 If the final content of a legal command was determined or specified by algorithms, the decision-making process would be de-humanized. This procedural shift might affect the acceptability of decisions even if the outcome is – from the perspective of ideal legal reasoning – less flawed or “better” than in case of purely human decision-taking. 60 Indeed, empirical research has shown that humans might care more about procedure than about outcome.75 However, it is not self-evident that people tend to accept

59

70 Cf Koepke/Robinson, Danger ahead: Risk assessment and the future of bail reform, (2018) 93 Washington Law Review 1725; Reporter (Note), Bail Reform and Risk Assessment: The Cautionary Tale of Federal Sentencing, (2018) 131 Harv. L. Rev. 1125. 71 Kleinz, Das automatische Finanzamt, https://algorithmenethik.de/2018/03/27/das-automatische-finanzamt/ (last viewed: 9 November 2019). 72 Cf. Positionspapier des Gesamtverbandes der Deutschen Versicherungswirtschaft eV: “Verwendung von Algorithmen in der Versicherungswirtschaft” https://www.gdv.de/resource/blob/32438/0f255d9a31f6454e90e33c5c6c52f314/nutzung-von-algorithmen-stellungnahme-download-data.pdf (last viewed: 9 November 2019). 73 Other objections, like, e.g., data privacy issues, have sufficiently been highlighted in the discourse. See with regard to privacy concerns Porat/Strahilevitz, supra Part 1.A. 74 The word “legitimacy” is therefore used here in the tradition of empirical legitimacy-inquiries (emblematic, e.g., Gilley, The Right to Rule. How States Win and Lose Legitimacy, Columbia University Press 2009). 75 Tyler, Why People Obey the Law, Princeton University Press 1990.

132

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:11 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

algorithmic decisions to a lesser extent than human ones. One important reservation against the acceptance of algorithmic decisions relates to the deliverance of reason which is a central aspect of procedural fairness and may be distorted by a lack of transparency resulting from algorithmization.76 Another human characteristic that we presume to be widespread is the expectation to 61 be heard by a human decision-maker.77 Indeed, what people care about may not so much be to be treated in a particularized way, but in a personalized way that algorithms cannot provide: People might not want to be the object of a personalized legal decision as the sum of endless proxies, but as them. They may want their personality to be considered in an integral way. The intuition of another human being and their instinct, as well as some sort of fellow-human empathy is probably seen to be indispensable if the deliverance of justice seeks to achieve broad acceptance among the human subjects of the law. b) Constitutional concerns with regard to personalization. The constitutional 62 concerns to which we would now like to turn are not only related to algorithms as such, but rather to the phenomenon of personalization – which, of course, largely (even though not exclusively)78 presupposes algorithm-based data-analysis. Obviously, our constitutional considerations must remain rather cursory in this context. We will concentrate on some aspects which appear to be worthwhile from a Private Law perspective. In that context, we will outline concerns with regard to separation of power (aa), democracy (bb), and equality (cc). aa) Separation of powers: possible power shifts away from the parliament and the 63 judiciary. In several ways, the personalization of legal norms would influence the power architecture within a state. The technical aspect of algorithmization, for example, tends to cause a shift of power towards experts.79 But also the hyper-tailoring of rules, from a legal perspective, would require the power architecture to find a new balance. The concrete form of shifts depend on who will control the algorithms that produce the personalized law. If it is the parliament, it will occupy a quasi-administrative and quasiadjudicative power in that it would draft not only the general rules, but provide a solution to the individual case – as is conventionally the task of the administration and the judiciary in a modern legal system. If, however, parliament only enacts general guidelines,80 and the algorithms are elaborated by computer scientists affiliated to the administration, the executive branch will grow in power. Given current expert-affiliation, this development is more probable. If private actors elaborate the algorithms, this will imply a new form of privatization. Power shifts away from parliament would be particularly problematic given its core position in every democratic system. Indeed, with regard to the integration of Germany into the European Union, the German Constitutional Court held that the parliament must not become an “empty capsule”.81 76

Pointing to the opacity of algorithms Waldman, supra (fn. 11), at 613, 614, 618 et seq. See, e.g., Tyler, Why People Obey the Law, Princeton University Press 1990, 116 et seq. 78 Busch, infra Part 4.N. 79 Generally critical about the dominance of scientific experts cf Jasanoff, Subjects of reason: goods, markets and competing imaginaries of global governance, (2016) 4 (3) London Rev of Int Law 361; cf also Waldman, supra (fn. 11), at 613, 627. 80 This is the vision of Casey/Niblett, supra Part 1.C. 81 Cf BVerfG, 15 December 2015 (order), Az. 2 BvR 2735/14, with commentary by Bender, Einzelfallbezogener, menschenwürderadizierter Grundrechtsschutz im Rahmen der Identitätskontrolle, (2016) ZJS 260, 261 et seq.; BVerfG, 12 October 1993 (judgment), Az. 2 BvR 2134/92, 2 BvR 2159/92 (Maastricht), juris (mn. 93). A similar problem arises in case the state as such cannot impose its decisions because it is a failing state (cf Risse/Stollenwerk, Legitimacy in Areas of Limited Statehood, (2018) 21:1 Annual Review of Political Science 403). 77

Grigoleit/Bender

133

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 64

Even though the lack of concreteness with regard to the personalization project makes it extremely difficult to determine whether indeed the parliament would lose power and who will profit most from the development, one tendency seems to be obvious: the judicial branch will lose power – with regard to its rule creation as well as its rule application function. First, personalizing legal commands will affect the rule creation function of judges. Indeed, the productive and innovative function of the judiciary will be lost to some extent. In a system that operates with general legal norms, judges do not only apply the law, but also create law. Standards, in particular, can be seen as provisions by which the legislator delegates power to the judiciary, because standards have to be specified in each single case.82 The personalization of legal commands would shift this specification task to those individuals who are in control of the design of algorithms. Therefore, the “rulification”,83 that is the process of replacing standards by personalized, ex ante created rules, affects the power of the judiciary. Second, judicial review, i.e. the application of constitutional law or European Union law, might lose effectiveness: The algorithmized rules may not be transparent to private actors (lack of individual transparency84), so that plaintiffs are in a weaker position to challenge norms and submit them to judicial review. Even if the rule as such could be decoded, it might be hard to get to know all proxies and quasi-motives used by an algorithm.85 Judges will therefore have difficulties to perform their task of judicial review, because it will be extremely complicated to reconstruct all aspects that played a role in the creation of the rule.

bb) Democracy: dissolution of publicity. The lack of the above-mentioned individual transparency also has a public dimension which leads to an elementary problem with regard to democracy. Indeed, the dissolution of publicity and public transparency86 threaten the democratic organization of society. The law is supposed to be contested in a public forum that functions as a “marketplace of ideas”87. The link between publicity and democracy has been underlined – again – by the German Constitutional Court especially when it controlled the European integration process and limited the shift of sovereignty to the EU level, arguing that the lack of democracy in Europe is – besides other things – linked to a lack of a European publicity.88 66 It is certainly true that not every legal provision can be effectively discussed in the public forum. But this does not justify the systematic surrender of public transparency. In fact, some norms are widely discussed in public, especially if they involve distributive issues (like organ donation89), and most other norms are or could potentially be – at least – discussed in a community of legal scholars and practitioners. Even such a qualified open discourse has a crucial democratically legitimizing force – which would at least be greatly reduced by a deeply-rooted algorithmization of the law. Taking a

65

82

Cf supra (fn 17). Cf also Casey/Niblett, supra Part 1.C. 84 Cf supra (fn 61). 85 Casey/Niblett, supra Part 1.C, are less critical and propose a sort of judicial algorithm-review. 86 See Sunstein, Foreword, in: Alemanno/Sibony (eds.), Nudge and the Law. A European Perspective, Hart Publishing 2015, ix (“transparency and public scrutiny”); Hacker, supra (fn. 61), at 308 et seq.; Thaler/Sunstein, Nudge. Improving Decisions About Health, Wealth, and Happiness, YUP 2008, 243 et seq., 245 (“publicity principle”). 87 Douglas (dissent), in: United States v Rumely, 345 US 41 (1953); Brandenburg v Ohio, 395 US 444 (1969). 88 Cf, e.g., BVerfG, 12 October 1993 (judgment), Az. 2 BvR 2134/92, 2 BvR 2159/92 (Maastricht), juris (mn. 108); BVerfG, 9 November 2011 (judgment), Az. 2 BvC 4/10, 2 BvC 6/10, 2 BvC 8/10, juris (mn. 81); BVerfG, 30 June 2008 (judgment), Az. 2 BvE 2/08 et al. (Lissabon), juris (in particular mn. 249 et seq. regarding public discourse and democracy; see also mn. 346 et seq.). 89 This is a widely used example in the context of default rule personalization, cf Sunstein, supra (fn. 21), at 4 et seq.; Porat/Strahilevitz, supra Part 1.A. 83

134

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

D. The Law between Generality and Particularity

pragmatic philosophical approach, public transparency and the presumption of an open discourse should be part of the legitimizing legal argumentation as regulative ideas90 – just like Dworkin’s one-right-answer-thesis91 with regard to judicial decisions can be seen as such a regulative assumption.92 cc) Equality: discrimination and dissolution. A structural personalization of the 67 legal system would also involve issues of equality. In the context of default rules, we have mentioned the difficulty to justify some classifications that come along with the personalization of legal norms. Those justification problems are particularly severe when the algorithmic approximation is based on objectionable discrimination factors, especially race. In those cases, algorithmic classifications would have to meet elevated justificatory burdens, which in the United States are determined by the standards of strict93 or intermediate94 scrutiny. But beyond those justification-problems, personalized law could significantly change 68 the structure of the principle of equality itself, as it does not only demand treating like cases alike (first prong of the principle of equality), but also to treat different cases differently (second prong of the principle of equality).95 In that sense, a structural personalization of the law might – at first glance – foster the principle of equality, because its second prong could be realized more accurately.96 However, the first prong of the principle of equality, to treat like cases alike, may lose its entire scope of application. If everything is special, nothing is equal. This would complete the shift towards a “society of singularities”,97 in which only the particular case matters. In such a revolutionarily personalized legal system, the role-model of the equal citizen tends to be lost. It is exactly this role-model upon which liberal constitutions are based. In the German constitutional context, for instance, an explicit assurance of generality can even be found in article 19 I 1 of the Fundamental Law (GG), in which a generality-requirement is put up for statutes which limit certain fundamental rights.98 But even where this provision may not be applicable – especially if we think about a system in which the statutes remain general, but are absorbed by a personalized governance through administrative acts – the essence of the principle of equality, i.e. its dignity-dimension,99 guaranteed through article 19 II GG,100 requires some generality, because to uphold the essence of the principle of equality also requires to uphold the essence of its first prong. Kant, Kritik der reinen Vernunft, 2nd edn., 1787, B 672. Cf Dworkin, Taking Rights Seriously, HUP 1977, 14 et seq. 92 Cf Canaris, Richtigkeit und Eigenwertung in der richterlichen Rechtsfindung, (1993) 50 Grazer Universitätsreden, 23 et seq. 93 Strict scrutiny has to be seen in the context of United States v Carolene Product Co., 304 US 144 (1938) (“footnote 4”). In the context of race, it is usually traced back to Korematsu v United States, 323 US 214 (1944), but was fully developed in McLaughlin v Florida, 379 US 184 (1964) (“most rigid scrutiny”). 94 Cf Craig v Boren, 429 US 190 (1976). It has to be noted, however, that the “important governmental objective” in some later cases involving gender was called “exceedingly persuasive justification”, which seems as if heightened intermediate scrutiny was applied, cf, e.g., US v Virginia, 518 US 515 (1996). 95 For this observation cf also Hacker/Petkova, supra (fn. 12), at 1, 7 et seq. With regard to BVerfG and CJEU jurisprudence cf BVerfGE 110, 141, 167; cf also Kischel in: Epping/Hillgruber (eds.), Beck’scher Online-Kommentar. GG, 41st edn., C.H. Beck 2019, Art. 3 mn. 16. 96 Hacker/Petkova, supra (fn. 12), at 1, 7 et seq. with a view to private use of Big Data. 97 Reckwitz, Die Gesellschaft der Singularitäten: Zum Strukturwandel der Moderne, Suhrkamp 2017; arguing for a transfer of this concept into a legal concept Busch, infra Part 4.N. 98 Cf, e.g., Remmert in: Maunz/Dürig (eds.), GG-Kommentar, Volume. III, 85th edn., C.H. Beck 2018, Art 19 I mn. 14 et seq. 99 Bender, Einzelfallbezogener, menschenwürderadizierter Grundrechtsschutz im Rahmen der Identitätskontrolle, (2016) ZJS 260, 264. 100 Remmert in: Maunz/Dürig (eds.), GG-Kommentar, Volume. III, 85th edn., C.H. Beck 2018, Art. 19 II mn. 18. 90 91

Grigoleit/Bender

135

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/04_H.C. Grigoleit - P. Bender, The Law between Generality and Particularity. Chances and Limits of Personalized Law.3d from 30.10.2020 11:35:12 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

V. Conclusions 69

The goal of this contribution is threefold. A first idea was to structure the discourse of personalization around fundamental pairs of opposite notions (part II). A second objective was to examine the phenomenon of personalization – according to a fieldspecific, or evolutionary approach – in particular areas of the law (part III): In torts, for example, we showed that changes will mainly be of a factual nature and we discussed the interaction between personalization on the one hand and rules and standards on the other. In contracts, the necessary normative element in default rules puts serious conceptual and constitutional limits to an empirical, preference-based model of hypertailoring. In contrast, personalizing disclosure duties bears a significantly more promising potential. A third objective was to take the overall-perspective of what we call revolutionary approaches and explore – from a rather abstract vantage point – potentials and limits of structural personalization of the legal system as a whole (part IV). We would suggest that a shift towards particularity – making use of algorithm-based data analysis – will provide more specific and better fitting results wherever the law refers to the resolution of factual issues, especially with regard to probabilities. However, in other respects profound changes in the structure of the law will be confronted with serious reservations. In particular, if one considers algorithmic norm- and legal decision-production, we expect that people will be reluctant to renounce the human factor in crucial decision-making and to accept purely automated decisions. Furthermore, the phenomenon of personalization tends to bring about shifts within the system of separations of powers, as it may strengthen the influence of experts and especially weaken the judiciary. There will be democratic challenges, too, related to the deficits of publicity and of public transparency. Last – but not least – comprehensive personalization may raise equality concerns, as it may come along with (unjustifiable) discriminations and even altogether dissolve the principle of equality as it has been enshrined in the constitutions of modern legal systems. Besides those normative reservation, we repeatedly came across practical limits of personalization, which exacerbate normative objections. On the whole, from the practical perspective, we would suggest that the technology inspired personalization project might over-estimate the calculability of the human decision-making process – an optimism that is familiar and might be derived from the EAL movement.

136

Grigoleit/Bender

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:23 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law: A Critique Throughout the history of law, legal work has been the domain of individual creative 1 work in the sphere of arguments and, in modern history, the monopoly of trained lawyers. With the rise of artificial intelligence and Big Data, this monopoly of creative legal intelligence has become an arena of challenge. One of the recently contested fields is the structure and design of legal norms. A well-known distinction in legal theory is the one between “rules” and “standards” or “principles”. Leaving aside the specific, philosophically heightened meaning given to this distinction by Ronald Dworkin and his followers,1 the difference between both types of legal regulations ultimately turns on a formal difference associated with specific costs and trade-offs.2 Rules can be described as being more “precise”, “detailed”, or “clear” than standards. This entails higher ex ante costs in their drafting, which are balanced by stronger ex ante incentives for compliance and, thus, lower ex post costs of law enforcement. Jurists usually talk of “legal certainty” to describe this effect. Vague standards, on the other hand, show the reverse pattern of costs and benefits. While they are cheap to draft ex ante, they cannot offer the same guidance and efficient regulation of behavior ex ante as detailed rules and can therefore be expected to generate higher ex post costs of law enforcement. Another way to describe this double trade-off is by using Rudolf von Jherings juxtaposition of “formal” versus “substantive realisability”:3 The former is the germane virtue of rules, associated with high levels of legal certainty, yet inseparable from inequities generated by the inevitable over- and under-inclusiveness of strict rule enforcement. Standards, on the other hand, show the reverse properties of “substantive realisability”, i.e., a low measure of legal certainty but a high potential for equitable enforcement in each and every single case. Against this background, the development of novel algorithmic and Big Data 2 procedures, artificial intelligence and “Legal Tech” as tools of personalized, or, to use a recently coined, more precise synonym, “granular”4 legal design and legal administration seems to hold the promise that the well-known trade-off between rules and standards will disappear if it is confronted with the potential of tailoring legal rules to single persons and legally relevant situations on an individual, case-to-case basis. The promise of granular law amounts to no less than a fully unprecedented style of legal 1 Dworkin, Taking Rights Seriously, 1977, 22; Alexy, Theorie der Grundrechte, 1986, 71; see also Auer, Richterbindung und Richterfreiheit in Regeln und Standards, in: Schumann (ed.), Gesetz und richterliche Macht, 2019. 2 Ehrlich/Posner, An Economic Analysis of Legal Rulemaking, 1974, 3 J Legal Stud, 257; Kaplow, Rules versus Standards: An Economic Analysis, 1992, 42 Duke L J, 557; Kaplow, A Model of the Optimal Complexity of Legal Rules, 1995, 11 JLEO, 150; Casey/Niblett, supra Part 1.C; for a recent rendering, see also Morell, Rechtssicherheit oder Einzelfallgerechtigkeit im neuen Recht des Delistings, 2017, 217 AcP, 61 (65). 3 Jhering, Geist des römischen Rechts auf den verschiedenen Stufen seiner Entwicklung, Volume 1, 5th edn., 1891, 51; cf Auer, Materialisierung, Flexibilisierung, Richterfreiheit, 2005, 47; Auer, supra (fn. 1); Morell, supra (fn 2), at 65. 4 For terminology, see Porat/Strahilevitz, supra Part 1.A; Ben-Shahar/Porat, supra Part 1.B (further distinction between “crude” and “granular” personalised default rules; granularity as a matter of degree). In the present chapter, “personalisation” and “granularization” are understood broadly synonymously as both pointing towards a maximum of individual precision of legal rules.

Auer

137

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:23 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

design: The “crude”, “one-fits-all” design of today’s law seems ripe for a revolution if it becomes technically feasible to integrate the legally relevant factual aspects of each and every individual and situation into legal design on a digital basis. Granular law, so the promise goes, will combine the advantages, but not the respective disadvantages of rules and standards into one single, hybrid form of legal normativity. Law will become both more predictable and more equitable if it can be reformulated in granular form on a digital basis.5 This also has far-reaching institutional consequences. In one recent article, Anthony J. Casey and Anthony Niblett have even argued that granular “microdirectives” will not only level out the conventional distinction between rules and standards, but will ultimately lead to no less than the end of adjudication as a necessary mediator between abstract rules and concrete cases.6 3 The goal of the present chapter is to challenge the promises made by the proponents of algorithmic granular law. This challenge does not entail rejecting algorithmic methods as such or “Legal Tech” in a broad sense as necessary and indispensable future tools of legislation and adjudication. Rather, the following argument will turn on the structural optimism entailed by the novel visions of granular law. Against Cass Sunstein, I will argue that granular law, at least for the broad majority of legal subjects, will not be “the wave of the future”.7 Rather, a closer look will reveal a deep structural connection between the concepts of granular law and novel styles of liberal paternalist regulation which, from the point of view of a classic liberal understanding of the rule of law, casts serious theoretical doubt on both. 4 My argument will proceed in five steps. First, I intend to challenge the view put forward by the proponents of granular law that the personalization of legal standards, even in its most fine-grained form, will be the end of typical situations and stereotyped personal identities as the objects of legal regulation (I.). Second, I investigate the kind of typification resulting from Big Data-driven granular law. I argue that legal granularization will likely reinforce discriminatory and constitutionally banned stereotypes and will thus lead to a resurgence of legal discrimination (II.). On that basis, I identify some of the areas of the law where a granularization of legal rules seems to be most promising. I argue that these areas share a specific social and political function which is closely linked to the rise of consumerism. This is also where the structural connection with the concepts of libertarian paternalism or “nudging”, as coined by Richard Thaler and Cass Sunstein, comes into play (III.). This leads to reflections about the rule of law and its purpose in modern regulatory legal systems (IV.). The final part of my argument will return to the concept of adjudication in a rule-based system of law and the philosophical problem of rule-following which, as I argue, will not be rendered obsolete by granular law (V.). Throughout my argument, I will retain a descriptive, not a normative angle. In particular, I will not touch upon the serious normative concerns of privacy raised by granular law.8 I will also use the concept of personal “autonomy” just in the limited sense explained below.

I. The Inevitability of Legal Typification 5

One of the claims put forward by the advocates of granular law is that granular law will reduce and perhaps altogether abolish legal distinctions between different groups 5 For such optimism, see, e.g., Sunstein, Deciding by Default, 2013, 162 U Penn L Rev 1 (10); Porat/ Strahilevitz, supra (fn. 4). 6 Casey/Niblett, supra (fn. 2). 7 Sunstein, supra (fn. 5), at 10. 8 See, e.g., Porat/Strahilevitz, supra (fn. 4); Sunstein, supra (fn. 5), at 54.

138

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:23 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

of persons on the basis of generalized, stereotypical descriptions as commonly found in the current legal system.9 Granular law will, for instance, allow for the possibility of addressing not all “car-drivers”, or “husbands”, or “wives”, but rather each and every single member of any legally relevant social subgroup without referring to general and stereotyping attributes. This claim seems plausible at first glance given the promise that Big-Data-driven, algorithmic search procedures already provide the technical tools for singling out each and every individual for legal purposes according to legally relevant behavior and personality traits. In this vein, it seems obvious that the “one-fits-all” solutions or coarse-grained, stereotyped distinctions made by the current legal system to identify legally relevant groups of addressees will ultimately be rendered obsolete by granular, much more precise structures of legal regulation. From this angle, the optimal state of the law seems to be one where the granular ideal of specific rules for each and every person and situation comes as close to realization as possible. At a closer look, however, the linear correlation implied by this argument – the 6 more personalized the law, the better – is not necessarily true. One straightforward way to challenge it is on economic grounds. It can and has actually been successfully shown that more granularization does not necessarily imply more efficiency in rulefollowing and adjudication. In fact, the opposite can be true. One example is the personalization of the negligence standard in tort law. A convincing case can be made that personalization of negligence law will only lead to mixed results depending on different dimensions of personal negligence such as individual skill and individual risk. It is even likely that a general negligence standard with only limited pockets of personalization very close to the actual state of the law is the most efficient solution for an actual liability regime.10 Moreover, this critique implies an even more fundamental challenge to the possibility 7 and fruitfulness of individually granulized law. As the example of the negligence standard shows, it has to be questioned whether individual personality traits can and should in fact be the dominant and centrally relevant feature for the purpose of legal regulation. In many cases, tailoring a personal legal rule for each and every single individual is neither necessary nor possible for the goal of efficient incentivisation or regulation, because the structure of the social pattern relevant for the pertinent legal design does not require any distinction between single individuals. Put differently, there are relevant cases where the additional information to be achieved by distinguishing between individuals is irrelevant for the purposes of the law. It follows that the movement towards granular law will likely not be the end of typification as a means of legal regulation, i.e., the need to distinguish between generally defined social groups as opposed to individual subjects as the addressees of legal rules. One way to put this is as an insight about the structure of social knowledge. It is a 8 fallacy to assume that the more individualized the members of modern societies regard themselves, the more diverse they also become and that this need be reflected by the law. In fact, the opposite is much more likely and should be considered: Human beings are not significantly different in relevant aspects regarding the effectiveness of legal or social control. Even the most individualized modern societies show surprising patterns 9

For this expectation, see Porat/Strahilevitz, supra (fn. 4); Ben-Shahar/Porat, supra (fn. 4). This conclusion is reached by Ben-Shahar/Porat, supra (fn. 4). For the efficiency of the general negligence standard (“reasonable man standard”; “Learned-Hand-Formula”), see, classically, United States v. Carroll Towing Co., 159 F. 2d 169 (2d Cir. 1947); cf Cooter/Ulen, Law and Economics, 6th edn., 2013, 197; Schäfer/Ott, Lehrbuch der ökonomischen Analyse des Zivilrechts, 5th edn., 2012, 184; Kötz/Wagner, Deliktsrecht, 13th edn., 2016, 54 at para. 114. 10

Auer

139

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:23 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

of conformity among social sub-groups. These patterns contain the deepest riddle of sociology, namely, why and how social structures and patterns of social conformity are possible at all under the condition of individual freedom to depart from the pattern at will at any time. 9 An interesting example for the pervasiveness of social conformity is consumer behavior. It is a common observation that individuals do not behave more individualistically with an increasing number of market options. The more market options are available, the more consumers in fact behave alike and regroup in social groups defined by consumer tastes which are remarkably predictable and surprisingly coarse-grained.11 The deeper insight behind this observation is that the structure of social knowledge does not turn on individualization at all. Knowledge about social structures relevant for legal design is derived from the typical behavior not of individuals but of groups, even if they are so small as to be almost granular. This insight works both ways: Individual persons are generally not a relevant part of social knowledge because it is neither necessary nor possible to distinguish between individuals to gain information about social structures. Relevant information about social structures will generally not depend on the individual person, but rather on statistical data on the typical behavior of groups, even if they are defined by a multitude of fine-grained criteria. The generality of social knowledge structurally excludes the individual person. 10 This insight is highlighted by the epistemological structure of knowledge generated by Big Data. In theories of granular law, algorithmic Big Data profiling acts as a necessary epistemic source for personalizing rules over large groups of society.12 What is truly novel and striking about information generated by Big Data, however, is that it does not require individual data in order to generate precise statistical predictions on individual behavior or relevant personality traits. Some Big Data mining algorithms allow to predict individual characteristics such as race or sexual orientation with greater exactness from mere data patterns, i.e., impersonal, statistical distributions of frequency read through the lens of a skillful combination of search criteria, than by directly using the personal data of the targeted groups if only the data volume used for the search is large enough.13 The latter condition is always fulfilled in the modern data economy due to the exponential growth of the worldwide data volume available for Big Data searches to which we all contribute on a daily basis through our own online habits.14 Against this background, it is not far-fetched to argue that the very centerpiece of Western enlightenment culture, the individual person, is losing her literally in-divisible quality by becoming increasingly “dividual”, a calculable artefact of the modern data economy, to 11 This insight has even been the object of popular art. A remarkable example is the urban street style photography series “Exactitudes.com” by Ari Versluis and Ellie Uyttenbroek which displays a documentation of dress styles worn in metropoles like Rotterdam, Milan, New York, and Moscow between 1998 and 2014. The similarities between the individual wearers are surprising and offer evidence of the complex tension between the need of modern individuals to distinguish themselves from one another and yet to belong to defined social groups, demarcated by subtle and yet striking borders of style. See http://www.exactitudes.com. 12 Porat/Strahilevitz, supra (fn. 4); cf also Kobayashi/Ribstein, Law’s Information Revolution, 2011, 53 Ariz L Rev 1169; Katz, Quantitative Legal Prediction – or – How I Learned to Stop Worrying and Start Preparing for the Data-Driven Future of the Legal Services Industry, 2013, 62 Emory L J 909; McGinnis/ Wasick, Law’s Algorithm, 2014, 66 Fla L Rev 991. 13 See, e.g., Rudder, Dataclysm: Love, Sex, Race, and Identity – What Our Online Lives Tell Us about Our Offline Selves, 2015; cf Porat/Strahilevitz, supra (fn. 4). See also the recent debate on Big Data in election campaigns (“Cambridge Analytica”); e.g., Grassegger/Krogerus, Ich habe nur gezeigt, dass es die Bombe gibt, Das Magazin No. 48, 3 December 2016, https://www.tagesanzeiger.ch/ausland/europa/diesefirma-weiss-was-sie-denken/story/17474918. 14 See, e.g., Porche/Wilson/Johnson/Tierney/Saltzman, Data Flood: Helping the Navy Address the Rising Tide of Sensor Information, 2014, 4 (further references).

140

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:23 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

use a term aptly coined by Gilles Deleuze.15 In a dystopic view of a future culture of BigData-generated granular law, individual right thus degenerates into “dividual law”, addressing human beings as the mere results of algorithmic pattern matching and no longer as indivisible and autonomous beings. The individual person is lost behind a smokescreen of searchable data which allow for a more precise prediction of individual behavior over large numbers than even the direct human cognition of the individuals involved can provide. In the end of this digital dystopia, individuality loses its relevance as a category for the purposes of legal control when we have reached a point where algorithms virtually know human beings better than they do themselves. This epistemological background has to be kept in mind when considering Big Data 11 patterns as a source for personalized law. The promise of granular law described above – i.e., that the rise of information technology will make it possible and meaningful to distinguish between single individuals for purposes of efficient legal design – becomes elusive when confronted with the necessarily statistical, “dividual” structure of social knowledge generated by Big Data search algorithms. It is important to note that even the most sophisticated contemporary models of algorithmic profiling build on statistical pattern matching, that is, on probabilistic distinctions between social groups which necessarily depart from the single individual as the measure of legal design. One example is the use of the Five Factor Model, better known as “Big Five”, as a 12 means of personality profiling in the design of granular norms, as suggested by some of their proponents. The “Big Five”, namely, extraversion, neuroticism, agreeableness, conscientiousness, and openness, are widely regarded as the state of the art in psychological personality profiling.16 As such, they beg for recognition in the design of personalized law wherever individual personality traits play a legally relevant role. In this vein, Ariel Porat and Lior Strahilevitz have argued that personality profiling based on the “Big Five” backed up by social media usage data can serve as a scientifically sound way of identifying legally relevant personality traits.17 In particular, the “Big Five” seem to provide the necessary conceptual link between patterns of human behavior observable in social media usage and scientifically discernable, stable personality traits which arguably provide a legitimate basis for legal distinctions. According to Porat and Strahilevitz, it is possible to predict 33 % of the variation in extraversion, 26 % of the variation in neuroticism, and 17 % of the variation in individual conscientiousness on the basis of Big Data analyses of Facebook usage patterns,18 while conscientiousness itself can predict up to 19 % of the individual variation of likelihood to become an organ donor.19 Taking both insights together means that it is possible to infer from Big Data patterns that any single individual is 19 % more likely to become an organ donor if he or she scores high on conscientiousness, while the latter score can be predicted with 17 % accuracy from his or her social media usage. It cannot be questioned that these and comparable findings are statistically highly 13 significant. However, their purely statistical quality again highlights that granular law must ultimately fall short of its promise that legal design can and should in fact account for the specifics of each and every individual person and situation. A lawmaker occupied with the design of a default rule on the issue of organ donation would, for instance, raise the question how the law should be designed for a person who belongs to 15

Deleuze, Postscript on the Societies of Control, 1992, 59 October 3 (5). See only McCrae, The Place of the FFM in Personality Psychology, 2010, 21 Psychological Inquiry 57 (further references); cf Porat and Strahilevitz (fn. 4) 1434. 17 Ben-Shahar/Porat, supra (fn. 4). 18 Id., 1439. 19 Id., 1442. 16

Auer

141

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

the statistical group described above. If the person is likely to score high on conscientiousness, what should his or her personal default rule on the issue of organ donation be? Or, put differently: What do we actually know when we know that a certain person likely belongs to a group whose members will do something with a certain statistical likelihood? For the purposes of legal design, a level of statistical predictiveness for a certain behavior of far below 100 % or even 50 % seems insufficient as an empirical basis for justifying legal distinctions which draw clear-cut lines between groups that cannot be demarcated equally clearly in the social sphere. 14 More importantly, this argument ultimately does not turn on the magnitude of statistical likelihoods at all. In many cases, it is not necessary to predict individual behavior with or near 100 % certainty in order to design a meaningful and significant default rule. It may in fact be fully unnecessary or even counter-productive for the law to distinguish between social groups in order to accommodate all interests involved, even if those groups are separated by near-certain statistical likelihoods in their behaviors or preferences. One example is the negligence standard in tort law as described above.20 In most cases, a single standard provides just the right level of incentivisation for careful behavior irrespective of personal care levels or risk appetites, and all this without any further need for personalization or granularization. If one concedes this point, it is inevitable to accept that typification is and will remain a necessary part of legal design even under the conditions of Big Data profiling. Legal design cannot and should not avoid typification. Legal rules depart from the properties of the individual person with epistemic necessity. Algorithmic Big Data procedures offer no way out of this epistemic structure even though they promise more powerful insights into individual personality profiles than any technology has been able to offer before. As mentioned before, individual behavior can be predicted more precisely on the basis of Big-Data-based profiling mechanisms than on the basis of the human reason of the very persons involved. Nonetheless, this somewhat disturbing insight does not change the fact that the epistemic structure of Big Data findings is and remains merely statistical and, as such, knowledge about social groups or collectives. From this epistemic reason, it follows that algorithmically generated granular law will most likely not reach a state where its design will allow a truly personal, one-by-one representation of each and every individual person within a society. But we would not want them to reach such a state in the first place, either.

II. The Problem of Algorithmic Discrimination If typification is inevitable as a means of legal design, the focus of the question shifts to whether granular law will lead to better and more useful distinctions between social groups with regard to legal purposes than the one-fits-all solutions or coarse-grained distinctions made by current law. This entails the further question whether granular law will help to overcome or, on the contrary, reinforce undesirable discrimination against minorities by introducing novel, supposedly scientifically backed distinctions between algorithmically distinguishable sub-groups of society. 16 An interesting example is the case of marital surnames.21 The usual legal default regarding marital name law in Western countries is a non-discriminatory one-fits-all 15

20

Cf supra (fn. 10). Sunstein, supra (fn. 5), at 25; Porat/Strahilevitz, supra (fn. 4); both with further reference to Emens, Changing Name Changing: Framing Rules and the Future of Marital Names, 2007, 74 U Chi L Rev 761. 21

142

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

solution which allows both husbands and wives to retain their premarriage surnames, unless other name choices, again strictly non-discriminatorily open to both sexes, have been preferred. This default, however, sticks for the great majority of men, but not for women. In the United States, the overwhelming majority of women change their surnames upon marriage (90 %), while only very few men do so.22 If one were to design a personalized default rule on marital names considering name change preferences, a “crude” version would, therefore, simply amount to splitting up the rule along gender lines. But such a design of family law would most likely be ruled unconstitutional in many Western countries because it discriminates against women and reinforces constitutionally banned gender stereotypes.23 In fact, the abolition of gendered family name laws was celebrated as a hard-won victory against patriarchal family law traditions in countries such as Germany where it took decades as well as numerous judgments by the Federal Constitutional Court to implement egalitarian constitutional values into a formerly strongly anti-egalitarian field of law.24 Against that background, the proponents of granular law have proposed that granular 17 default rules in areas such as marital name law can make way for new, scientifically precise distinctions between individuals or very small, meaningful social groups without recourse to undesirable stereotypes. In particular, Ariel Porat and Lior Strahilevitz have suggested that “crude personalized default rules that are dependent on mere stereotypes are undesirable, but granular personalized rules based on hard data and sound science may be desirable.”25 They offer the following examples for their point that it may be undesirable to split up the marital names default rule along gender lines, but that it might be highly desirable to create default rules for precisely defined social sub-groups demarcated by habits and consumer tastes alone without recourse to the category “gender”: “Suppose it turned out that Caucasian women who regularly shop at Wal-Mart, frequently dine at Cracker Barrel, dropped out of college, and are marrying spouses with similar characteristics adopt their husbands’ surnames 98 % of the time but that Asian American women who have a master’s degree in education, subscribe to the Vegetarian Times and Mother Jones, and take yoga classes adopt their husbands’ surnames only 7 % of the time. Would it be normatively undesirable for the state to adopt as a default rule the assumption that Caucasian women with these characteristics would see their surnames changed upon marriage but the Asian American women would not? Imagine if the data showed that 88 % of male, vegan, Prius drivers with PhDs in philosophy adopt their wives’ surnames upon marriage. Why not flip the default for these husbands to a name change unless they opted out?”26 Leaving aside the strong social stereotypes conferred with each of these examples, it 18 should be noted that they are, in interesting contrast to all other statistical data used by Porat and Strahilevitz, merely made up as hypothetical constructions. And this is not just an accident. In fact, it indicates a deeper problem rooted in the statistical construc22

Emens, supra (fn. 21), at 785–86. For a strong case on sociological grounds against the desirability of state rules increasing the likelihood of female name change, see Emens, supra (fn. 21), at 774–85; cf also Sunstein, supra (fn. 5), at 34; Porat/Strahilevitz, supra (fn. 4). 24 For Germany, see, in particular, the Supreme Constitutional Court ruling BVerfG, 05.03.1991–1 BvL 83/86 and 1 BvL 24/88, BVerfGE 84, 9; see also Dethloff/Walter, Abschied vom Zwang zum gemeinsamen Ehenamen, 1991, NJW 1575. In the USA, a similar development occurred during the 1970s; see Dunn v Palermo, 522 SW2d 679 (Tenn 1975); Emens, supra (fn. 21), at 772–73 (further references). 25 Porat/Strahilevitz, supra (fn. 4). 26 Id. 23

Auer

143

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

tion of exactly these examples. On the basis of actual statistical research, it is highly unlikely to find such great differences in individual variation of name choices between sub-groups of American husbands and wives as hypothesized here. The reason is that no empirical research on that matter can detach itself from the basic probability of marital name change measured across the entire population of men and women within the whole society. If the basic probabilities reported above (men: negligible, women: 90 %) are taken into account, statistical research will likely show that gender remains the single most significant indicator to predict marital name change, while other, statistically dependent factors such as consumer preferences will only have minor influence on its probability and will only in rare cases allow the prediction of its reversal.27 19 For this reason, it is questionable whether unconstitutional discrimination can be avoided by replacing discriminatory criteria such as gender or race by supposedly neutral granular distinctions on the basis of personal tastes or consumer preferences. The latter may effectively serve to hide undesirable discrimination behind a smokescreen of supposedly random, free-willed personal tastes and habits. Yet, it seems impossible to avoid at least indirect recourse to the hard, constitutionally banned discriminatory criteria precisely because they are highly significant for personal and social profiling in many far-reaching contexts such as voting, education, employment, or housing patterns. In fact, this is the very reason why these criteria are undesirable and even legally banned as justifiable grounds for discrimination among social groups in egalitarian societies which subscribe to the principle of equal opportunities for each of its members. Against this background, granular law offers no way out of the normative and constitutional problem of unequal treatment, but will likely reinforce it by replacing direct through indirect discrimination on the basis of algorithmically generated criteria which are innocent only at first glance, whereas, at a closer look, it is impossible to circumvent the basic probabilities of hard discrimination on the basis of statistically dependent secondary criteria. To take up Porat and Strahilevitz’s example once again, it looks like a promising departure from gender discrimination in marital name law to avoid “crude”, discriminatory distinctions along gender lines by shifting to granular, apparently autonomy-related criteria like “preference for yoga classes”. However, the resulting empirical data on name change preferences will likely be methodologically faulty in terms of the statistical sciences because the shift of criteria has no influence on the underlying basic probability of name change measured across the entire population of men and women. This basic probability will likely be reinforced and reflected by the dependent variable that someone who happens to like yoga classes tends to be a woman. That means that, in the end, gender will remain the single most significant predictive factor for marital name change. Replacing hard discriminatory criteria by granular descriptions of consumer behavior thus amounts to a mere game with statistical correlations which inevitably leads back to the hard, constitutionally banned criteria such as gender and race. 20 There exists, therefore, a considerable risk that Big Data-driven granularization of legal norms law will ultimately reinforce and petrify undesirable stereotypes and 27 This insight is reflected by actual statistical information on marital name change distinguishing between specific sub-groups of women. Even among female Harvard graduates of the class of 1980, an additional advanced degree like a Ph.D. or an M.D. only led to a reduction of about 25 % in the probability of name change upon marriage, while each year of marriage delay accounted for a 1 % decline, and each year of delay in having children was related to a 1.3 % decline. Overall, “the fraction of all U.S. college graduate women who kept their surnames upon marriage rose from about 2 to 4 percent around 1975 to just below 20 percent in 2001.” See Goldin/Shim, Making a Name: Women’s Surnames at Marriage and Beyond, 2004, 18 J Econ Persp 143 (144, 158–59); cf Emens, supra (fn. 21), at 787.

144

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

discrimination against historically excluded groups by providing a novel and supposedly scientific basis for their different treatment. This amounts to the real danger of a slippery slope into arguing straightforwardly for the permissiveness of new legal distinctions on the basis of traditionally banned criteria if these criteria are only “validated” by sufficient new algorithmic evidence. As soon as one embraces the beneficial effects of statistical modelling in legal design, there seems to be no good reason to exclude even hard discriminatory criteria from legal design if they are in fact among the statistically significant criteria for predicting individual behavior (which they are). As a consequence, algorithmic and especially Big Data-based forms of granular legal design seem to provide no less than a scientific validation for the social desirability of discrimination. The burden of argument shifts from the necessity to justify discrimination to the direct opposite of having to provide reasons for equal treatment. Porat and Strahilevitz straightforwardly concede this point by arguing that “most people would probably prefer an algorithm that knows their race and gender and, as a result, more accurately predicts their preferences over a system that excludes their race and gender from consideration and consequently provides them with less accurate default rules.”28 Whether or not this is true, it should be considered what kind of society will 21 ultimately result from such considerations and whether we are willing to live in it. That a renaissance of discrimination on a supposedly scientific basis would be socially desirable is not a novel claim, but in fact a staple of conservative and neo-conservative political thinking. One does not even have to look very closely to observe the close structural similarities between Porat and Strahilevitz’s argument and the observation that, even in Western democracies, a significant number of men and women resists – consciously or not – political activism and institutional reform into the direction of gender mainstreaming.29 Regardless of whether or not that is the case as a matter of empirical fact, this argument, turned normatively and backed up with a supposedly scientific basis, amounts to a straightforward naturalistic fallacy which cannot be justified by reference to individual preferences without falling back behind the intellectual standards of decades, if not centuries of anti-discrimination discourse. We can, of course, accept or even pursue such a course of action as a matter of social policy. But we should then know just what we are doing.

III. The Scope of Granular Law and the Rise of Consumerism The danger of reinforcing discrimination does not, however, preclude the granular 22 personalization of the law in areas where this is clearly beneficial or even a necessary part of the very purpose of a given legal field. Insurance law and social security law are examples for such fields. Within their scope, personalized rules and contract terms are already an accepted standard today. Big Data-driven granularization will probably not amount to a conceptual revolution in such fields, but will only provide expanded technical opportunities for the administration of their already highly personalized legal structures. This background should be kept in mind when engaging with the recent debate on granular law. This debate, too, does not pertain to the whole legal system, but is in fact centered around a small number of areas of the law where the benefits of granularization appear to be most promising. These areas, which are also the dominant 28

Porat/Strahilevitz, supra (fn. 4). See, e.g., Cockburn, In the Way of Women: Men’s Resistance to Sex Equality in Organisations, 1991; Rantalaiho/Haiskanen (eds.), Gendered Practices in Working Life, 1997; Wittman, Looking local, finding global: Paradoxes of gender mainstreaming in the Scottish Executive, 2010, 36 Rev Int’l Stud 51 (66–70). 29

Auer

145

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

examples in the recent discussion, are default rules, liability rules, and rules about disclosures in fields such as consumer law, shareholder protection, and medical malpractice, with these field of law and pertinent legal rules obviously overlapping.30 23 When one looks more closely, however, the discussion is in fact centered around even more restricted issues. In particular, not all areas of default rules and liability rules seem equally likely to profit from personalization. For instance, landlord-tenant law or even negligent tort law are much less likely candidates for personalized rules than insurance law or social security law. An interesting example to understand the difference between fields governed by legal defaults where personalization is a promising course of legal policy and others where it is less likely so is, again, the law of marriage. Porat and Strahilevitz argue that legal defaults in marriage law are not likely to profit from personalization because “nearly everything associated with marriage entails undoing a default choice. The default choice is to remain single. Once one decides to get married, the default choice is not to serve food at the wedding, to forgo flowers, to wear pajamas during the ceremony (or no clothing at all!), and to send no thank-you notes after receiving gifts. In short, defaults are not really relevant in these high-stakes settings.”31 Marriage is, in other words, a legal and social practice wherein individual choice is supremely important, but cannot or should not be anticipated by legal defaults: Choice matters, but legal defaults do not. But why is that so? The answer is less obvious than it may seem at first glance given the universal promise that granularization will reduce transaction costs and maximize the efficiency of individual choice within the scope of legal defaults, irrespective of what kind of choice is made. If this promise were viable, it would be a logical consequence to design a fully personalized family law which, for instance, opts individuals characterized by certain personality traits into marrying at a given age and, by default, also proposes a reasonable number of suitable marriage partners plus the complete wedding arrangement. On its face, it may seem absurd to ask why the law does not offer such opportunities given the fact that the necessary technical tools to realize them are already well-established, e.g., on dating platforms. At a closer look, however, it seems inevitable to ask questions such as this one to distinguish the fields where default personalization as a matter of law is appropriate from others where this is not the case. 25 What, then, makes some fields of the law more and others less likely candidates for personal default rules? A tentative way of answering this question could be the following: The fields where personalized defaults are considered are fields where the law promotes or even reduplicates individual preferences without, however, offering the possibility of exercising deep, meaningful, consequential levels of personal autonomy with a true impact on human life. In order to understand this, it is useful to start with the premise that, as a general rule, there is usually no need for the law to reduplicate personal preferences. If the law wants personal preferences to govern in areas such as contract law, it has two choices. One way is to generally abstain from legislative regulation and to leave the regulatory task to the private parties concerned. The other way is to offer impersonal default and sometimes impersonal mandatory rules. For the 24

30 See, e.g., Porat/Strahilevitz, supra (fn. 4); Ben-Shahar/Porat, supra (fn. 4); Sunstein, supra (fn. 5), at 11–17. 31 Porat/Strahilevitz, supra (fn. 4). Surprisingly, Porat and Strahilevitz disagree on this point with Sunstein, who claims that marriage law counts among the fields where “the choice of the default rule is exceedingly important”. See Sunstein, Choosing Not to Choose. Understanding the Value of Choice, 2015, 7.

146

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:24 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

justification of such impersonal legal regimes, it is important to understand that the law fulfils a social function on its own which does not necessarily depend on or correspond with individual preferences. Another way to put this is that there are legitimate legal goals other than minimizing transaction costs.32 There are numerous cases where the law accepts existing or even imposes additional transaction costs in order to pursue its own goals. An example is the prevention of individually harmful decisions by imposing legal formalities in certain contract types such as real estate sales or suretyships.33 These examples show that the law does not generally have to offer good reasons for providing general rules, even if such rules create additional transaction costs. With regard to the project of granular law, this means that the burden of argument shifts to the side of its defenders: There has to be a good reason for the law to depart from general rules, and the mere avoidance of transaction costs does not necessarily provide such a reason. Against this background, a tentative hypothesis for describing the plausibility of legal 26 personalization in some fields as opposed to others might be the following: There seem to be fields in the modern society where individual choice matters in the very restricted sense that people do not actually exercise meaningful autonomy when they are choosing, but choice is highly important nonetheless as a matter of construction of the autonomous self as the basis of our society. Put differently, the more the law imitates or reduplicates personal preferences through personalized law, the more it serves the modern ideology of autonomy without having to offer spheres of truly meaningful autonomy at the same time. This insight links the debate on personalized default law to the rise of consumer law in a broad sense: Personalized defaults are a specific tool of modern consumerism insofar as they construe and condition legal subjects as consumers who perceive themselves as autonomous subject but whose autonomy is in fact reduced to the deficient freedom of choosing between alternative consumption styles. This insight also reveals an important connection between personalized law and the 27 ongoing debate on “nudging” as a regulatory tool of “libertarian paternalism”, as prominently proposed by Richard Thaler and Cass Sunstein.34 According to Thaler and Sunstein’s definition, a nudge is “any aspect of the choice architecture that alters people’s behavior in a predictable way without forbidding any options or significantly changing their economic incentives.”35 Nudging thus seems to promise a non-intrusive way of “improving decisions about health, wealth and happiness” by subtly influencing people’s choices through the design of “choice architectures”. Since, as Thaler and Sunstein have argued, there is no neutral or necessary way of presenting alternatives offered for choice, it is the regulators’ responsibility to frame a set of given alternatives in a way which takes into account the scientifically relevant cognitive biases of choosers in order to improve the quality of their decisions without impairing their individual autonomy.36 Whether or not this claim is justified, cannot be discussed in the present chapter. 28 From a liberal legal and ethical standpoint, it is highly objectionable that the manipulative paternalism exerted by deliberately “tricking” consumers into desired behaviors 32 For an overview over the multiple goals of private law legislation including, but not restricted to, efficiency goals, see Wagner, Privatrechtsdogmatik und ökonomische Analyse, in: Auer et al (eds.), Privatrechtsdogmatik im 21. Jahrhundert. Festschrift für Claus-Wilhelm Canaris, 2017, 281 (293–95). 33 Germany: §§ 311b (1), 766 BGB; see also Wagner, Aufgabenübertragung auf Notare – Kosten und Nutzen, in: Preuß (ed.), Aufgabenübertragung auf Notare, 2011, 69. 34 Classically Thaler/Sunstein, Nudge. Improving Decisions about Health, Wealth and Happiness, 2008; subsequently expanded by Sunstein, Why Nudge? The Politics of Liberatarian Paternalism, 2014; Sunstein, supra (fn. 31). 35 Thaler/Sunstein, supra (fn. 34), at 6. 36 Sunstein, supra (fn. 31), at 5–7.

Auer

147

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

though nudging should be justified by or, in fact, is only hidden behind the threadbare argument that nudging does not infringe upon autonomy because it does not reduce the eligible options of choice.37 This discussion is not, however, the most interesting point about nudging with regard to its relevance for granular law. The interesting issue for the present purpose is the wide area of overlap between fields where personalized defaults are discussed and where nudging appears to be a viable regulatory option. This is not to say that granular law and nudging are necessarily congruent. In fact, it is obvious that they are not: Nudging does not necessarily turn on the modern digital methods of granularization, just as, conversely, the idea of granular law does not necessarily depend on nudging techniques. 29 Yet, as the proponents of nudging as well as of granular law agree, the design of default rules is one of the most important tools of nudging.38 Well-designed legal defaults tend to “stick”, that is, they exercise considerable effect on the outcome of individual choices by triggering well-known behavioral biases such as individual inertia and risk aversion – at the same time one of the most common forms of nudging.39 Thus, it seems obvious that granularization may even heighten that “sticky” quality wherever legal defaults are effective. According to Cass Sunstein, defaults offer advantages and should be preferred to the alternative of “active choosing” mainly in three situations: First, where the context is “confusing, technical, and unfamiliar”, second, when “people would prefer not to choose”, and finally, when “learning is not important”.40 Moreover, impersonal defaults should be chosen only under the further condition that “the population is not heterogeneous along any relevant dimension”, whereas relevant heterogeneity does not make an unequivocal case for personalized defaults but may instead be a good argument against any regulation by default.41 If these criteria are not met, especially when “people would actually prefer to choose”,42 active choosing, according to Sunstein, is generally preferable to a default rule. 30 For the purposes of the present chapter, it is striking that these criteria can also serve as a precise description of the situations where personalized law seems to be most promising. As argued before, personalized rules seem to be most appropriate where the law promotes free choice or indeed conveys a prima facie impression of free choice, but the choice made is inherently paradox in the sense that it overuses the concept of autonomy: The choice of better fine print in consumer contracts can hardly be understood as a meaningful expression of autonomy. It is important to understand that this is not just because choosing is not individually rational in such a situation. Rather, the problem with the expression of autonomy in such situations lies deeper. Choice, in the cases of choosing fine print or information duties in consumer contracts, is a way of conceptualizing the social normativity of welfare capitalism in the language of autonomy. It is another way of expressing that autonomy, as a moral principle, is unable to justify individual decisions in a world where the possibilities for individual expression through consumer decisions have become boundless. Information defaults in consumer contracts provide an example on the point: It is a well-known paradox that more information does not necessarily lead to more efficient choices. In fact, the reverse is true: Information overload may lead to worse 37 For a critique along these lines, see McCrudden/King, The Dark Side of Nudging: The Ethics, Political Economy, and Law of Libertarian Paternalism, in: Kemmerer/Möllers/Steinbeis/Wagner (eds.), Choice Architecture in Democracies: Exploring the Legitimacy of Nudging, 2016, 75 (104–124); cf also Waldron, It’s all for Your Own Good, The New York Review of Books, 9 October 2014. 38 Sunstein, supra (fn. 31), at 25; cf Sunstein, supra (fn. 5), at 1; Porat/Strahilevitz, supra (fn. 4). 39 Sunstein, supra (fn. 31), at 25. 40 Id., 18. 41 Id., 18–19. 42 Id., 18.

148

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

choices or, indeed, to no choice at all.43 This is where information defaults usefully come into play. It is, however, a misconception to understand such defaults as a tool of maximizing the rationality of individual choices. Rather, they should be seen as a smokescreen against admitting that people in fact do not exercise autonomy in a meaningful sense in the world of modern consumer law. Again, the insight to take away is that the debate on legal personalization should not 31 be taken at face value. Rather, it should be read as a symptom of the declining state of Western liberal legalism. It does not seem necessary to add that the global future holds more than enough challenges for the concepts of choice and preferences understood to be the vital bases of liberal legalism today.

IV. Regulation and the Rule of Law Notwithstanding the critique up to this point, many arguments in favor of granular 32 law specifically turn on the point that personalized legal norms improve the legal system’s adherence to individual freedom and choice in such legal fields where fostering those values seems effective and desirable. If, however, one advocates free choice and personal autonomy as the core values of a moral system which also embraces an essentially liberal Enlightenment concept of law, one should also remember what the function of law is in that very system. A short answer would be that it is the purpose of the law to regulate individual behavior. In short, law is regulation. But the reverse is not equally true: Not all regulation is law, nor should it be. Law is the normative ultima ratio used by a society to govern conflicts or to allocate goods according to general rules. Thus, a vital part of the liberal concept of law in the philosophical tradition of Western Enlightenment is its design in the form of general rules. The claim that the generality of the law, i.e., the undiscriminating design and 33 application of legal rules regardless of personal differences of the legal subjects, is a necessary precondition of the rule of law can be found throughout Enlightenment philosophical thinking about law. Immanuel Kant sums up this idea in his well-known concept of right, which he defines as “the sum of the conditions under which the choice of one can be united with the choice of another in accordance with a universal law of freedom” under direct reference to the concept of a universal law.44 Kant thus draws the conclusion from the earlier treatment of the matter by Jean-Jacques Rousseau, who had expressed the clear conviction in his “Social Contract” that there can be no political general will of the contractually united citizenry directed to a particular object: “When I say that the object of the laws is always general, I mean that the law considers the subjects as a body and their actions in the abstract, never any man as an individual or any particular action. Thus the law can very well enact that there will be privileges, but it cannot confer them on any one by name; the law can create several Classes of Citizens, it can even specify the qualifications that entitle to membership in these classes, but it cannot nominate this person or that for admission to them; it can establish a royal government and a hereditary succession, but it cannot elect a king or name a royal family; in a word, any function that relates to an individual object does not fall within the province of the legislative power.”45 43 See only Jolls/Sunstein/Thaler, A Behavioral Approach to Law and Economics, 1998, 50 Stan L Rev 1471 (1533–36). 44 Kant, The Metaphysics of Morals, 1797, Introduction to the Doctrine of Right, § B, 6:230, Denis (ed.), Gregor (trans.), 2nd edn., 2017, 27. 45 Rousseau, The Social Contract, 1762, II.6.6, Gourevich (ed. and trans.), 2nd edn., 2019, 69.

Auer

149

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 34

But the roots of this idea are older still. One does not have to embrace the metaphysical ethos of Kantianism or the republican logic of Rousseauvian etatism in order to arrive at exactly the same conclusion: That the law, if it be called law and not privilege, power, or immunity, presupposes generality and applies indiscriminately to all citizens. At the very beginning of modern Enlightenment philosophy, one and a half centuries before Kant and Rousseau, we find the reason given from the very opposite empiricist and materialist angle. Thomas Hobbes argues that human beings, understood as modern legal subjects, are physically and mentally equal and, as such, indiscriminative in any relevant sense for the purposes of law: “Nature hath made man so equall, in the faculties of body, and mind; as that though there bee found one man sometimes manifestly stronger in body, or of quicker mind then another; yet when all is reckoned together, the difference between man, and man, is not so considerable, as that one man can thereupon claim to himselfe any benefit, to which another may not pretend, as well as he. For as to the strength of body, the weakest has strength enough to kill the strongest, either by secret machination, or by confederacy with others, that are in the same danger with himselfe.”46

Whether one follows Kant, Rousseau, or Hobbes, the idea is always the same: For the concept of law as developed in the Enlightenment tradition, human beings are equal. Thus, it follows that legal rules are general rules and that the generality of the law functions as a core aspect of the rule of law. This insight is also reflected by important constitutional principles common to the Western constitutional law tradition, notably equal rights and anti-discrimination clauses. The core importance of equal rights clauses in modern Western constitutionalism further highlights the importance of an egalitarian design of legal rules as the centerpiece of the rule of law. In short: The concept of a general law is the very reason why the rule of law is called the “rule of law” and not the “rule of regulation”. Modern legislators are well-advised to not obscure the difference more than necessary. 36 Again, this is not to say that modern legislation cannot or should not produce new types of law in the form of personalized regulation. To take such a path of development may, however, ultimately lead to a point where the entire concept of law is transformed from a well-established ultima ratio system of general norms to govern social conflict into an increasingly fine-grained system of personal control of each and any aspect of everyday life. This might indeed, at some point, include the dystopian view of a fully personalized law of family relationships as hypothesized above. Going down this road is equivalent to abandoning the concept of law as shaped by Enlightenment reasoning. It entails the development of an increasingly totalitarian bureaucratic society which conditions and commodifies its members by means of personal, paternalistic, and antiegalitarian regulation. This is no less than the state of “governmentality” as described by Michel Foucault.47 A society of governmentality delegalizes the rule of law and replaces it by ubiquitous technocratic and bureaucratic mechanisms of social governance. 37 Against this background, many of the regulatory problems discussed today in economic terms in the context of legal personalization, like adverse effects, crosssubsidizing, strategic behavior, or freeloading, can also and should in fact be seen as the harbingers of the specific culture of governmentality increasingly pervading the modern society.48 Such effects should not simply be taken for granted as inevitable, 35

46

Hobbes, Leviathan, 1651, Ch. XIII, Tuck (ed.), 1996, 86–87. For an explanation and further references, see Litowitz, Postmodern Philosophy and Law, 1997, 73–75. 48 For a useful discussion of these economic side effects, see Porat/Strahilevitz, supra (fn. 4). 47

150

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

matter-of-fact economic side-effects of law making. Rather, they should be understood as direct consequences of a specific legal culture which seems to have forgotten what the rule of law is all about: A general law for all people. Again, the pertinent question is not whether or not we can depart from this classic understanding of the rule of law in favor of a culture of governmentality. Perhaps we can, because we simply prefer the latter (by what we consider to be our capacity for autonomy). But then, again, we should know just what we are doing.

V. Granularization and the Problem of Rule-Following There is at least one reassuring final point to be made. Against their proponents’ 38 claim, granular norms will most certainly not be the “death” of rules and standards, that is, they will render obsolete neither the household distinction between rules and standards or principles, nor the necessity of legal adjudication through a court system.49 In the current legal system, a separate system of adjudication through judicial courts apart from the legislative power is necessary because the language of law is sufficiently open to leave broad leeway for legal interpretation and judicial discretion. It is important to note that this condition will also be fulfilled in a system of personalized and granular norms. In fact, there will likely be no less judicial discretion under granular law than in the present system of impersonal legal rules. The reason for this is a classic feature of rule-following widely discussed in the philosophy of language: There is no such thing as simply following a rule. It is important to understand that this point does not turn on the relative vagueness or personal precision of a given rule or standard. The insight that granular law will not make judicial discretion disappear is, in other words, fully independent from the theoretical distinction between rules and standards and the question discussed above whether or not this distinction will be rendered obsolete by granular micro-directives. Likewise, the economic trade-offs associated with the choice between rules and standards will likely have no effect at all on the unaltered existence of judicial law-making even in a strongly granularized legal system. Consider, again, Casey and Niblett’s statement of the economic trade-off between the 39 respective costs associated with rules and standards: “When lawmakers enact laws today, they must choose between using rules and using standards to achieve a desired goal. This choice requires a trade-off between certainty and calibration. Rules provide certainty through clear ex ante statements of the content of the law. But rules are costly to design because lawmakers must, at the outset, identify and analyze all the various scenarios to which rules might apply. Rules can also be imprecise and error prone. Because they are defined ahead of time, they can be poorly calibrated to the events as they actually occur. Standards, on the other hand, are adjudicated after the fact. As a result, lawmakers avoid high up-front design costs. Moreover, when applied after the fact, standards can be precisely tailored or calibrated to a specific context as it actually arose. But they also generate ex ante uncertainty because regulated actors do not know up front whether their behavior will be deemed by the adjudicator to comply with the standard.”50 As said above, Casey and Niblett continue by arguing that increasing personalization 40 of the law will make this trade-off disappear and replace the choice between rules and 49 50

For these claims, see, again, Casey/Niblett, supra (fn. 2). Id., 1–2.

Auer

151

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

standards by a new form of law, the “micro-directive”, which they hypothesize as a combination of the advantages, but not the disadvantages of both rules and standards. Micro-directives are supposed to be, on the one hand, as clear as rules by providing “a clear instruction to a citizen on how to comply with the law” in each and every situation, while, on the other, they are adaptable even to unforeseen situations just like standards.51 Future technology will allow to tailor and communicate micro-directives to each and every individual in real time by using novel predictive and communicative means. Thus, with micro-directives looming at the horizon, Casey and Niblett see nothing less than the eternal problem of judicial discretion solved: “Imagine a world where lawmakers enact a catalog of precisely tailored laws, specifying the exact behavior that is permitted in every situation. The lawmakers have enough information to anticipate virtually all contingencies, such that laws are perfectly calibrated to their purpose – they are neither over- nor under-inclusive. Now imagine that when a citizen in this world faces a legal decision, she is clearly informed of exactly how to comply with every relevant law before she acts. This citizen does not have to weigh the reasonableness of her actions; nor, does she have to search for the content of a law. She just obeys a simple directive. The laws at work in this world are not traditional rules and standards. Instead, they take a new form that captures the benefits of both rules and standards without incurring the costs. This new form – we call it the micro-directive – is the future of law.”52 “In this way, micro-directives will turn hundreds or thousands of context-specific machine-generated rules into simple directives that are easy to understand and follow. The law controlling a particular scenario may take into account hundreds or thousands of factors; but the individual will receive a simple command like a red or green light.”53 41

In short: All this, appealing as it may sound, is not going to work. The standard argument against the idea that the openness of the law and judicial freedom will disappear with micro-directives can be found in Ludwig Wittgenstein’s treatment of rule-following.54 Wittgenstein argues that there is no such thing as simply following a rule because no rule can be read without a meta-rule which states how the rule is supposed to be read. Since the same applies to any possible meta-rule as well, any attempt at following a rule inevitably leads into an infinite rule-regress, meaning that the possibilities of interpreting the correct way of following a particular rule are boundless. And this insight is, again, fully independent of the issue of vagueness and, thus, from the distinction between rules and standards. Wittgenstein’s argument applies indiscriminately, irrespective of whether one follows a standard or a particularly “precise” rule: “A rule stands there like a sign-post. – Does the sign-post leave no doubt open about the way I have to go? Does it shew which direction I am to take when I have passed it; whether along the road or the footpath or cross-country? But where is it said which way I am to follow it; whether in the direction of its ringer or (e.g.) in the opposite one? – And if there were, not a single sign-post, but a chain of adjacent ones or of 51

Id., Abstract. Id., 1. 53 Id., 12. 54 Wittgenstein, Philosophical Investigations, 1958, §§ 82–87, §§ 198–242. For a comment, see Kripke, Wittgenstein on Rules and Private Language, 1982, 7; for a recent perspective from legal theory Kuntz, Recht als Gegenstand der Rechtwissenschaft und performative Rechtserzeugung, 2016, 216, AcP 866 (892–95). 52

152

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:25 3B2 9.1.580; Page size: 160.00mm  240.00mm

E. Granular Norms and the Concept of Law

chalk marks on the ground – is there only one way of interpreting them? – So I can say, the sign-post does after all leave no room for doubt. Or rather: it sometimes leaves room for doubt and sometimes not. And now this is no longer a philosophical proposition, but an empirical one.”55 Again, Wittgenstein was not the only philosopher to see this point. A similar 42 argument was made by H.L.A. Hart in his famous discussion of the supposedly simple rule “no vehicles in the park”.56 Hart aptly points to the surprising pockets of openness left by even such a comparably simple rule applied in an everyday situation: “A legal rule forbids you to take a vehicle into the public park. Plainly this forbids an automobile, but what about bicycles, roller skates, toy automobiles? What about airplanes? Are these, as we say, to be called ‘vehicles’ for the purpose of the rule or not? If we are to communicate with each other at all, and if, as in the most elementary form of law, we are to express our intentions that a certain type of behavior be regulated by rules, then the general words we use – like ‘vehicle’ in the case I consider – must have some standard instance in which no doubts are felt about its application. There must be a core of settled meaning, but there will be, as well, a penumbra of debatable cases in which words are neither obviously applicable nor obviously ruled out. These cases will each have some features in common with the standard case; they will lack others or be accompanied by features not present in the standard case. Human invention and natural processes continually throw up such variants on the familiar, and if we are to say that these ranges of facts do or do not fall under existing rules, then the classifier must make a decision which is not dictated to him, for the facts and phenomena to which we fit our words and apply our rules are as it were dumb. The toy automobile cannot speak up and say, ‘I am a vehicle for the purpose of this legal rule,’ nor can the roller skates chorus, ‘We are not a vehicle.’ Fact situations do not await us neatly labeled, creased, and folded, nor is their legal classification written on them to be simply read off by the judge. Instead, in applying legal rules, someone must take the responsibility of deciding that words do or do not cover some case in hand with all the practical consequences involved in this decision.”57 The important insight to take away from Wittgenstein’s and Hart’s treatments of 43 rule-following is that every factual situation is infinitely more complex than even the greatest complexity of rules can anticipate. Again, this insight is fully independent from the theoretical difference between rules and standards. In particular, the economic trade-off between the respective costs and benefits of rules and standards is not pertinent to the issue at all. The infinite factual complexity will remain just the same, regardless of whether individuals are confronted with an abstract or with a specific, highly personalized norm: There will always remain infinitely many options of reinterpreting both the norm as well as the underlying factual situation in ways not anticipated by the norm-giver. It is, therefore, highly unlikely that granular “micro-directives” will replace the common forms of rules and standards in future, nor will the need to distinguish between the ex-ante application of a norm and its adjudication after the fact disappear. Wittgenstein, supra (fn. 54), at § 85. Hart, Positivism and the Separation of Law and Morals, 1958, 71 Harv L Rev 593 (607); see also Fuller, Positivism and Fidelity to Law – A Reply to Professor Hart, 1958, 71 Harv L Rev 630 (662–63). From the abundant discussion, see only Schlag, No Vehicles in the Park, 1999, 23 Seattle U L Rev 381 (further references). 57 Hart, supra (fn. 56), at 607. 55 56

Auer

153

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/05_ Auer, Granular Norms final 19.3d from 30.10.2020 11:35:26 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 44

In fact, flexible and case-by-case administrable standards, rather than rules, will likely play an increasingly important role with the rising relevance of algorithmic procedures within the legal process.58 But this unquestionably important development will be neither the “death” of rules and standards nor the end of adjudication. In fact, this utopian suggestion has had many predecessors in earlier fantasies of technocratic totalitarianism such as Jeremy Bentham’s famous “Panopticon”.59 We should be grateful that such a dystopic scheme has never been implemented successfully up to now and will likely not work in the future either. Nonetheless, the possibilities of Big Data and Legal Tech in adjudication have just begun. They will be indispensable for future law making, and a primary goal for present-day legislators and adjudicators should be to use and develop these technologies responsibly. To paraphrase Oliver Wendell Holmes, the man – or the woman – of the future will be the lawyer who is also a master of Legal Tech and the algorithmic sciences, and be it only a master of their limits.60 58

This much is conceded by Casey/Niblett, supra (fn. 2). Bentham, Panopticon, or, The Inspection-House, 1787. Ironically, the fantasy of “seeing it all” as the dystopia of totalitarian regulators of all time is, in its core, the misguided fruit of Enlightenment rationalism. 60 Cf, famously, Holmes, The Path of the Law, 1897, 10 Harv L Rev 457 (469). 59

154

Auer

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:46 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia: Normative Typification and Granular Norm’s Informational Differentiation Alle sozialen, nationalen, konfessionellen Vorurteile sind ja schließlich nur die Wirkungen dieses typischen Denkens. Die Fähigkeit, stets zu individualisieren, ist das Zeichen höchster Bildung. […] Jede Formulierung eines Typus trägt daher stillschweigend die beiden Klauseln: “in der Regel” und “rebus sic stantibus” in sich. Georg Jellinek, Allgemeine Staatslehre, 3. Aufl. (Häring, Berlin 1914), 36 and 41, note 1.

I. More acts or more words: negotia, pragmata, activities Which number is higher, that of the words or that of the actions that the mind is 1 capable of imagining? It is an old story. The classical rhetoric represents this question as: verba or res?1 The Digest of Justinian shows no doubt: 2 “Natura enim rerum conditum est, ut plura sint negotia, quam vocabula”.2 More negotia than words: yet, what does negotia mean? An English translation of the 3 Digest reads: “For it arises from the nature of things, that there are more business transactions than terms to designate them”.3 Negotia as “business transactions”. If there are more words than things, may we say 4 that nomina sunt consequentia rerum again?4 To the well-tempered jurists, to their cultivated obedience, to their imagination 5 inclined towards the necessary goodness of sovereignty, legal naming – the supreme act of nomination – seems like a mystery. In the 14th century Baldo degli Ubaldi wrote: “Amplius quaedam sunt nomina generalis impositionis ut bos et asinus et ista sunt immutabilia quasi naturalia […] Quaedam misterio iuris sunt introducta […] nomina autem magistralia quibus non est datum certum misterium [...] item nomina sacralis designationis ut Petrus bene possunt mutari si non captiosa sint mutatio”5 1

Eggs, Res-verba-Problem, in: Ueding (ed.), Historisches Wörterbuch der Rhetorik, 7, 2005, 1200–1310. See D. 19, 5, 4 (Ulpianus, lib. XXX ad Sabinum): Mommsen/Krüger (eds.), Digesta Iustiniani Augusti, 1870, 575. 3 Scott, The Civil Law, V, 1932, 108. A 19th century anonymous Italian translation renders “negotia” with “affari” (business, transactions): Corpus juris civilis nella sua miglior lezione secondo gli studi più recenti, VII, 1885, 1284. 4 Montorzi, Tra retorica ed enciclopedia. L’ontologismo linguistico del giurista medievale, 2006, 9 Rechtsgeschichte, 58: working about nomina juris and their etymologies, in a seemingly objective manner, the glossators manipulated the texts, according to their purposes. 5 degli Ubaldi, Ad tres priores libros decretalium commentaria, 1578, 2 verso, Prooemium, no. 2 (emphasis added). About the impositio nominis: Cigada, Nomi e cose. Aspetti semantici e pragmatici delle strutture nominali, 1999, 74 et seq. 2

Femia

155

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:46 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

Naming as the dark core of legal authority. Some names are unchangeable; other may be changed only bona fide. The symbiosis of naming and mystery, however, is a collapse into the irrationality. This way the typification problem – the inquiry about the forces of the linguistic representation in the regulative processes – remains obscure. 7 It is better to came back to the Digest and work about the meaning of negotia in the aforementioned quotation. Negotium is not, obviously, the Rechtsgeschäft (legal transaction); Roman law did not know the modern use of this category.6 Negotium – as we are told by Lorenzo Valla in his Dialecticae disputatones (1438–1439) – translates the Greek pragma, that is activity. 8 Valla talks about the polysemy of res (thing) – “quae res autem ad omnia spectat praedicamenta” – and deals with negotium, res and pragma along the same lines: 6

“Et apud rhetores: quod omnis oratio necesse est habeat et rem et verba: quodque eadem constat ex iis quae significantur, et iis quae significant, rebus et verbis. Quae verba quidem appellantes signa, dicunt omnia constare et rebus et signis. […] Neque mirandum est, tot significata unam vocem habere specialia, cum omnia significata generaliter contineat. Magis, ut sentio, commodum nomen, quam illud Graecorum πράγμα quod solet transferri negotium: licet, et nos pro re aliquando dicamus negotium […] Et quale forsan illud Ulpiani: Natura enim rerum inductum7 ut plura sint negotia quam vocabula”.8 “The same goes for orators because every oration needs to contain a thing and words both, and also because a speech consists of what is signified and what signifies – things and words. Hence words are called ‘token of things’, and some call them ‘signs’, claiming that everything consists of words and signs (whence comes the word ‘signify’ and then ‘signification’). […] And no wonder one word has so many special significations since it includes them all comprehensively. ‘Thing’ works better (I feel) than the Greek pragma, which is usually translated as ‘activity’, though we too sometimes say ‘activity’ for ‘thing’ […] This also may be what Ulpian means: “it was actually established in the nature of things that there are more activities than words””.9 9

Negotia, pragma, res, in accordance with the res-verba division. The last passage cited by Valla is the same extract from Ulpian:10 Valla translates negotia into pragma, which

6 Calasso, Il negozio giuridico. Lezioni di storia del diritto italiano, 1967, 5 et seq.; Cappellini, Negozio giuridico (storia), in: Sacco (ed.), Digesto delle discipline privatistiche (sezione civile), XII, 1995, 106. 7 Inductum instead of conductum is attested for example in Corpus iuris civilis iustinianei cum commentariis Accursii, scholiis Contii et D. Gothofredi lucubrationibus ad Accursium, I, 1627, col. 1866 (here, however, a side comment takes into account the different lectio). 8 Valla, Disputationes Dialecticae, 1541, I, 1, 16, numbers 3 and 9. See Justinian’s Constitutio Tanta, § 18: “Sed quia divinae quidem res perfectissimae sunt, humani vero iuris condicio semper in infinitum decurrit et nihil est in ea, quod stare perpetuo possit (multas etenim formas edere natura novas deproperat), non desperamus quaedam postea emergi negotia, quae adhuc legum laqueis non sunt innodata.” (italics added). A translation in Watson (ed.), The Digest of Justinian, I, 1998, XLIX: “Now things divine are entirely perfect, but the character of human law is always to hasten onward, and there is nothing in it which can abide forever, since nature is eager to produce new forms. We therefore do not cease to expect that matters will henceforth arise that are not secured in legal bonds”. See Cancelli, La codificazione dell’edictum praetoris. 259–260; Ferrini, Manuale di Pandette, 3rd edn. 1908, 143, fn. 2. 9 Valla, Dialectical Disputations, I, 1, ed. and transl. by Copenhaver/Nauta, 2012, 27, 29 and 31. 10 Pringsheim, Animus donandi, 1921, 42 Zeitschrift der Savigny-Stiftung für Rechtsgeschichte, Romanistische Abteilung, 280: “Es scheint, als hätten die Byzantiner bei ihrem Fortschreiten zum Begriffe des Rechtsgeschäftes das klassische negotium benutzt und es zwischen pactum und contractus eingestellt: das negotium ist kein echter, kein benannter Kontrakt, aber es ist doch klagbar”.

156

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

for Valla’s translators counts as activity: That is almost different from Samuel Scott’s ‘business transaction’. A second, distinguished, translation of Digest makes an interesting choice: a complex 10 syntagma, via the concept of type: “For it is implicit in the nature of reality that there are more types of transactions than names for them”.11 Plura negotia quam vocabula becomes ‘more types of transaction’ than ‘names’ (not: 11 word). How useful is ‘type’ in our way of legal worldmaking? Is it true in legal mind too, that “knowing is as much remaking as reporting”?12 Typification of human activities, of our πράγματα or negotia in broad sense, is a mirror or a filter, the name for setting the frame or for building the very structure of reality?

II. Two ways of grasping reality: taming the chaos with Emilio Betti and Tullio Ascarelli Regulation is a reduction of complexity. The norm selects infinite ways of reality, 12 arranging them according to typical models. One of the leading Italian jurists of the 20th century, Emilio Betti, gave a classic 13 definition (1935): “A necessary moment in the device of legal norms is, therefore, an abstracting process of configuration by types: in short, typification. Indeed, social life is not subject to legal regulation in its entirety, but only in relation to the problems of organization and arrangement that it presents: accordingly, it is not subject to legal regulation in the complexity and concreteness of its infinite manifestations, but in the restricted character that matters to law for its relevance to those problems. Hence, it is necessary to simplify, by distinguishing in facts the relevant from the irrelevant issues. It is also necessary to summarize and classify social phenomena in categories, which enhance their accessibility to the provided regulation. In particular, human conduct is regarded and valued by law according to the generic provision of the way it ordinarily appears in social life, and thus considered according to those, which are deemed to be its normal characters and typical features, without taking into account all the other specific circumstances of individual facts. Of course, such simplification and classification by types cannot be separated from a certain amount of arbitrariness, and implies such a deformation of the phenomenon as it actually is, in the natural abundance of its real features. Yet, this deformation, or rather transfiguration, lies in the very logic of legal process, as well as – just to draw on obvious similarities – in that of the artistic, technical, scientific process (of the classificatory sciences). Law is essentially form: a form, that, in order to dominate and permeate the substance of social life which it overlaps, must necessarily start configurating this life according to its mould and direct it through its valuations”.13 11

Watson (ed), supra (fn. 8), at 117. Goodman, Ways of Worldmaking, 1988, 22: “Furthermore, if worlds are as much made as found, so also knowing is as much remaking as reporting. All the processes of worldmaking I have discussed enter into knowing. Perceiving motion, we have seen, often consists in producing it. Discovering laws involves drafting them. Recognizing patterns is very much a matter of inventing and imposing them. Comprehension and creation go on together”. 13 Betti, Diritto romano, I, 1935, 5–6 (my translation). This is the original passage: “Momento indispensabile nel congegno della norma giuridica è, quindi, un procedimento astrattivo di configurazione per tipi: in breve, di tipizzazione. Invero, la vita sociale costituisce materia di regolamento 12

Femia

157

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

Legislation is an abstracting process; typification is the essence of legal norm. Cases, each of them with infinite characteristics, come within the legal system neither as organic multitude nor as single entities, through a comprehensive description. Language does grasp the individual neither as a whole nor as life unity. The only way – Betti says – is a radical transfiguration by typification. Law as a form implies law as an ongoing process of typification of external world. 15 An almost contemporary prominent Italian scholar was Tullio Ascarelli. His legal theory has more and more influenced Italian law. In the early Thirties, he worked out the theory of negozio indiretto (indirect transaction)14 to solve the problems that arise from the gap between legal forms and social life. He built up the image of “legal inertia” (inerzia giuridica): social life constantly creates new needs, which cannot be met by structures in the pre-existing legal forms.15 For the sake of tradition, the legal order encourages “misoneism”16 – and yet we must construct the development using the language of stasis.17 Twenty-four years later, Ascarelli transcends the feeble balance embedded into the concept of legal inertia and deepens this point in an original theory of interpretation as an act of continuous creation of order; but, and this is the point, the dynamic of law is achieved through a typology of social reality, that is constructed by the interpreter. 14

giuridico non già nella sua interezza, ma solo con riguardo ai problemi di organizzazione e di composizione che presenta: non già, per conseguenza, nella complessità e concretezza delle sue infinite manifestazioni, ma sotto il circoscritto profilo che interessa il diritto per le sue attinenze con quei problemi. Di qui la necessità così di semplificare, sceverando nei fatti i lineamenti rilevanti da quelli irrilevanti per il diritto, come di riassumere e di classificare i fenomeni sociali in categorie che li rendano meglio accessibili al regolamento che loro si destina. In particolare, il comportamento umano viene contemplato e valutato dal diritto in base alla previsione generica del modo com’esso si presenta d’ordinario nella vita sociale, e quindi considerato secondo quelli che si ritengono suoi caratteri normali e lineamenti tipici, non già con specifico riguardo a tutte le circostanze concrete del caso individuale. Certamente tale semplificazione e classifica per tipi non va disgiunta da una certa dose di arbitrio, e importa una tal quale deformazione del fenomeno qual è in realtà, nella naturale ricchezza de’ suoi lineamenti concreti. Ma si tratta di una deformazione, o piuttosto di una trasfigurazione, che è nella logica stessa del procedimento giuridico, come – per richiamare ovvie analogie – in quella del procedimento artistico, tecnico, scientifico (delle scienze classificatorie). Gli è che il diritto è essenzialmente forma: forma che, per dominare e per permeare la materia della vita sociale cui si sovrappone, deve necessariamente cominciare dal configurarla secondo il suo stampo e dall’orientarla nella direttiva delle sue valutazioni”. In the quoted passage Betti uses both ‘configuration’ and ‘classification’ (by types); therefore, he represents the functioning of the legal system as ‘transfiguration’. Among the many meanings of ‘figura’ in its historical development, Eric Auerbach remarks on the Greek ‘typos’: apart from its plastic sense as ‘imprint’, “typos was also significant for figura because of its suggestion of universality, exemplarity, and law […]. This in turn contributed to effacing the distinction between it and forma – which of course had only been a subtle one at best”. Auerbach, Figura, 1938, in: Id., Time, History, and Literature, 2014, 68 (italics in original). See Porter, Disfigurations: Erich Auerbach’s Theory of Figura, 2017, 44 Critical Inquiry (1), 80–113. The young Betti spoke only about ‘classfication’: see Betti, Letter to Benedetto Croce, December 30, 1916, edited in Nietsch, Il giudice e la legge, 2012, 309–310: the ‘giudizio giuridico’ (the act of judging something under the law, the so-called ‘qualification’ according the legal system) is a ‘giudizio di classificazione’ (classification judgement: ‘judgement’ in the sense of the act of thinking). 14 Alpa/Zeno-Zencovich, Italian Private Law, 2007, 173: “A transaction is indirect when the parties conclude it with the intention of bringing about by oblique means the outcome of a different transaction. The means adopted by the parties are said to fall outside the scope of their intended outcome”. 15 Ascarelli, Il negozio indiretto e le società commerciali, in: Studi di diritto commerciale in onore di Cesare Vivante, I, 1931, 23‐98. 16 Ascarelli, supra (fn. 15), at 26. 17 In Il negozio indiretto Ascarelli does not use ‘type’ nor ‘typification’ as a tool for shaping the historical movement. He spoke in terms of law’s organic development. The legal transaction, the Rechtsgeschäft has, in the case of indirect use too, its typical causa and the new interests count as ‘motives’: Ascarelli, supra (fn. 15), at 42.

158

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

“Indeed, interpretation is and is not the interpreted element. Interpretation is a construction and a reconstruction, which explains, develops, limits and essentially modifies it; relating back to the interpreted element at all times while still modifying it. […] Because every single norm is expressed in words and every single norm refers to one legal ‘fattispecie’.18 So, the interpreter constantly creates a typology of social reality as a function of the application of the norm, just as he organizes a hierarchy of norms depending on their application. And the interpreter’s hopes, traditions and beliefs are asserted in this work of creating and ordering; through the ordering of norms and the typological reconstruction of reality. So we can set the typological ordering of reality, as a function of the application of norms, against the regula juris which simply summarizes a set of law provisions. The norms would not be able to be interpreted and applied without this ordering”.19 The types are not the act of judging made by the transfigurative power of law (Betti); 16 they overflow from interpretation of reality, as a sort of a weak-constructive world’s ‘natura’. Typification, therefore, solves the regulative problem: how to manage the infinite qualitative difference between verba and negotia, between facts (expressing a social communicative action and an economic transaction)20 and their representations (product by irritation, the system theory says) according the inner discursive potentialities of the legal norms. The very normal chaos of life becomes a bundle of words; our incommensurable 17 individualized shaping is nothing but an action of continual type-reframing.21 Type is construction by selection, an open-ended closure; type is the by-product of systemic differentiation.22 Betti sees the limits, the type as inescapable form; Ascarelli, on the 18 ‘Fattispecie’ is a ‘hypothetical fact situation’: see Merryman, The Italian Style I: Doctrine, 1965–66, 18 Stanford Law Review, 49. 19 Ascarelli, Antigone and Portia, [1955] (trasl. by C. Crea), 2015, 1 The Italian Law Journal, 175–176 (I have slightly modified the translation). See Crea, What Is to Be Done? Tullio Ascarelli on the Theory of Legal Interpretation, 2015, 1 The Italian Law Journal, 181‐205. This is the original passage: “L’interpretazione appunto è e non è il dato interpretato; ne è una costruzione e una ricostruzione che spiega, sviluppa, restringe, sostanzialmente modifica; sempre riconducendosi al dato interpretato eppur sempre modificandolo. […] Ché ogni norma si esprime in parole e ogni norma si riferisce a una fattispecie. Perciò l’interprete continuamente costruisce una tipologia della realtà sociale in funzione dell’applicazione della norma, così come ordina gerarchicamente le norme in funzione della loro applicazione. E in questa costruzione e in questo ordinamento si fanno valere le convinzioni, le tradizioni, le speranze dell’interprete; appunto attraverso l’ordinamento delle norme e la ricostruzione tipologica della realtà. Alla regula juris che meramente riassume una normativa possiamo così contrapporre l’ordinamento tipologico della realtà in funzione dell’applicazione delle norme, ordinamento indipendentemente dal quale le norme non potrebbero essere interpretate ed. applicate”: Ascarelli, Antigone e Porzia, [1955], in: Id., Problemi giuridici, I, 1959, 12. 20 Teubner, In the Blind Spot: The Hybridization of Contracting, 2006, 8 Theoretical Inquiries in Law, 51–71. 21 Individualization “covers a complex, manifold, ambiguous phenomenon, or more precisely a social transformation”; “the very conditions which encourage individualism produce new, unfamiliar dependencies: you are obliged to standardize your own existence”: Beck/Beck-Gernsheim, The Normal Chaos of Love, 1995, 7. 22 Luhmann, System as Difference, 2006, 13 Organizations, 40: “Information is information only if it is not just an existing difference; it is information only if it instigates a change of state in the system. This is the case whenever the perception (or any other mode of input one might have in mind) of a difference creates a difference in the system. Something was not known; then information arrives, namely that these, and none other, are the facts of the matter. […] A difference that makes a difference! In this case as well,

Femia

159

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

contrary, stresses the construction, the type as social process. Two ways for taming the chaos: typification as object or as relationship.23

III. End of the journey among the concepts’ penumbra. From type to typification, and from typification to dissemination The problem, now, is if type and typification work identically. The latter is the process, the former its product. But the process owns forces that the reception of typediscourse into the legal theory during the 20th century has almost entirely lost. 19 The debut of type in legal theory established a categorical opposition to concept. The concept was intended as a structure with rigid boundaries, the type with shifting ones.24 The type had noble origins, in logical thinking;25 and obscure origins in the so-called völkisch belief of National Socialism (Nazism).26 Its function was to introduce a difference inside the legal system:27 the traditional area (Begriff/ concept), in which the logical subsumption is the method of applying the legal rule to the case; and an alternative space where there is no ‘subsumption’ but ‘imputation’ (Zuordnung).28 20 The distinction (in turn, hyper-conceptual) between concept and type, has now only a historical value. That something can be regarded as a schema, a model, a frame, or (mediating a different couple in philosophy) as ‘type’ of a multitude of possible ‘tokens’29 does not depend on the intrinsic qualities of the rule nor on the inner structure of the rule itself, but rather on the relationship between this something and all the other norms. 21 The normative language works upon a communicative continuum. It is built with fragments that can be assumed in a lot of ways: as abstract ideas, reworking of concrete experiences, realistic figures, symbols, narratives. The operation of these fragments within the normative language has more constraints of coherence than the ordinary language does. Man/woman, for example, are concepts, if we are within a normative system that imposes the demarcation between discrete, finite entities; but, if we include a principle of gender self-determination in the normative system, male/female must be interpreted as ideal polarities which do not meet (in principle) any physical reality: they are guidelines for the decision, not names that can be given to real objects. If the word rested on the irreducible opposition male/female, then ‘transgender’ would not be a concept, but a nonsense. If, on the contrary, breaking these oppositions becomes a constitutional question of liberty, then sex/gender is not a set of discrete entities, but a 18

the question of how a theory arrives at its first difference remains unanswered. One begins with a difference and, interestingly, ends with a difference. Information processing in its entirety takes place between an initial difference and a difference that emerges during, and as a consequence of, the process.” 23 Sebald, Typik und Semantik, 2009, VI Arhe, 194: “Dieser Ansatz einer Definition stellt meines Erachtens einen Ansatzpunkt für die Entwicklung eines über Schütz hinausgehenden Typenbegriffes dar: nicht (nur) der Gehalt an Bestimmungen macht einen Typus aus, sondern die jeweiligen Relationen von einzelnen Bestimmungselementen. So verstanden hat ein Typus keinen gegenständlichen Charakter, sondern einen formal-relationalen” (discussing the ideas of Alfred Schütz). 24 Strache, Das Denken in Standards. Zugleich ein Beitrag zur Typologik, 1968, 19 et seq. 25 Heyde, Typus. Ein Beitrag zur Typologik, 1952, 50 Studium generale, 235–247. 26 Especially in Larenz’s legal thinking: Kokert, Der Begriff des Typus bei Karl Larenz, 1995, 84 et seq. See Larenz, Über Gegenstand und Methode völkischen Rechtsdenkens, 1938, 45; Hüpers, Karl Larenz: Methodenlehre und Philosophie des Rechts in Geschichte und Gegenwart, 2010, 177 et seq. 27 Leenen, Typus und Rechtsfindung, 1971, 25 et seq. 28 Larenz, Methodenlehre der Rechtswissenschaft, 2nd edn. 1969, 440. 29 Wetzel, Types & tokens. On abstract object, 2009, 1 et seq.

160

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

field of intertwined biological and cultural data.30 ‘Breaking’ is here the transformative operation into the categorical polarity; breaking is, therefore, a new typification built through de-conceptualization. The increased complexity produces a growing differentiation: The more complex, the 22 more differentiated is the social system, the more complex, the more differentiated is the legal system.31 Distinguishing between ‘core’ and ‘penumbra’32 within legal concepts is not enough. The problem is not to demarcate easy from hard cases, certainty from vagueness; the legal system in the global society must produce not objects, tools for regulation, but processes, distinguishable courses of actions for shaping and reshaping words and ideas in rules.33 Typification removes types; but the ongoing process for the production of the social- 23 adequate regulative complexity knows no end. The regulative process applies to itself the distinction between type and typification, object and process; the meta-reflective process of regulation by typification disseminates its products: it does not produce norms yet, but practices of regulative senses. The disseminated senses recollect themselves as a variety of signs for values: signs organized themselves in set of narratives about values, narratives narrate themselves as metanarratives. At the end of the story, signs flow in a new set of types and typification restarts: a circular autopoietic recursiveness.

IV. Big data: quantities make a qualitative shift in nomogenesis Legislating is a complex species of deciding. Making a decision with a rule, and not 24 directly deciding, shifts the focus: not on the assignment of right and wrong, but on the ratio of the act of assignment lies the core of power. Each practice, deciding and legislating, implies data processing.34 The process of data for legislation is more complex (for the sake of simplicity, we assume to hear that even case law is a form of legislation: it shifts in the same way the epistemic strategies of the regulator from the act to the ratio of deciding). Legislating needs a lot of information; it generates risks of under- and overinclusiveness.35 Legislation, however, balances this inconvenience with a lower cost in the administration of courts activity and in a more stable production of social order. The costs and benefits of the governance by regulation foster – again in the self- 25 reflexive way – a dislocation (dislocation, rather than allocation) of normative powers. Not only a “decentralized law for a complex economy”,36 but a social dissemination of normative powers in order to fit the communicative spread of regulatory functions: from business transactions to every act of construction of the private sphere (private lives are made by contracts).37 Big data subvert the ordinary way of legislative informa30 Butler, Gender Trouble: Feminism and the Subversion of Identity, 2nd edn., Routledge, New York 1999, 9–11. For the critics of gender dichotomy see Meadow, “A Rose Is a Rose”. On Producing Legal Gender Classifications, 2010, 24 Gender & Society, (6), 814‐837. 31 Complexity is more ambiguous than system theory’s autopoiesis may represent: Webb, Exploring System Boundaries, 2013, 24 Law and Critique, 131‐151. 32 Hart, Positivism and the Separation of Law and Morals, 1958, 71 Harvard Law Review, 607. 33 The (de)construction process is nonlinear: Finchett-Maddock, Nonlinearity, autonomy and resistant law, in: Murray/Webb/Wheatley (eds.), Complexity Theory and Law: Mapping an Emergent Jurisprudence, 2019, 213–233. 34 McGinnis/Wasik, Law’s Algorithm, 2014, 66 Florida Law Review, 991–1050. 35 Sunstein, Problems with Rules, 1995, 83 California Law Review, 992–993. 36 So to cite the essay of Cooter, Decentralized Law for a Complex Economy, 1993, 23 Southwestern University Law Review, 443‐451. 37 The dissemination strategy brings to nonlinear creative dynamic of multiple legalities. Even if the metaphoric use of science theory can be misleading: Kellert, Extrascientific Uses of Physics: The Case of

Femia

161

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

26

27

28

29

tion processing. The optimal set of information can now be tailored according to each record, to each past behaviour of the consumer. Big data allow an infinite variation in the information’s flow; among the mass of information, the rulers can select, one case at time, the individualized one: to each contract its own information set, a paradoxical mass-customization.38 Contract is always a tremendous name; the common law even knows its “sanctity”:39 it contains the narrative of human interaction: cooperation intertwined with conflict.40 The global economy has increasingly developed its function as private legislation; standard contracts make a – gentle in theory, but savage in content, in practice – dictatorship mediated by illusory consent.41 Therefore, that is the question: is the huge increase in the available information for rulers – for those who do have the power to legislate on private life by imposing contracts – a turning point in the nomogenesis? More precisely: do the two shifts in the contemporary history of legislation (the globalization of mass government through structures of private ordering, among all contracts;42 the availability of big data processing for the mass customizations of consumer contract) cause a qualitative turn in the autopoiesis of law? From quantity to quality? Some see in this process a positive revolution.43 The quantitative change becomes a qualitative shift in the regulation, which radically modifies the default rule concepts and the technique consisting in the intensive use of mandatory rules.44 The last development of this shift in the structures of law data processing is the concept of “dynamic rule”: “Dynamic rules are rules that automatically change without intervention by the rule giver according to changes in future conditions that the rule itself comprehensively and accurately fixes. As computation increases, it becomes easier to add complex conditions, both because these conditions can be continually monitored and because the application of the new rule can be more readily calculated.”45

30

Is the dynamic rule the end of typification, the sign of a radical change in the autopoiesis of law? The legislation theory seems to run now not top-down, but bottomup; not descending from a central authority which governs by the scriptural power to make norms, but from a network of private authorities which create infinite differentiation. Is there no typification in mass customization? Or, conversely: where the mass is, there lies the law in the shadow of typification?

Nonlinear Dynamics and Legal Theory, 2001, 68 Philosophy of Science, 455–466. About legality, “alegality” and chaos see Lindahl, Fault Lines of Globalization. Legal Order and the Politics of A-Legality, 2013, 184–186. 38 Busch, The future of pre-contractual information duties: from behavioural insights to big data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, 2016, 221–240. 39 Parry, The Sanctity of Contract in English Law, 1959, 1 et seq. 40 For a reflective use of the category, Radin, The Deformation of Contract in Information Society, 2017, 37 Oxford Journal of Legal Studies, (3), 505–533. 41 Kessler, Contracts of Adhesion – Some Thoughts About Freedom of Contract, 1943, 43 Columbia Law Review 640. 42 Rachlinski, Bottom-Up versus Top-Down Lawmaking, 2006, Cornell Law Faculty Publications Paper 918. 43 Porat/Strahilevitz, supra Part 1.A pages 9–17; Kobayashi/Ribstein, Law’s Information Revolution, 2011, 53 Arizona Law Review, 1169‐1220. 44 Ben-Shahar/Schneider, The Failure of Mandated Discourse, 2011, 159 University of Pennsylvania Law Review, 647‐749. 45 McGinnis/Wasik, supra (fn. 34), at 1039.

162

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

V. Nomogenesis at the intersection point between normative technique and informational limit An answer to the previous question must be sought in the intersection between 31 normative technique and informational limit. This point of intersection is the genesis of the normative.46 The theory that the law is concerned only with the average behaviour stems from the epistemic limit. The tendency to the average is not the essence of law,47 but only a contingent consequence of the epistemic limit. The limit moves (because the costs of epistemic transactions decrease: where ‘epistemic transaction’ means the act by which each decision-maker access to the processed information), the quality of regulation increases. The type has played an important role in dealing with the epistemic limit in the pre- 32 digital era. As Winfried Hassemer exemplarily writes: “The type transcends the system where it is defined, for it refers to the reality outside this system. This reference is peculiar, because the profile and the substance of the reality, which it refers to, are not available before the reference occurs, but are created alongside the reference itself. In this way, it is precisely the type that produces what is a necessary condition for its understanding.”48 Typification is a form of movement of the categorial thought (a movement of self- 33 deconstruction); the dynamic norm is the ultimate form of this metamorphosis of nomogenetic structures. From this point of view, typification is only a primitive function of a dynamic-primitive norm, because, in the norm defined with the type, variations in meaning are neither expressly indicated neither ruled neither formally recognizable. Type is a tool to manage a certain complex of information. In the process known as 34 rulification of standards (principles)49 a series of typifications out of legal precedents and typical cases created by scholars, overcomplicate the simple provision of the standard, turning it into a catalogue of possible solutions. The interpretation of the standard is reduced (it changes) into the case being traced back to a point in the “catalogue”. The rulification of standards by (sub-)typification of discrete areas of intervention for the standard itself is a sign of an intermediate process of adaptative morphogenesis of the legal system to the increasing information.

46 Norms do have informative function [McAdams, The expressive Powers of Law: Theories and Limits, 2015, 136–168] and state legislation is not the only regulative force: Ellickson, Forceful Self-Help and Private Voice: How Schauer and McAdams Exaggerate a State’s Ability to Monopolize Violence and Expression, 2017, 42 Law & Social Inquiry (1), 49–59. Law’s informative function, however, is not the quality of law and law only. The informative function depends on the amount of information: if the information (and the ability to process it) is too much or to little, so legal norms become unfit to make decisions. 47 On the contrary, Betti, supra (fn 13). 48 Hassemer, Tatbestand und Typus. Untersuchungen zur strafrechtlichen Hermeneutik, 1968, 112 (my translation). 49 Schauer, The Tyranny of Choice and the Rulification of Standards, 2005, 14 Journal of Contemporary Legal Issues, 803–814; Id., The Convergence of Rules and Standards, 2003, New Zealand Law Review, 303–328. The rule is not the possible applicative results of the standard, but the necessary element for its enforcement.

Femia

163

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

VI. The loss of informational innocence 35

36

37

38

39

For centuries, law has worked as if there was a single central unit (the common, the sovereign-divine will, the Volksgeist, the consensus omnium, the consuetudo, the people as a political body, the legislator embodied in Parliament) which processes information. The legal system has dealt with processed information as they were environment, natural systemic resources; it has considered public opinion as a place for producing and developing information, the place where the elements of several experiences take political form. On the contrary, if information and media hype are themselves intended as products to be subject to regulation, variation and critique, then the issues of legitimization of political power change. Political power can be now seen, firstly, in the assessment/ transformation of production/communication processes (and not only in government processes). While information increases and differentiation occurs (proliferation, molecular growth of producers/diffusers), legal science loses its claim to innocence with regard to the matter of thought. The selection of information itself (the data processed in the development of rules, also, and above all, through private ordering, by contracts) is a matter of regulation (a reflective regulation of nomogenesis). Private legislators (the law firms that elaborate the global boilerplates, for instance), therefore, are liable for the collection, selection and the flow of information, which is built within its own way of thinking; they are also forced to deconstruct any pre-existing category (because all the categories need a cross-fertilization with new informative flows). In order to develop this research programme, it is useful to distinguish between: – Dynamic legal norm (Type) = norm-programme/principle (norms that establish scopes and accurately indicate the single elements that, when changing, trigger a change in the content of the additional norm that must be produced); – Granular legal norm (Token) = norms-result/rules (concrete products of informative selection). The relation between dynamic and granular norms is not logical, but critical. We hardly find the extension of this possible revolution in the forms of regulation; yet, we do understand its dangers very well.50 The digitalization of regulation processes throws the relation between interpretation and production of law into crisis.51 50 The main danger in the sacralization of the hegemonic praxis. See Fischer-Lescano/Christensen, Auctoritatis Interpositio: How Systems Theory Deconstructs Decisionism, transl. M. Everson, 2012, 21 Social & Legal Studies (1), 104 (against Carl Schmitt): “an auto-legitimation for legal praxis through legal praxis”. 51 Gunther Teubner elucidates this in an exemplary fashion: “The strict binary relation 0–1 which in the real world was limited to the legal code in the strict sense of legal/illegal, is now extended in the virtual world to the legal programs, to the whole ensemble of substantive and procedural structures that condition the application of the binary code. This excludes any space for interpretation. Have you ever tried to discuss with your computer the interpretation of its commands? To click or not to click – that is the question when it comes to accept the standard contracts in the digital world. Normative expectations, which traditionally could be manipulated, adapted, changed, are now transformed into rigid cognitive expectations of inclusion/exclusion of communication. In its day-to-day application the internet code lacks all the subtle learning abilities of law. Any informal micro-variation of rules through new facts and new values is excluded. Arguments do not play any rule in the range of code-application. They play their role of course in the programming of the code, but lose their power in the permanent activities of rule application, implementation and enforcement. Thus, informality, as an important countervailing force to the formality of law, is reduced to zero. The digital “code” knows of no exception to the rules, no principles of equity, no way to ignore the rules, no informal change from rule-bound communication to

164

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

In order to address this difficulty of interpretation – which means difficulty of 40 reflection – the legal system must introduce self-reflection into the legal system into itself as a phenomenon of digital-informational constitutionalization. This is the path: (i) Self-critique of the dynamic norm (systematic connection between dynamic norm and legal principles within the legal order). • Critique of the algorithm at the core of the nomogenesis (for example, it is discriminatory, hampers economic competition etc.). • Breakdown of hidden typifications at the heart of the definition of the algorithm [type = form of human interaction/typification = use of recurrent clusters within non typified (yet simply more abstract) human interactions]. (ii) Critique of the granular norm. • Suspension of its application until the end of the critical phase. Or rather: feedback after the application (ex post control of application results). • Assessment of the extent/adequacy of the informative set. (iii) Liability (compensation for damages): • of who has defined the optimum level of information (based on the information on the information: metainformation – liability for negligent metainformation) • of who has stored the information • of who has updated it. (iv) Modification/confirmation (or: invalidation/convalidation) of the granular norm.

VII. Norms on the move The real concept that results from the application of the granular norm might be 41 more or less close to a type. Three categories can help us to understand a contract regulated via granular norms: fluid transitions (German: fließende Übergänge) – transfigured legal transaction (German: nachgeformtes Rechtgeschäft) – hybrid. In the fluid transitions,52 a specific figure moves from one polarity to the other. The 42 two extremes are the two distinct types – their polarity is the heuristic tool to understand the figure of the particular case, which is set at a certain point of the distance between opposite types. In the transfigured legal transaction53 an original figure (a type of legal transaction) 43 has been altered, although the transformation does not necessarily indicate a path from a known polarity to another, known as well.54 The old transforms itself, it is still unknown whether the metamorphosis will lead to the definition of a new type. political bargaining or everyday life abolition of rules. No wonder that such a loss of “reasonable illegality” in the cyberworld nurtures the myth of the hacker, who with his power to break the code becomes the Robin Hood of cyberspace”: Teubner, Horizontal Effects of Constitutional Rights in the Internet: A Legal Case on the Digital Constitution, 2017, 3 The Italian Law Journal, 203–204. 52 See Kluge, Empirisch begründete Typenbildung: Zur Konstruktion von Typen und Typologien in der qualitativen Sozialforschung, Springer Fachmedien, Wiesbaden 2013, 31 et seq. 53 Rabel, Nachgeformte Rechtsgeschäfte: Mit Beiträgen zu den Lehren von der Injurezession und vom Pfandrecht, 1906, 27 Zeitschrift der Savigny–Stiftung für Rechtsgeschichte, Romanistische Abteilung, 290–335; and 1907, 28, 311–379. 54 About “typological operations”, see Lazarsfeld, Some Remarks on the Typological Procedures in Social Research, 1937, 6 Zeitschrift für Sozialforschung, 138: “The word “type” in current social science literature is used either to describe standards developed from one attribute by serial operations or to designate attribute combinations developed from more than one attribute by typological operations. […] The main one is the reduction of an attribute space to a system of types. Three kinds of reduction were distinguished: the functional, the arbitrary numerical and the pragmatic. The latter one is the most frequent and most important in empirical research; its inversion is called substruction. Substruction

Femia

165

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

In the hybrid55 the combination of elements belonging to known categories (concepts or types) is potentially infinite. Here, the dynamic element (the passage, the transition, the metamorphosis) does not matter; indeed, what matters is the static element: mixed legal transactions. Hybridization can even involve hard-core categories: public/private, contracts between opposite parties/company and then networks. Hybridization erodes and/or confirms hybridized categories. 45 It is easier to trace the framework of the hybrids than that of the fluid transitions or transfigured legal transactions. Hybrids do not include consideration of movement: they are considered in a moment when the transformation process has been done, concluded in a finite figure (although this figure, compared to the hybridized categories, is so deviant that it appears monstrous). The other two figures (fluid transitions, transfigured legal transactions), instead, include the movement in their own concept (they are a kind of Hegelian figures – fragments of a phenomenology of the legal transactional spirit which is still to come). Where movement is, it is impossible to establish a framework. The framework should not be framework of the movement (this sets the path, the transition speed, its limits, the persons entitled to create it, recognize and employ it). 46 Dynamic norms are, instead, paradoxical framework on the move; factors of constant hybridization of law. Type is only a tool for framing subject to informational constraint. 44

VIII. Les communications & les commerces Type dissolves itself in typification; typification also dissolves itself in a dissemination of dynamic (and granular) norms. All those transitions must be critically thought. 48 At the very beginning of the modern contractual theory, Jean Domat said: 47

“Les règles du Droit sont des expressions courtes & claires de ce que demande la justice dans les divers cas;56 La matière des conventions est la diversité infinie des manières volontaires dont les hommes règlent entre eux les communications, & les commerces de leur industrie, & de leur travail, & de toutes choses, selon leurs besoins.”57 Expressive forms of justice short and clear; infinite substance for communication and commerce, human needs. Verba and nomina sustain each other in law’s informational regulation making. 50 In 1928 the philosopher of science Paul Oppenheim drawn a distinction – in his Die natürliche Ordnung der Wissenschaften – between static and dynamic subsum49

consists in matching a given system of types with that attribute space and that reduction from which it could have originated logically. This substruction of an attribute combination to a given system of types permits one to check the omissions or overlappings in this system and points the way to its practical applications”. 55 Teubner, Digitale Rechtssubjekte? Zum privatrechtlichen Status autonomer Softwareagenten, 2018, 218 Archiv für civilistische Praxis, 196 et seq. 56 Domat, Lois civiles dans leur ordre naturel, Livre préliminaire, Tit. I, Sect. I, I, Jean Baptiste Cognard, Paris 1689, 5. 57 Domat, 1689, Première partie, Livre I, T. I, Sect. I, III, 64. See Première partie, Livre I, T. I, Sect. I, VII, 66: “De ces différentes sortes de conventions, quelques-unes sont d’un usage si fréquent & si connû par tout, qu’elles ont un nom propre: comme la vente, le loüage, le prêt, le dépôt, la société, & autres: & il y en a qui n’ont pas de nom propre, comme si une personne donne à quelqu’un une chose à vendre à un certain prix, à condition qu’il retiendra pour luy ce qu’il pourra en voir de plus. Mais toutes les conventions soit qu’elles ayent, ou n’ayent point de nom, ont toûjours leur effet, & elles obligent à ce qui est convenu” (idibem, note t: “Natura enim rerum conditum est, ut plura sint negotia, quàm vocabula”).

166

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

tion.58 This work prepared the well-known study (written with Carl Gustav Hempel) about the concept of type, published in 1936. In this book about typology, the internal division within the concept of subsumption (static/dynamic) turns in a radical differentiation in two functions (or conceptual forms), corresponding to classification (aut…aut, either…or) and graduation (more or less). “Unsere Untersuchungen sind durch Überlegungen angeregt, die P. Oppenheim in einer früheren Veröffentlichung angestellt hat. Dort wird eine Unterscheidung zwischen “statischen” und “dynamischen” Subsumptionen eingeführt, die eng mit der oben erwähnten Unterscheidung von klassifikatorischen und abstufbaren Begriffsformen zusammenhängt; statische Subsumptionen sind nämlich solche, die nach dem klassifikatorischen Schema “entweder-oder“, dynamische solche, die nach dem Schema “mehr-minder” erfolgen. Oppenheim hat in seinem Buch unter den Hinweis auf die methodologische Bedeutung der dynamischen Subsumptionen eine eingehendere Untersuchung dieser Unterscheidung als wünschenswert bezeichnet und dabei auf die mathematische Logik als geeignetes Hilfsmittel verwiesen. − Eine solche eingehendere Untersuchung wird nun in der vorliegenden Schrift am Beispiel der typologischen Begriffsbildung durchgeführt.”59 In this passage, we can read the starting point of development of the 20th century 51 (continental?) legal theory, from subsumption to balancing60 and beyond. More than other sciences, law works with types: typification precedes every definition; typification, also, occurs in every act of subsumption.61 Image or sign; type and then concept: by 58 Oppenheim, Die natürliche Ordnung der Wissenschaften, 1926, 221 ss. Oppenheim distinguishes two art of subsumtion and then two definitions, static and dynamic: “Einerseits hat man eine scharfe Definition für ein Ding; andererseits ist die Subsumption unscharf. Letztere Tatsache steht also nicht im Widerspruch zu ersterer. Man darf demnach eine Definition nicht lediglich deshalb ablehnen, weil die Subsumption zu Zweifelsfällen Anlaß gibt, mit anderen Worten, weil die Subsumptionsgrenzen fließend sind. Wir nennen deshalb diese Art von Subsumption zum Unterschied von der statisch dynamisch” (p. 222, italics instead of original letter spacing). 59 Hempel/Oppenheim, Der Typusbegriff im Lichte der neuen Logik, 1936, 8: “Our inquiries are stimulated by the reflections that P. Oppenheim has presented in a past publication. Here is introduced a distinction between static and dynamic subsumption that closely attach itself to the classificatory and gradual conceptual forms; the static subsumptions are, in fact, those that follow the classification schema “either...or”; the dynamic the ones that follow the schema “more...less”. In his book Oppenheim speaks for a more in-depth analysis about the methodological sense of the dynamic subsumptions and therefore indicates the mathematical logic as optimal tool. Such a detailed inquiry in our work is developed through the model of typologic formation of concepts” (my translation). See, too, p. 79: “Die Bevorzugung von Reihen, die durch zwei “Pole” begrenzt sind und nicht (wie z. B. die den meisten physikalischen Begriffen zugrunde liegenden Ordnungen) nach einer oder nach beiden Seiten ins Unendliche sich erstrecken, erklärt sich zunächst wohl daraus, daß in der Entwicklung der Begriffsbildung die bipolaren Reihenordnungen bestimmte unzulängliche Zweiteilungen ablösen; von der ursprünglichen Aufteilung eines Gebietes in zwei scharf getrennte Klassen bleibt oft noch die terminologische Auszeichnung gewisser, meist nicht genau festgelegter Gruppen “extremer” Fälle als “polarer” Gegensätze erhalten, auf die dann die alten Klassenbezeichnungen übergehen. Auf der Stufe dieser ordnenden typologischen Systeme lassen sich nun zwei Hauptanwendungsformen der Typenbegriffe unterscheiden: eine dem logischen Charakter der ordnenden Begriffsbildung entsprechende ordnende und eine aus der Stufe der einstelligen Prädikatbegriffe übernommene klassifizierende Form” (italics in original). 60 Alexy, On Balancing and Subsumption. A Structural Comparison, 2003, 16 Ratio Juris (4), 433–49. 61 Oppenheim, supra (fn. 58), at 120; “Im Grunde besteht eine Typisierung bei jeder juristischen “Subsumption“, und zwar in doppeltem Sinne: Denn erstens mußte dem Akt der Subsumption die Definition vorausgehen, welche ihrerseits, was keine Erläuterung bedarf, eine scharfe Typisierung darstellt; zweitens finden wir eine nicht minder scharfe Typisierung, wenn der “Fall” der Definition “subsumiert” wird. Diese Tendenz zur Typisierung wird noch dadurch unterstrichen, daß Fiktionen und Analogien zu ihrer Erleichterung herangezogen werden”.

Femia

167

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

those way we think under our limited capabilities to grasp the reality.62 The huge computing power of big data processing makes the balancing metaphor largely too simple to compete with the computational model of infinite differentiation. 52 From “dynamic” subsumption to typification;63 and then, from typification to dynamic/granular rule. The type encourages binary partitions; the algorithm of the dynamic rule transcends this dual structure; it may have a high number of variables. “Dynamic rules can provide a solution to the problem of legislative inertia or regulatory ossification. Dynamic rules are rules that are tied directly to real world empirical data, so that they automatically update as the data to which they are tied changes. Dynamic rules can therefore increase the ability of rules to adapt to continuously changing circumstances rather than await another legislative decision to adapt. […] Rather than requiring legislators or regulators to look at empirical data, using dynamic rules could bypass regulators altogether by placing the collection and analysis of data at the heart of the regulatory system. Instead of setting up a fixed rule or a schedule of rule changes, rulemakers would create an algorithm. The algorithm could be fed information from existing sources of data (such as economic information), or could be fed data from an information-gathering system set up by the rulemaking body. As the data reflecting real-world information changes, the algorithm would alter the rule. Rule makers would oversee the regulating process, but the algorithm would process and update the actual regulations. This process would fit with the trend of regulators moving towards “meta-regulation”: the regulation of the regulation process itself.”64 53

In close relation with the personalized default rule introduced by Cass Sunstein,65 Christoph Busch traces a vivid picture of a ‘learning law’ – and we can point a stimulating analogy with the lernendes Recht model of the system theory, in WillkeTeubner fashion.66 “Personalized disclosure should be conceived as a dynamic and ‘learning’ system in the sense that the content of the information can change over time. In such a dynamic system the relevance of the information can continuously be improved. In addition, the information provided can be adapted to changing circumstances in the life of the consumer and intra-personal changes in consumer preferences. If, for example, the consumption pattern indicates that the consumer is pregnant or has developed an intolerance to gluten, appropriate health warnings can be displayed more visibly than 62 In a later development, under the influence of Rudolf Carnap, Carl Gustav Hempel distinguished classificatory, comparative and quantitative concepts; see Hempel, Fundamental of Concept Formation in Empirical Science, 1952, 54: “But with growing emphasis on a more subtle and theoretically fruitful conceptual apparatus, classificatory concepts tend to be replaced by other types, which make it possible to deal with characteristics capable of gradations. In contrast to the “either...or” character of classificatory concepts, these alternative types allow for a “more or less”: each of them provides for a gradual transition from cases where the characteristic it represents is nearly or entirely absent to others where it is very marked. There are two major types of such concepts which are used in science: comparative and quantitative concepts”. 63 Properly speaking, Oppenheim says, all definition is dynamic, because every subsumption of a real thing is open-ended. “Wahrscheinlich ist sogar streng genommen jede Subsumption von wirklichen Dingen unter einem Begriff dynamisch”: Oppenheim, supra (fn. 58), at 224. 64 McGinnis/Wasik, supra (fn. 34), at 1040 and 1042. 65 Sunstein, Deciding by Default, 2013, 162 U Penn L Rev, 44 et seq. 66 Willke, Stand und Kritik der neueren Grundrechtstheorie, 1975; Teubner, § 242 (Grundsatz von Treu und Glauben), in: Brüggemeier (ed.), Alternativkommentar zum Bürgerlichen Gesetzbuch, Bd. 2 (1980), 32–91.

168

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

before. The learning process of the system could be complemented by asking the consumer for feedback on the helpfulness of the information provided.”67 Learning by progressive typification.68 The norms classify; the classifications 54 differentiate themselves by type; the typifications spread out into taxonomies69 ever more complicated up to a molecular spreading into a myriad of micro-taxonomies: the granular norms works as the application of a finite set of micro-taxonomies to single cases, each of them again recursively typified until the next informational flow occours. To think the Becoming is a tremendous way: logical thinking is movement on a 55 “thought-surface”.70 Recursus ad infinitum produces antinomies; recursiveness is the form of autopoiesis. At the beginning of 20th century, the French mathematician Henry Poincaré reflects 56 about infinite series. “La logique formelle n’est autre chose que l’étude des propriétés communes a toute classification […] et c’est à cela que se réduit toute la théorie du syllogisme. Quelle est alors la condition pour que les règles de cette logique soient valables? C’est que la classification adoptée soit immuable.”71 The use of infinity ends in nonsense:

57

“Est-il possible de raisonner des objets qui ne peuvent pas être définis en un nombre fini des mots ? Est-il possible même d’en parler en sachant de quoi l’on parle, et en prononçant autre chose que des paroles vides? Ou au contraire doit-on les regarder comme impensables. Quant à moi, je n’hésite pas à répondre que ce sont de purs néants. Tous les objets que nous aurons jamais à envisager, ou bien seront définis en un nombre fini de mots, ou bien ne seront qu’imparfaitement déterminés et demeureront indiscernables d’une foule d’autres objets ; et nous ne pourrons raisonner congrûment à leur endroit, que quand nous les aurons distingués de ces autres objets avec lesquels ils demeurent confondus, c’est-à-dire quand nous serons arrivés à les définir en un nombre fini de mots.“72 The cognitive openness of type is rulified in the algorithmic sequences of dynamic 58 norm. The infinite is closed into system – it is its recursiveness: the infinite processing of unending infrasystemic normogenesis. The algorithm is here a sequence of clusters, each of them elaborate a type of data 59 and influences the final (in our terminology, ‘granular’) norm. But this number, so high as it may, is even a finite ones. Again, as Roman law Digest teaches us, we have more negotia than vocabula. Who use the algorithm, has the power. Constitutionalized private law counterbalances this new strategic form of private ordering. 67

Busch, supra (fn. 38), at 235–236. Collier/LaPorte/Seawright, Putting Typologies to Work: Concept Formation, Measurement, and Analytic Rigor, 2012, 65 Political Research Quarterly (1), 217–232. 69 “When only one fundamentum divisionis is considered, a classification scheme is produced […] Classes need not be at the same level of generality, and may be ordered. When several fundamenta are jointly considered, a typology is produced. [...] When several fundamenta are considered in succession through a series of intensional classifications, a taxonomy is produced”: Marradi, Classification, Typology, Taxonomy, 1990, XXIV Quality and Quantity (2) 129. 70 Oppenheim, Die Denkfläche. Statische und dynamische Grundgesetze der wissenschaftlichen Begriffsbildung, 1928, 33. 71 Poincaré, La logique de l’infini, 1909, 17 Revue de métaphysique et de morale, 461. 72 Poincaré, supra (fn. 71), at 478. 68

Femia

169

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

Typification and big data processing in granular legal norms amount to an internal transcendence of the legal system. Type is now not the opposite of concept; instead, it is the brick which the internal processes of algorithmic shaping of granular norms are made of. The granular legal norm, therefore, are highly individualized but not apt to a single act of decision. Even the granular norm does have inner indefinite elements: types. This indefiniteness is its salvation. 61 The contractual communication spreads overall; but overall is constrained by many private agencies of regulation. Big data revolution shows dramatically the clash between worldwide informational powers and the highly fragmented everyday life of each of us. Private regulators tracks (or can acquire another’s tracking) individuals: the processing of information is thought to improve the quality of goods or services which are offered to consumers, but its root cause is after all the economic growth, the profit of the companies enforcing the mass customization. That is why the granular norm must be first examined from the perspective of its impact (potential and real) on the ways of contracting. And here the need to imagine precontractual liability, as well. 60

IX. Politics or Algorithmics 62 63

64

65

66

The main problem lies, however, in the change of the channels of political communication. Traditionally, the epistemic limit which denotes legislation (norms are written because sufficient information will never be available to always decide on a case by case basis, each time anew) has the form of a prohibition of the algorithmic metaregulation of the informative processes which underpin the development of norms. According to this received scheme, the information at the heart of law’s algorithm flows freely in the open field of social communication and no one, but the Sovereign, may select them. All this depends on the assumption that law is a form of politics – politics is not the existential decision of the community, it is not the decision on the style of life of citizens which occurs in material conflict resolution (this is called jurisdiction), but it is the abstract and general decision, the provision of scopes in the form of rules, the semiotic strengthening of values through the position/scripturalization of principles. Law as a form of politics means basic equivalence of political with legislative power (and radical scandal involving any political assessment in the administration/application of law). The algorithmic form, introduced into the dynamic regulation, radically restricts legislative power: while making the dynamic rule, the algorithmic form imposes responsibility upon who defines constants and variables, classes of possible solutions, connecting factors between solutions and variations, minimum and maximum levels of variation, units of measure that subdivide the spaces between minimums and maximums into finite parts; it also imposes the accountability for monitoring the adequacy, reasonableness and proportionality of results. Dynamic regulation is the ultimate accomplishment of legislation transforming itself into administration. We have seen that counterbalance lies in the critique of data processing. Consumers must be allowed to discuss the criteria for selecting and assessing elements that produce normative meaning in practice:73 thus, not only the result (the granular norm), but also the matrix (the dynamic norm). Once again, the political guarantee is based on technical decisions. 73 Femia, Vertragsbeobachtende Verträge, in: Lomfeld (ed.), Die Fälle der Gesellschaft. Eine neue Praxis soziologischer Jurisprudenz, Mohr, Tübingen 2017, pp. 151–162.

170

Femia

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/06_P. Femia, Logopoeia- Normative typification and granular norms informational differentiation. 6.3.2019.3d from 30.10.2020 11:35:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

F. Logopoeia

Technique – borrowing the words form Ezra Pound – is “the trampling down of 67 every convention that impedes or obscures the determination of the law, or the precise rendering of the impulse”.74 Our efforts for the constitutionalization of dynamic/granular nomogenesis is the 68 composition of law and impulses. Therefore, law’s gaze at the normative dissemination, at the metamorphosis of typification techniques, at their infinite fragmentation into a multitude of variables, each of these being capable of causing a different normative content, is not a fight with words (logomachy), but a production with words (logopoeia). Language – Ezra Pound again – “is a dance of the intelligence among words and ideas and modification of ideas and characters”.75 An unending chain of sense: logopoiesis, autopoiesis, constituzionalization. Nomina, 69 of course; but activities, but constructions for freeing typification out of repetition compulsion. Novelty through type, not sameness. 74 Pound, Prolegomena, 1912, The Poetry Review, February, 73: “Technique. – I believe in technique as the test of a man’s sincerity; in law when it is ascertainable; in the trampling down of every convention that impedes or obscures the determination of the law, or the precise rendering of the impulse”. 75 Pound, Others, 1918, The Little Review, March, 57: “Logopoeia or poetry that is akin to nothing but language, which is a dance of the intelligence among words and ideas and modification of ideas and characters”.

Femia

171

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies: Liberal, Neoliberal and Socialist Perspectives I. Granularization: a consistent outcome of a neoliberal trend 1

2

3

4

5

6

Classical liberal law was established around the notion of general and abstract rules as opposed to the law of the ancien régime with its personalized norms and its chaos of local practices.1 Classical liberal thought prized general rules able to entrust every human being with a right to freedom insofar “(…) as it can coexist with the freedom of every other in accordance with a universal law”.2 The positive characteristics of an ideal liberal legal system revolve around two poles: universalizability, which evokes especially Kant’s thought, and calculability, which is the feature exalted by Max Weber. Contemporary legal thought, especially that inspired by the economic analysis of law, celebrates rules boosting efficient behaviour, with the aim of maximizing overall welfare. Given that each situation, examined from the viewpoint of its capacity to produce welfare, has its own peculiarities, rules tend to be specific, rather than general. This tendency towards an increasing fragmentation of rules, is supported by the centrality attributed to the phenomenon called “transaction”, broadly defined as “(…) the transfer of a good or service across a technologically separable interface”3 and considered as the typical means by which individuals can increase their well-being (without, hopefully, harming others). In the resulting vision, which we qualify as neoliberal, the whole reality is construed as a mere sum of interactions (more or less voluntary or involuntary) between unencumbered individuals. As the “common good” is identified with the maximization of overall welfare, and the main tool for increasing welfare is identified with the implementation of efficient transactions, the success of private transactions is assumed to be coincident with the achievement of the common good. The purpose of law, in this neoliberal vision, can be substantially limited to that of making feasible, and possibly encouraging, efficient interactions (transactions) whilst discouraging the others. Rules are consequently shaped according to what is needed for the success of the corresponding transactions, and the legal system inevitably moves toward fragmentation, because, in the end, each transaction has its peculiarities.4 1 In 1789, Mirabeau remarked that France was an “aggregate of divided people” (Crettez/Deffains/ Musy, Legal Centralization: A Tocquevillian View, 2018, 47 The Journal of Legal Studies, 295). 2 Kant, Die Metaphysik der Sitten, 1797, 237. 3 Williamson, The Economics Institutions of Capitalism, 1985, Free Press. A transaction is not necessarily voluntary so it appears to refer substantially to any economically relevant interaction. 4 Often the focus seems to be on the interacting subjects (their ability to pay attention, their skilfulness, the levels of negligence of which they are capable, etc.) rather than on the transaction. The attention apparently paid to the parties rather than to the transaction, does not conflict with the illustrated tendency to assume the transaction as the elementary datum from which to start. In fact we do not observe a valorisation of the general and fundamental characteristics of the persons (their autonomy, their Kantian ability of Willkuer and Wille, their overall dignity, their needs, etc.) but rather a concern exclusively focused on the attitudes that the subject exhibits with reference to a specific transaction. The discipline applicable to, say, a consumer transaction, is not based on an analysis of all the characteristics

172

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

The neoliberal vision differs not only from the Fordist-Keynesian vision that historically preceded it, but also from the classical liberal vision. The different concept of efficiency and of the relationship between state and market is especially meaningful. Nineteenthcentury classical liberals predicated the separation of the respective spheres of competence between the state and the spontaneous market order, and thought that whole system’s efficiency could be guaranteed, at a general level, by the operation of the “invisible hand”. The liberals of the Fordist-Keynesian period thought (with obvious and clamorous differences according to the periods and the different currents of thought) that the limits of the market can be overcome by cunningly steering the great economic aggregates. The neo-liberals on the other hand believe that problems of market malfunctioning can be solved by the governance of the single transactions. Resorting to a simple comparison we could say that Fordist-Keynesian liberals believed that markets, spontaneously clearing, or appropriately guided by a wise public intervention, are able to basically ensure fair bargaining and efficient transactions, whereas the neo-liberals overthrow this relationship and argue instead that the important thing is to facilitate the implementation of any possibly efficient transaction. The idea is that a sum of particular efficiencies can only result in global efficiency. In financial markets, the management of credit and money supply, once considered as the main goal, has been displaced by the governance of investor-intermediary transactions (MIFID and so on); the control of total consumption has been replaced by the meticulous regulation of the transactions of individual consumers with individual merchants (the various directives on consumer rights). The enormous power of managers (with its dubious legitimacy) that so troubled Berle & Means,5 is no longer a source of concern, and the attention has been shifted to the transaction between shareholders and managers concern for the overall structure of the markets, which worried Mason and Bain,6 disappears, and the focus has turned to the single practices and “transactions”, examined no longer from the viewpoint of their overall effects on market structure, but from that of the presumed effects that each, considered in isolation, can have on the so called consumer welfare. Transactions are considered as the basic unit of analysis and the so called transaction costs as the theoretical operating tool for classifying, distinguishing, aggregating, etc., the various types of transactions, in order to design rules suitable for overcoming the costs of each type. This idea explains, among other things, the progressive fragmentation of the legal system and the multiplication of increasingly special disciplines linked to the specificities of individual transactions. The same idea encourages a further personalization of the rules. Now that the use of big data promises to eliminate or, at least, greatly reduce, information cost (considered as the factor preventing an almost infinite multiplication of rules) this neoliberal trend actually possessed by the subjects that stipulate it, but on inferences about what an average transaction on a consumer market may imply in terms of features of the deal and of the individuals involved in it. The same subject (e.g. a plumber) is in many respects subjected to very different disciplines, depending on whether he makes deals with consumers, another plumber, a large company, a bank or a financial intermediary, etc. The polar star guiding all reasoning is always the success of the single transaction (achievement of the maximum gain at the lowest costs). 5 Berle/Means, The Modern Corporation and Private Property Macmillan, 1932, New York. On general features of the neoliberal legal “style” see Denozza, Conclusioni: lo sitle giuridico neoliberalòe e ilsuo superamento, in Sacchi e Toffoletto (eds) Esiste uno “stile giuridico” neoliberale?, Atti dei seminari per Francesco Denozza, Giuffrè, Milano, 2019. 6 Bain, Economies of scale, concentration, and the condition of entry in twenty manufacturing industries, 1954, The American Economic Review, 44(1), 15–39; Mason, The current status of the monopoly problem in the United States, 1949, Harvard Law Review, 62(8), 1265–1285; Mason, Market Power and Business Conduct: Some Comments, 1956, The American Economic Review, 46(2), 471–481.

Denozza/Maugeri

173

7

8

9

10

11

12

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:00 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

towards particularization seems to find new possible developments, and new areas of application. 13 Granularization, understood as “a shift from impersonal law based on the widespread use of typifications to a more personalized law based on “granular legal norms” that are tailored to the individual addressees”7 might be the next step of the neoliberal obsession with the efficiency of single transactions. 14 A general and abstract evaluation of a possible evolution from impersonal to personalized law is very difficult. The problems arising from rule granularization may be very different in different contexts and the viewpoint from which the phenomenon can be judged may be different too. In any case, in this chapter we make some general comments on the possible “costs” (other than information costs) of granularization (next section), on its consistency with the liberal tradition (section III), and on its efficiency (section IV). We conclude (section V – VI) by examining problems of justice arising from a specific possible application of legal granularization techniques: the case of cross-subsidy (probably the case in which issues surrounding granularization appear in the most emblematic way).

II. The costs of granularization: the many shortcomings of algorithmic governmentality 15 16

17

18

19

Personalizing may have huge costs, even in a world in which (some) information can be obtained very cheaply. Granularization still has to resort to generalizations. Compared to more traditional law, granularized rules simply multiply the different groups of individuals to which the rules apply and the characteristics defining the groups. This increased selectivity in the application of rules relies on an increasing number of factual connections established (thanks to Big Data) between certain visible characteristics of given individuals (or situations) and some relevant qualities of the same individuals (or situations). These connections are mostly based on statistical generalizations. Resorting to statistical generalizations in order to define the facts to which a given rule applies may create problems. This is especially true in cases in which available data are unable to establish that (not just the majority) but all individuals possessing certain characteristics, and therefore belonging to a certain group, actually possess also the relevant quality (a quality therefore attributable, at the statistical level, to a group as a whole, but not necessarily to every member, as in the case of the presumption that women have a physical strength inferior to that of men, or other similar presumptions).8 Problems become even greater in the case of big data, which usually show relations not of causation but of correlation. Statistical generalizations may be, depending on the circumstances, objectionable even where based on causal relations, but they become much more so, where based on simple correlations. The legitimacy of generalization based not on causation but on simple correlations, may appear highly questionable in many contexts.9 7 Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in Mak/Tai/Berlee (eds.), Research Handbook on Data Science and Law, Edward Elgar 2018, 409. 8 Another example: not all airline pilots suffer decreased alertness when they pass the age of 60. Thus a rule which compels all commercial pilots to retire at that age might be questionable; see Schauer, Generality and equality, 1997, Law and Philosophy, 16(3), 279, 283. 9 In this sense especially Feinberg, The Moral Limits of the Criminal Law: Harm to Others, 1984. Oxford University Press, 201. For a critique of the Feinberg’s thesis, Rüegger, La discrimination statistique entre pertinence et arbitraire, 2007, Revue de philosophie économique, 8 (1), 73, 86.

174

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

An example can help illustrate the point. Take the prohibition of driving under the influence of alcohol. This may be considered a statistical generalization based on causation. It is a statistical generalization because we know that the majority of those who drive under the effect of alcohol are more accident prone, but we are not absolutely sure that all drunks drive worse than sobers. It is however a generalization based on a causal connection because we know that alcohol abuse has short-term effects which include vision impairment, lack of coordination, memory lapses, etc. Factors that all have a causal influence on driving ability and that we know for sure are at work in any case of drunkenness. The prohibition is therefore based on the fact that all members of the group share a characteristic (being drunk) able to cause the undesired effects.10 Correlation, on the other hand, connects two variables having only a linear association with each other.11 Imagine for instance that Big Data inform us that car accidents are often caused by people who have just watched a western movie. Should we consider this information a valid basis for legitimizing a ban on driving after watching “Stagecoach”? From the viewpoint of the legitimacy and effectiveness of legal rules the difference between correlation and causation must not be underestimated. For example, statistical studies observe a correlation between wealth and driving abilities12 (it seems that rich people drive worse than poor people) or between obesity and debt delinquency (estimated as up to 20 % higher among the obese than the nonobese).13 May we infer that a driving ban for the wealthy, or on obtaining credit by the obese, is equally justifiable and appropriate, as the prohibition of driving drunk? The obvious difference is that in the cases of the wealthy or the obese the prohibition would affect individuals whose characteristics are not the cause of the result we want to avoid (respectively bad driving and debt default) and are correlated to it only by statistical associations casually observed in diverse unchecked and unconsidered contexts. The risk of errors is multiplied by the fact that we can have an infinite set of more or less nonsensical14 or spurious15 correlations, which is not the case with causation Other problems arise when individuals can easily adapt their behaviour to the needs of preventing unwelcomed correlations. People aware of the fact that police found a correlation between crime rates and tattoos, may choose not to get tattoos in order to avoid exposure to suspicion. Whereas, people not aware of this fact shall bear unknown and unintended consequences of their decision to get tattoos. In both cases their freedom is severely restricted.16 10

Maitzen, The ethics of statistical discrimination, 1991, Social Theory and Practice, 17(1), 23. As is well known, correlation does not imply, and does not prove, causation. Inferring causation from correlation is rather considered as a logical fallacy (cum hoc, or post hoc, ergo propter hoc). 12 Preston, The Rich Drive Differently, a Study Suggests, The New York Times, 12 August 2013. 13 Guthrie/Sokolowsky, Obesity and Household Financial Distress (14 October 2014). Available at SSRN: https://ssrn.com/abstract=1786536. 14 Such as the high positive correlation between birth rate and number of storks for a period in Britain or the negative correlation between birth rate and road fatalities in Europe over a number of years, see Haig, Spurious correlation, in: Encyclopedia of Measurement and Statistics Thousand Oaks, 2007 Sage. 15 Moosa, Blaming suicide on NASA and divorce on margarine: the hazard of using cointegration to derive inference on spurious correlation, 2017, Applied Economics, 49(15), 1483. 16 Obviously a similar adaptive behaviour can be induced also by rules based on genuine causal relationships. Prohibition of driving drunk limits my freedom of drinking alcoholic beverages. Apart from a judgment of opportunity (preventing people from drinking may be considered more appropriate than preventing people from getting tattoos) the difference is that the amount of causal relationships that can be established is limited by the natural laws of which causality is concretization. As noted, the possibility of finding correlations is instead almost infinite. So, given that a correlation has been 11

Denozza/Maugeri

175

20

21 22

23 24

25

26 27

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 28

29

30 31

32

33

34

35

36

Another problem arises from the fact that increased personalization inevitably implies an increased number of specific provisions, increased difficulties in distinguishing cases and in subsuming each under the appropriate provision and, ultimately, increased legal uncertainty. The historical experience to date is not reassuring with respect to the idea that detailed regulation of each situation17 could be the right path to increase certainty and calculability (in a Weberian sense). No attempt to anticipate clear regulation for every situation has ever come even remotely close to success. A web of many detailed rules, each taking into account the peculiarities of a group of cases, can create more uncertainty than a system with general rules equally applicable to an entire community. Does Big Data offer a chance to change this secular trend? We do not think so. The uncertainty of the real world is not computational, as it is for example in chess where the context (the set of the rules, the positions of the pieces, etc.) is well known and the problems arise from the difficulties of calculating all the possible schemes. We live in an uncertain context, whose laws largely elude us. We have to cope not with limitations of human computational ability, but with the fact that any estimation of future consequences is beset by uncertainty we do not think that big data will radically change this situation. Cases of prediction failures based on information collected via big data are far from negligible. Probably the most famous failure is “Google Flu”, but we think that also the sub-prime mortgage crisis provides some important lessons. The complex methods used for weighting and calculating the risk attached to mortgage bonds, far from providing exact forecasts, fuelled the speculation that led to the crisis. One of the causes is probably that in a social, as distinguished from a natural, environment, complexity and uncertainty are continuously increased by the very human action the effects of which we try to predict. The problems arising from the uncertainty of the environment combine with the problems arising from the opportunism of individuals. In a world in which individuals are caged in a set of personalized rules that regulate every detail of their life, a loophole seeking mentality can be predicted to be very common. All individuals will try to exploit legal uncertainties, created by the complexity of a continuously changing world, aiming to profit in their own interest. We also have to consider that the solution of many relevant problems requires information and evaluations that even big data are unable to provide. This difficulty represents both a problem in itself, and a factor that might undermine the sense and legitimacy of projects of rule personalization. For example, from the viewpoint of improving deterrence, differentiation can appear, in principle, very useful. Each individual has their own characteristics, and sensitivities, and normally what may be a strong deterrent for some people, may not necessarily deter another at the same level. If we could personalize all elements on which deterrence depends, we could reach a great result. Problems, especially of legitimacy, immediately arise where we are not able to implement a generalized and reasonable consideration of each relevant characteristic. Everyone can justifiably ask why a negative (for that person) differentiation has been established between margarine consumption and divorce rate in Maine, if I want to look like a reliable prospective husband to a (living in Maine) fiancée, should I avoid consuming margarine? 17 These problems are anything but new and we can recall what Savigny (The Vocation of Our Age for Legislation and Jurisprudence, The Lawbook Exchange, Ltd., 2002, 38) said about codification: “this undertaking [deciding each case by a corresponding provision of the code] must fail, because there are positively no limits to the varieties of actual combinations of circumstances.”

176

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

carried out, whilst another feature that could imply a positive differentiation has not been considered. For example, a skilful but poor driver could maintain that, if we want to improve deterrence, long before thinking of a complicated personalization of the required levels of care, we should start with the much simpler task of differentiating punishment, for example in accordance with the wealth of the culprit. In fact, we can easily suppose that a fixed pecuniary fine discourages rich people much less than poor people. Therefore, we should start by increasing the punishment for unskilled wealthy drivers, rather than increasing the level of care required by the skilful poor. The unskilled rich driver could however object that some poor people may be 37 judgment proof and that imprisonment or physical punishment are more egalitarian sanctions. But even physical punishment does not guarantee an adequately modulated deterrence. We know that the same period spent in prison can cause very different suffering to different people. The point is that delivering a punishment for each individual that would induce the 38 exact level of desired deterrence, faces the challenge of accurately measuring the effects that a certain punishment can have on a certain individual and therefore of establishing the consequent level of deterrence. We can easily imagine that a given fine deters poor people more than rich people, but we do not know how much, and in proportion to what. Income, wealth, greed, avarice etc.? Measurement of each of these factors may be very problematic. In this context a project of personalization advocated in the name of a more 39 articulated and effective deterrence, and yet limited to the only factors believed measureable, would have little credibility and legitimacy. Every disadvantaged individual would legitimately protest against the fact that other individuals escape the personalization that they must undergo.

III. Liberal general principles v. neoliberal “granularized” rules In a (classical) liberal vision, rules must chiefly act generally – that is impersonally – and abstractly – that is for an unknown number of future instances, regardless of the specific features of each case. Rules should relate to individual conducts, not social states, and their generality should enables us to live our lives in accordance with our plans as responsible persons. In this sense generality contributes to our autonomy. From a different viewpoint, a stable legal system, with few general rules, makes an important contribution to the predictability of the state’s action and the behaviour of our fellow citizens, thereby making rational calculation possible. More recently, contemporary liberal thinkers have underlined other virtues of general and abstracts rules. Hayek, for example, exalted the use of dispersed knowledge that is made possible by a system of rules establishing few general prohibitions and otherwise leaving individuals free to employ their personal knowledge in search of alternative ways to solve their problems.18 It is evident that detailed rules, aimed not at prohibiting, but at conforming the behaviour of individuals to certain imperatives, reduce the set of possible choices and consequently the opportunities to exploit personal knowledge. Another factor which has been considered is divisiveness. With relative positions clearly identified, different participants may place differing evaluations on alternative sets of rules, and agreement may become impossible.19 18

Hayek, The use of knowledge in society, 1945, The American economic review, 35(4), 519–530. Buchanan/Congleton, Politics by principle, not interest: Towards nondiscriminatory democracy, 2006, Cambridge University Press. 19

Denozza/Maugeri

177

40

41

42

43

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives 44

45

46

47

48

49

The assumptions and values underpinning the appreciation of a legal system with said features, imply a specific picture of the individuals and of the social environment. The conception of the legal subject is that of a Kantian autonomous self and of a Weberian rational actor. The social environment is that of competitive markets, well regulated by the iron laws of supply and demand. The overall picture is that of a homogeneous society with standardized transactions and human interactions that can be generalized and abstracted into rules”.20 This image certainly does not fit well with the features of current society and is no longer acceptable. Therefore the question arises: are classical liberal values (or at least some of them) no longer valid in our variegated society and are we ready to accept even those consequences of rule granularization that are openly in conflict with them? For instance, some personalization based on statistical correlations could have the paradoxical effect of resurrecting something akin to the relevance of status. The fact of being born with certain characteristics (obviously no longer of nobility but, perhaps, biological) determines the special rules to which the individual is for a life time subjected. Are we ready to accept this sort of “regression” (instead of advancing “from status to contract” it would be a backward path “from contract to status”)? Other obvious problems arise in cases of profiling based on correlations disparately connected with age, race, sexual orientation or other characteristics and behaviours included among the so called “protected characteristics”.21 Personalization can often imply an unequal treatment which is formally based on “innocent” correlation, but in fact leads to a discrimination disproportionately damaging individuals belonging to a protected group.22 These issues urge reflection on a possible general argument against the project of exploiting cheap information to substitute general rules with differentiated treatments, arising from the basic liberal idea that not every piece of information can legitimately be used to distinguish among different individuals.23 Supplying persons with an authentically respectful treatment, may imply abstaining from evaluating some of their variable capacities. “In some even if not in all respects the just society is one in which differences among individuals are often and desirably suppressed in the service of both equality and community”.24 This argument (strictly connected with a liberal concept of justice as impartiality) provides a principled objection against a proliferation of differentiated treatments. Each piece of information used for differentiating among individuals should be carefully 20 Horwitz, The transformation of American law, 1870–1960: The crisis of legal orthodoxy, 1992, Oxford University Press, 219. 21 In every liberal legal system rules are present preventing any discrimination based on some criteria. The criteria may not be the same in each legal system, but usually discriminations based on gender, race, religion etc., are considered objectionable, not only in the case of direct discrimination against people who have “protected characteristics”, but also when apparently neutral criteria have the effect of disadvantaging people who share these characteristics. 22 “Big Data platforms enable racial profiling in subtle and invisible ways by targeting home address and other characteristics as a proxy for race”. Gumbus/Grodzinsky, Era of big data: Danger of discrimination, 2016, ACM SIGCAS Computers and Society, 45(3), 118.; Newman, How Big Data Enables Economic Harm to Consumers, Especially Low Income and Other Vulnerable Sectors of the Population, 18 Journal of Internet law, 11 December 2014. 23 Carter, Respect and the Basis of Equality, 2011, 121 Ethics 538. 24 Schauer, Profiles, probabilities, and stereotypes, 2009, Harvard University Press, 300. Or, in different words: “Antidiscrimination policy is not only about assuring equal treatment to equals, but also about assuring that specific differences among individuals should be ignored”, Zarsky, Understanding discrimination in the scored society, 2014, 89 Wash. L. Rev., 89, 1375,1382.

178

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:01 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

scrutinized from the viewpoint of the respect due to persons and to their freedom of exhibiting qualities, and of making choices, disapproved by others. In this perspective a thorny question is that of the importance given to self-owner- 50 ship. In many cases the use of some information for better differentiating among people with different characteristics may end up in a sort of “expropriation” of their qualities or at least of the freedom to pursue the economic advantage that derives from their individual traits. For example, respect of self-ownership might prevent the “expropriation” of the economic advantage that skilled drivers enjoy in a system of undifferentiated standards of care. Concluding on this point, we think that neoliberal proponents of granularization 51 should carefully reflect on the possible illiberal consequences of an unchecked exploitation of information provided by big data.

IV. Is granularization efficient? Abstraction and totality in neoliberal thought As we noted, in classical liberal thought the overall efficiency of the system is produced on a general level by the invisible hand and the related market equilibrium. In the Keynesian-Fordist view efficiency is (also or above all) the result of a state’s interventions which appropriately control and, if necessary, correct the main economic variables. In neoliberal thinking the overall efficiency of the system depends on the efficiency of the individual transactions. As already noted, the idea is that adding together efficient transactions can only produce efficient results. It is obviously not easy to make a generalized statement on whether the neoliberal vision is more credible than the classical or Keynesian-Fordist one. We put forward a couple of observations. On the factual level, we think that the myopic view of transactions as isolated elements, abstracted from the totality to which they belong, has had many negative consequences, causing, among other things, a disastrous misunderstanding of the evolution of the problems of corporations and financial markets25 (thus contributing to the outbreak of the financial crisis) and an equally ruinous misunderstanding of the sense and function of competition and antitrust law26 (thus contributing to a huge increase in inequality). On the theoretical level, we think that the main weakness of the neoliberal approach derives from the distorted perspective engendered by focusing on a specific transaction, abstracted from the totality to which it belongs. The evaluation of single interactions in order to establish which (in the given situation) is the most efficient transaction, requires a cost-benefit analysis that takes into account the alternatives considered as viable and the positive and negative values attached to each. Both elements (alternatives and values) however depend on a much wider context than that visible when observing the specific interaction we are focused on. In this case abstraction leads to myopia: an excessive focus of attention hides the alternatives that a broader view would reveal. In the celebrated Coasean example of the bargaining between farmer and cattleraiser27 the overall efficiency of the result depends not only on the absence of transac25

Denozza/Stabilini, Principals vs Principals: The Twilight of the Agency Theory, 2017, 3 Italian LJ,

511. 26 Denozza, The future of antitrust: concern for the real interests at stake or etiquette for oligopolists? in Orizzonti del Diritto Commerciale, Rivista telematica, 1/2017. 27 Coase, The Problem of Social Cost’(1960), 1988, 3 Journal of Law and Economics, 1.

Denozza/Maugeri

179

52 53

54

55

56

57

58

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

tion costs, but on a lot of assumptions regarding the global context in which the deal takes place. If, for example, we, from a general social viewpoint, take into account health, or ecological, costs associated with excessive consumption of meat, the agreement reached by the two neighbours may be beneficial for them, but not for society as a whole. The same may happen if the price of meat is not an efficient one (for example, because there are imperfections in the market). 59 Let us consider another example. If we focus not on the isolated interactions between skilled drivers, unskilled ones, pedestrians, and so on, but on the problem of car accidents in general, we can immediately discover that the main way to reduce overall harm, is not that of boosting the efficient behaviour of each agent and, maybe, increase driving ability at high speeds, but simply that of prohibiting the circulation of cars able to reach a speed higher than a given low level.

V. Granularization and cross subsidy Many hypotheses of proposed granularization have to do with phenomena that can be classified as “cross subsidy”. Cross-subsidy is usually noticed in consumption contexts and defined as the situation arising “ (…) when two consumers impose different costs on a service provider but are charged the same price – and the excess funds from one are used to make up the shortfall for the other”.28 The phenomenon however occurs also in other circumstances29 and can be defined more generally as the situation where individuals in a group are charged at a disproportionately high cost (compared to the benefit they receive) in order to subsidize lower prices for another group. 61 Cases of cross-subsidy are present in a lot of different situations.30 For example, in contract law, where a party (say, the employer) is obliged, by a mandatory rule, to provide something (say, maternity benefit) for which all counterparties (the employees) pay a price (in terms of a wage reduction able to cover, totally or partially, the costs incurred by the employer), despite the fact that some of them appreciate the benefit less, or do not appreciate it at all. 62 Many other private law problems can be read with the lens of cross-subsidy. For example, with reference to tort law, the choice of having a general diligence standard or a standard “finely tuned in each case to fit the individual circumstance of each tort defendant”31 immediately evokes, among others, also problems of cross-subsidy. We can indeed imagine the context regulated by tort law as a situation in which each of us, in order to participate in social interaction (for example: road traffic) and its benefits, has the duty to contribute, according to some criteria (for example: diligence) to a fund to restore harms inevitably created by interferences. In this perspective, the problem, posed 60

28 Brooks/Galle/Maher, Cross-Subsidies: Government’s Hidden Pocketbook, 2017, 106 The Georgetown L. J., 1229. 29 As is well known, problems of cross–subsidy occur in some very dramatic situations, as when life and health are involved, and the possibility of genetic discrimination comes to the fore. Is it fair that all of us contribute to a common health care system to the same extent, despite the fact that some of us are more likely to get sick and therefore burden much more on the overall costs of the system? See Joly/Feze/Song/ Knoppers, Comparative approaches to genetic discrimination: chasing shadows?, 2017, 33 Trends in Genetics,(5), 299; Wauters/Van Hoyweghen, Global trends on fears and concerns of genetic discrimination: a systematic literature review, 2016, 61 Journal of human genetics, (4), 275. 30 Quillen, Contract Damages and Cross-Subsidization, 1987, 61 S. Cal. L. Rev., 1125. 31 Logue/Avraham, Redistribution Optimally: Of Tax Rules, Legal Rules, and Insurance, 2002, 56 Tax L. Rev., 157.

180

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

especially by Ben-Shahar & Porat,32 of personalizing negligence law can be understood as a problem of cross subsidization between skilled and unskilled injurers. If we imagine that the minimization of accidents requires a given amount of diligence and of related costs (we assume here that the amount of costs remains unchanged, thus we do not face problems of efficiency, but only of distribution of a fixed sum of costs), a diligence criterion fixed at an equal level for everyone, benefits people (skilled drivers) who are able to reach this level in cheaper ways, at the expenses of other people (less skilled drivers), who have to spend more to reach the same result. In this case skilled drivers are “subsidized” by unskilled drivers. Appropriately differentiating the duty of care, while keeping the amount of the restoration fund and its costs unchanged, can distribute costs equally and eliminate the subsidization. Usually cross-subsidization raises problems both of efficiency and justice. From an efficiency point of view, we have individuals who enjoy a benefit for which they do not pay the full cost. So there is a potentially dangerous misalignment between the distribution of the benefit and the distribution of the corresponding cost. From another viewpoint, cross subsidy equalizes certain parameters (say, the benefit) but has differentiated effects on others (say, the cost). In the example of the choice of the standard of diligence, a uniform rule equalizes the required level of diligence and differentiates the costs incurred by skilled drivers from that incurred by unskilled ones. A personalized rule equalizes the costs, but differentiates the level. Often a tension arises between formal equality (all the individuals involved pay the same price or pay according to a given criterion which is the same for all) and substantial discrimination (individuals pay the same price but receive different benefits, or pay according to the same criterion which in fact has differentiated results). Therefore, deciding about cross-subsidy is not a matter of choosing whether to apply, or not, a principle of strict equality, but a matter of applying the appropriate justice criteria that, from time to time, establish what the kind of (un)equal treatments are, according to the circumstances, fairer (or, at least, less unfair) than others. So we must give up on resolving the issue of cross-subsidy once and for all. Rather, we need an analysis of the criteria that in each individual case may suggest the pros and cons of cross – subsidy. A preliminary investigation of this subject is oulined in the next and last section of this chapter.

63

64

65

66

VI. What’s wrong, if anything, with cross-subsidy? Before trying to explore the pro and cons, from a justice viewpoint, of the proposed 67 increase in the personalization of rules, especially in order to prevent cross-subsidy, we point out that many cases are excluded from our examination. First of all, cases in which the problem concerns not the justice of the end (the way people are differentiated), but the opportunity of the means (such as where cross subsidy is contrasted with taxation)33 and cases in which cross subsidy is criticized not as such, but because, in the given circumstances, it produces results inconsistent with reasonable ends (as in the cases of so called regressive cross subsidy). We also exclude the cases (many of which we 32

Ben-Shahar/Porat, supra Part 1.B. See for example: Posner, Taxation by regulation, 1971, The Bell Journal of Economics and Management Science, 22–50; Brooks/Galle/Maher, Cross-Subsidies: Government’s Hidden Pocketbook, 2017, available at SSRN: https://ssrn.com/abstract=3050674, and Logue/Avraham, supra (fn. 31), at 157, maintaining that some sorts of cross-subsidization are more efficient than the tax‐and‐transfer alternative. 33

Denozza/Maugeri

181

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

68

69

70

71 72

have already mentioned above) where the issue is not (or not only) the result (i.e. the treatment – equal or differentiated – of the two groups of cases) but the criterion used to distinguish a group from the other (such as in cases of dubious correlations and statistical generalizations, where differentiation is based on a quality attributable, at the statistical level, to a group as a whole, but not necessarily to every member).34 Cases of profiling sensitive to protected characteristics like gender, race, religion, etc. are also excluded. As we have already noted, cases like these require consideration of many peculiar factors and can thus be left aside here. In general, arguments against cross-subsidy can be raised, from a justice viewpoint, by stressing the lack of reciprocity that features the positions of the individuals involved. Basically, what characterizes cross-subsidy is the fact that some receive more, and others less, than they contribute. This fact seems immediately unjust if we share a concept of justice understood as a sort of self-interested reciprocity.35 This concept requires that persons participating in a cooperative enterprise contribute and benefit in a proportionate way, viz., in a measure that rational individuals, freely contracting their participation, would consider in their self-interest to accept. Reciprocity is here understood in a sense very close to the sense the term has when used with reference to market exchange. In this vision, relationships in human societies are conceived as a gigantic system of market exchanges, in which individuals should receive from fellow members of society benefits the value of which is more or less equal to the value of what they give to them. The fact that people get much more, or much less (compared to their contribution), appears in this perspective blatantly unjust. Better distinctions, differentiated treatments (in which the reciprocity between what is given and what is received can be better guaranteed) and, therefore, reduction or elimination of cross-subsidy, all appear as tendential improvements, both in terms of efficiency and in terms of justice. To this vision we can compare a different, in a sense socialist,36 perspective. Let’s start with a theoretical critique of the notion of justice as reciprocity. Justice as reciprocity does not consider that the capacity to contribute is socially determined. In order to establish that an individual A contributes X, we must have previously decided that X belongs to this individual. The conclusion that A, having contributed X, is therefore entitled to Y (something more or less equivalent to X) is in fact, long before 34 In this case, at least where we can rely on data with a high level of accuracy, efficiency reasons might justify the generalization, even if it implies an unfair treatment of a small percentage of false positives (for example, the case in which you are not getting your loan under fair conditions, and do pay a higher interest rate because many people in your neighbourhood defaulted, Helbing, Big Data Society: Age of Reputation or Age of Discrimination?, 25 September 2014, available at SSRN: https://ssrn.com/abstract=2501356 or http://dx.doi.org/10.2139/ssrn.2501356. From a justice viewpoint the generalization may however appear, even in this case, objectionable at least according to a justice criterion which rules out the legitimacy of compensating the harm unfairly caused to some people, with a greater benefit secured to other people. If we consider not utility and efficiency, but fairness and rights, resorting to a criterion that is only sound statistically may be highly disputable (Maitzen, supra (fn. 10), at 23–45). 35 Buchanan, Justice as reciprocity versus subject-centered justice, 1990, Philosophy & Public Affairs, 227, 229. 36 We qualify the views exposed in this section as socialist because they prioritize the social aspects of the problems and reject, or radically change the meaning of some notions central in liberal thought (as the notion of reciprocity). However, we do not intend to exclude that conclusions akin to that reached here can be equally supported by some kind of liberal reasoning. For example by appropriately developing the Rawlsian theory of veil of ignorance or the idea of reciprocal altruism (instead of self-interested reciprocity). Zabdyr-Jamróz, The veil of ignorance and solidarity in healthcare: finding compassion in the original position, 2015, Diametros: Internetowe Czasopismo Filozoficzne – Diametros: An Online Journal of Philosophy, (43).

182

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

G. “Granularization” and Cross-Subsidies

judging equivalence, based on an implicit judgement about the fact that X belongs entirely to A, and to nobody else. If, in examining a given case, we fail to make a critical assessment of the second judgment, and therefore take for granted that everything that is somehow in the legitimate sphere of influence of an individual belongs to him or her exclusively, without any form of debt towards the society of which they are a member, we implicitly approve the underpinning scheme of social cooperation that in fact determines who can contribute and what. The problem, however, is precisely that of deciding whether a system of social cooperation which gives something to someone and something else to someone else, is just. Justice as reciprocity, which considers only the contributions and their values, is useless for establishing the justice of the scheme of social cooperation on which the ability to contribute depends.37 This fact contributes to shed light on the difficulties arising from the problem of self-ownership examined in section III. The point is that deciding on the fairness of rules which imply that people (drivers, consumers, etc.) skilled, wealthy, educated, genetically healthy, etc. subsidize (or are subsidized by) unskilled, poor, uneducated, or genetically unhealthy people, requires us to have already decided that the economic value of these characteristics belongs to people who exhibit the characteristics in question. An idea, viz. “that individuals should be free to pursue the economic advantage that derives from any of their individual traits, including their proneness to disease and disability”,38 which is anything but uncontroversial. In fact, individual abilities often are a consequence of luck (as in the case of natural abilities), or are socially determined, as in the case of skilfulness, education, control over resources, etc. From a justice viewpoint, the relevance given to subjective qualities loosely linked to effort and merit, is obviously objectionable. Besides, individuals normally behave in a given environment that they did not individually create, and that they cannot individually change. Under these conditions, the same notions of reciprocity (and of individual moral responsibility) should be elaborated in a more sophisticated way, taking into account conditioning imposed by the way in which certain social activities are organized.39 We believe that the view of justice as reciprocity can be countered by a notion of justice in which abilities and contribution on one hand, and needs on the other, operate separately, in the sense that the level of the former (abilities and consequent contribution) shall not mechanically affect the level of the latter (satisfaction). In each case, we should therefore resort to two separate evaluations. An evaluation which considers abilities and behaviour of the individuals involved, but not from the viewpoint of the resulting contribution, but from the viewpoint of the effort put in place. Effort is, admittedly, a much more difficult factor to measure than outcome. It is however a more egalitarian criterion, in the sense that the possibility of employing some effort is within everyone’s reach more than the possibility of producing given outcomes. 37 Lister, Reciprocity, relationships, and distributive justice, 2013, 39 Social theory and practice, (1), 70, 88: “dependence of reciprocity on prior rights and duties means that we cannot ground principles of social justice on the duty of reciprocity”. 38 Daniels, Insurability and the HIV epidemic: ethical issues in underwriting, 1990, The Milbank Quarterly, 497. 39 That is true not only from a justice viewpoint, but also from a practical perspective. For example, considering car accidents as simply the result of carefully balanced individual decisions about levels of care, specific precautions and personal prudence, loses sight of a lot of relevant socially defined phenomena, such as, for example, the overall effects, statistically inevitable, of a rule increasing the maximum speed allowed, effects that no individual driver can counteract by simply decreasing her own speed (that is, in any case, often not sufficient to eliminate accidents and even responsibility).

Denozza/Maugeri

183

73

74

75

76

77

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/07_F. Denozza - M. Maugeri, Granularization and Cross-SubsidiesLiberal, Neoliberal and Socialist Perspectives , 3.06.2019.3d from 30.10.2020 11:36:02 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 2. Critique and Theoretical Perspectives

On the other hand, we should evaluate the importance of the need to be satisfied in the given context. Cases in which cross-subsidy enables satisfaction of fundamental interests (as in the example of maternity benefits) obviously deserve different consideration from cross-subsidizations favouring the satisfaction of rather trivial needs. 79 Obviously even effort and need are socially influenced parameters and, as such, they do not provide mechanical solutions. 80 In the end we cannot do without political evaluations. The choice of prioritizing criteria based on effort and need rather than other criteria, remains nevertheless significant. Assuming effort and need as the two independent criteria for evaluating fairness of cross-subsidy, radically changes the favoured viewpoint, entailing replacement of reciprocity with solidarity. 81 The final question is: why should individuals accept to contribute in a certain measure to the creation of a social benefit that they will enjoy in the same measure as other individuals, who contributes less or nothing at all? The obvious answer is that each of us in different situations and in different periods of our lives is at times stronger and at times weaker. It is thus perfectly rational that one helps others when strong and lucky, while is helped by others when weak and less fortunate.40 It is a perfectly reasonable decision, even if the sum of the aids received does not equal, for each individual and for each situation, the aids given. Surety and solidarity are the values which support this attitude. 78

40 For example, in certain types of contractual relationships I could be a good choosing consumer in youth, and a poor choosing in oldness (and elsewhere vice versa). Is it unfair that in my youth I pay to cross-subsidy a withdrawal right for older people, while in the old age I am in turn subsidized by others?

184

Denozza/Maugeri

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

PART 3 PERSONALIZATION IN CONTRACT, CONSUMER AND TORT LAW H. ‘Granular Legal Norms’ in the Financial Services Trade I. The advent of a digital law The impact of new technologies on the trade of financial services is already 1 considerably high today and is likely to increase in the future. Robot trading and artificial intelligence systems are responsible for a growing share of financial transactions, even when executed by retail investors.1 In the high-speed trading of financial products, contracts are often concluded on the basis of algorithms, which elect the terms to be offered to or accepted by the other party.2 It would be hard to deny that these new techniques for entering into a contract and setting its terms will challenge many of the tenets of private law;3 they could eventually undermine the paradigm of contract as an agreement based on the parties’ free will.4 Unprecedented and demanding challenges are thus cast upon law and upon lawyers. 2 Particularly, the risk of new kinds of discrimination appears,5 e.g., with regard to the 1 Under European law, an express definition of ‘algorithmic trading’ is provided by art 4, para 1, fn. 39), European Parliament and Council Directive 2014/65/ of 15 May 2014 on markets in financial instruments and amending Directive 2002/92/EC and Directive 2011/1/EU [2014] OJ L173/349 (hereinafter also referred to as MiFID II). This provision stipulates that in algorithmic trading ‘a computer algorithm determines individual parameters of orders such as whether to initiate the order, the timing, price or quantity of the order or how to manage the order after its submission, with limited or no human intervention; and does not include any system that is only used for the purpose of routing orders to one or more trading venues or for the processing of orders involving no determination of any trading parameters or for the confirmation of orders or the post-trade processing or executed transaction’. ‘Algorithmic trading’ is purportedly ruled by art 17 MiFID II, which is complemented by arts 18–20, Commission delegated Regulation (EU) 2017/565 of 25 April 2016 supplementing Directive 2014/65/EU of the European Parliament and of the Council as regards organisational requirements and operating conditions for investment firms and defined terms for the purposes of the Directive [2017] OJ L 87/1 (hereinafter also referred to as MiFIR). 2 According to recital 61 MiFID II, ‘high frequency algorithmic trading’ occurs when ‘a trading system analyses data or signals from the market at high speed and then sends and updates large number of orders within a very short time period in response to that analysis. In particular, high-frequency algorithmic trading may contain elements such as order initiation, generating, routing, and execution which are determined by the system without human intervention for each individual trade or order, short-time frame for establishing or liquidating positions, high daily portfolio turnover, high order-totrade ratio intraday and ending the trading day at or close to a flat position’. 3 For an overall account, see Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in Mak/Tjong Tjin Tai/Berlee (eds.) Research Handbook in Data Science and Law, Edward Elgar Publishing 2019, 408 et seq.; Hacker, Personalizing EU Private Law: From Disclosures to Nudges and Mandates, (2017) 25 European Review of Private Law 651. 4 Twigg-Flesner, Disruptive Technology – Disrupted Law? How the Digital Revolution Affects (Contract) Law, in: De Franceschi (ed.), European Contract Law and the Digital Single Market: The Implications of the Digital Revolution, Intersentia 2017, 21 et seq. See also the essays collected in Grundmann (ed.), European Contact Law in the Digital Age, Intersentia 2018. 5 Wagner/Eidenmüller, Down by Algorithms? Siphoning Rents, Exploiting Biases, and Shaping Preferences: Regulating the Dark Side of Personalized Transactions, (2019) 86 University of Chicago Law

Sirena

185

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

3

4

5

6

7

prices set by suppliers on e-commerce platforms.6 Privacy concerns inevitably arise as well, regarding the massive use of data analytics for monitoring individuals and groups.7 Furthermore, digital contracting reflects novel asymmetries of information and power.8 The development of marketing and contracting techniques that is driven by new technologies eventually points to the abandonment of law, which is silently assumed to represent a nuisance to everyday life and to be a potential hurdle to trade. It would be hard to deny that law is based upon techniques of social control that incur a high cost both for parties to a litigation and for society as a whole. This high cost is due mainly to the inner characteristics of legal procedures for settling social conflicts (formalism) and to the role played by actors entrusted with their management (judges, attorneys, etc.). The involvement of these legal professionals makes such procedures considerably lengthy and costly; furthermore, the resulting judgments are necessarily uncertain since human interpretation of law is – by definition – subjective and, therefore, mutable if not unpredictable; the more human decision-makers are involved, the less predictable judgments become. More generally, law is moulded on the remnants of a pre-modern and analogic rationality, which, although secularized through the Enlightenment of the late 18th century, was bequeathed to it by religion. This rationality model is based on the invocation of non-negotiable principles that are to be applied to any single case through a bundle of formalized rituals, whose observance vouches for the justice of judgments. For modern law, it is the means that are paramount and that legitimize judgments, irrespective of a purported goal; conversely, no goal can determine whether the ‘right’ decision was reached in any single case. By contrast, digital rationality underpinned by new technologies legitimizes any solution on the basis of a purported goal, whilst the procedures do not matter as such; all that is required is that they shrink to the minimum. If confronted with the alternative between 0 and 1, i.e., the bulk of digital rationality, the panoply of formalistic devices that are deployed by jurists to tackle social problems and sort them out looks like a repertoire of magic spells cast by a medieval alchemist. A central issue of the ‘disruptive’ impact of new technologies on law is that they purport to do without lawyers,9 in order to eliminate the cost and the uncertainty resulting from their subjective interpretation. To that extent, it is dealt with a wellknown phenomenon that affects each layer of social and professional experience: new technologies perform a general ‘disintermediation’ of the access to services and immaterial goods (including knowledge), which become more and more directly attainable by their final users. The development of what may be called an algorithmic law – or a digital law, or also cyber law or robot law – is (i) aimed at devising new procedures enabling businesses Review 581; Bar‐Gill, Algorithmic Price Discrimination. When Demand Is a Function of Both Preferences and (Mis)perceptions, (2019) 86 University of Chicago Law Review, 217; Busch/De Franceschi, supra (fn. 3), at 423 et seq. 6 Resta, Digital platforms and the law: contested issues, [2018] Media Laws, issue 1, 231. From an antitrust perspective, see Maggiolino, Personalized Prices in European Competition Law, Bocconi Legal Studies Research Paper No. 2984840 (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2984840) accessed 30 May 2019; Maggiolino, I Big Data e il diritto antitrust, Egea 2018; Maggiolino, Big data e prezzi personalizzati, (2016) 23 Concorrenza e mercato, numero speciale, Big Data e concorrenza, 95. 7 Busch, Implementing Personalized Law: Personalized Disclosures in Consumer Law and Data Privacy Law, (2019) University of Chicago Law Review 309; Busch/De Franceschi, supra (fn. 3), at 422 et seq. 8 Daly, Private Power, Online Information Flows, and EU Law: Mind the Gap, Hart 2016. See also the essays collected in Devolder (ed.), The Platform Economy. Unravelling the Legal Status of Online Intermediaries, Intersentia 2019. 9 Verstein, Privatizing Personalized Law, (2019) 86 University of Chicago Law Review 551.

186

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:19 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

and consumers to execute transactions and (ii) designed to be fully transparent, reliable and cost-saving. The purpose of such law is annulling the risk of deceit, opportunistic behavior, etc.; its goal is sweeping legal professionals away10. In the essence, however, removing legal professionals means removing law as such. Without judges and lawyers committed to its interpretation and application, law is destined to wither away and become an empty shell of procedures pointing to nowhere. The illusion of instrumentalizing technology as a more suitable and efficient means to apply law (code is law) soon turns out to be a gradual but unstoppable replacement of law with technology (law is code).11 The ‘computational turn’ is thus not going to change the law;12 it will instead eventually sweep it away. The announcement that rules and standards are (almost) dead represents a step towards that end. Initially, predictive and communicative technologies were depicted as a means to personalize the duty of care upon which civil liability is based. Particularly, it has been advocated that the pattern of tort should be shaped upon a personalized standard of diligence, which should reflect not the ‘reasonable person’ of any non-distinct member in a given pool (encompassing, for instance, the average diligence of doctors in cases of malpractice), but the ‘reasonable you’ of each individual.13 That step taken, however, it is inevitable to take another one in the same direction, and yet another, until, as has already been prophesized, law could (or should) be replaced by micro-directives, communicated to every user simultaneously through their online devices.14 Even if all that could (or should) happen under the control of a public agency, or official, it would be hard to deny that a shift of power would occur from law to technology, and, therefore, from legal professionals to engineers, data scientists, etc.

8

9

10

11

II. The trend towards the personalization of private law: from the ‘average consumer’ to the ‘images of the consumer’ Since World War II, the traditional typification of legal rules has been increasingly 12 questioned, thus triggering the trend toward a gradual personalization of law. A fundamental tenet of juridical and political culture of the Enlightenment was 13 enshrined in the generality and abstractness of the State’s law, which was thus empowered to overcome the legal particularism that had characterized the feudal order of the Ancien Régime. The State’s law was predicated on being general because it had to be applied to all citizens irrespective of the social classes they belonged to; the State’s law was predicated on being abstract because it had to be applied to any case irrespective of when and where a contract was entered into or a wrong was committed. The precept of equal treatment was therefore paramount (treat like cases alike). From a political perspective, the State was thus entrusted with the monopoly of law 14 production through the enactment of legislative measures, these being aimed at achieving a rational mediation between non-negotiable principles of natural law and the 10 Casey/Niblett, Framework for the New Personalization of Law, (2019) 86 University of Chicago Law Review 335. 11 Path-breaking was Lessig, Code and Other Laws of Cyberspace, Basic Books 1999. 12 Hildebrandt/de Vries (eds.), Privacy, Data Process and the Computational Turn: The philosophy of law meets the philosophy of technology, Routledge 2013. 13 Ben‐Shahar/Porat, supra Part 1.B. For further reference, see Busch/De Franceschi, supra (fn. 3), at 417 et seq. 14 Casey/Niblett, supra Part 1.C.

Sirena

187

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:19 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

15

16

17

18

19

political needs of civil society, as advocated by Hegel.15 In a State under the rule of law (Rechtsstaat), generality and abstractness of law ensure the equal treatment of citizens by the State and guarantee that general decisions will be taken through standardized procedures – and not captured by the interests of individuals or groups. Generality and abstractness of law was explicitly categorized and advocated by Kant, who, together with Hegel, may for good reason be acknowledged as the initiator of modern philosophy of law. Kant posited the ‘general principle of law’,16 according to which: ‘law is any action that, either directly, or through its formulation, is able to make the arbitrary liberty of each individual coexist with that of any other according to a universal regulation’.17 Such a universal regulation was laid down by Kant himself in the following terms: ‘Act in the outside world so as the free avail of your judgment may coexist with the liberty of anyone else’.18 That way, law could be set out as a ‘pure’ concept, although directed to practice, i.e., to its application to cases experienced in the outside world.19 Kant’s philosophy of law is undeniably the starting point and the source of inspiration for the ‘pure theory of law’ developed by Kelsen. Together with the milder version elaborated by Hart, the ‘pure theory of law’ is the cultural manifesto of legal positivism, what was to become the predominant doctrine of general theory and philosophy of law during most of the 20th century. According to Kelsen’s depiction, law is a system of general and abstract rules that are to be styled as hypothetical propositions (legal norms); the ‘pureness’ of his theory is given not only by the strict seclusion of law, deemed to be isolated from religion, ethics, and any other field of human knowledge, but also by its insensitivity towards the particular issues of each case. Within this conceptual framework, therefore, the legal subject is any undifferentiated member of society, meeting the same characteristics of generality and abstractness that are paramount for legal norms.20 Legal subjects are classified by law into categories, just like legal facts are; for example, individuals are divided into those of age that hold the capacity to act, and those that are underage that do not hold such capacity. This typification has also a general and abstract character, because it is not sensitive to the concrete traits of each legal subject that is thus classified; for example, a minor is generally devoid of the capacity to act even if she is endowed with discernment and expertise exceeding those of a person of age, and vice versa. The capacity to act is achieved after attaining a certain age (e.g., eighteen years) that is the same for any individual and that disregards her specific characteristics of education, intelligence, experience, etc. Inevitably, however, typifications resorted to by private law tend to become increasingly refined over time. With regard to the division between legal subjects who are under age and those of age, for instance, German law came to accept that the capacity to act is progressively gained by individuals while growing up, so that also a minor is able to exercise some rights to personality;21 furthermore, the newly enacted § 105a BGB Hegel, Grundlinien der Philosophie des Rechts (first published 1821) § 211. Kant, Die Metaphysik der Sitten, first published 1797, Akademieausgabe von Immanuel Kants Gesammelten Werken, 230, § C: Allgemeines Princip des Rechts. 17 ibid: ‘Eine jede Handlung ist Recht, die oder nach deren Maxime die Freiheit der Willkür eines jeden mit jedermanns Freiheit nach einem allgemeinen Gesetze zusammen bestehen kann’. 18 Kant, supra (fn. 16), at 231, § C: ‘handle äußerlich so, daß der freie Gebrauch deiner Willkür mit der Freiheit von jedermann nach einem allgemeinen Gesetze zusammen bestehen könne’. 19 Kant, supra (fn. 16), at 205, Vorrede: ‘der Begriff des Rechts [ist] […] ein reiner, jedoch auf die Praxis (Anwendung auf in der Erfahrung vorkommende Fälle) gestellter Begriff’. 20 In this line of reasoning, see Falzea, Il soggetto come fattispecie, Giuffrè 1939. 21 Wolf/Neuner, Allgemeiner Teil des bürgerlichen Rechts, 11th edn., C.H. Beck 2016, 128 et seq. 15 16

188

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:19 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

stipulates that even one who is devoid of the capacity to act (like a minor) can perform legal transactions of everyday life (Geschäfte des täglichen Lebens).22 After the reform of French law of contract and obligations passed in 2016–201823, a similar rule is provided by art. 1148 Code civil24. During the 20th century, moreover, the shift from a liberal State to a welfare State 20 changed the understanding of the principle of equal treatment, which became aimed not only at formal but also at substantive equality.25 It thus prompted the tendency towards a ‘materialization’ of (private) law (Materialisierung)26, the purpose of which is to have account of the socio-economic or relational weakness of the addressees of legal rules. Under German law, § 138, para 2, BGB traditionally mandates the voidness of a contract by means of which one contracting party exploits the situation of need, inexperience, economic distress or mental weakness of the other, inducing the latter to give, or promise to give, a consideration that is grossly disproportionate to the performance27. Furthermore, a wide application of general clauses, like that of good faith, allowed German courts and scholarship to expand the law, adding solutions induced by moral values that require a ‘contextualization’ and a ‘flexibilization’ of rules enacted by legislators.28 A renowned judgment issued by the German constitutional court (Bundesverfassungsgericht) did not refrain from declaring void a guarantee surety that was issued by a dependant of the debtor who, due to her young age and her inexperience, was deemed to have taken a risk unbearably high relative to her patrimonial and financial situation.29 In the framework of the European Union’s law, most directives governing contract 21 law are targeted at business-to-consumer transactions, which are thus legally differentiated from peer-to-peer transactions, be they amongst consumers or businesses. European law aims more at the establishment and the functioning of the single market 22 than at the protection of consumers as weaker contracting parties.30 Therefore, the division of businesses and consumers is not of a social or economic nature, but is merely relational, and the subjective scope of the norms is shaped on the concrete relation that is entered into by the contracting parties. The same legal subject can conclude some contracts while acting like a consumer and others while acting like a professional; provided that she acts like a consumer, she can enter into a contract with another consumer or with a 22 See Herrler, sub § 105a, in Staudingers Kommentar zum BGB, de Gruyter 2017; Spickhoff, sub § 105a, in Münchener Kommentar zum BGB, C.H. Beck 2018. 23 See the essays collected in Cartwright/Fauvarque-Cosson/Whittaker (eds.), La réécriture du code civil: le droit français après la réforme de 2016, Société de législation comparée 2018; and those collected in Bien/Borghetti (eds.), Die Reform der französischen Vertragsrechts. Ein Schritt zu mehr europäischer Konvergenz?, Mohr Siebeck 2018. 24 See Dehayes/Genicon/Laithier, Réforme du droit des contrats, du regime general et de la prevue des obligations: Commentaire article par article, 2nd edn., Lexis Nexis 2018, 266 et seq. 25 Habermas, Faktizität und Geltung: Beiträge zur Diskurstheorie des Rechts und des demokratischen Rechtsstaats, Suhrkamp 1992, 498. 26 Canaris, Wandlungen des Schuldvertragsrechts – Tendenzen zu seiner “Materialisierung, (2000) 20 Archiv für die civilistische Praxis 273; Wagner, Materialisierung des Schuldrechts unter dem Einfluss von Verfassungsrecht und Europarecht – Was bleibt von der Privatautonomie?, in: Blaurock/Hager (eds.), Obligationenrecht im 21. Jahrhundert, Nomos 2010, 13 et seq. In the same vein, see Medicus, Abschied von der Privatautonomie im Schuldrecht?: Erscheinungsformen, Gefahren, Abhilfen, O Schmidt 1994. 27 See Fischinger, sub § 138, in: Staudingers Kommentar zum BGB, de Gruyter 2017; Armbrüster, sub § 138, in: Münchener Kommentar zum BGB, C.H. Beck 2018. 28 Marietta Auer, Materialisierung, Flexibilisierung, Richterfreiheit (Mohr Siebeck 2005). 29 BVerfG, 5.8.1994–1 BvR 1402/89. 30 However, see Mak, The Consumer in European Regulatory Private Law, in: Leczykiewicz/Weatherhill (eds.) The Images of the Consumer in EU Law. Legislation, Free Movement and Competition Law, Hart Publishing 2016, 381 et seq., who distinguishes between European general law, targeted at a ‘rational consumer’, and European regulatory private law, aimed at protecting the consumer as the weaker party (or ‘Calimero consumer’).

Sirena

189

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:19 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

professional. The legal rules governing contracts change in accord not only with the abstract type of legal subject, but also with the concrete aim that is pursued by each of the contracting parties while performing the transaction. This technique is streamlined by the regulatory function that characterizes European contract law;31 the latter is in fact deprived of much of the traditional autonomy it enjoys in national laws and is instead carved out to achieve some institutional goals of the European Union, starting with the establishment and the functioning of the European single market.32 23 Although showing a considerable degree of novelty towards traditional contract law, however, European contract law is based on a typification technique that does not challenge tremendously the generality and abstractness of legal norms, instead combining them with the subjective quality of the contracting parties. 24 The scope of the rules belonging to European private law was determined by the European Court of Justice (ECJ) through the concept of the ‘average consumer’,33 defined as a consumer that is ‘reasonably well-informed and reasonably observant and circumspect’.34 This definition was subsequently crystallized in the directive enacted by the European Union in the field of contract law. Only sporadically is European legislation instead addressed to particular groups of consumers: the most important case of this kind is that of the Unfair Commercial Practices Directive (UCPD).35 25 Pursuant to Art. 5, paragraph 2, lit. (b), UCPD, the assessment of unfairness of commercial practices is treated differently when it is targeted to ‘a particular group of consumers’, instead of the ‘average consumer’.36 31 Amongst many, see Grundmann, Privatrecht und Regulierung, in: Auer/Grigoleit/Hager and others (eds.), Privatrechtsdogmatik im 21. Jahrhundert: Festschrift für Claus-Wilhelm Canaris zum 80. Geburtstag, de Gruyter 2017, 907 et seq.; Brownsword/van Gestel/Micklitz (eds.) Contract and Regulation. A Handbook on New Methods of Law Making in Private Law, Edward Elgar 2017; Comparato/Micklitz/ Svetiev, The regulatory character of European private law, in: Twigg-Flessner (ed.), Research Handbook on EU Consumer and Contract Law, Edwar Elgar Publishing 2016, 35 et seq.; Mezzanotte¸ L’appartenenza come tecnica di regolazione (a proposito di “Regulatory Property Rights”), in [2016] Rivista critica di diritto privato 635; Svetiev, European Regulatory Private Law: From Conflicts to Platforms, in: Purnhagen/Rott (eds.), Varieties of European Economic Law and Regulation, Springer 2014, 153 et seq.; Micklitz, The Visible Hand of European Regulatory Private Law-The Transformation of European Private Law from Autonomy to Functionalism in Competition and Regulation, (2009) 28 Yearbook of European Law 3. See also the essays collected in Micklitz/Svetiev (eds.), A Self-sufficient European Private Law: A Viable Voncept?, in [2012] EUI Working Paper Law, issue n 31; and those collected in Micklitz/Svetiev/ Comparato (eds.), European Regulatory Private Law – The Paradigm Tested, in EUI Working Paper Law, 2014, issue n 04. Among the monographs, see Hellgarth, Regulierung und Privatrecht: Staatliche Verhaltenssteurung mittels Privatrecht und ihre Bedeutung für Rechtswissenschaft, Gesetzgebung und Rechtsanwendung, Mohr Siebeck 2016; Poelzig, Normdurchsetzung durch Privatrecht, Mohr Siebeck 2015. 32 Schmid, Die Instrumentalisierung des Privatrechts durch die Europäische Union. Privatrecht und Privatrechtskonzeptionen in der Entwicklung der Europäischen Integrationsverfassung, Nomos 2010. 33 Weatherhill, Who is the average consumer?, in: Weatherhill/Bernitz (eds.), The Regulation of Unfair Commercial Practices under EC Directive 2005/29, Hart Publishing 2007, 115 et seq.; Mak, The “average consumer” of EU law in domestic litigation: Examples from consumer credit and investment cases, in [2012] Tilburg Law School Legal Studies Research Paper Series, issue n. 4, 5. 34 Case 210/96 Gut Springenheide GmbH and Rudolf Tusky v Oberkreisdirektor des Kreises Steinfurt – Amt für Lebensmittelüberwachtung [1988] ECR I-4657; Case 385/01 Commission v Spain [2003] ECR I13143, fn. 53. Already earlier, the ECJ had clearly mentioned the ‘reasonably circumspect consumers’ as a point of reference for applying European private law: Case 470/93 Verein gegen Unwesen in Handel und Gewerbe Köln eV v Mars GmbH [1995] ECR I-1923, para 24. 35 European Parliament and Council Directive 2005/29/EC of 11 May 2005 concerning unfair businessto-consumer practices in the internal market and amending Council Directive 84/450/EEC, Directives 97/ 7/EC, 98/27/EC and 2002/65/EC of the European Parliament and the Council and Regulation (EC) No 2006/2004 of the European Parliament and the Council [2005] OJ L 149/22. 36 Cartwright, The consumer image within EU law, in: Twigg-Flessner, supra (fn. 31), at 199 et seq.; Poncibò, The average consumer, the unfair commercial practices directive and the cognitive revolution, (2007) 30 Journal of Consumer Policy 21.

190

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:19 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

Moreover, Art. 5, paragraph 3, UCPD acknowledges that commercial practices may be ‘likely to distort the economic behavior only of a clearly identifiable group of consumers who are particularly vulnerable to the practice or the underlying product because of their mental or physical infirmity, age or credulity’. The paradigm of the ‘vulnerable consumer’, which had already been used by the European legislature in directives governing universal service with regard to the supply of energy and telecommunication,37 was thus introduced into European private law.38 This given, it should nevertheless be pointed out that this more nuanced approach does not detract the assessment of unfairness of commercial practices from being conducted along the standard of the ‘average consumer’, which is, however, to be identified with regard to the specific group at which such practices are targeted or with regard to the group of vulnerable consumers. Subsequently, a (limited) reference to the ‘vulnerable consumer’ was made by the Consumer Rights Directive (CRD)39. Recital 34 CRD sets out that, in providing information with regard to arrangements by means of which the consumer is to pay a deposit to the trader, the latter ‘should take into account the specific needs of consumers who are particularly vulnerable because of their mental, physical or psychological infirmity, age or credulity in a way which the trader could reasonably be expected to foresee’. Nevertheless, it is immediately afterwards specified that: ‘taking into account such specific needs should not lead to different levels of consumer protection’. More recently, the European legislature has addressed the ‘vulnerable consumer’ in order to encourage her access to services that are deemed essential for participating in the internal market and benefiting from its advantages. With regard to the retail banking market in particular, payment accounts with basic features must be offered to clients so as to include also ‘unbanked vulnerable consumers’.40 Such varieties of legislative references led scholarship to advocate in favor of spelling out different ‘images of the consumer’ (Verbraucherleitbilder),41 which could (and should) be based on different traits of behavior or personality characterizing individuals. Attention was therefore drawn to vulnerable consumers; hasty consumers; consumers with inferior bargaining power; and uninformed consumers.42 Along a different taxonomy, the following consumer images were to be set out: the fully informed consumer; the information seeker; the passive glancer; the snatcher; the irrational consumer; and the consumer without choices.43 37 Reich/Micklitz, Economic law, consumer interests, and EU integration, in: Reich/Micklitz/Rott/ Tonner, European Consumer Law, 2nd edn., Intersentia 2014, 46 et seq. 38 Reich, Vulnerable consumers in EU Law, in: Leczykiewicz/Weatherhill (eds.), supra (fn. 30), at 139 et seq. 39 European Parliament and Council Directive 2011/83/EU of 25 October 2011 on consumer rights amending Council Directive 93/13/EEC and Directive 1999/44/EC of the European Parliament and of the Council and repealing Council Directive 85/577/EEC and Directive 97/7/EC of the European Parliament and of the Council [2011] OJ L 304/64. 40 European Parliament and Council Directive 2014/92/EU of 23 July 2014 on the comparability of fees related to payment accounts, payment account switching and access to payment accounts with basic features [2014] OJ L 257/214. In detail, see art 18, para 4, and art 20, para 1, as well as recitals 3, 46, 48, 49, and 54. 41 See the essays collected in Leczykiewicz/Weatherhill (eds.), supra (fn. 30); and those collected in Klinck/Riesenhuber (eds.), Verbraucherleitbilder. Interdisziplinäre und europäische Perspektive, de Gruyter 2015. Furthermore, see Schüller, The definition of consumers in EU law, in Devenny/Kenny (eds.), European Consumer Protection: Theory and Practice, CUP 2012, 123. 42 For some references, see Grundmann, Targeted Consumer Protection, in Leczykiewicz/Weatherhill (eds.), supra (fn. 30), at 225 et seq. 43 Wilhelmsson, Twelve Essays on Consumer Law and Policy, University of Helsinki 1996, 105 et seq.

Sirena

191

26

27

28

29

30

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

III. The discourse on granular legal norms (particularly with regard to the duties of disclosure provided by European contract law) 31

32

33

34

35

The advent of Big Data analytics, as well as new predictive and communicative technologies, is seen as prompting a radical shift from a ‘crude’ typification to a ‘granular’ personalization of law,44 a change that would tremendously accelerate the trend towards a lessening and eventual abandonment of the generality and abstractness of legal norms. Unlike other threads of ‘contextualization’ of law (e.g., the doctrine of Materialisierung),45 the design of ‘granular norms’ would be prospectively based on a combination of individual characteristics that do not (necessarily) have a social dimension. Therefore, it would not be aimed at protecting social classes or groups, but at differentiating to the utmost amongst single individuals and emphasizing the idiosyncratic preferences of each. The expected goal is not only to inform private parties (especially consumers) more accurately, but also to drive their behavior so as to more effectively comply with the policies implemented by legislators;46 in this line, the theory of ‘granular norm’ is aligned with some tenets of ‘liberal paternalism’ that advocate for a new theory (and design) of law based on ‘nudging’.47 As to mandatory norms,48 the process of personalization seems not to question the existence of legal prohibitions or impositions as such, but points to design and implementation of a process of continuous and automatic adjustment of their content, so as to ascertain how they should be applied in any discrete case. This scenario poses many novel questions as to the limits of civil, criminal, and administrative liability, and even leads one to wonder whether the divide between rules and standards is doomed to be overwhelmed by the advent of ‘granular norms’.49 The theory of ‘granular norms’, however, has been developed particularly with regard to mandatory norms that do not provide a prohibition, or imposition, of a substantive nature (e.g., stipulating the voidness of contractual terms), but to those that enshrine information rules. Most typically, this is the case of rules stipulating the duty of a professional to disclose information to a (potential) client or customer, especially if a consumer.50 Imposing disclosure duties on traders (vis-à-vis consumers) is a technique of regulating markets that has been widely resorted to by European legislators.51 This disclosure 44

Porat/Strahilevitz, supra Part 1.A. See para II. 46 Hacker, supra (fn. 3), at 651. 47 Hacker, Nudge 2.0 – The Future of Behavioural Analysis of Law, in Europe and Beyond, (2016) 24 European Review of Private Law 297. For an insightful account of behavioral perspectives on private law, see Micklitz/Sibony/Esposito (eds.), Research Methods in Consumer Law: A Handbook, Edward Elgar 2018; Hacker, Verhaltensökonomik und Normativität, Mohr Siebeck 2017. 48 Ben‐Shahar/Porat, Personalizing Mandatory Rules in Contract Law, (2019) 86 Chicago University Law Review 255. 49 See para I. 50 Even where the provider of financial services is obliged to abstain from trading (particularly due to conflicts of interests), disclosure remains available as an alternative to her and, that way, the doors to contract are still opened. Significantly, many a national law interprets the European rule of ‘disclosure or abstain’ in the sense that if the provider of financial services does not disclose a conflict of interests and the contracting client does not waive such conflict, the contract concluded between them is nevertheless valid, although the provider may be liable for damages. 51 For an insightful overview, see Busch, The future of pre-contractual information duties: from behavioural insights to big data, in: Twigg Flessner, supra (fn. 31), at 224 et seq. 45

192

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

strategy has been long and sharply criticized.52 Particularly, it has been pointed out that a great amount of standardized information engenders an informational overload that proves not only useless, but even harmful to consumers, and therefore it has been advocated that the disclosure technique should be abandoned as such.53 According to a milder version, disclosures should be abandoned solely insofar as undifferentiated, but their defects could be cured by the advent of ‘granular legal norms’:54 in other words, these authors argue for the adoption and implementation of ‘smart disclosures’.55 In fact, information duties could, and should, be rejuvenated and adapted to the insights from behavioral and psychological research,56 which have drawn increasing attention to the problems of bounded rationality and bounded attention.57 These techniques would allow a curing of the defects of information overload in that they would reduce the information flow to consumer and increase the salience (i.e., the cognitive accessibility) of key information.58 It is, however, to be wondered whether the argument for a personalization of 36 disclosures might be conveniently conveyed through the theory of ‘granular norms’ As a matter of substance, personalizing disclosures of information means a 37 personalization of duties of disclosure in respect of their content, not the rules governing them. To put it differently, personalizing disclosures of information has little in common with embarking on empirical research by means of guinea pigs, or similar experiments, aiming to ascertain the default rules which would better mimic the individual preferences of the different addressees of a norm.59 By contrast, personification of information disclosures is pursued by obliging the more informed party, i.e. the business, to inquire of the less informed party, i.e. the consumer, about her personal characteristics and, consequently, to tailor information which most suits her profile and to present her with this tailored information. Therefore, the personalization of legal norms pertains to legislation and is a job to be done by legislators, while the personalization of information disclosures pertains to business of conduct rules and is to be done by the parties themselves or, more accurately, by the more informed side, i.e., the business. In the context of European contract law, uniform information is not properly aimed 38 exclusively at the protection of the weaker party, or her awareness of contacts terms; rather, and maybe even to a greater extent, it also seeks to create a ‘level playing field’ for businesses of many different home countries, each of them compliant with a different national law.60 To some extent, therefore, uniformity is paramount for the 52 Bar‐Gill/Ben‐Shahar, Regulatory Techniques in Consumer Protection: A Critique of European Contract Law, (2013) 50 Common Market Law Review, 109. 53 Ben‐Shahar/Schneider, More Than You Wanted to Know, Princeton UP 2014, 5 et seq., 110; Ben‐ Shahar, The Myth of the “Opportunity to Read” in Contract Law, (2009) 5 European Review of Contract Law 1; Ben‐Shahar/Schneider, Coping with the failure of mandated disclosure, (2015) 11 Jerusalem Review of Legal Studies 83–93; Marotta-Wurgler, Even more than you wanted to know about the failures of disclosure, (2015) 11 Jerusalem Review of Legal Studies 63. 54 Porat/Strahilevitz, supra (fn. 44). 55 Bar‐Gill, Defending (smart) disclosure: A comment on More Than You Wanted to Know, (2015) 11 Jerusalem Review of Legal Studies 75. 56 Helleringer/Sibony, European Consumer Protection Through the Behavioral Lens, (2017) 23 Columbia Journal of European Law 621 et seq. 57 Busch, supra (fn. 51), at 230. 58 Busch, supra (fn. 51), at 231; Hacker, supra (fn. 3), at 666 et seq. 59 Porat/Strahilevitz, supra (fn. 44). For further reference, see Busch/De Franceschi, supra (fn. 3), at 420 et seq.; Hacker, supra (fn. 3), at 670 et seq. 60 Grundmann, Information, Party Autonomy and Economic Agents in European Contract Law, (2002) 39 Common Market Law Review 269.

Sirena

193

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

39

40

41

42 43 44

establishment of a single European market, given that it is confronted with twenty-eight national markets. In the field of regulated financial services, particularly, European law promotes uniformity of information to overcome the national fragmentation of markets and legal systems.61 From the viewpoint of private enforcement, uniformity of information enables consumers to compare on the same basis offers by suppliers of different nationalities and to assess and select the preferable one; it is intended to remedy adverse selection by consumers and moral hazard by professionals. From the viewpoint of public enforcement, uniformity of information is paramount for supervision by public authorities on the market, particularly when it stretches over transparency of contracts and fairness of conduct by traders. With regard to credit services, the most important devices that European private law applies to convey uniform information are the following: 1) pre-contractual advertising and contractual indication of the annual percentage rate of charge (APR) for any credit arrangement, to represent the total costs of the credit, including any fees, expenses, etc.;62 APR constitutes a comprehensive score for the credit arrangement that must be calculated according to the mathematic formula provided in Annex II of both the Consumer Credit Directive and the Mortgage Credit Directive; 2) a standardized document that spells out the items of information indicated in these Annexes: in the Consumer Credit Directive such document is titled the Standard European Consumer Credit Information (SECCI);63 in the Mortgage Credit Directive it is titled the European Standardised Information Sheet (ESIS).64 Yet, uniform information must be accompanied by some personalized disclosures. Particularly, the Mortgage Credit Directive stipulates expressly that pre-contractual information is to be ‘personalized’;65 particularly, it has to ‘include adequate specific risk warnings, for instance about the potential impact of exchange rate fluctuations on what the consumer has to repay and, where assessed as appropriate by the Member States, the nature and implications of taking out a security’.66 Moreover, the ESIS must have a user-friendly structure and indicate an illustrative amortization;67 it must be drawn up with simple and understandable language.68 With regard to investment and insurance services, the provider must inform (potential) clients about conflicts of interest.69 After summarizing some key-points of the European law on financial services, one may wonder whether the plea for personalized information is really of core importance for improving and sharpening consumer protection (or the functioning of the market). A tentative answer should move from the assumption that information about financial 61 For an overall view, see Mak, supra (fn. 30), at 314 et seq.; Colaert, Investor Protection in the Capital Market Union, in: Busch/Avgouleas/Ferrarini (eds.), Capital Market Union in Europe, OUP 2018, 342. 62 European Parliament and Council Directive 2008/48/EC of 23 April 2008 on credit agreements for consumers and repealing Council Directive 87/102/EEC Directive 2008/48/EC [2008] OJ L 133/66 (hereinafter also referred to as the Consumer Credit Directive), art 5; European Parliament and Council Directive 2014/17/EU of 4 February 2014 on credit agreements for consumers relating to residential immovable property and amending Directives 2008/48/EC and 2013/36/EU and Regulation (EU) No 1093/2010 [2014] OJ L 60/34 (hereinafter also referred to as the Mortgage Credit Directive), art 11. 63 Consumer Credit Directive, annex II. 64 Mortgage Credit Directive, annex II. 65 Mortgage Credit Directive, art 14, para 1. 66 Mortgage Credit Directive, recital 22 and art 11. 67 Mortgage Credit Directive, recital 40. 68 Mortgage Credit Directive, recital 41. 69 MiFID II, art 23; European Parliament and Council Directive (EU) 2016/97 of 20 April 2016 on insurance distribution [2016] OJ L 26/19 (hereinafter also referred to as IDD), art 28.

194

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

services, however simplified and/or personalized it may be, is too complex for clients, even if professionals. This assumption explains one of the reasons why these services may be provided exclusively by authorized intermediaries that are also subject to a system of supervision by public authorities, thereby ensuring not only their financial stability (prudential supervision), but also the transparency of their contracts and the fairness of their commercial practices. In fact, authorized intermediaries also play the role of honest brokers who explain information and make it more digestible to clients, especially when consumers. It is therefore fully understandable that the European legislature aimed to achieve the personalization not of information as such, but of financial services contracts.70 To pursue this goal, European legislators applied two main types of devices: 1) with regard to investment and insurance services, pre-contractual tests of suitability or appropriateness of the contract in relation to the profile of the (potential) client or customer;71 2) with regard to credit services, there is a pre-contractual assessment regarding the creditworthiness of the (potential) client.72 The movement towards achieving a personalization of financial services contracts is specifically intertwined with the nature of such services. In fact, a fiduciary relation arises between the provider of financial services and its (potential) clients.73 Generally, clients are not sufficiently acquainted with financial instruments or credit agreements and, therefore, they are de facto compelled to rely on the competence and professionality of the providers whom they trust. On the other side, the providers compete on the market by calling for the trust of their clients, who can make a choice based only on the competence and professionality of authorized intermediaries. The fiduciary nature of the legal relation between the parties implies that the provider of financial services has a duty of loyalty and confidence towards the client. This duty, which is added to the bundle of obligations provided for by the common law of agency between the parties, has been ‘translated’ into a corpus of ‘conduct of business rules’, which have been laid down by self-regulatory organizations or public authorities responsible for the supervision of financial markets and traders of financial services.

45 46

47

48

49

IV. The personalization of financial services 1. The know-your-customer rule in investment services Amongst the conduct of business rules that characterize the performance of invest- 50 ment services, one of the most important is that mandating the provider of such services to act in the best interest of the client.74 Up until the Great Depression of 1929 (Black Tuesday), the broker was granted full 51 autonomy as to the choice of the investment that the client was to underwrite. The 70 On the personalized contracts at which European law on financial services aims, see Imbruglia, La regola di adeguatezza e il contratto, Giuffrè 2017. 71 See paras IV.1. and IV.2. 72 See para IV.3. 73 Plato‐Shinar/Weber, Three Models of the Bank’s Fiduciary Duty, (2008) 2 Law and Financial Markets Review, 422 et seq. 74 MiFID, art 24, para 1, and recitals 91–97. See Enriques/Gargantini, The Overarching Duty to Act in the Best Interest of the Client, in: Busch/Ferrarini (eds.) Regulation of the EU Financial Markets. MiFID II and MiFIR, OUP 2017, 85 et seq.

Sirena

195

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

52

53

54

55

56

seriousness of this historical financial crisis taught, however, a lesson that had to be learnt by legislators and policymakers, obviously starting with those based in the United States of America. Thus, the rule was created according to which the broker could give advice on or perform only those financial services that she found suitable to her client’s profile (suitability rule). After being enacted by the United Kingdom (particularly through the 1986 Financial Services Act), the best interest rule spread to other legal systems and eventually became one of the tenets of European law on investment services.75 Pursuant to the rules provided by MiFID II,76 the suitability test is required for the performance of investment services; for the other investment services listed in Annex I of the directive, a test of appropriateness is required instead. Both tests were streamlined in accord with the subjective categories into which the European legislature subdivided clients:77 the main division is that between retail and professional clients, to which a further category was added, that of eligible counterparties. Moreover, professional clients are further subdivided into clients that are professional per se and clients that are elective professionals, i.e., those non-professional clients who, under given conditions, are allowed ‘to waive some of the protections afforded by the conduct of business rules’.78 The least protected category of clients under MiFID II is that of eligible counterparties, which may be chosen by each Member State among the following subjects: investment firms, credit institutions, insurance companies, undertakings for collective investments in transferable securities (UCITS) and their management companies,79 pension funds and their management companies, other financial institutions authorized or regulated under Union law or under the national law of a Member State, national governments and their corresponding offices (including public bodies that deal with public debt at national level), central banks and supranational organizations.80 Transactions concluded between investment firms and eligible counterparties are not generally subject to the suitability test,81 unless the contracting eligible counterparty requests, either on a general form or on a trade-by trade basis, to be treated as a client whose business with the investment firm is subject to the suitability test.82 Compared to MiFID I, MiFID II significantly raised the degree of legal protection provided to eligible counterparties. This is due to the fact that financial crisis has in fact shown that also non-retail clients are exposed to significant risks when contracting with investment firms, since they do not always have the capacity to assess the characteristics of financial products.83 For that reason, MiFID II extended to eligible counterparties certain duties of information and reporting of the contracting investment firm, which MiFID I had provided only towards retail and professional clients. A peculiar case is that of municipalities and local public authorities, which suffered high financial losses due to swaps contracts they were advised to enter into in order to 75 See ESMA, Final Report. Guidelines on certain aspects of the MIFID II suitability requirements, 28 May 2018, ESMA35–43-869. 76 MiFID II, art 25. 77 Kruithof, A Differentiated Approach to Client Protection: The Example of MiFID, in: Grundmann/ Atamer (eds.) Financial Services, Financial Crisis and General European Contract Law: Failure and Challenges of Contracting, Wolters Kluwer 2011, 105 et seq. 78 MiFID II, annex II, part II.1. Particularly, this may be the case of public sector bodies, local public authorities, and municipalities, as well as of private individual investors. 79 About UCITS products, see Mak, supra (fn. 30), at 330 et seq. 80 MiFID II, art 30, para 2. 81 MiFID II, art 30, para 1. 82 MiFID II, art 30, para 2, subpara 1. 83 MiFID II, recital 104.

196

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:20 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

make fast cash (instead of reducing the exchange rate risk). A large litigation between such public entities and the contracting banks thus arose in the United Kingdom and several other member States of the European Union (credit default swaps litigation). For this reason, such public entities have been taken out of the category of eligible counterparties, nor are they considered as professional clients unless, as already pointed out, they request to be treated as such (elective professionals). Apart from eligible counterparts, the division of clients between the categories of professional and retail is paramount; an investment firm’s information duties are differentiated accordingly, being, of course, broader and deeper in respect of retail clients.84 The suitability test is foreseen in the event the services to be provided are investment advice and portfolio management. This test requires that the investment firm involved obtains from the client the necessary information regarding: 1) the client’s financial experience and knowledge; 2) her financial situation, including her ability to bear losses; 3) her investment objectives, including her risk tolerance.85 However, the duty of the investment firm involved to obtain such information is relaxed if the service is to be provided to a professional client. In that case, in fact, it is to be generally assumed that such a client possesses the necessary level of financial experience and knowledge regarding the envisaged products, transactions, and services;86 particularly, if it is deal with a client that is a per se (and not an elective) professional, it is also to be generally assumed that she is financially able to bear any related investment risk.87 A specific report is due by investment firms providing investment advice to a retail client; the report is to include an outline of the advice given and how the recommendation provided is suitable for the client.88 Before the transaction is concluded, the investment firm is, on a durable medium, to provide the retail client with a statement on suitability specifying how that advice meets her preferences, objectives, and other characteristics.89 Where such information is not provided, an investment firm is prohibited from recommending investment services or financial instruments to a (potential) client;90 the same applies in the event none of the services or instruments are suitable for the client.91 For other investment services, a test of appropriateness is foreseen whereby the investment firm involved is required to have obtained solely the necessary information regarding the client’s financial knowledge and experience.92 However, an investment firm is entitled to assume that a professional client has the experience and knowledge needed to understand the financial risk at stake.93 By contrast, none of the ‘know-your-customer’ tests have to be applied to ‘execution only’ transactions, which consist of the mere reception and transmission of client orders (whether or not accompanied by any of the ancillary services specifically listed in the 84

MiFID, art 24; MiFIR, arts 44–51. MiFID II, art 25, para 2. 86 MiFIR, art 54, para 3, subpara 1. 87 MiFIR, art. 54, para 3, subpara 2. By contrast, as stated in MiFID II, annex II, part II.1., elective professional clients will not ‘be presumed to possess market knowledge and experience comparable to that’ of per se professional clients. 88 MiFIR, art 54, para 12, subpara 1. 89 MiFID, art 25, para 6, subpara 2, and recital 82. 90 MiFIR, art 54, para 8. 91 MiFIR, art. 54, para 10. 92 MiFID II, art 25, para 3. 93 MiFIR, art 56, para 1, subpara 2. 85

Sirena

197

57

58

59

60

61

62

63 64

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:21 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

directive).94 In order to be exempted, however, such services must fulfil three criteria: 1) they involve non-complex financial instruments, like shares and simple bonds;95 2) they are provided at the initiative of the client; 3) the latter must be informed that no ‘knowyour-customer’ obligations apply. 65 However, it should be observed that the suitability regime tends to be pervasive, since European and national supervising authorities, including the Committee of European Securities Regulators (CESR), put a strong emphasis on individual advice and define it in a very broad way.96 Particularly, even spot advice on a single product, where an ‘execution only’ transaction is accomplished, is considered by regulatory authorities to qualify as individual advice such that the suitability regime is deemed applicable.97 66 The ECJ has spelled out that the exclusion of suitability and appropriateness tests prior to the performance of investment services must be understood as exceptional since the tests constitute the bulk of MiFID II regulations; instances where such tests are excluded are therefore to be considered under strict and narrow interpretation.98

2. The know-your-customer rule in insurance services 67

Also regarding the performance of insurance contracts, the European legislature has provided tests on suitability and appropriateness that are aligned with those required in the field of investment services; also stipulated are reporting duties to customers.99

3. The know-your-customer rule in credit banking The rules enacted by the European Union with regard to contracts for credit services are centered upon an assessment of creditworthiness that must be conducted and communicated to the (potential) client before she is bound by any credit agreement. 69 The Consumer Credit Directive of 2008 assigned great importance to creditworthiness assessment, which aims to hinder irresponsible commercial lending practices by creditors.100 Similarly, major emphasis was put on such assessment by the Mortgage Credit Directive of 2014101, which was adopted in the wake of the big financial crisis102: the subprime loan vicissitudes on the U.S. banking market had in the meantime taught the lesson that extending credit when the value of the mortgage that serves as a 68

94

MiFID II, art 25, para 4. An indicative list of non-complex financial instruments is given by MiFID II, art 25, para 4, lit. (a). The criteria for assessing which financial instruments are to be considered as non-complex are spelt out by art. 57 MiFIR. For indications of compliant implementation, see ESMA, ‘MiFID II Supervisory briefing. Appropriateness and execution-only’, 4 April 2019, ESMA35-36-1640. 96 See also MiFID II, recital 85, in that it excludes that a service may be considered as provided at the initiative of the client insofar as ‘the client demands it in response to a personalized communication from or on behalf of the firm to that particular client, which contains an invitation or is intended to influence the client in respect of a specific financial instrument or specific transaction’. 97 CESR, ‘MiFID complex and non-complex financial instruments for the purposes of the Directive’s appropriateness requirements’, 3 November 2009, CESR/09–559. 98 Case 604/11 Genil 48 SL and Comercial Hostelera de Grandes Vinos SL v Bankinter SA and Banco Bilbao Vizcaya Argentaria SA, ECLI:EU:C:2013:344. For a comment, see Grundmann, The Bankinter Case on MIFID Regulation and Contract Law, (2013) 9 European Review of Contract Law 267. 99 IDD, art 30. 100 Atamer, Duty of Responsible Lending: Should the European Union Take Action?, in: Grundmann/ Atamer (eds.), supra (fn. 77), at 179; Osuji, Responsible Lending: Consumer Protection and Prudential Regulation Perspectives, in: Fairweather/O’Shea/Grantham (eds.), Credit, Consumers and the Law: After the Global Storm, Routledge 2017, 62 et seq. 101 Anderson/Arroyo Amayuelas (eds.), The Impact of the Mortgage Credit Directive in Europe: Contrasting Views from Member States, Europa Law Publishing 2017. 102 Schmidt/Esplugues/Arenas (eds.), EU Law after the Financial Crisis, Intersentia 2016. 95

198

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:21 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

collateral is below the threshold required to cover the debt of the borrower, or when the overall burden of the debt is disproportionate to the borrower’s overall financial situation, means creating a risk that may spread to the overall financial system, and in any case risks undermining the trust of consumers in such a system.103 In order to promote practices of responsible lending, therefore, the Mortgage Credit Directive applies two main devices: 1) an appropriate evaluation of residential immovable property;104 2) an assessment of creditworthiness that is made prior to the conclusion of any credit arrangement.105 Furthermore, European legislators have required the Member States to support financial education of consumers, an idea that equally serves the purpose of promoting responsible lending practices by creditors.106 Under the Mortgage Credit Directive, assessment of responsible lending must be conducted by taking into consideration all necessary and relevant factors that can influence a consumer’s ability to repay the credit over her lifetime.107 In particular, in order to rank a potential borrower as creditworthy, it does not suffice that the value of the residential immovable property exceeds the amount of the credit; even less compelling is the assumption that such property’s value will increase.108 Furthermore, the capacity of the consumer to transfer part of the credit risk to a third party should not lead the creditor to ignore the conclusions of the creditworthiness assessment.109 The source of information on which creditworthiness is to be assessed is the potential borrower herself,110 although the creditor has the duty to verify the information appropriately, including through reference to independently verifiable documentation when necessary.111 In order to conduct such an evaluation, the creditor has the right to access credit databases, operated either by private credit bureaus or by credit reference agencies, and public registers.112 With regard to the Consumer Credit Directive, the ECJ stated: ‘Notwithstanding the pre-contractual information which must be provided […], the consumer may, before entering into the credit agreement, still need additional assistance in order to decide which credit agreement is the most appropriate for his needs and his financial situation’.113 Although all obligations envisaged are pre-contractual and no indication about the chronological order of their fulfillment was given by the legislature, it was held that ‘the assessment of creditworthiness means that the adequate explanations provided need to be adapted and that those explanations must be communicated to the consumer in good time before the credit agreement is signed, without this, however, requiring a specific document to be drawn up’.114 103

Mortgage Credit Directive, Recital 3. Mortgage Credit Directive, art 19 and recital 26. 105 Mortgage Credit Directive, art 18 and recitals 55–61. 106 Mortgage Credit Directive, art 6 and recital 29. Even if it is dealt with just one article, it is significant that it has been placed in an autonomous chapter, the second, of the directive; this choice of the European legislature is due to the importance it attributes to the financial education of consumers. See Patti, L’educazione finanziaria e la direttiva 2014/17/UE (sui contratti di credito ai consumatori relative a beni immobili residenziali), [2015] Contratto e impresa 1423. 107 Mortgage Credit Directive, art 18, para 1, recital 55. 108 Mortgage Credit Directive, art 18, para 3, recital 55. 109 Mortgage Credit Directive, recital 57. 110 Mortgage Credit Directive, art 18 and recital 58. 111 False or inaccurate information as such does not entitle the creditor to terminate the contract (Mortgage Credit Directive, recital 58), unless it is demonstrated that the consumer knowingly withheld or falsified it (Mortgage Credit Directive, art 20, para 3, subpara 2). 112 Mortgage Credit Directive, art 21 and recital 59. 113 Case 449/13, CA Consumer Finance SA v Ingrid Bakkaus, Charline Bonato (née Savary) and Florian Bonato, ECLI:EU:C:2014:2464, para 41. 114 Case 449/13, supra (fn. 113), at para 41. 104

Sirena

199

70

71

72

73

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:21 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law 74

75

76

77

78

79

80

Although both European directives on credit services clearly set out that the creditor is obliged to give ‘adequate information’ to a potential client,115 they did not go so far as stipulating that the creditor is required to provide advice to the consumer. In particular, therefore, the creditor is not obliged to advise the consumer as to the credit agreement that is best suited to her interests, nor does the creditor have a duty to refrain from entering into a credit agreement that proves inappropriate in light of the consumer’s interests (e.g., when it puts the consumer in a situation of over-indebtedness). Advice is governed by the credit service directives as a financial service in its own right.116 Such an obligation of advising the consumer had been provided by the first proposal for the Consumer Credit Directive,117 but it was later abandoned in the modified proposal of 2005.118 This amendment was explained by the European Commission by saying that, in response to a request by the banking sector and some of the Member States, it was compelled to declare that it is the consumer who is ultimately responsible for her final decision whether or not to enter into a given credit agreement.119 The duty of the creditor is, therefore, to put the consumer in a position to adequately assess the advantages and disadvantages of a credit agreement and to strike the appropriate balance. This framework is consistent with the rules enacted by European legislators with regard to financial services. The MiFID II directive designates advice as an autonomous service;120 the same applies to insurance services.121 Nevertheless, the ECJ came to the conclusion that the Consumer Credit Directive does not prevent a national legislation (like the Belgian one) from stipulating that the creditor and the credit intermediary are obliged to ascertain amongst the credit agreements that they usually offer or intermediate the type and amount of credit most suitable (i) to the overall financial situation of the consumer at that moment and (ii) to the objective pursued by the contract.122 Moreover, according to article 18, paragraph 5, lit. (a), of the Mortgage Credit Directive, the creditor is to refrain from entering into a credit contract where the result of the creditworthiness assessment indicates that the consumer is not likely to meet in the required manner the obligations prospectively arising from the credit agreement. On the contrary, the Consumer Credit Directive ‘does not contain any provision regarding the course of action to be taken by the creditor in case of doubts as to the creditworthiness of the consumer’. Based on its Recital 26 it has been tentatively assumed that, where the assessment of creditworthiness is negative, the creditor has to warn the consumer but is not obliged to refrain from entering into a credit arrangement with her.123 Yet, the ECJ came to the conclusion that the Consumer Credit Directive does not prevent a national legislation (like the Belgian one) from imposing such an obligation on the creditor ‘if he cannot reasonably take the view, following the check of the 115

Consumer Credit Directive, art 5, para 6; Mortgage Credit Directive, art 16 and recital 48. Mortgage Credit Directive, art 22 and recitals 63–65. 117 Proposal for a Directive of the European Parliament and of the Council on the harmonisation of the laws, regulations and administrative provisions of the Member States concerning credit for consumers, COM (2005) 443 final. 118 Amended proposal for a Directive of the European Parliament and of the Council on credit agreements for consumers amending Council Directive 93/13/EC, COM (2005) 483 final. 119 Case 377/14 Ernst Georg Radlinger and Helena Radlingerová v Finway a.s., ECLI:EU:C:2015:769, para 64; Case 58/18 Michel Schyns v Belfius Banque SA, ECLI:EU:C:2019:467, para 34. 120 MiFID II, art 24. 121 IDD, art 20. 122 Case 58/18, supra (fn. 119), at paras 35–36. 123 Rott, Consumer credit, in: Micklitz/Reich/Rott/Tonner (eds.), supra (fn. 37), at 199; Mak, supra (fn. 30), at 322. 116

200

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:21 3B2 9.1.580; Page size: 160.00mm  240.00mm

H. ‘Granular Legal Norms’ in the Financial Services Trade

consumer’s creditworthiness, that the consumer will be able to fulfill the obligation arising from the proposed agreement’.124

V. Some final remarks In summation, the trend fostered by European law towards a personalized contract is dependent on the fiduciary obligation which is characteristically assigned to the traders of financial services; it impinges on the duty to act in the best interest of the client.125 Accordingly, the same trend is not to be broadened to general contract law, where no duties of confidence or trust bind one contracting party, even if a professional, towards the other one, even if a consumer. Although most European countries may wish to adhere to good faith requirements in negotiations and the performance of a contract, this cannot imply an obligation to act in the best interest of the other party and to advise the latter prior to and during the performance of contract. The starting point is that European contract law has been designed and enacted as a technique to overcome the fragmentation of national markets and to construct a single market. From a constitutional viewpoint, the conferral of legislative competence to the Union on private law is aimed at the establishment and functioning of the internal market and solely to this extent does the European Union have an undisputed competence to adopt legislative measures for the approximation of national laws, at least where the ordinary legislative procedure is followed126. In other words, the primary goal of European contract law is that of creating a ‘level playing field’ for businesses and, therefore, to remove national hurdles and barriers to cross-border transactions. It is undeniable that consumer protection should be conceived against the backdrop of a continuum of weaker parties,127 which may eventually encompass small and medium-sized enterprises. However, the justifications for such protection must be consistent with the foundations of European contract law, a framework that is fashioned not only by the multi-levelled interplay with the national laws of Member States, but also by its institutional goal of securing the establishment and functioning of an internal market (article 114 and article 26 TFEU).128 As has been demonstrated129, the weakness of consumers as such is deemed to justify consumer protection only insofar as it creates the risk of a failure of internal market; in that case, European legislators and policymakers are called upon to enact appropriate legislative regulations that aim at neutralizing the risk. By contrast, if consumers are affected by individual and subjective weaknesses – on account of cognitive errors and bias (like most factors vitiating contractual will)130 – justification for legislative regulation is not self-evident. Instead, each measure of the kind must be supported and justified by mandatory reasons of public good that have to be demonstrated by the European legislator. 124

Case 58/18, supra (fn. 119), at para 49. See para IV.1. 126 Cf Gutman, The Constitutional Foundations of European Contract law: A Comparative Analysis, OUP 2014, 277 et seq. and Kaupa, The Pluralistic Character of the European Economic Constitution, Hart 2016. 127 Grundmann, supra (fn. 42), at 224. 128 See para II. 129 Grundmann, supra (fn. 42), at 224. 130 Patti, ‘Fraud’ and ‘Misleading Commercial Practices’: Modernising the Law of Defects in Consent, (2016) 12 European Review of Contract Law 307. 125

Sirena

201

81

82

83

84

85

86

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/08_P. Sirena, Granular legal norms in the financial services trade, 08.07.2019.3d from 30.10.2020 11:36:21 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law 87

88

89

90

91

Insofar as cognitive errors and bias expose the consumer to the risk of an economic loss, the loss is going to be suffered by that individual herself and no regulatory intervention by European legislators would be justified. Different is the case where cognitive errors and bias expose the consumer to the risk of an existential loss. A loss may be deemed existential not only when health and human life or other inviolable rights of the person are endangered, but also when a threat is posed to interests, albeit economic interests, that may affect other existential aspects of the consumer.131 This is generally the case of financial services – not only because they may well create the danger of an economic loss much higher than that which may be incurred day-to-day and therefore affect the consumer for the rest of her life, but also because this kind of service may involve pensioning funds and other resources ensuring financial security at later stages of life. Secondly, to better serve the purpose of personalizing financial services contracts, greater personalization of information rules (particularly of disclosure) may be welcome, but it is not likely to represent a significant innovation. In fact, information is too complex to be personalized in a way that can actually enable the client (particularly the consumer) to master it. The role of intermediaries as ‘honest brokers’ is still paramount. Therefore, such intermediaries should be explicitly obliged to provide in the precontractual stage fitting advice for any prospective clients. Particularly, they should be held liable in damages in the event clients enter into a contract that creates overindebtedness. 131

202

Grundmann, supra (fn. 42), at 232.

Sirena

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics? The Case of Consumer Law* I. Clarification and Argument “Imagine a world where lawmakers enact a catalog of precisely tailored laws, specifying the exact behavior that is permitted in every situation. The lawmakers have enough information to anticipate virtually all contingencies, such that laws are perfectly calibrated to their purpose – they are neither over- nor underinclusive. Now imagine that when a citizen in this world faces a legal decision, she is clearly informed of exactly how to comply with every relevant law before she acts. This citizen does not have to weigh the reasonableness of her actions, nor does she have to search for the content of a law. She just obeys a simple directive. The laws at work in this world are not traditional rules and standards. Instead, they take a new form that captures the benefits of both rules and standards without incurring the costs. This new form – we call it the microdirective – is the future of law.” This is the introductory passage of Casey and Niblett’s The Death of Rules and 1 Standards.1 What the authors call micro-directives is perhaps the most radical form of personalized law that one could currently imagine. Translated into consumer policy and law, the scenario looks like this:2 Predictive and communication technology replaces the existing consumer law composed of rules and standards. The interplay between technology and consumer law occurs in four steps, (1) these technologies will transform consumer policy objectives into rules to achieve the objective, (2) they will identify the specific rules applicable in concrete circumstances, (3) they will translate specific rules into micro-directives on how the consumer can comply with the law, and (4) they will communicate that micro-directive to the consumer. The authors are not envisaging a normative scenario and are aware of the political problematique – the increasing role of the executive, the decreasing role of the judiciary, the philosophical and ethical issues, not least the “horror juris”. They are launching a double-edged wake-up call. The troublesome scenario runs like this: business does and will do what is possible, legal, and efficient. The more appeasing scenario goes like this: human beings have to do the programming.3 There are less provocative scenarios for a personalized law; these carefully weigh what 2 is technologically feasible against what is societally desirable. Porat and Strahilevitz investigate the future of personalized default rules, underlining that data privacy is perhaps the most important barrier which prevents personalized law.4 Ben-Shahar and * I would to thank my colleagues from the Sachverständigenrat für Verbraucherfragen, the Advisory Council for Consumer Affairs at the Federal Ministry of Justice and Consumer Protection. I could not have written this paper without the support of Giovanni Sartor, Bologna/EUI, and Przemyslaw Palka, Yale, who taught me what I know about AI, digitalization and big data analytics. Responsibility for errors remains mine alone. 1 Casey/Niblett, supra p. 1; Casey/Niblett, A Framework for the New Personalization of Law, University of Chicago Law Review, 68, 2019, 255-282. 2 Casey/Niblett, supra Part 1.C, at Part 1 B p. 10. 3 Casey/Niblett, supra Part 1.C, at Part 1 B p. 10. 4 Porat/Strahilevitz, Personalizing Default Rules and Disclosure with Big Data, Michigan Law Review, 112, 2014, 1417-1478. Part 1.A. at p. 12-17.

Micklitz

203

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

Porat trace the implications of personalized law on mandatory rules.5 Consumer law plays a prominent role in terms of personalized law, be it in the form of consumer as patient or as buyer making use of the right to withdraw or the right to information. Are we on the move towards personalized consumer law or is it just a potential scenario? Wagner and Eidenmüller6 call for regulatory safeguards to tame the incoming tide of personalized law, Elkin-Koren and Gal7 warn that incentives for businesses to collect data might be fading and therewith personalized law. 3 I am trying to capture the insinuated development in the move from typification, via granularization to personalization. All three deserve to be clarified. Typification sounds like a category from the past. Legal generalizations and typifications – already known in Roman law and later in codified private laws – relate back to practical reason.8 Granularity or granularization does not have such a clear-cut meaning. Granular norms are the result of an ever finer, ever more sophisticated fragmentation of the legal system. Initially consumer law was meant to follow rather established typification patterns, defining a consumer image, a consumer contract, and a consumer procedure, although adjusted to the particularities of consumer law, aiming at protection of the weaker party. Attempts to build a self-standing typology different from traditional private law failed.9 Over the last thirty years not only private law but also consumer law have become ever more fragmented. Legislators and behavioral economics, however, remained bound to a desperate search for typification. By and large they have failed to link different patterns of consumer images to different types of information requirements, remedies, or procedures. Consumer law is falling apart into ever more granular rules. 4 Granularization of consumer law precedes personalization. Granularization is the bridge between typification and personalization. Personalization and personalized law are the opposite of typification. In the digital economy and the digital society, each consumer may get their own law, tailored in line with individual preferences. Individual justice through technology seems to be on the horizon. In typified and granularized consumer law the default rules are tied to the distinction between the responsible, the confident, and the vulnerable consumer. In personalized consumer law the default rule is not the result of a decision made by law or the legislator, but the result of an algorithmic decision based on individual preferences, such as the degree to which a consumer is risk-averse or cares about the environment. These preferences may even 5 Porat/Ben-Shahar, Personalised Mandatory Rules in Contract Law, University of Chicago Law Review, 68, 2019, 255–282, University of Chicago Coase-Sandor Institute for Law & Economics Research Paper No. 855, University of Chicago, Public Law Working Paper No. 680, https://papers.ssrn.com/sol3/papers. cfm?abstract_id=3184095. 6 Wagner/Eidenmüller, Down by Algorithms? Siphoning Rents, Exploiting Biases, and Shaping Preferences: Regulating the Dark Side of Personalized Transactions, University of Chicago Law Review, 68, 2019, 581–609. 7 Elkin-Koren/Gal, Personalised Law, The Chilling Effect of Governance-by-Data on Data Markets, University of Chicago Law Review, 68, 2019, 403–431, warn to take data collection as a given. 8 Busch/De Franceschi, Granular’ Legal Norms: The End of Typification?, fn. 9, refer to Jhering, Geist des römischen Rechts auf den verschiedenen Stufen der Entwicklung, Part 2, Volume 1 Leipzig, 1854, 45 where he refers to the problem of ‘formal feasibility’ (formale Realisierung) of legal norms and Auer, Materialisierung, Flexibilisierung, Richterfreiheit: Generalklauseln Im Spiegel Der Antinomien Des Privatrechtsdenkens, Tübingen, 2005, 46; Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in: Mak/Tjong Tijn Tai/Berlee (eds.), Research Handbook on Data Science and Law, Edward Elgar 2018. 9 From a legal theoretical perspective Calais-Auloy, president of the Commission de refonte du droit de la consommation, Proposition pour un nouveau droit de la consommation, rapport de la commission de la refonte du droit de la consommation au secrétaire d’État auprès du ministre de l’Économie, des Finances et du Budget chargé du Budget et de la Consommation, 1985 ; from a sociological perspective, Roethe, Der Verbraucher: Rechtssoziologische Betrachtungen, Baden-Baden: Nomos, 2014.

204

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

differ in between persons that are grouped together under a particular consumer image. Two further clarifications are necessary. Preferences have to be distinguished from individual characteristics. Only individual characteristics make each and every person unique. Personalized law built around preferences is not the law of the “person”, but rather what we disclose about our preferences on the internet. What is not known about “me” but what is needed for the algorithmic decision is accomplished from proxies. Therefore personalized law is a composite of my own preferences and the preferences that all those who belong to my “group” have voiced. The “oil” that allows for personalization of consumer law is Big Data analytics, but this in turn is dependent on the availability of “big data”. Big data is characterized by the three Vs: volume, velocity, and variety.10 The warning by Elkin-Klein and Gal is worth recalling, as personalized law depends on ever more and ever more sophisticated data on the person and on the law.11 The term “big data analytics” refers to techniques to extract that value from these data sets.12 This is what the paper is all about – the opportunities and the risks that result from the potential enshrined in big data analytics. However, I do not intend to go down the road of personalized law, on how far 5 personalization reaches or could reach. Two concerns dictate against personalization. The first has to do with the European approach to data privacy and the degree to which the GDPR may monitor and control the data economy.13 The second reaches deeper. To what extent has the “personality” to be shielded against data-driven personalization of the law, a process driven by private actors not by the law,14 not to mention the political economy of AI and machine learning?15 This is left for a separate paper. However, I argue that typification – even in a more granular form than the “legal subject”16 – helps to maintain the “moral integrity of the law”.17 Here is the argument: 6 If granular norms are the end of typification, then big data analytics are the means of re-typification. Can big data analytics be used to reduce granularization for the benefit of a refined typified law? Can big data analytics contribute to finding rationality in the thicket of rules? Could this be a viable alternative to the idea that big data analytics renders it theoretically possible to implement each and every rule through personalized law, without questioning the purpose, content, and legitimacy of these rules? The argument is developed in two steps. The first serves to give shape to granularized 7 and personalized consumer law prior to and after big data analytics. The stock-taking documents the validity of Casey and Niblett’s claim. The legislator has gone far in granularizing consumer law. Personalization appears to be the logical consequence. In the second step I will outline the potential of big data analytics to re-typify consumer law along different consumer images: by looking into the perspectives of better lawmaking, by adjusting regulation of consumer law to the ‘real problems’ of consumers and by reducing the complexity of information rights and cross-border consumer 10 De Mauro/Greco/Grimaldi, A Formal definition of Big Data based on its essential Features, 2016, Library Review, 65: 122–135. 11 Elkin-Koren/Gal, supra (fn 8). 12 Palka/Lippi, Big Data Analytics, Online Terms of Services and Privacy Policies, on file with the author. 13 Elkin-Koren/Gal, supra (fn 8). 14 But who personalizes is all too often forgotten, Verstein, Privatising Personalized Law, University of Chicago Law Review, 68, 2019, 551–580. 15 Benkler, The Role of Technology in Political Economy, in Law and Political Economy, 2018 https:// lpeblog.org/author/ybenkler/. 16 Demogue, La notion de sujet de droit, Revue Trimestrielle de Droit Civil 8, 1909, 611–655 (611–631); Kennedy/Belleau, La place de René Demogue dans la généologie de la pensée juridique contemporaine, Revue Interdisciplinaire d’Etudes Juridiques, 2006, 163–211. 17 Allan, Dworkin and Dicey: The Rule of Law as Integrity, 1988, OJLS 8 (2): 266–277.

Micklitz

205

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

litigation – through big data analytics. Whether such a scenario could become reality remains to be tested in legal scholarship.

II. From Typification to Granularization prior to Big Data Analytics 8

Three different modes of granularization of European consumer law can be distinguished: differentiation of the consumer image; transformation from consumer law 1.0 to consumer law 2.0; and the rise of technical standards, codes of conduct and standardized terms giving shape to consumer law 2.0 below the radar of formal lawmaking. These different modes of granularization do not result in personalized legal relations. However, they demonstrate that consumer law governed by typification is vanishing.

1. Consumer images The rise of consumer law in the Western democracies has been built on the idea that the consumer is “the weaker party”. Consumer law therefore meant protecting the weak against the strong, the “little” consumer against the “big” supplier. The decline of national consumer law and policy from the late 1970s onwards goes hand in hand with the rise of European consumer law and policy. The EU changed the outlook from consumer protection law to consumer law without protection.18 The average European consumer was the well-informed, circumspect responsible consumer, needed to complete the internal market. Gradually the EU realized that such a strong normative consumer image does not match the varieties of consumer images in the Member States. After fierce debates the EU adopted a diversified consumer image in the Directive 2005/ 29, with the “vulnerable” consumer complementing the “average consumer”.19 10 From this moment on the consumer could no longer be equated with being weak. There were variations of weaknesses: the responsible consumer, the confident consumer, and the vulnerable consumer. In my opinion for the German Juristentag in 2012 I argued that the threefold distinction of the consumer image can be associated with different degrees of protection, different degrees of justice, and different rights and remedies.20 Such a model – said to be enshrined in consumer law de lege lata – typifies the consumer and ties legal consequences to the different consumer statuses. 9

2. Consumer law 2.0 11

The second form of granularization results from the move from consumer law 1.0 to consumer 2.0, from the transformation of the consumer society into the service society21 and now the digital society.22 The consumer law of the 1950s to the 1980s implicitly or explicitly starts from the sale of goods as a blueprint for regulatory action. The service society shifted the focus from the sale of goods to the sale of services, from markets in 18 Micklitz, The Expulsion of the Concept of Protection from the Consumer Law and the Return of Social Elements in the Civil Law: A Bittersweet Polemic, 2012, 35 Journal of Consumer Policy 283–296. 19 Wilhelmsson, in: Howells/Micklitz/Wilhelmsson, European Fair Trading Law: The Unfair Commercial Practices Directive, Ashgate, 2006, ch. 3. 20 Micklitz, Do Consumers and Business need a New Architecture for Consumer Law? A Thought Provoking Impulse, Yearbook of European Law, Volume 32, No. 1, 2013. 21 Rosecrance, Rise of the Trading State: Commerce and Conquest in the Modern World, Basic Books 1986. 22 Micklitz/Reisch/Joost/Zander Hayat (eds.), Verbraucherrecht 2.0 – Verbraucherrecht in der digitalen Welt, VIEW Schriftenreihe, Nomos Verlagsgesellschaft, 2017.

206

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

goods to the market in services – legally speaking to regulated markets: telecoms, energy, finance, and transport. The consumer turns into the customer. This move already has its own language. Servicification describes an economy and a society where ownership loses importance and where the use and usability of products depends on access to a service, which is usually subscription based. This is the bridge to the digital society which adds a new layer to the development of consumer law that cuts across goods and services. Regulatory activities, if there are any, have not yet fully realized that digitalization 12 breaks down the barriers between the consumer and the service society. Whether digitalization reduces granularity or increases granularity remains to be demonstrated. The laws which follow the transformation waves of the economy and society could still be understood as a modern form of typification. The consumer image of the consumer society was the weaker party, whereas the consumer image in the service society takes at least three forms, namely the responsible, the confident, and the vulnerable, each of whom could in theory get their typified consumer law. What will be the image of the consumer in the digital economy? So far consumer law has not gone beyond the three images. However, behavioral economics and behavioral sciences demonstrate through empirical evidence that ever finer-grained images would be needed to match the particular needs of the consumer. Big data analytics looks like the tool to implement them. However, big data analytics has two sides, as shown below.

3. Contractual underworld The third form of granularization results from the increasing importance of non- 13 binding rules below the level of formal laws. These rules complement consumer law – technical standards, codes of conduct, and standardized terms. Each of these different techniques have in common that they are not binding. They contradict the idea that contracts can be typified through the legislator. The private regulators behind the rules introduce an extremely broad range of variations of what a contract should be all about. The old distinction between sales and services is complement away after miles from the regulatory practices that govern the contractual underworld. The so called New Approach, adopted in 1985, promoted the use of technical 14 standards as a tool to build and complete the Internal Market.23 These technical standards shape the level of conformity to which products have to comply so as to be marketed in the EU.24 The EU undertook various efforts to extend the New Approach to services. In practice, standardization of services goes hand in hand with the shaping of regulated markets in the service society.25 A prominent example is the establishment of the Banking Union. In the field of services the line between contractual obligations agreed upon between the parties and non-binding obligations resulting from technical standards becomes blurred and ever more difficult to distinguish.26 Codes of conduct, in particular when connected to Corporate Social Responsibility, 15 affect the legal relationship between the consumer and the supplier in various ways.27 23 Council Resolution of 7 May 1985 on a new approach to technical harmonization and standards, OJ No. C 136, 4.6.1985, 1. 24 Pelkmans, The New Approach to Technical Harmonization and Standardization, JCMSt. 1987, 249. 25 Busch, DIN-Normen für Dienstleistungen – Das europäische Normungskomitee produziert Musterverträge, in: Neue Juristische Wochenschrift (NJW), 2010, 3061–3066. 26 Micklitz, Services Standards: Defining the Core Elements and Their Minimum Requirements, study commissioned by ANEC, 2007, http://www.anec.eu/attachments/ANEC-R&T-2006-SERV-004final.pdf. 27 Beckers, Enforcing Corporate Social Responsibility Codes: On Global Self-Regulation and National Private Law, Hart Publishing 2016.

Micklitz

207

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

These are new regulatory tools which promote granularization. They concretize – directly or indirectly – the binding regulatory framework for the consumer, the service and the digital society. They enlarge the variety of rules considerably, to an extent where the sheer quantity of rules reaches beyond manageability.

4. Consumerization of consumer law Consumer law is paradigmatic for the explosion of norm production. The body of European consumer law has grown from the mid/late 1980s from scattered highly targeted problem-solving to ever more comprehensive and voluminous rules that have gradually but steadily penetrated economic law.28 Secondary EU law was initially rather small in scope and short on words. Over time the respective EU consumer rules have become longer and longer, ever more detailed and ever more directionist. If one adds to the body of consumer law the rules on regulated markets (telecoms, energy, transport, finance) which also cover – inter alia – consumer interests, the list gets ever longer and interferes in the realms of totally different legal fields: telecoms, energy, banking and finance, transport.29 The variety of services in regulated markets does not fit into the 19th century distinction of obligations de résultat (Werkvertrag, contracts to manufacture) and obligations de moyen (Dienstvertrag, service contracts).30 17 The EU legitimizes the explosion through its mandate to complete the Internal Market at a ‘high level of consumer protection’ (Article 114 TFEU). Market-building goes hand in hand with enhancing consumer protection through law. The EU insinuates that the position of the consumer in the Internal Market can be improved through ever more granularized rules that follow the fragmentation of (regulated) markets and/or global value chains. Granularization as promoted through the European Digital Market31 no longer follows a consumer policy guided by a particular consumer image that would allow for typification. Like an octopus, consumer law takes a grip on ever broader areas of the digital and the analogous economy. The Dworkinian judge is more often than not faced with the problem of making sense out of nonsense, to concretize vague rules, to solve contradictions in legislative acts, to give shape to responsible, confident, and vulnerable consumers, sometimes for the better, sometimes for the worse. Servicification has led to agencification; public officials in regulatory agencies are confronted with a plethora of rules that they can hardly administer.32 In their daily practice they (must) behave like free-standing judges. Judicial review is the exception to the rule. Digitization and big data analytics upgrade private parties as regulators and even as (law) enforcers. In the digital economy and society, the regulator is not necessarily the legislator, but the 16

28 Schulze/Zimmermann, Europäisches Privatrecht – Basistexte, 5th edn. 2016, have divided their collection into three categories, Unionsrecht (Law of the Union), Einheitsrecht (uniform law), Gemeinsame Rechtsgrundsätze (Common Principles). The collection comprises 900 pages, not all are on consumer law, but nearly all are relevant for consumers. The collection would be much longer if the editors had also included the relevant rules on regulated markets. 29 The Florence School of Regulation is a perfect place to study the different silos, each built out of lawyers, economists and technicians as far as needed, from legal practice, from lawyers, companies, and academics. 30 This is the big deficiency of the Study Group project on services. The group did not look into regulated markets, at least not systematically, see Barendrecht/Jansen/Loos/Pinna/Cascao/and van Gulijk, Principles of European law. Service Contracts, Sellier European Law Publishers, 2007. 31 Shaping the Digital Market, https://ec.europa.eu/digital-single-market/en/policies/shaping-digitalsingle-market. 32 Scholten/van Rijsbergen, The Limits of Agencification in the European Union, 2014, 15 German Law Journal 1223–1256.

208

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

private party, which translates in our current daily life into the big five: Apple, Amazon, Facebook, Google, Microsoft.

III. From Granularization to Personalization through Big Data Analytics Is big data analytics able to reduce the complexity of consumer law rules or is it just 18 doing the contrary, increasing complexity where personalized law appears as a kind of logical consequence? Is big data analytics shifting the power from public to private regulation? Can big data analytics help to re-enthrone the public legislator, enabling the legislator to adopt rules and standards, rules on clear verdicts and standards which are more than just regulatory labels, impossible to enforce? In order to answer these questions it is necessary to distinguish between digitalization and big data analytics in regard to their impact on granularization. Digitalization is claimed to work in one direction only – increasing granularization and hollowing out typification. Big data analytics operates in two directions: Casey and Niblett predict the death of rules and standards. Porat and Strahilevitz and Ben-Shahar and Porat elaborate the contours of personalized law through big data analytics for identifying individual preferences and linking them to the existing body of rules, thereby promoting individual justice. Big data and big data analytics, however, can also be used to re-empower the legislator, to reinstall the law and to save its “integrity”.

1. Digitalization Digitalization enhances granularization through “cost free” services, through disloca- 19 tion of access and dislocation of services and last but not least through the key role of intermediaries. All four phenomena are well known but the implications for legal systems remain under-researched and the consequences for consumer law are not fully thought through. Consumers are “paying” for access to internet services with their data. Palka argues 20 that the internet service providers have introduced a new kind of contract which cannot be subsumed under existing contract law schemes. It is an “as if” relation, which follows the logic of the internet but not that of offer and acceptance or – in the common law world – consideration. Both parties behave as if they have concluded a contract, whilst this might legally not (yet) be the case. “As if” relations create barriers in the control of standard contract terms.33 Consumers know that they are paying with their data, but do not want to know what exactly happens to their data – even if worried about leaving personal data to the supplier. The potential price of data is hard to estimate.34 Whether consumers have a legal right to their data – the “my data rhetoric” – is controversial.35 33 Palka, Terms of Service Are Not Contracts: Beyond Contract Law in the Regulation of Online Platforms, in: Grundmann (ed.) European Contract Law in the Digital Age, Intersentia 2018. 34 Palmetshofer/Semsrott/Alberts, Der Wert persönlicher Daten – Ist Datenhandel der bessere Datenschutz?, study commissioned by the SVRV, Open Knowledge Foundation Deutschland e. V., 2017, http://www.svr-verbraucherfragen.de/wp-content/uploads/Open_Knowledge_Foundation_Studie.pdf. 35 Critical statement by the European Data Protection Supervisor EDPS Opinion 8/2018 on the legislative package of A New Deal for Consumers, 6 October 2018 on data as counter-performance; equally critical Kerber, Digital Markets, Data and Privacy: Competition Law, Consumer Law and Data Protection, GRUR Int. 2016, 639; Kerber, A New (Intellectual) Property Right for Non-Personal Data? An Economic Analysis, GRUR Int. 2016, 989; for the opposite position representative data property as a civil right, Fezer, Repräsentatives Dateneigentum, Ein zivilgesellschaftliches Bürgerrecht, Konrad-Adenauer Stiftung 2018.

Micklitz

209

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

Uncertainties abound which can hardly be solved with legal tools designed to decide over the existence of a contract.36 21 The major internet providers – the big five – are located in the United States. This has implications for jurisdiction, for the applicable law, and for execution of judgments outside the EU. The Brussels Regulation, the Rome I and Rome II Regulation provide particular rules on consumer protection, on the right of the consumer to sue the contracting partner at the consumer’s place of residence, and on the applicability of their national law subject to the most favored protection principle, not least through the possibility to execute judgments across EU borders under relatively simple requirements. However, in practice there is little evidence that individual consumers are suing one of the big five at their place of residence, thereby successfully relying on their national rights. The big five try to impose on the consumer, if not the place of jurisdiction then the law where they have their business seat in Europe or in the USA. Consumer organizations have tried in vain to get choice of law clauses in standard terms prohibited Europe-wide through an action for an injunction.37 The promising outlook for pro-consumer rules unfolded limited practical effect. International private law is of crucial importance in terms of the conceptual and theoretical relationship between different legal orders. However, international private law is not a reliable legal source to enforce consumer rights against the big five. There are simply too many legal technical problems to be overcome. 22 Digitalization brought platforms into a prominent position. Platforms understand themselves as intermediaries like brokers and agents.38 They do not take responsibility for the quality and correctness of the information they transmit. Only under exceptional circumstances may they be held liable, that is, if they behave and appear as a contracting partner of the service requested by the consumer.39 So far legal research has not made a cross-cutting effort to study the role of intermediaries, their responsibilities, and their liabilities in the EU.40 Intermediaries are discussed in relation to their role in the relevant markets, mainly in the financial markets or in housing. The rise of platforms has triggered discussion on whether platforms should be held liable beyond the existing threshold in the e-commerce Directive. The European Law Institute (ELI)41 drafted a proposal which advocates a kind of subsidiary liability. However, even if Google or Amazon would be liable, international private law rules apply in the case of a conflict. Taking all the different phenomena together, digitalization promotes granularization, provokes legal uncertainty, and hollows out typification. 36 This is much more fully developed in the Opinion of the SVRV in consumer law 2.0. http://www.svrverbraucherfragen.de/en/wp-content/uploads/sites/2/Report-1.pdf. 37 ECJ Case C-191/15 Amazon ECLI:EU:C:2016:612, http://curia.europa.eu/juris/document/document. jsf;jsessionid=720B5BE6F59D6AAFEF85EC579BE6C811?text=&docid=182286&pageIndex=0&doclang= EN&mode=lst&dir=&occ=first&part=1&cid=3421077. The Israeli Supreme Court came to a similar solution with regard to choice of law clauses of Facebook. 38 Micklitz/Adam, Information, Beratung und Vermittlung in der digitalen Welt, Working Paper 6/ 2016 of Sachverständigenrat für Verbraucherfragen, December 2016, http://www.svr-verbraucherfragen. de/wp-content/uploads/SVRV_WP06_Information_Beratung_Vermittlung.pdf, 122 pages; shortened version in Micklitz/Adam, Verbraucher und Online Plattformen, in: Micklitz/Reisch/Joost/Zander Hayat (eds.), Verbraucherrecht in der digitalen Welt, Nomos Verlagsgesellschaft, VIEW Schriftenreihe, 2017, 45–102. 39 Reference from the German Supreme Court to the ECJ in September 2018 on the liability of Youtube for property rights infringements. http://juris.bundesgerichtshof.de/cgi-bin/rechtsprechung/document.py? Gericht=bgh&Art=pm&pm_nummer=0150/18. 40 Sartor, Providers Liability: from the E-Commerce into the Future; In-Depth Study for the European Parliament, Director General for Internal Policies IP/A/IMCO/2017-07 PE 614/179. 41 https://www.europeanlawinstitute.eu/projects-publications/current-projects-feasibility-studies-andother-activities/current-projects/online-platforms/.

210

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

2. Big data analytics What if companies use their data on consumer preferences, personalize advertising, 23 information, advice, prices and quality and even provide the ground for a personalized law? So far the reality looks different, a conclusion which is of relative importance only, in light of the speed of technological development. Three types of research could be invoked: The first is a project on personalized prices, 24 undertaken by the Advisory Council for Consumer Affairs at the Federal Ministry of Justice and Consumer Protection (SVRV).42 The SVRV mandated and managed the project. The second is the ARTSY project commissioned by the Berkman Klein Centre at Harvard University and executed by the European University Institute on an AI observatory.43 It documents inter alia how business uses artificial intelligence44 in seven sectors of the economy: finance and insurance, information services, energy and “smart solutions”, retail, autonomous vehicles, healthcare, and legal services. For each sector analyzed the gains for businesses stemming from the deployment of AI were studied, as well as challenges for consumers, and third party effects. The third source is the report on Consumer-friendly scoring of the Advisory Council of Consumer Affairs,45 which refers to an empirical research series on the current relevance of scoring in Germany. The findings at the time of writing are as follows. In an extremely fast moving 25 research field these findings have to be tested against ongoing research: There is little evidence on personalized pricing, some on personalized advertising, personalized information and personalized advice. Despite so far limited evidence, there is lively discussion on the pros and cons for the individual consumer and for competition law.46 Scoring is most developed in the credit business and is on the rise in health care and in the telematics tariffs of health insurance. Suppliers do not (yet) use machine learning techniques.47 The “score” is personalized in that each consumer has a credit score, which is used by banks, by insurers, by landlords, and by employers as one decisive element in whether or not to offer credit, insurance, a tenancy, or an employment contract. The score sets a benchmark for decision making as defined by the business. Suppliers use proxies when they do not have personal data or when the accessible data are not sufficient to define a personal score. To what extent is the score personalized? Only businesses could fully answer the question if they were willing to disclose the relevant information on the score to the consumer and/or the supervisory authorities. 42 Schleusner/Hosell, Expertise zum Thema “Personalisierte Preisdifferenzierung im Online-Handel”, (Opinion on personalised pricing in online trade), 2016, http://www.svr-verbraucherfragen.de/wp-content/uploads/eWeb-Research-Center_Preisdifferenzierung-im-Onlinehandel.pdf. 43 Micklitz/Jabłonowska/Kuziemski/Nowak/Pałka/Sartor, Consumer law and artificial intelligence Challenges to the EU consumer law and policy stemming from the business use of artificial intelligence, EUI Working Paper Series, Law 2018/11, 89 pages. 44 Ibid., 7 ‘When speaking of “artificial intelligence” we refer to a socio-technological practice of companies (or other actors) using machine learning tools to generate computer-readable knowledge out of big amounts of data, and further use that knowledge to optimize certain processes and undertake new types of actions, for example to predict consumer (individual/group) behavior, influence it, take decisions, etc. The entities performing these tasks are called artificial agents. Depending on the type of agent or its task, the exact legal challenges and regulatory responses might be very different in content and in form.’ 45 https://www.svr-verbraucherfragen.de/en/wp-content/uploads/sites/2/Report-2.pdf. 46 Bar-Gill, Algorithmic Price Discrimination When Demand Is a Function of Both Preferences and (Mis)perceptions, University of Chicago Law Review 86, 2019, 216–254, stressing the risks but also pointing to the opportunities; see also Gillis/Spiess, Big Data and Discrimination, University of Chicago Law Review, 68, 2019, who discuss the tension between existing laws which prohibit discrimination and algorithmic pricing decisions offers. 47 However, the Edinburgh Centre of Consumer Credit is currently testing ML techniques in scoring.

Micklitz

211

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

China aims at developing a social score for each citizen, which can then be used to steer their behavior in the economy and society.48 As the project is driven by the Chinese government one might wonder whether and to what extent Chinese business will have access to data – probably only as far as the consumer’s behavior in the economy can be used for wider social and political purposes.49 26 Big data analytics allows collection of all sorts of data on consumer preferences which could be used for personalized advertising and sales promotion strategies, for adjusting quality and prices. There is a dark side but there is also a bright side. The new digital world provides endless opportunities to discriminate against consumers because of their social status and their willingness to pay. Cass Sunstein speaks of the right not to be manipulated.50 Companies may find and exploit correlations between a consumer’s preferences and their propensity to react to ads, offers, and other messages, and consequently to address differently consumers – rich-poor, young-old, male-female, hetero-homo, open-closed personality. Such scenarios are in line with Caplovitz’s51 ‘The Poor pay More’. However, another scenario is also possible, namely the one stressed at the beginning of this paper. Personalized law allows one to imagine a world in which the rich pay for the poor, in which the advantages of a competitive legal order are not necessarily reaped by the stronger and more circumspect consumer. In this perfect world each and every individual would pay the price they can afford to pay in relation to their income. It remains to be seen whether the EU General Data Protection Regulation could serve as a useful tool for a fair balance between the conflicting scenarios.52

IV. Big Data Analytics in Law Making and Law Enforcement There is not yet much big data in European consumer law making and in consumer law enforcement, let alone big data analytics. However, there is considerable need for both in improving law making and law enforcement. So far there is an enormous mismatch in what the legislator – or, more broadly speaking, the three democratic powers – know about the consumer and what business knows about their preferences. Paraphrasing Whitman,53 the discourse on personalized law in the USA is driven by consumerism, while the upcoming debate on personalized law in the EU is guided by producerism. 28 The following sets the scenario for what might be possible if European politics and European legal scholarship is ready to swim against the tide. There is potential in the use of big data analytics, if properly and responsibly applied.54 For consumer law, consumer lawyers, and perhaps for law and lawyers, the age of big data analytics offers opportunities to investigate where the law is positioned in between all these rules and 27

48 Sithigh/Siems, The Chinese Social Credit System: A Model for Other Countries? EUI Department of Law Research Paper No. 2019/01; also https://www.chinalawtranslate.com/seeing-chinese-social-creditthrough-a-glass-darkly/?lang=en. 49 On the political role of consumer policy and consumer law in China, Trentmann, The Empire of Things, 2016. 50 Sunstein, Is there a right not to be manipulated?, https://www.ucl.ac.uk/pals/events/2020/jan/thereright-not-be-manipulated-cass-sunstein-harvard-law-school. 51 Caplovitz, The Poor Pay More, 1967. 52 Hacker, Teaching fairness to artificial intelligence, Existing and novel strategies against algorithmic discrimination under EU Law, 55 CMLRev 1143–1186, 2018. 53 Whitman, Consumerism Versus Producerism: A Study in Comparative Law, 117 Yale L.J. 407. 54 Lippi/Contissa/Jabłonowska/Lagioia/Micklitz/Palka/Sartor/Torroni, The Force Awakens: Artificial Intelligence for Consumer Law, Journal of Artificial Intelligence Research, 2020, Volume 67, 169–190.

212

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

whether and to what extent typification through big data analytics could turn into a political and societal alternative to business-driven personalized law. A crucial twist in perspective is needed though. Personalized law focuses on individual consumer preferences, while typification through big data analytics has to take collective consumer problems as the starting point. This is a shift from the individualist perspective to one that takes into account the collective dimension of consumer law and policy; it is also a shift in disciplines and in politics, from economic efficiency to social effectiveness. Individual preferences relate to behavior; in the current dominance of behavioral economics the yardstick for measuring preferences risks being reduced to economic efficiency. Socio-legal societal effectiveness might clash with economic efficiency. Both have their part to play. The focus on consumer problems lays the emphasis on the societal side, on effectiveness, on socio-legal research.

1. Evidence based policy through big data analytics In terms of making consumer law, the currently dominating rhetoric of “evidence- 29 based consumer policy” springs to mind.55 The European Union – more specifically the European Commission – is at the forefront to “improve” law making through “evidence”. The current political design is enshrined in the “Refit programme”,56 the latest version of what was previously termed “better regulation”57 or “smart regulation”.58 These initiatives find their origin in the Lisbon Agenda 200059 and the move towards “Governance” in 2002.60 The latter generated the notion of “impact assessment”. All these activities are united in one and the same goal, namely to make consumer law more efficient – the cost-benefit dimension – and more effective – the societal dimension. Both objectives imply that law making is linked to the consumer, their needs, their problems, and that regulatory tools fit. The politics behind evidence-based consumer policy deserves a separate analysis.61 The ‘Consumer Refit’ – the revision of six major consumer law directives – has 30 produced an enormous amount of data. A rough calculation shows that the consumer refit initiative has produced some 3500 pages in less than two years.62 If big data could be equated with “tonnage”, then “Consumer Refit could be regarded as a milestone”. What is missing, however, is big data analytics. Despite all the bombastic rhetoric and the considerable economic and human resources put into it, the Consumer Refit looks rather old fashioned. It is not designed for using big data analytics in European consumer law making. The potential alternative insinuates that the existing body of law by and large covers 31 the relevant consumer problems. In the most ambitious perspective consumer law making should start with a systematic evaluation of consumer problems in the three 55 https://ec.europa.eu/info/policies/consumers/consumer-protection/evidence-based-consumer-policy_en. 56 https://tem.fi/en/refit-programme-of-the-european-commission. 57 https://ec.europa.eu/info/law/law-making-process/planning-and-proposing-law/better-regulationwhy-and-how/better-regulation-guidelines-and-toolbox_en. 58 https://www.oecd.org/regreform/policyconference/46528683.pdf. 59 http://www.europarl.europa.eu/summits/lis1_en.htm. 60 https://ec.europa.eu/europeaid/european-governance-white-paper_en. 61 Micklitz/Villanueva, REFIT or Rethink – The Politics of EU research – A Grand Misunderstanding?, in: van Schagen/Weatherill (eds.), Better Regulation in EU Contract Law: The Fitness Check and the New Deal for Consumers, Hart Publishing, 2019, 37–59. 62 All information is freely available on the EU Commission website on the Consumer Refit – A new deal for consumers, https://ec.europa.eu/info/law/law-topic/consumers/review-eu-consumer-law-newdeal-consumers_en; on Consumer Refit see the contributions in van Schagen/Weatherill, supra (fn. 62).

Micklitz

213

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

societies; the consumer society, the service society, and the digital society. “Problems” are enshrined in consumer complaints, expressed towards peers, towards business, towards consumer organizations, consumer agencies, ADR bodies, and courts. The EU did not succeed in setting up a European wide standardized system for the collection of consumer complaints and/or of statistics on consumer litigation.63 The Eurobarometer serves as a substitute for getting a general overview.64 However useful it might be, shaky methodological ground undermines its viability. The Eurobarometer can only be as good and reliable as the national data on which it is based.65 In the Refit exercise the European Commission asked for a definition of consumer problems and for a crosscheck of the existing body of rules against consumer problems. However, the overall design of the project and the time left to execute the study in nine months throughout the (then) 28 Member States did not allow a serious evaluation of consumer problems in the fields under scrutiny. 32 One has to go to Brazil and China to study the potential of a nation-wide standardized electronic collection of consumer complaints for law making and law enforcement purposes.66 Pushing one single button in the websites of the Brazilian Ministry of Justice or the Chinese consumer protection ministry suffices to identify not only what the most important consumer problems are, but also in what area of law they are located and what kind of potential remedies were or could be used. Electronic systems are part of a blaming and shaming culture. As there are no data protection laws the authorities67 get to know the names of the consumers and those of the companies. Both are identifiable and both can be ranked: consumers with regard to their vulnerability, and companies with regard to the number and type of infringements. The two countries demonstrate the potential and the problematique of big data analytics. It goes beyond this paper to discuss the implications for their national laws and enforcement, let alone the dark side of data protection regulation, which prevents effective law enforcement.

2. Shortcomings of statistics 33

However, deficiencies do not necessarily result from lack of data protection alone.68 Relying on electronically reported consumer complaints might simply not be enough to identify “relevant” consumer problems. The sheer volume of complaints voiced might serve as an indicator, but consumers may suffer from problems they do not (yet) even know of, or problems they do not voice although these problems might be important. Consumer organizations as watchdogs may be in a better position to more accurately define consumer problems. Similarly, national consumer agencies or even businesses 63 On the basis of the ODR (Online Dispute Resolution) Regulation 524/2013 the European Commission has developed a complaint format which is available online https://ec.europa.eu/consumers/odr/ main/?complaintType=1&event=main.complaints.new. 64 http://ec.europa.eu/commfrontoffice/publicopinion/index.cfm. 65 Germany – just like in most other Member States – does not operate a nation-wide electronic consumer complaint system. Complaints are collected by the consumer organizations at the level of the German Länder. Not even the 16 models are fully harmonized. There are equally no statistics on dispute settlement or on consumer litigation in courts. Germany can only report ‘soft data’ which then find their way into Eurobarometer. 66 I am not aware of any publication to refer to. However, I had the chance to study and in limits to test the workability of the electronic complaint systems in Brazil and China during study visits in the last five years. 67 de Azevedo Cunha, Market Integration through Data Protection: A EU-Mercosur analysis, EUI phd 2011. 68 Whitman, The Two Western Cultures of Privacy: Dignity versus Liberty, 2004, Faculty Scholarship Series, Paper 649, http://digitalcommons.law.yale.edu/fss_papers/649.

214

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:36 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

themselves are collecting and analyzing data which reach beyond quantitative statistics.69 At the end of the day, however, it is the consumer who has to be given a voice. Designing and defining consumer problems is in itself a highly political issue. 34 Consumer Refit stands out as a deterrent example for path dependency. Consumer problems are by and large equated with those enshrined in the current body of European consumer law. However, it might well be that the bulk of rules the EU has adopted do not match what consumers need. Do consumers receive information when they need it? What about post-contractual information obligations? Or – to remain within the Consumer Refit – where is the link between existing European consumer law and the ambitious ideal of a circular economy?70 Big data analytics can play a useful tool in law making if lawyers are ready to engage 35 with computer scientists in order to cut through the thicket of consumer rules in order to find “the law” which currently all too often contains only “rules”.71 Consumer law enforcement could equally benefit from the systematic use of big data analytics, both with regard to individual and to collective enforcement.72 Here civil society and enforcement authorities tie in. Co-operation between academics from law and computer science is not sufficient. The raw material has to come from consumers themselves, from civil society, and from public authorities. Big data analytics may pave the way for building a consumer law that starts from the individual, moves to the commonality of consumer problems in order to initiate a re-typification of consumer law along the lines of different consumer images, and then link the different images to the appropriate mandatory rules and necessary remedies.

V. Prospects for big data analytics in consumer law Porat and Strahilevitz advocate experimental research.73 They have projects in mind 36 that test the feasibility of personalized law. Whilst this is definitely a way to go, big data analytics could also be used to strike down over-complex substantive consumer law rules and enforcement mechanisms. This chapter should be read as a proposal for initiating research. Information rules are a prominent candidate for both, for testing personalized law – the Ben-Shahar and Porat/Strahilevitz strand of the discussion – and for testing the (here) advocated re-typification of granular information duties. Testing big data analytics in law enforcement requires focus on a concrete case. Dieselgate provides deep insights into the overlap between common EU rules on rights, remedies, procedure, and institutions on the one hand and, on the other, deviating national rules on rights, remedies, procedure, and institutions. Consumers, consumer organizations and consumer agencies in the different EU Member States have mobilized their national tools in receiving compensation for the damage consumers suffered from.74 69 ISO 9001 Auditing Practices Group Guidance on Customer Complaints, https://committee.iso.org/ files/live/sites/tc176sc2/files/documents/ISO%209001%20Auditing%20Practices%20Group%20docs/ Auditing%20to%20ISO%209001%202015/APG-CustomerComplaints2015.pdf. 70 EU Package on the Circular Economy http://ec.europa.eu/environment/circular-economy/index_en.htm; critical of such a change in light of 500 years of consumer history, Trentmann (n 50), in his conclusions. 71 What is the difference between rules and law? A full answer is beyond the scope of this paper. 72 Lippi/Palka/Contissa/Lagioia/Micklitz/Sartor/Torroni, CLAUDETTE: an Automated Detector of Potentially Unfair Clauses in Online Terms of Service, Artificial Intelligence and Law, Springer Nature BV, 2019, Volume 27, Issue 2, 117–139. 73 Ibid. Conclusions. 74 BEUC, the Bureau des Organisation des Consommateurs, Volkswagen Four Years Down the Road, 2019, https://www.beuc.eu/publications/beuc-x-2019-050_report_-_four_years_after_the_dieselgate_ scandal.pdf.

Micklitz

215

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:37 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

1. Information rights and obligations European consumer law is by and large based on the information paradigm. This goes back to the famous declaration by President Kennedy in 1962, which made its way into EU consumer law programs and the UN guidelines on consumer protection. There is strong disagreement on the use and usefulness of information rights as an integral part of consumer policy – theoretically, conceptually, and empirically.75 Advocates of consumer protection argue that a consumer law based on information idealizes the normative image of the circumspect consumer, while behavioral economists demonstrate the dysfunctional effects of disclosure rules. My concern is different, though connected. I am interested in the information overload and the role that big data analytics could play in reducing complexity. 38 EU rules on consumer information can be broken down into different layers. Article 169 TFEU grants consumers a constitutional right to information. This right is often backed by reference to the EU Charter of Fundamental Rights if not the European Convention on Human Rights. Secondary consumer law could be broken down first and foremost into information duties related to the pre-contractual stage, which begins with advertising and sales promotion and leads via rules on the invitatio ad offerendum to contract-independent information obligations and later to contract-specific information rules. The idea behind this looks convincing at first sight – the nearer the moment of conclusion of the contract approaches, the more and the more precise information the consumer needs to make a rational decision. So far, so good. However, European regulations and directives have reached a degree of sophistication where efforts to achieve systematization and typification are doomed to fail – namely on product- and service-specific labelling, on commercial practices broken down into misleading actions and misleading omissions, on consumer information rights that cover all kinds of sales and service contracts, on consumer information rights dealing with the modalities of contract conclusion (direct and distant selling) and – last but not least – on contract specific information rights (package tours, time sharing, telecom, energy, transport, credit and finance). 39 Information rights and duties are granular norms par excellence. In the early days legal academics tried to classify and categorize information duties, not least with regard to the potential remedies that a potential infringement might entail. The analysis of information duties has immigrated into ever more fragmented legal fields, advertising and sales promotion, distant and direct selling, respectively along the line of the specialization in different types of contracts. There is even an element of resignation, on the part of the legislator, that transposes EU information rights and duties without raising too many questions about their overall usefulness. As a consultant I was involved in supporting the now “new” Member States as to the transposition of EU consumer law into their national legal systems. Today neighboring states that still hope for a closer relationship – if not membership – have consented to adapting their national laws as part of association agreements. A short anecdote provides deep insight. The president of the Georgian parliamentary committee wanted to understand why more or less the same information duties (her words) form an integral part of Directive 2005/29/EC on Unfair Commercial Practices and of Directive 2011/83 on consumer rights. What might be explainable to a well-trained European consumer lawyer might be hard to understand in legal orders that are still struggling with the socialist legacy. The solution the 37

75 Simitis, Schlagwort oder Rechtspinzip, Nomos 1976; Ben-Shahar/Scheider, More Than You Wanted to Know, The Failure of Mandated Disclosure, Princeton University Press 2014.

216

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:37 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

president advocated was: We do not understand what EU law says, but as the Association Agreement requires transposition, we have to do it. Sidestepping for a moment – what kind of law can emerge from such a legal transplant? There is also resignation if not disinterest on the side of legal academics. The EU 40 legislator follows the once chosen path of the information paradigm. Neither the European Parliament nor the Member States are raising the crucial question: What are all these rules for? Are they useful? Do they effectively protect consumers? What are the costs and benefits? More provocatively: are they just producing useless costs for both business and consumers? The Consumer Refit does not take such questions or similar ones seriously enough, although it could easily have been part of the agenda.76 The harshest attack against the use and usefulness of information rights stems from behavioral economics, with particular emphasis on disclosure rights in financial services.77 European behavioral research has not gone that far yet.78

2. Reducing the complexity of information through big data analytics What about the following: building up a data file in which all EU consumer 41 information rights and duties from all relevant areas are inserted, then setting a legal benchmark against which all these legal rules can be measured?79 Information rights and duties are more or less fully harmonized. The body of law is big enough to build a data file. The benchmark could be taken from empirically defined consumer information 42 needs and help to limit information rights and duties to what is “relevant” and for what there is/should be a “remedy”. The empirical data do not need to be collected from a European wide electronic complaint system. It seems highly unrealistic that such a system will come into existence in the near future. A simpler and less costly solution is available: starting from heuristics, hermeneutics, and qualitative research, jointly undertaken, by lawyers, sociologists, psychologists and economists. The requirements for getting such a project going are the following: raising enough 43 money to build a data file and to undertake this urgently needed empirical research. Big data analytics could turn into a useful tool to reduce complexity and maybe to come up with a set of “standards” that pass the threshold of “effectiveness” and “efficiency”. Such “standards” might be less detailed and – most importantly – adjusted to the three consumer images: responsible, confident, and vulnerable. Provided it is possible to reduce complexity and to define standards, big data analytics would reduce granularization and allow for typification.

3. Complexities of consumer law enforcement First and foremost the EU itself has no enforcement powers in the core field of 44 consumer protection. The Treaty of Rome drew a clear line between the making of 76 For a thorough analysis of the Consumer Refit, van Schagen, The Fitness Check of EU Consumer Law and the Impact Assessment for the New Deal for Consumers, in: van Schagen/Weatherill (eds.), Better Regulation in EU Contract Law: The Fitness Check and the New Deal for Consumers, Hart Publishing, 2019, 93–124. 77 Ben-Shahar/Schneider, supra (fn. 76); Bar-Gill, Seduction by Contract: Law, Economics, and Psychology in Consumer Markets, OUP 2012, with a review by Hugh Collins, 2014, 77 Modern Law Review 1030. 78 Sibony, Data and arguments: empirical research in consumer law, in: Micklitz/Sibony/Esposito (eds.), Research Methods in Law and Consumer Law, Elgar 2018, ch. 5. 79 Busch, Implementing Personalized Law: Personalized Disclosures in Consumer Law and Data Privacy Policy, (2019) University of Chicago Law Review 68, 309–331 discussing in particular the opportunities of personalized information.

Micklitz

217

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:37 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

supranational law, subject to the list of enumerated powers, and enforcement – which remained in the hands of the Member States. There are only two exceptions: competition law and the common agricultural policy, where the European Commission holds enforcement powers. It is plain that the EU has enlarged its influence in particular through market- and sector-related agencies. However, with the exception of the ECB and ESMA none of these agencies has enforcement powers. 45 This does not mean that the EU did not exert influence through harmonization as to how enforcement of consumer law is structured and organized in the EU.80 The EU granted consumers individual rights through secondary EU law. In the last ten years the number of references by national courts to the CJEU has dramatically increased. Today, more than 100 judgments cover core areas of consumer law, with unfair terms control playing a prominent role.81 If one understands EU law as constitutional law, due to its primacy and direct effect, then the CJEU’s role is outstanding in the most literal sense. No other national constitutional court comes even close to the CJEU’s record. The judgments have not yet been fully analyzed, in particular not with regard to their procedural dimension.82 Below the level of access to courts through the development of European rights, remedies, and procedure, the EU has managed to establish a new layer of individual enforcement through the mandatory introduction of Alternative Dispute Resolution Bodies.83 So far the EU has been less successful in introducing collective rights for consumers. The only collective remedy so far is the action for injunction, although a much more ambitious project on collective compensation is currently in the pipeline.84 At first glance it seems as if the Member States remain free to decide whether collective rights are enforced through consumer organizations and/or public agencies. Over time, however, and in particular through Regulation 2006/2004 on Co-operation in Transborder Enforcement, the EU managed to put pressure on Member States to set up consumer law agencies. The market and sector related EU-based agencies are responsible not only for guaranteeing the functioning of the respective markets, but also for taking the collective interests of consumers into account. This has led to administrative enforcement of European private law, largely below the radar of public awareness, occasionally outside formally granted powers.85 46 However, enforcement of consumer law is not only divided between the EU and the Member States, but is also broken down into public vs. private enforcement, into courts vs. ADR, into individual vs. collective. Taking these different levels together it is close to impossible to gain a full overview as a legal scholar on EU law and the law of the (now) 27 Member States, law in the books and – all the more – law in action. What can be the 80 Cafaggi/Micklitz (eds.), New Frontiers of Consumer Protection – the Interplay between Private and Public Enforcement, Elgar 2009. The contributions cover the whole range of rights, remedies, procedures and institutions. 81 European private law journals provide regular overviews, European Review of Private Law, European Contract Law Review, Zeitschrift für Europäisches Privatrecht. B. Kas and myself delivered the most comprehensive analysis in the EWS – Europäische Zeitschrift für Wirtschaft- und Steuerrecht, unfortunately in German though, Rechtsprechungsübersicht zum Europäischen Vertrags-und Deliktsrecht (2014–2018) – EWS 2018, 181–219 and 241–300 (Part I and Part II). 82 della Negra, The Uncertain Development of the Case Law on Consumer Protection in Mortgage Enforcement Proceedings: Sánchez Morcillo and Kušionová, 2015, 52 CML Rev 1009–1032. 83 Hodges/Benöhr/Creutzfeldt-Banda, Consumer ADR in Europe, Hart 2012. 84 Art 5 of the Proposal of the European Commission on a Representative Action, Brussels, 11.4.2018 COM (2018) 184 final introduces an opt-in mechanism and an opt-out for small value claims; Voet, Where the Wild Things are, Reflection on the State and future of European Collective Redress, 2017 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2913010. 85 Hodges, The Reform of Class and Representative Actions in European Legal Systems, A New Framework for Collective Redress in Europe, Hart Publishing 2009 with examples.

218

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:37 3B2 9.1.580; Page size: 160.00mm  240.00mm

I. De- or Re-typification through Big Data Analytics?

role of big data in consumer law enforcement? Can big data help to realize a holistic perspective on consumer law both in the books and in action? Can big data help to reduce complexity to the benefit of consumers? Is big data the tool that allows us to find transnational answers to transnational problems?

4. Reducing complexity of enforcement through big data? The following might look like “horror juris” or as an achievable political goal, 47 depending on one’s perspective. As usual in ideologically loaded debates, the truth might well be somewhere in between. Together with the Lisbon Summit 2000 and the White Paper on Governance in 2002, 48 the European Commission terminated its sponsorship of the so-called European Consumer Law Group (ECLG). This group had operated successfully for quite some 20 years as an advisor and commentator to the European Commission. Each Member State was presented by one academic and one representative of a national consumer organization. The European Commission replaced the ECLG through subject-related expert groups that were called together on an ad hoc basis. Thereby the European Commission obtained the necessary flexibility to select experts. Later the selection process was formalized, submitted to a procedure with open calls and published lists with names. Consumer organizations and more broadly the Consumer Law Community turned out to be the losers. The only remaining voice was BEUC, the Bureau des Organization des Consommateurs. Due to BEUC’s imagination and the European Commission’s half-hearted willingness to compensate the consumer organizations for the loss of ECLG, the Consumer Law Enforcement Forum (CLEF) was created, later renamed as COJEF – the Consumer Justice Enforcement Forum.86 The Forum does not have a stable infrastructure and is dependent on project related funding. Over the last ten years CLEF/COJEF have developed a new strategy to overcome segmentation of national law enforcement structures. Instead of using the European regulatory framework on transborder litigation, rudimentarily enshrined in Directive 2009/22 on injunctions combined with the Brussels Regulation on jurisdiction and the Rome I regulation on contract and the Rome II regulation on torts, the national consumer organizations invented the “co-ordinated action model” under the tutelage of BEUC.87 The organizations agreed on an action plan, developed a joint strategy and then took action within their national enforcement schemes, though co-ordinated in time, subject and strategy. This is reduction of legal complexity through practice. One of the fields of action is Dieselgate. Can big data help in improving comparability 49 on both the factual and the legal side? On the factual side are two interconnected questions: to what extent did Volkswagen use the same advertising strategies in Europe and were these different from the company’s approach in the USA? The other question concerns whether the cars were defective in the meaning of the EU 1999/44 Consumer Sales Directive. On the legal side, in some countries consumers are fighting on their own. More prominently, however, consumers are seeking the shelter of a public body or a consumer association to defend their interests. The overall reluctance of the EU, in line with many Member States, against the US class action is well-known. The majority of Member States rely on an opt-in mechanism. In these countries a plaintiff is needed who collects claims from individual consumers, bundles them and sues Volkswagen in the form of a collective action. Provided there were resources available to build up a 86

http://cojef-project.eu. Best described in Durovic, The Apple Case Today: Factual and Legal Assessment, 2016, EUI Working Paper EUI-ERC No 3. 87

Micklitz

219

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/09_Micklitz, De- or Re-typification through Big Data Analytics The Case of Consumer Law, 24.02.3d from 30.10.2020 11:36:37 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

data file on Dieselgate litigation across borders, and provided this data file contained the relevant facts, the relevant laws in all the Member States and cases already decided by national courts – could big data analytics help to reduce complexity in defining the common denominators among factual patterns? This seems highly likely. Out of the thousands of complaints, it would be possible to identify business strategies, patterns of factual defectiveness, and patterns of factual misleading advertising. But even further – would it be possible to identify patterns of individual remedies in consumer claims, and/ or collective remedies which are deduced out of the many variations of Member-State laws in books and now in action and which comply with the differing needs of the three consumer groups – the responsible, the confident, and the vulnerable?

VI. Big Data Analytics and Re-typification 50

Is current European consumer law, like the Allgemeines Preussisches Landrecht (General State Laws for the Prussian State), an attempt to regulate each and every detail of consumer protection? The Allgemeines Preussisches Landrecht contained 19,000 provisions, as each and every detail was meant to be regulated. It took more than a hundred years before this vast body of rules was replaced by the German Civil Code. Today’s consumer law comes closer to the Preussische Landrecht than to an abstract civil code. Big Data analytics can be used in two directions. The first is to instrumentalize the preferences of consumers for shaping personalized laws. If such an approach works it could make the existing body of overcomplex European consumer law manageable and enforceable across all the differences at EU level and national level in how enforcement is organized. The second option is to use big data analytics to reduce the complexity of consumer law and consumer law enforcement. This paper stresses the need to seriously study and explore the feasibility of a re-typfication of consumer law through Big Data Analytics.

220

Micklitz

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:51 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalization of the Law and Unfair Terms in Consumer Contracts I. Introduction Behavioral law and economics literature has shown that information disclosure is not 1 able to remedy market failures resulting from failing information and the ‘signingwithout-reading-problem’.1 In general, a procedural approach to unfair terms is unsuccessful because consumers simply do not spend time in reading even if they could be able to understand the legal implications of a clause.2 Also obtaining the information is considered of little use for contracting parties who usually believe that they have no alternative but to accept the standard term.3 It turns out that a paternalistic intervention through a substantive control on the 2 terms is the most appropriate way to combat market failures.4 On this regard, it is often stated that the control over standard terms should not correspond to an imposition of mandatory model terms, but should be shaped as a flexible tool in order to secure a certain degree of variety depending on the different market sectors. An assumption of law and economic theory is that in a contract efficient terms are the ones that the parties would have added if they had negotiated.5 A connected basic finding is that if terms that the parties would have added to the contract after a negotiation are deemed unfair based on an external fairness criterion and not on the welfare of the parties, the results will be distorting for competition and damaging for consumers. It is difficult to dispute the aforementioned assumptions. The problem that standard 3 terms pose is that private actors, who make use of them, are not interested in augmenting the welfare of the system. They want primarily to enhance their benefits in the market in imposing standardized terms, able to provide for certainty and advantages in cases of disputes with their contracting parties. Thus, there is the need of an external control, aimed to reduce the imbalance between the parties’ rights and 1 See., i.a., Rakoff, Contracts of Adhesion: An Essay in Reconstruction, 1983, 96 Harv. L. Rev. 1174; Bebchuk/Posner, One-Sided Contracts in Competitive Consumer Markets, in: Ben-Shahar (ed.), Boilerplate: The Foundation of Market Contracts, Cambridge University Press 2007, 3; De Geest, The signingwithout-reading problem: An analysis of the European Directive on unfair contract terms, in: Schäfer/ Lwowski (eds.), Konsequenzen wirtschaftsrechtlicher Normen Festschrift für Claus Ott, Gabler 2002, 213–35; Ben‐Shahar, The myth of the “opportunity to read” in contract law, (2009) 5 European Review of Contract Law 1. See also Faure/Luth, Behavioural Economics in Unfair Contract Terms Cautions and Considerations, [2011] Journal of Consumer Policy 337–58. 2 This is confirmed, i.a., by the Italian and the Dutch legislation: see Hondius, Unfair Contract Terms and the Consumer: ECJ Case Law, Foreign Literature, and Their Impact on Dutch Law, (2016) 24 European Review of Private Law 457, at 461. 3 Cf. Hatzis, An Offer You Cannot Negotiate: Some Thoughts on the Economics of Standard Form Consumer Contracts, in: Collins (ed.), Standard Contract Terms in Europe: A Basis for and a Challenge to European Contract Law, Kluwer Law International 2008, 43, 45: “If we think about it, we will realize that there is no such thing as negotiating a contract if you are a consumer”. 4 See Grundmann, Targeted Consumer Protection, in: Leczykiewicz/Weatherill (eds.), The Images of the Consumer in EU Law. Legislation, Free Movement and Competition Law, Hart 2016, 223, 238, who refers that in the context of standard contracts “no information rule could give the consumer the opportunity to know properly the content of those standard contract terms at a price/effort which is roughly similar to the one which the professional (a very regular user) of those terms has to invest”. 5 Hatzis, ibid.

Patti

221

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:51 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

duties arising out of the contract. Such control mechanisms exist all over the world, especially with respect to B-2-C relationships, where legislators and judges intervene to eliminate unfair terms from the contract. The main type of controls apply to contracts formed over the internet.6 4 The aforementioned area of consumer contract law becomes more and more important and, within the European context, new forms of enforcement are discussed7 in order to augment the rules’ level of effectiveness.8 Even if it does not seem yet the core of the studies on personalization of the law, the latter could have a significant impact on the way in which the unfairness control of clauses may be exercised by Courts and Authorities and augment its efficiency. In fact, at least in the European context, default rules are the main basis used to assess whether a term is unfair. Therefore, it seems worth to investigate if a “personalization of the law” or the creation of “microdirectives”9 could have an impact in the way in which problems related to standard form contracts and consumer protection are tackled.

II. The setting within the European context 1. The theoretical framework 5

In his very known contribution on standard contractual terms, Ludwig Raiser has indicated default rules as a valuable parameter to assess the unfairness of a term, given the fact that in the legal system they could not be considered an order created by chance, but “Justice” in the sense of an objectification of the community’s idea of justice.10 The problem was tackled by the German author in the following way. Starting from the obvious assumption that in individual contracts, default rules can be set aside, the question posed was weather such a derogation should be considered possible also through standard form contracts. The answer to Raiser’s relevant question depends on the type of default rules and on the way in which the default rules are derogated. Some default rules have just the aim to fix a general rule available for the whole society and thus a derogation provided by standard terms only in extreme cases turns in an abuse. In other cases, the default rule represents what the society considers as a balanced regulation of a contractual relationship.11 An additional parameter for the assessment of an abuse could be the existence of a huge difference between the default rule and the standard term. According to this 6 See Hillman/Rachlinski, Standard-Form Contracting in the Electronic Age, (2002) 77 N.Y.U. L. Rev. 429, 456–60; and for comparative remarks Maxeiner, Standard-Terms Contracting in the Global Electronic Age: European Alternatives, 2003, 28 Yale J. Int’l L. 109. With respect to the application of the Unfair Terms Directive, see Loos/Luzak, Wanted: a Bigger Stick. On Unfair Terms in Consumer Contracts with Online Service Providers, (2016) 39 Journal of Consumer Policy 63. 7 See Steennot, Public and Private Enforcement in the Field of Unfair Contract Terms, (2015) 23 European Review of Private Law 589–619, pointing out that since consumers are in most cases not aware of (the possibility to invoke) rules on unfair contract terms and consumer protection associations often have only limited financial means to apply for injunctions, private enforcement mechanisms can in themselves not realize consumer protection from unfair contract terms. 8 See Sørensen, In the Name of Effective Consumer Protection and Public Policy!, (2016) 24 European Review of Private Law 791–822, presenting a model which consists of four steps that include the specific ideas behind consumer protection provisions as well as the interaction between the principle of effectiveness and principle of equivalence and elaborating the idea of regarding consumer protection provisions as (European Union (EU)) public policy rules. 9 The abovementioned terminology is adopted by Casey/Niblett, supra Part 1.C, at 99–101. 10 Raiser, Das Recht der allgemeinen Geschäftsbedingungen, Hermann Gentner Verlag 1961, 293–295. 11 According to the theory of Raiser Gerechtigkeitsgehalt from default rules can differ.

222

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:51 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

criterion, a term that fully conflicts with the contents of a default rule could be considered abusive and unfair. This assumption is also nowadays considered appropriate, especially in the German 6 legal thinking.12 Outside the EU, also in Swiss law default rules are intended as an apt benchmark able to protect the interests of the parties.13 The reason is that they were adopted through a democratic procedure with the aim to fulfill the expectations of the society and provide for a fair and just regulation of relationships between privates.14 Of course, the default rules that are used by the national court as a benchmark to assess whether a contract term is unfair or not are impersonal default rules which are not customized on a particular category of consumers. Thus, the described methodological approach to standard terms is not based on a 7 personalization of rules according to big data findings nor on a sharp evaluation of the economic consequences attached to the application of the rules. Such a theoretical approach is consistent with the rules adopted on a supranational level by the Unfair Terms Directive 93/13.15 A one-size-fits all solution, hinged on the notion of consumer,16 which has often been criticized by scholars. As it was recently affirmed, “technologies associated with big data, prediction algorithms, and instantaneous communication reduce the costs of discovering and communicating the relevant personal context for a law to achieve its purpose, the goal of a well-tailored, accurate, and highly contextualized law is becoming more achievable”.17 Such a personalized law could influence consumer protection and counterbalance the high personalization techniques of businesses, who are able to exploit consumers in using their data.18

2. The European methodology It is not possible to provide a comprehensive account of all the problems involved in 8 the European approach to unfair contract terms. Thus, three crucial areas in which default rules play a role in the application of the substantive control of clauses and its effect will be put under scrutiny. a) The scope of application of the judicial control. For the purposes of the present 9 examination, it is important to observe that in the Unfair Terms Directive two provisions limit the scope of application of the unfairness control. First, the Directive expressly excludes a review of terms, which reflect mandatory statutory or regulatory provisions and the provisions or principles of international conventions.19 Moreover, 12

See infra 2.b). Schmid, AGB und die Rolle des dispositiven Recht, in: Brunner/Schnyder/Eisner (eds.), Allgemeine Geschäftsbedingungen nach neuem Schweizer Recht, Schulthess 2014, 206; Frei/Jung, Revised Control of Unfair Terms in Swiss Law – Consumer Protection by Competition Law? (2015) 4 Journal of European Consumer and Market Law 165, 166: “if a certain contract is governed by specific legal provisions, these provisions should still serve as a referral system to assess the fairness of the standard terms”. 14 Schmid, ibid., 208. See generally Möslein, Dispositives Recht. Zwecke, Strukturen und Methoden, Mohr Siebeck 2011. 15 Council Directive 93/13/EEC of 5 April 1993 on unfair terms in consumer contracts, OJ L 95, 21.4.1993, 29–34. 16 See generally Leczykiewicz/Weatherill, The Images of the Consumer in EU Law, in Id. (fn. 4), at 1. 17 Casey/Niblett, Framework for the New Personalization of Law, (2019) 86 U. Chi. L. Rev. 333, 335. See also Busch/De Franceschi, Granular legal norms: big data and the personalization of private law, in: Mak/Tjong Tjin Tai/Berlee (eds.), Research Handbook in Data Science and Law, Edward Elgar, 2018, 408. 18 Wagner/Eidenmüller, Down by Algorithms? Siphoning Rents, Exploiting Biases, and Shaping Preferences, (2019) 86 U. Chi. L. Rev. 581; See also Mak, Contract and consumer law, in: Mak/Tjong Tjin Tai/Berlee, supra (fn. 17), at 17, 35–37. 19 Art. 1(2) Unfair Terms Directive: “The contractual terms which reflect mandatory statutory or regulatory provisions and the provisions or principles of international conventions to which the Member 13

Patti

223

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:51 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

the Directive excludes from the review the main subject matter of the contract and the adequacy of the price and remuneration, in so far as these terms are in plain intelligible language.20 10 The first exemption relies on the idea that in presence of a mandatory provision or a provision that would apply between the parties by virtue of law unless otherwise agreed upon,21 the national legislature has intervened to strike a balance between the positions of the parties. Therefore, an additional substantive control of the clauses is not required, except in cases in which the national law conflicts with European primary law.22 11 The second exemption, concerning the subject matter of the contract and the price, is the result of the compromise sought by the Directive, which aims to protect the consumers’ rights and to comply with the principles of a free market economy.23 Before the enactment of the Directive, two German authors affirmed the need to introduce the exemption and criticized a proposal of the EU Commission, which extended the control also on core terms and even in the presence of a negotiation.24 They observed that: “Since individual autonomy is a principle recognized by the legal systems of all the Member States, and one of considerable importance in a free market economy …, it goes without saying that such a step should not be taken lightly”.25 With reference to the core terms, they declared that “The relationship between the price and the goods or services provided is determined not according to some legal formula but by the mechanisms of the market. Any control by the courts or administrative authorities of the reasonableness or equivalence of this relationship is anathema to the fundamental tenets of a free market economy. It would partially abrogate the laws of the market and hence prevent the offerers of goods or services from acting in accordance with those States or the Community are party, particularly in the transport area, shall not be subject to the provisions of this Directive”. In the Recital n. 13 of the Directive it is further indicated that: “Whereas the statutory or regulatory provisions of the Member States which directly or indirectly determine the terms of consumer contracts are presumed not to contain unfair terms; whereas, therefore, it does not appear to be necessary to subject the terms which reflect mandatory statutory or regulatory provisions and the principles or provisions of international conventions to which the Member States or the Community are party; whereas in that respect the wording ‘mandatory statutory or regulatory provisions’ in Article 1 (2) also covers rules which, according to the law, shall apply between the contracting parties provided that no other arrangements have been established”. For an interpretation of the provision, see especially ECJ, 21 March 2013, Case C-92/11, RWE Vertrieb AG v. Verbraucherzentrale NordrheinWestfalen e.V., ECLI:EU:C:2013:180. 20 Art. 4(2) Unfair Terms Directive: “Assessment of the unfair nature of the terms shall relate neither to the definition of the main subject matter of the contract nor to the adequacy of the price and remuneration, on the one hand, as against the services or goods supplies in exchange, on the other, in so far as these terms are in plain intelligible language”. The exemption does not affect secondary terms which concern the method of price calculation and price adjustment terms: see Nebbia, Unfair contract terms in European law: a study in comparative and EC law, Hart 2007, 124. 21 Micklitz, Unfair terms in consumer contracts, in: Reich/Micklitz/Rott/Tonner (eds.), European Consumer Law, 2nd edn., Intersentia, 2014, 136. 22 Ibid. 23 See generally Ibid., 138; Hesselink, Unfair Prices in the Common European Sales Law, in: Vogenauer/Gullifer (eds.), English and European Perspectives on Contract and Commercial Law: Essays in Honour of Hugh Beale, Hart Publishing 2014, 231; Dellacasa, Judicial review of “core terms” in consumer contracts: defining the limits, (2015) 11 European Review of Contract Law 152, 158; Rott, Unfair contract terms, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Edward Elgar 2016, 287, 293–96; Heirman, Core terms: interpretation and possibilities of assessment, (2017) 6 Journal of European Consumer and Market Law 30–34. 24 See Brandner/Ulmer, The Community Directive on Unfair Terms in Consumer Contracts: Some Critical Remarks on the Proposal Submitted by the EC Commission, (1991) 28 Common Market Law Review 656. 25 Ibid. 652.

224

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:51 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

laws; the consumer would no longer need to shop around for the most favourable offer, but rather could pay any price in view of the possibility of a subsequent control of its reasonableness”.26 As it will be seen, the two described exemptions and their underlying rationales could 12 nowadays be questioned in the light of the massive use of big data. Data could permit to develop a personalized set of mandatory rules and give to businesses the possibility to adopt personalized pricing policies, which could harm consumers’ interests.27 New technologies give to businesses the opportunity to exploit consumer through the application of prices that reflect their willingness to pay. At any rate, the two exemptions provided by the Unfair Contract Terms Directive are in line with the concept described by Raiser, who primarily referred to default rules as a basis for the assessment of unfairness of a term. In fact, the two exemptions operate in fields where the parameter provided by default rules may not be applied. b) The unfairness test. In Article 3(1) the Unfair Terms Directive 93/13 provides for 13 a general test, referring to ‘good faith’ and to a ‘significant imbalance in the parties’ rights and obligations arising under the contract’.28 It took almost 20 years until a national Court referred to the ECJ a preliminary ruling concerning the clarification of those concepts.29 The opportunity to interpret Article 3(1) was given by the case Mohamed Aziz.30 The 14 ECJ developed a two-steps approach and, first of all, stated that: “in order to ascertain whether a term causes a ‘significant imbalance’ in the parties’ rights and obligations arising under the contract, to the detriment of the consumer, it must in particular be considered what rules of national law would apply in the absence of an agreement by the parties in that regard. Such a comparative analysis will enable the national court to evaluate whether and, as the case may be, to what extent, the contract places the consumer in a legal situation less favourable than that provided for by the national law in force. To that end, an assessment should also be carried out of the legal situation of that consumer having regard to the means at his disposal, under national legislation, to prevent continued use of unfair terms”.31 This view coincides with the approach adopted by Member States’ national laws, in 15 particular in the German legal system and, outside of the EU, by Switzerland. For instance, the German Bürgerliches Gesetzbuch (BGB) provides that “An unreasonable disadvantage is, in case of doubt, to be assumed to exist if a provision … is not compatible with essential principles of the statutory provision from which it deviates”.32 26

Ibid. 656. Gilo/Porat, The Hidden Roles of Boilerplate and Standard-Form Contracts: Strategic Imposition of Transaction Costs, Segmentation of Consumers, and Anticompetitive Effects, (2006) 104 Mich. L. Rev. 983, 988–93; Wagner/Eidenmüller, supra (fn. 18), at 585–588. 28 Art. 3(1) Unfair Terms Directive: “A contractual term which has not been individually negotiated shall be regarded as unfair if, contrary to the requirement of good faith, it causes a significant imbalance in the parties’ rights and obligations arising under the contract, to the detriment of the consumer”. 29 See Gavrilovic, The Unfair Contract Terms Directive through the Practice of the Court of Justice of the European Union: Interpretation or Something More?, (2013) 9 European Review of Contract Law 180–193, discussing the practice of the Court of Justice of the European Union concerning implementation and interpretation of 1993 Unfair Contract Terms Directive into national legislation of Member States. 30 ECJ, 14 March 2013, case C‐415/11, Mohamed Aziz v Caixa d’Estalvis de Catalunya, Tarragona i Manresa (Catalunyacaixa). See the comment of Iglesias Sánchez, Unfair terms in mortgage loans and protection of housing in times of economic crisis: Aziz v. Catalunyacaixa, (2014) 51 Common Market Law Review 955–974. 31 ECJ, Mohamed Aziz, para 68. 32 § 307, para. 2, No. 1 BGB. The English translation is available on https://www.gesetze-im-internet.de/ englisch_ 27

Patti

225

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

The second step outlined by the ECJ is to determine under what circumstances such a “significant imbalance” is “contrary to the requirements of good faith”. On this issue, the ECJ concluded: “the national court must assess for those purposes whether the seller or supplier, dealing fairly and equitably with the consumer, could reasonably assume that the consumer would have agreed to such a term in individual contract negotiations”.33 The ECJ seems to refer to a case-by-case assessment.34 It means that the business cannot simply pursue his own interests of contractual efficiency against the legitimate expectations of the consumer.35 At any rate, not every derogation from national law amounts to an act which is “contrary to the requirements of good faith”. The Advocate General in the case Mohamed Aziz correctly pointed out that “in many cases parties have a legitimate interest in organising their contractual relations in a manner which derogates from the [rules of national law]”.36 17 The problem is understanding how the second step must be provided by national courts. A good example of the application of the second kind of assessment is the UK Supreme Court decision Beavis vs. ParkingEye37 concerning a payment due in cases of overstay in a parking place.38 Through an explicit reference to the decision Aziz of the ECJ, the UK Supreme Court affirmed that in situations like the one envisaged by the case, consumers interested in parking two hours for free in a shopping mall would have reasonably agreed to the term providing a payment for the overstay. The case shows that some consumers may wish for an onerous terms to be included in their contracts, as long as they are compensated for that through a lower price. The judgement of the UK Supreme Court may be seen as a demonstration that judges should not only use fairness criteria in order to assess the terms but economic criteria to assess the market setting where the particular transaction took place.39 In addition, the judgment of the UK Supreme Court takes into consideration the particular sector of car parking and the behavior of consumers. The aforementioned judgement came under criticism, especially because it was not clear on which basis the judges could express their knowledge on consumer behavior and on the relevant market sector. Big data and the use of statistical sciences may represent an important tool to understand if a consumer would have accepted a particular term in an individual negotiation. 18 In addition, legislators have developed indicative and non-exhaustive lists of terms, which may be regarded as unfair.40 The content of the lists may be understood as an application of the unfairness test. They provide guidance in practice and are able to 16

bgb/englisch_bgb.html#p0939. In original language the provision states: “Eine unangemessene Benachteiligung ist im Zweifel anzunehmen … wenn eine Bestimmung mit wesentlichen Grundgedanken der gesetzlichen Regelung, von der abgewichen wird, nicht zu vereinbaren ist”. On the provision, see Wurmnest, § 307, in: Münchener Kommentar zum BGB 8th edn., C.H. Beck, 2019, mn. 64–69, who states that the prohibition affects only derogations who “die tragende Gedanken des gesetzgeberischen Gerechtigkeitsmodells beeinträchtig[en]”. 33 ECJ, Mohamed Aziz, par. 69. 34 See also Micklitz/Reich, The Court and Sleeping Beauty: The Revival of the Unfair Contract Terms Directive (UCTD), (2012) 51 Common Market Law Review 771, 790. 35 Steennot, supra (fn. 7), at 589. 36 Opinion of Kokott, delivered on 8 November 2012, Case C-415/11, Mohamed Aziz v. Caixa d’Estalvis de Catalunya, Tarragona i Manresa (Catalunyacaixa), para 73. 37 Beavis v. ParkingEye [2015] UKSC 67. 38 “Failure to comply with the following will result in a Parking Charge of £85”: Beavis v. ParkingEye [2015] UKSC 67, para 91. 39 See generally Gilker, Case Note England and Wales, UKSC 4 November 2015, Cavendish Square Holdings BV v. Makdessi; ParkingEye Ltd v. Beavis, (2017) 25 European Review of Private Law 173–180. 40 See Art. 3(3) Unfair Terms Directive.

226

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

standardize the behavior of market players. Many Member states distinguish between a “black” and a “grey” list of unfair terms. The terms contained in the second list are only presumed to be unfair. As it will be seen, a more personalized approach to the problem of unfair terms driven by technology and data science may lead to abandon the use of fixed lists of terms which may be regarded as unfair. c) The consequences of unfairness. One basic feature of the mechanism of control 19 of standard terms developed in European legal systems, as for instance Italy and Germany, is that after a declaration of unfairness, default rules apply in order to fill the gap left in the contract.41 This very clear consequence of the declaration of unfairness, which has its roots in the German tradition, seems challenged by recent ECJ judgements in the light of the political need to deter professionals from insert unfair terms in their standard contracts.42 The legal base is given by article 6(1) of the Directive, according to which: “Member States shall lay down that unfair terms used in a contract concluded with a consumer by a seller or supplier shall, as provided for under their national law, not be binding on the consumer and that the contract shall continue to bind the parties upon those terms if it is capable of continuing in existence without the unfair terms”. The ECJ has held that a national provision which empowers a national court to 20 replace unfair terms with a modified (and fair) term is not compatible with article 6, par. 1, of the Directive.43 In particular, in Banco Español de Crédito, the ECJ held that: “Article 6(1) of Directive 93/13 cannot be understood as allowing the national court, in the case where it finds that there is an unfair term in a contract concluded between a seller or supplier and a consumer, to revise the content of that term instead of merely setting aside its application to the consumer”.44 In the light of the used wording, a ‘flexible approach’, entitling the Court to amend or alter the unfair contractual term (e.g. the reduction of a penalty as it is provided in some European legal systems) is certainly not compatible with the Directive. In the case Asbeek Brusse, the ECJ has provided an analogous interpretation in stating 21 that the national court is required to exclude the application of that clause in its entirety with regard to the consumer. In principle, the contract must continue to exist without any other amendment than the deletion of the unfair term. The obligation to remove the unfair term is justified by the aim of the Directive. If a national court would be able to revise the content of an unfair contract term, this would weaken the dissuasive effect on businesses. The goal is to deter businesses from using unfair terms. The possibility to merely revise the content of unfair terms implies that businesses would use unfair terms without any risk. 41 The German BGB contains a specific provision on this regard in § 306, para. 2: “To the extent that the terms have not become part of the contract or are ineffective, the contents of the contract are determined by the statutory provisions”. (The English translation is available on https://www.gesetze-iminternet.de/englisch_bgb/englisch_bgb.html#p0939). In original language the provision states: “Soweit die Bestimmungen nicht Vertragsbestandteil geworden oder unwirksam sind, richtet sich der Inhalt des Vertrags nach den gesetzlichen Vorschriften”. 42 See in general Cafaggi/Iamiceli, The Principles of Effectiveness, Proportionality and Dissuasiveness in the Enforcement of EU Consumer Law: The Impact of a Triad on the Choice of Civil Remedies and Administrative Sanctions, (2017) 25 European Review of Private Law 575, 584–588. 43 ECJ, 21 January 2015, Joined Cases C‐482/13, C‐484/13, C‐485/13 and C‐487/13, Unicaja Banco, SA v. José Hidalgo Rueda and Others, paras 28–32; ECJ 30 April 2014, case C‐26/13, Árpád Kásler and Hajnalka Káslerné Rábai v. OTP Jelzálogbank Zrt, paras 76–79; ECJ, 30 May 2013, case C‐488/11, Dirk Frederik Asbeek Brusse and Katarina de Man Garabito v. Jahani BV, par. 54–60; ECJ, 14 June 2012, case C‐618/10, Banco Español de Crédito v. Joaquín Calderón Camino, para 69–73. 44 ECJ, Banco Español de Crédito v. Joaquín Calderón Camino, para 71.

Patti

227

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

The problem which was not addressed in a clear manner is whether the business, after the deletion of the unfair term, still has the possibility to invoke default rules of national law. For example, if a penalty clause is unfair and thus not binding, is the business entitled to compensation according to default rules? Neither in Banco Español de Crédito nor in Asbeek Brusse the ECJ precluded explicitly national courts from applying supplementary provisions after a declaration of unfairness of a term. The only prohibition affected the amendment of the contractual term. However, in Unicaja the ECJ stated that the substitution of a supplementary provision of national law for an unfair term is limited to cases in which the invalidity of the unfair term would require the court to declare the contract invalid in its entirety, thereby exposing the consumer to disadvantageous consequences. With an ‘a contrario’ reasoning, it is possible to infer that it is prohibited for national courts to apply supplementary provisions of national law in the event where a contractual term has been considered unfair and the not binding character of the term does not lead to declare the nullity of the contract in its entirety. 23 The judgment makes clear that the objectives of Articles 6(1) of the Directive are realized when the national legislation determines that unfair contract terms have no effect at all, given that the business will not be able to rely on the unfair contract term in its relation to the consumer.45 This was also the conclusion of AG Trstenjak, who affirmed: “Article 6(1) of the directive must therefore be understood to mean that, after the unfair terms have been removed, the contract must continue in existence in unmodified form as to the remaining terms, if that is legally possible, which notionally precludes any replacement of terms or modification of the contract”.46 24 A national court is not precluded “in accordance with the principles of the law of contract, from deleting an unfair term and substituting for it a supplementary provision of national law (i.e. a national legal rule applicable to the issue governed by the term in the absence of other or contrary agreement) where this would enable “real balance between the rights and obligations of the parties to be restored” and where otherwise the invalidity of the unfair term would require the court to declare the invalidity of the contract in its entirety with disadvantageous consequences to the consumer.47 A key example could be found in the case of a term relating to the main subject matter of the contract, which fails the conditions set by Art. 4(2) that it must be “plain and intelligible” and which is found to be unfair and so not binding on the consumer; if a supplementary rule allows a court to govern the issue of the main subject matter, then reliance on it could rescue the contract from overall invalidity. The ECJ has further emphasized that this acceptance of the application of national rules in substitution for a contract term held unfair is limited to these particular circumstances.48 25 The ECJ’s interpretation makes clear that the dry application of default rules as contractual gap-fillers may not assure the achievements of the European Union’s political goals. In order to deter businesses from inserting unfair terms in the contract a more flexible approach is required. Also on this regard it is questionable whether big data and statistical sciences could be useful in order to create a personalized gap-filler, able to pursue the political goals of the European Union. 22

45

Steennot, supra (fn. 7), at 599. Opinion of AG Trstenjak, 14 February 2012, case C‐618/10, Banco Español de Crédito/Joaquín Calderón Camino, para 85. 47 ECJ, Unicaja Banco, SA v. José Hidalgo Rueda and Others, at 33; ECJ, Árpád Kásler and Hajnalka Káslerné Rábai v. OTP Jelzálogbank Zrt, paras 82–84; Unicaja at 33. 48 ECJ, Unicaja Banco, SA v. José Hidalgo Rueda and Others, 33. 46

228

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

3. A personalized approach? In the light of the methodology adopted by the ECJ, it is arguable that, with respect to 26 consumer law, a certain degree of personalization is recognizable in the way in which the most relevant provisions of the Unfair Terms Directive 93/13 are interpreted by the ECJ. The known judgment of the ECJ Aziz reveals how big data findings could possibly be useful in assessing whether a term is unfair or not, depending on the willingness of the consumer to accept the term under scrutiny in a hypothetical negotiation. The same is true in looking to the consequences of unfairness. In the latter field, the new technology could give the possibility to develop personalized gap-fillers, able to oppose the market power exercised through unfair terms. For both aspects of the European methodology default rules, which are impersonal in nature and aimed to reflect the sense of justice on which the legal system is founded, may not appear the best possible reference. The approach is certainly consistent with the theoretical background outlined before. Discussions on personalized law and granular legal norms may provide for new sources of inspiration in addressing issues related to unfair terms and default rules. In addition, given its scope of application, the Unfair Terms Directive may not hinder price discrimination policies, that could be detrimental for consumers and for the efficiency of the market.49

III. The role of personalized law Personalized law challenges the traditional relationship between unfairness control of 27 standard terms and default rules. As outlined before, one of the pillars of the unfairness test in consumer contracts are default rules, deemed to reflect the sense of justice within the society. The massive derogation from default rules provided by standard form contract requires an external intervention on the contract, which is paternalistic in nature. Personalized law, based on preferences of the contracting parties, strongly influences this assumption, because the creation of particular default rules could change the basic parameter of the unfairness test. On this regard, the first issue to evaluate concerns the impact of personalized default rules. Secondly, the different approach envisaged by the theory of mandatory default rules will be scrutinized. Finally, attention will be devoted to the practice of personalized pricing.

1. Personalized default rules According to Professor Porat and Professor Strahilevitz, the rise of Big Data could 28 determine the abandon of majoritarian default rules in favor of personalized default rules.50 Under their personalized default-rules theory, the authors affirm: “parties do not directly negotiate the terms of the contract but instead reveal information about their characteristics and traits, which in turn affects the contents of the default rules applied to them”.51 Such approach should “maximize social welfare generally, not just the welfare of the contracting parties”,52 because there would be the possibility to tailor 49 See in general Ben-Shahar, Algorithmic Price Discrimination When Demand Is a Function of Both Preferences and (Mis)perceptions, (2019) 86 U. Chi. L. Rev. 217. 50 Porat/Strahilevitz, supra Part 1.A. See also Geis, An Experiment in the Optimal Precision of Contract Default Rules, (2006) 80 Tulane L. Rev. 1109; Sunstein, Deciding by Default, (2013) 162 University of Pennsylvania Law Review 1; Ben-Shahar/Porat, supra Part 1.B, at 54–97. 51 Porat/Strahilevitz, ibid. at 16. 52 Ibid.

Patti

229

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

default rules according to the choices of particular groups of interest and, in this way, of reducing transaction costs. 29 The aforementioned contribution further explains the benefits that personalized default rules may bring in consumer law, which is seen as “the most natural field in which to apply the personalized default-rules approach. As we have explained, firms have an increasingly enormous amount of data on consumers’ preferences and characteristics, and they can use this data to tailor different default rules for their contracts. The parties can use this same data to settle disputes in the shadow of the law, and courts can use it for adjudicating unsettled disputes”.53 Professor Porat and Professor Strahilevitz declare that consumers “are generally aware of their characteristics and traits”, so that “they will find the personalized default rules more predictable than the impersonal default rules currently applied to their contracts”.54 The main example provided by the authors concerns a clause on “place of delivery”. It is convincingly argued that for a disabled consumer who uses a wheelchair the most efficient default rule would be the delivery at the buyer’s place, instead of the delivery at the seller’s place (which usually is the most efficient default rule in consumer contracts).55 30 The problematic aspect of the theory is that it does not take into account the economic advantages deriving from the standardization of contractual terms. Businesses are particularly interested in adopting standard terms in order to maximize their profits. Therefore, they certainly care about characteristics and traits of consumers, to raise more advantages in terms of applicable price.56 However, with reference to the juridical treatment of the contractual relationship it seems that they are particularly interested in consolidating a set of rules able to provide certainty and speed in cases of disputes with consumers. The transaction costs are usually not an issue, because consumers accept the contents of the contract on a “take-it-or-leave-it basis”, they are not even interested in trying to modify the contents of the contract.57 31 At any rate, personalized default rules could be of particular importance in order to create a personalized benchmark for the assessment of the unfair character of the term. The “two-step” approach outlined by the ECJ on the basis of Article 3(1) of the Unfair Terms Directive (referring to the “significant imbalance between the rights and duties” and the infringement of good faith)58 could be transformed in a single test in which the presumed unfair term is compared with the personalized default rule. In addition, if the term is declared unfair, the incomplete contract may be filled with a personalized default rule that would act as a personalized micro-directive able to provide for an adequate balance between the parties’ rights and duties.59

2. Personalized Mandatory Rules 32

A different approach was developed in a contribution of Omri Ben‐Shahar and Ariel Porat. Starting from the assumption that “people vary in the degree of protection they need and the cost of protection they can afford”, the authors propose the adoption of personalized mandatory rules.60 The argumentations of the authors seem to fit with 53

Ibid. at 24. Ibid. 55 Ibid. at 13. 56 See infra III.3. 57 See Rakoff, supra (fn. 1), at 1177. 58 See supra II.2.b. 59 The personalized gap-filler would work as a “self-driving contract”, according to the expression adopted by Casey/Niblett, Self-Driving Contracts, (2017) 43 J. Corp. L. 1. 60 Ben-Shahar/Porat, Personalizing Mandatory Rules in Contract Law, (2019) 86 U. Chi. L. Rev. 255. 54

230

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

situations that may occur in presence of standard contracts. They assume that “bad terms” may be contained in contracts because people do not notice or understand them, or because people cannot afford better ones. Such scenarios affect in particular transactions involving unsophisticated parties. That is why not everything should be left to “freedom of contract” and some basic protections should be nondisclaimable.61 Another indisputable point of departure of the contribution is that, on the one side, mandatory rules are effective because they assure a minimum bundle of rights that cannot be circumvented. On the other side, mandatory rules are risky because they could raise prices, shrink markets, or impose regressive cross-subsidies. On this basis, Omri Ben‐Shahar and Ariel Porat argue that “the service of legal 33 protection could be personalized to correspond to the predicted protective needs of contracting parties”.62 According to the authors, “if done properly, personalization could increase the benefits and reduce the unintended costs of mandatory law. Protective needs would be better addressed, and more consumers would be served”.63 The aforementioned proposals could be of great importance for the control of unfair 34 terms. The control mechanism developed in the European context relies on the idea of granting flexibility to contracting parties and, thus, respecting freedom of contract.64 Therefore, the benchmark to evaluate the unfair character of the terms are primarily the default rules and not every derogation of the latter, made through standard terms, amounts to a result which is prohibited by the law. If mandatory rules are personalized and adaptable to the contract according to the circumstances of the case there would not be anymore the need of solving issues related to unfair terms in referring to default rules. Also a list of terms regarded to be unfair, would not be useful. An unfair term would be a term contrary to a personalized mandatory provision. It would not be necessary to rely on default rules. As indicated before, default rules (personalized in nature) would nevertheless remain of importance as gap-fillers after the declaration of unfairness of a term. On closer inspection, the theory of personalized mandatory rules puts a strain on the 35 way in which European law tackles consumer issues.65 In fact, the application of mandatory consumer law follows the core notion of consumer, namely “a natural person, who is acting outside the scope of an economic activity (trade, business, craft, liberal profession)”.66 This rough status-based approach was recently criticized also in the European scholarship.67 Mandatory rules, as for instance withdrawal rights granted

61 Ibid. See also Bar-Gill, Seduction by Contract: Law, Economics, and Psychology in Consumer Markets, Oxford University Press 2012, 26–44. 62 Ben-Shahar/Porat, supra (fn. 60), at 256. 63 Ibid. 64 See Eidenmüller/Faust/Grigoleit/Jansen/Wagner/Zimmermann, Towards a revision of the consumer acquis, (2011) 48 Common Market Law Review 1077, 1087: “The device of court control over contract terms is a substitute for mandatory law that is less intrusive upon party autonomy and thus more in line with the founding principles of contract law. If generalized beyond the area of consumer transactions, it more or less obviates the need for mandatory rules”. 65 See also Bar-Gill/Ben-Shahar, Regulatory techniques in consumer protection: A critique of European consumer contract law, (2013) 50 Common Market Law Review, 109, 113–15. 66 Library of the European Parliament, The notion of ‘consumer’ in EU law, http://www.europarl. europa.eu/RegData/bibliotheque/briefing/2013/130477/LDM_BRI(2013)130477_REV1_EN.pdf. 67 See Eidenmüller/Faust/Grigoleit/Jansen/Wagner/Zimmermann, supra (fn. 64), at 1082: “Consumers do not form a distinct class of human beings at all, as the concept denotes a social role rather than a social group. In some contexts, everyone is a consumer. The suggestion that the vast majority of the population – consumers – suffers from congenital psychological defects is untenable. As human beings, consumers employ the same cognitive short-cuts and suffer from the same psychological biases as anyone else”; Bar‐ Gill/Ben‐Shahar, supra (fn. 65), at 113: “consumers are a heterogeneous group, with different preferences

Patti

231

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:52 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

to consumers, represent a significant cost for the businesses and create cross-subsidizations between consumer buyers.68

3. Personalized Pricing New technologies and the availability of data related to consumers enable businesses to provide for price discrimination with respect to the customers. Such practice occurs when a business charges a different price to different groups of consumers for identical or similar goods or services for reasons not associated with the cost of supply. The problem is that businesses pursue the aim of maximizing their profits without paying attention to consumers’ needs and their possible biases.69 In their paper on personalized mandatory rules, Omri Ben‐Shahar and Ariel Porat argue that personalized pricing is not per se a practice to avoid, if price discrimination is coupled with personalized mandatory rules.70 According to the authors, consumers willing to be protected more would pay more than others, willing to be protected less, the system would be more efficient and cross-subsidies would be avoided.71 The claim seems convincing, but price discrimination practices not only depend on costs related to the application of mandatory provisions, as for instance withdrawal rights granted to consumers. At any rate, problems exist in the actual legal landscape where mandatory rules apply in a uniform way and businesses have the possibility to provide for a discrimination in their prices. 37 Due to its scope of application, the Unfair Terms Directive seems not able to tackle the issue. As outlined before, the substantive control of contractual terms does not involve the so-called core terms, namely the terms concerning the price and the subject matter of the contract.72 It is questionable whether the ratio that inspired the European legislator more than twenty-five years ago is still sound in the light of the technological changes and the price discrimination practices. Gerhard Wagner and Horst Eidenmüller raise concerns of distributive justice and indicate that consumers may lose some of the total surplus that they enjoyed previously.73 They propose to address the issue through “a warning to the consumer that she is being offered a personalized price and, in addition, a right to indicate that she does not want to participate in a personalized pricing scheme proposed to”.74 Thus, an opt-out system that consumers could activate would be created. The authors admit that such an opt-out mechanism would create some negative effects, as the personalized price for people that “remain” in the personalized pricing scheme would be enhanced, but they believe that in the aggregate consumers will recoup some of the rents lost to businesses under full personalization.75 36

and different budgets”. See also Cafaggi, From a status to a transaction-based approach? Institutional design in European contract law, (2013) 50 Common Market Law Review 311. 68 Ben-Shahar/Posner, The Right to Withdraw in Contract Law, (2011), 40 J. Legal Stud. 115, 127–28, 144–45; Eidenmüller/Faust/Grigoleit/Jansen/Wagner/Zimmermann, supra (fn. 64), at 1097–1107; Smits, Rethinking the Usefulness of Mandatory Rights of Withdrawal in Consumer Contract Law: The Right to Change Your Mind, (2011) 29 Penn St. Int’l L. Rev. 671; Ben-Shahar/Porat, supra (fn. 60), at 256–58. 69 See Miller, What Do We Worry about When We Worry about Price Discrimination? The Law and Ethics of Using Personal Information for Pricing, (2014) 19 J. Tech. L. & Pol. 41, 45–47; Calo, Digital Market Manipulation, (2014) 82 Geo Wash L. Rev. 995,1010; Bar‐Gill, Algorithmic Price Discrimination When Demand Is a Function of Both Preferences and (Mis)perceptions, (2019) 86 U. Chi. L. Rev. 217, 224–27; Wagner/Eidenmüller, supra (fn. 18), at 585–88. 70 Ben‐Shahar/Porat, supra (fn. 60), at 265, arguing that each consumer would receive the individually optimal quality. 71 Ibid. 72 See supra II.2.a). 73 Wagner/Eidenmüller, supra (fn. 18), at 584. 74 Ibid. 75 Ibid. 592.

232

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:53 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

Of course, the problem needs further inspection. Nevertheless, it seems wrong to 38 address the issue of unfair price discrimination through classical remedies developed in European consumer law. Studies devoted to the practice demonstrate that personalized pricing is not necessarily and evil to combat, but that some limitations of contractual freedom are needed. In this sense, it seems that the only effective measure to control price discrimination schemes is evaluating the functioning of the algorithm that determines the prices and assure its reasonable and fair use.76 Moreover, it is possible to support the idea of providing personalized pricing caps, in order to avoid an overexploitation of consumers.77 If technology through big data permits to create personalized rules, than it could also permit to shape fair personalized prices.

IV. The enforcement The personalization of the law and the use of new technology may also impact on the 39 enforcement of the Unfair Terms Directive. In the last twenty years, important ECJ decisions have provided for a deviation from some rules of national civil procedure. The aim was to enhance the level of effectiveness of the Unfair Terms Directive in individual litigation, in accordance with art. 7(1) which states that “Member States shall ensure that, in the interests of consumers and of competitors, adequate and effective means exist to prevent the continued use of unfair terms in contracts concluded with consumers by sellers or suppliers”.78 In this regard, in order to guarantee the effectiveness of consumer rights, in several 40 cases the ECJ gave a consumer-friendly interpretation of EU law and conferred to the national courts specific powers to apply rules of its own motion in favor of consumers.79 In some judgments regarding the Unfair Terms Directive, the ECJ stated that a national court must also investigate of its own motion whether a term in a disputed contract concluded between a seller or supplier and a consumer falls within the scope of the directive.80 More in detail, with respect to the consequences deriving from a declaration of 41 unfairness, in the known decision Pannon, the ECJ held that “The national court is required to examine, of its own motion, the unfairness of a contractual term where it has available to it the legal and factual elements necessary for that task. Where it considers such a term to be unfair, it must not apply it, except if the consumer opposes that non-application”.81 For the purposes of the present contribution, the last part is the 76 See generally Chagal-Feferkorn, The Reasonable Algorithm, (2018) 1 U. Ill. J.L. Tech. & Pol’y 111, 116–17, trying to develop a “reasonable algorithm” standard applicable to non-human decision makers; Burk, Algorithmic Fair Use, (2019) 86 U. Chi. L. Rev. 283. 77 Bar‐Gill, supra (fn. 69), at 242–44. 78 See, i.a., Reich, The Principle of Effectiveness and EU Contract Law, in: Rutgers/Sirena (eds.), Rules and Principles in European Contract Law, Intersentia 2015, 46–49, 63; Cafaggi/Iamiceli, supra (fn. 42), at 575, 584–88. 79 See, ECJ, 27 June 2000, Joined Cases C-240/98 to C-244/98, Océano Grupo Editorial SA v. Rocío Murciano Quintero et al., para 29; ECJ, 26.10.2006, Case C-165/05, Elisa María Mostaza Claro v. Centro Móvil Milenium SL, para 38; ECJ, 4 June 2009, Case C-243/08, Pannon GSM Zrt. v. Erzsébet Sustikné Győrfi, para 25; ECJ, 6 October 2009, Case C-40/08, Asturcom Telecomunicaciones v. Cristina Rodríguez Nogueira, para 30–31. See also Trstenjak/Beysen, European consumer protection law: Curia semper dabit remedium?, (2011) 48 Common Market Law Review 95, discussing relevant case law. 80 ECJ, 9 November 2010, case C-137/08, VB Pénzügyi Lízing Zrt v. Ferenc Schneider, para 56; ECJ, 14 June 2012, case C-618/10, Banco Español de Crédito SA v. Joaquín Calderón Camino, para 44; ECJ, 21 February 2013, case C-472/11, Banif Plus Bank Zrt v. Csipai C. V., para 24. 81 CJEU, 4 June 2009, case C‐243/08, Pannon GSM Zrt./Erzsébet Sustikné Győrfi, para 35.

Patti

233

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:53 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

most important: the consumer may oppose the non-application of a term and that corresponds to a personalization of the treatment. In fact, consumers may choose whether to take advantage or not from the declaration of unfairness.82 The binding effect of the term is finally remitted on the consumer, who is considered the person in the best position to know which should be the outcome of the judge’s assessment. It is arguable that after the information given by the judge on her own motion about the unfairness of the term, perhaps through legal advice, the consumer is able to take a decision. In this way, the ECJ pays respect to the principle of party autonomy.83 42 In the light of the possibility to adopt technologies to control contractual terms, new enforcement mechanisms could be implemented. Some authors argue that it is possible to partly automate the process of abstract control of fairness of clauses in online consumer contracts.84 Certain tasks which currently need to be performed by humans could be automated in the process of abstract control of the clauses in consumer contracts. Through the use of algorithms it would be possible to highlight clauses that are detrimental to consumers.85 The creation of a dataset could then help to improve machine learning technology to detect potentially unfair terms.86 This is something which would primarily affect public and collective enforcement measures provided by authorities or consumer organizations.87 It is clear that, combined with the new proposals concerning the personalization of the law, such an enforcement mechanism would maximize the effectiveness of consumer protection and grant efficiency to the market.

V. Conclusion It is often stated that the control over standard terms should not correspond to an imposition of mandatory model terms, but should be shaped as a flexible tool in order to secure a certain degree of variety depending on the different market sectors. An assumption of law and economic theory is that in a contract efficient terms are the ones that the parties would have added if they had negotiated. In the light of the above, it is not surprising that default rules played central role for the assessment of the unfair character of a clause in the European tradition. 44 The proposals on personalization of the law may dramatically change the frame of reference. Flexibility will not anymore be granted by the possibility to derogate to a certain extend from uniform default rules according to the principle of good faith, but by personalized default rules shaped on the basis of particular features of the contracting parties that may work as micro-directives in a self-driving contract environment. On the other side, personalized mandatory rules will avoid the cross-subsidies and enhance the efficiency of the system. The latter topic should be connected to persona43

82 See Micklitz/Reich, The Court and Sleeping Beauty: The Revival of the Unfair Contract Terms Directive (UCTD), (2012) 51 Common Market Law Review 771, 781: ‘the consumer cannot be protected against his will; the national court has to give the consumer a chance to make his opinion heard, but need not wait for an initiative of the consumer. There must be some sort of minimal “contradictory proceedings” to be arranged by the national court allowing the consumer to oppose the non-application of the clause.’ 83 Steennot, supra (fn. 7), at 610. 84 Micklitz/Pałka/Panagis, The Empire Strikes Back: Digital Control of Unfair Terms of Online Services, (2017) 40 Journal of Consumer Policy 367. 85 Ibid. 377–83. 86 Ibid. 87 Cf generally Rott, supra (fn. 23), at 308–11.

234

Patti

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/10_F.P. Patti, Personalization of the Law and Unfair Terms in Consumer Contracts, 06.06.2019.3d from 30.10.2020 11:36:53 3B2 9.1.580; Page size: 160.00mm  240.00mm

J. Personalized Law and Unfair Terms in Consumer Contracts

lized pricing, a practice which in some cases can be detrimental for consumers. The adoption of new technologies shows that the exclusion of the substantive control of price terms can be repealed, because a paternalistic control is needed. Moreover, the effectiveness of the protection could augment in the future through the adoption of machine learning software able to detect unfair terms and to provide for personalized gap-fillers. All these discussions may appear excessively futuristic. Actually, they confirm that 45 technological developments may require significant modifications of consumer law.

Patti

235

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/11_Von Bar, Villa Vigoni 12 05 2017 GB.3d from 30.10.2020 11:36:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

K. Personalization of Tort Law? When the crisis of the VW group due to wrong measurements of exhaust gas emissions was currently in the news, I wanted to check the stock price of my VW shares on the internet. On the chosen website appeared an advertising of the first volume of my “Gemeineuropäisches Sachenrecht” which had just been published. I was so excited that I printed the page immediately, and I also went to see my assistants to show them that every person, who was checking the VW stock price, on that day would find my book. You can imagine their mild smile in response of my proudness. The advertisement was personalized; the day before I must have looked for comments and reactions to my book on the internet. So the advertising appeared only on my computer and nowhere else. The shame was the appropriate punishment for my vanity. But that is not all. The website operator or the person, who rented the advertising space, obviously had access to information about what I am doing at my writing desk. And this fact caused a really uncomfortable feeling. 2 In the future there will be many more problems than the betrayal of such little secrets. Future prospects predict the accumulation of data concerning persons and things in an incredible amount, and they will be compiled and used for purposes of all kind. There are already horror scenarios on the horizon: people with bracelets, which record their physical condition and emotional state as well as their consumption of medicine; people with glasses, which will indicate exhaustion and periods of stress. Cars will perhaps record every route they covered up to now, autonomously driven of course, so that the favourite restaurant of the owner, the holiday stays and the address of a boy or girl friend including the length of stay will always be available in the system. The professional career of a person, the working hours he did today, the contact details of the persons he met or with whom he communicated, all this information could be recorded to be available, to be read or to be heard. 3 This is still very much in the future. But some colleagues are already considering the question, which impact this development could have on tort law. Amazingly enough, they do not require a better protection against the flood of data, which infringe personal privacy, nor they require at least to classify the collection and compilation of data as a dangerous activity, for which a new and strict liability regime is required. The two American authors, Ben‐Shahar and Porat, who induced Christoph Busch to invite me to talk about the “personalization of tort law”, have a completely different intention. In their opinion, in consideration of the huge amounts of data, we can abandon the timehonoured “objective reasonable person standard” and “adopt instead a personalized subjective standard of care”.1 Objective negligence-based standards of care were a mere consequence of lack of information, and as in the near future there would be no lack of information any more, such standards would be redundant. 4 Even if I do not dispose of knowledge, not even rudimentary, about the technical basic principles of their idea, I dare to say, that, from a legal point of view, it seems incorrect to me. In the end, what both authors are aiming at seems to be a well-known matter: the tightening of the standard of care. Ben‐Shahar and Porat write: “The most favourable case for personalization is increasing the standard of care for risky injurers. 1

1

Ben‐Shahar/Porat, supra Part 1.B.

236

von Bar

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/11_Von Bar, Villa Vigoni 12 05 2017 GB.3d from 30.10.2020 11:36:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

K. Personalization of Tort Law?

(…) Also, increasing the standard of care for skilled injurers has many more pros than cons”. The latter appears halfway comprehensible to me, because people with a special expertise are already today obliged to use it. If they fail to do so, the borderline between negligent and intentional behaviour becomes rather thin. That is so because “a person causes legally relevant damage intentionally when that person causes such damage (…) by conduct which that person means to do, knowing that such damage, or damage of that type, will or will almost certainly be caused” (DCFR VI.–3.101(b)). And no one can hide behind a statutorily fixed standard of care when he has to understand in a real life situation that under the circumstances more care is required. You are not allowed to burn straw on your field, even if you respect the distance prescribed by law to the adjacent forest, if you realise, that the wind is much too strong on this day. However, less understandable is the concept of an increased duty of care for “risky 5 injurers”. First, the authors come back to a “standard” of care, which is a generalising normative principle. So not even the authors think in terms of “personalisation”; rather they still refer to a “standard”. A “standard”, however, in a strict sense is no longer consistent with their approach. From the viewpoint of a “personalised tort law” it is not important anymore what other people are doing or what they have to do; everyone is living in his own little microcosm of negligence law. Second, negligence cannot be examined without a comparison of the concrete behaviour of a person with the behaviour of another person, which only exists in theory. So it seems also inappropriate to still use terms like negligence, care or carelessness. In the absence of a reference point – the reasonably thinking and acting doctor, lawyer, car driver, et cetera – each negligence test will simply be obsolete. And should the analysis lead to the question: “what will happen, when we know everything?” then the answer will probably be: in this case we will need no law any longer and also no tort law. Finally (and on a much more banal level): The idea of standards of “care” for “risky injurers” appears extremely strange to me. Of course, a person, who suffers from diabetes, has to control his blood sugar level, before he or she will start a dangerous machine; a person, who is exhausted, has to practice reasonable self-monitoring before starting a car, so that he or she will notice the exhaustion. Nothing in that is new. But to develop care standards for criminals and persons with a huge potential for violence, i.e. the truly “risky injurers”, this is certainly not the task of the law. So what could be the purpose of collecting huge amounts of data in terms of liability 6 law? Even if, horribile dictu, a judge or a liability insurer could get a personality profile at the push of a button – what is the use of it? Data are not conducive to the decisionmaking process. The evaluation of carelessness always requires a normative assessment, and this assessment has to be done in each and every case. Data are data and not more. They do not provide an answer to issues of law. They are completely useless to ascertain whether or not the requirements of a certain tort are met. Of course this also applies to the small part of tort law, which is the subject of the theses: negligence. Even former judgments, where the act or omission of a person was qualified as negligent, do not have a reliable prejudicial effect. That is so because a specific situation does never recur. The data furore may allow for answers in certain matters of fact. In contrast, for the negligence judgment itself, it seems meaningless to me. It is important to note that in many cases today the analysis of negligence is not at 7 all relevant any longer. In their publication the authors consider their national law. A more comparative approach, however, would quickly demonstrate that their examples simply do not fit in other areas of the world, in particular not in continental Europe. In France e.g., the cases of gardien liability outnumber the cases of negligence liability by far. And also in Germany the liability for road traffic accidents follows a strict von Bar

237

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/11_Von Bar, Villa Vigoni 12 05 2017 GB.3d from 30.10.2020 11:36:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 3. Personalization in Contract, Consumer and Tort Law

regime, even if there may be specific provisions for the collision of two or more cars. Halter, gardiens, custodes and many other are not liable for having ignored the (or a) standard of care. They are quite simply liable because they controlled (secured by a liability insurance) a person or a property and the control went lost for some reason. Such a strict liability has often been introduced for technical developments where, due to lack of experience, it was not yet possible to establish with certainty, what is careful and what negligent. For this reason it is, in my view, exigent to introduce a special regime of strict liability for the collection, storage and the transfer of huge amounts of data. But it is absolutely dispensable to use data in order to backfill a “personalized” negligence judgment. 8 By the way: One right out of a sudden notices that a strict liability regime displays much more humanity than a sort of “personalized” liability. No one’s personality will be exploited, dissolved, screened. It is not important, what a person did, thought and felt. And that is why it remains a private concern and nobody else’s business even in terms of a pretended protection against damage. Victim protection can be organised in a much more intelligent and effective way, if it is not based on the well-nigh absurd and voyeuristic wish to know everything about everyone. 9 There are many parallels in the traditional method of determining negligence. A dentist cannot say that he has learning difficulties and forgets quickly, that he had had a dispute that day with his wife or that he did not feel well. As he cannot use such arguments to defend himself, he is not obliged to reveal his innermost. Liability occurs when a person “made a mistake”, but it does not matter whether he or she was “able to help it”. In my opinion it shall remain that way. In public life everybody should be able to rely on other people doing what one can expect from them in the circumstances of the given situation. 10 Individualized standards of care do not prevent accidents; they make them more likely. If everybody had to respect specific speed limits depending on his current personal state of health and mind, the participation in road traffic would amount to an intolerable risk. The clever machine which knows everything about the user, could perhaps send out warning signals: “Attention! This bicyclist is exhausted, this pedestrian has too high a blood pressure today!” But because of all the warning signals you would not find your way anymore. It is precisely the purpose of the concept of negligence that you can move freely without always looking back over your shoulder (or constantly observing your electronic devices). The objective standard of negligence is nothing more than a piece of guarantee of personal freedom, both for the potentially responsible person and for the potential victim. The generalization has nothing to do with a lack of information; it simply makes the exchange of information superfluous. And it protects the victim against being reproached of contributory negligence. Too much knowledge does not enhance the protection of victims but it reduces it. I fear that the concept of a “personalised” negligence liability is based on a fundamental misunderstanding. The fact that the currently existing law does not permit a precise prognosis concerning the outcome of a trial (because different judges, applying the general test, may come to different conclusions) reveals no drama. It is the expression of a law made by people for people. Desperate attempts to replace general framework directives by always new and always more concrete details are prone to failure. The number of subcategories is as infinite as life itself. This becomes apparent in many areas of the law which face, or seem to face, similar difficulties. If a judge is not able anymore to assess by virtue of his own personality and his own competence what is “unreasonable”, “inadequate”, “unethical”, society does not need him anymore. Only two generations ago this was still beyond dispute; our generation lost a lot. 238

von Bar

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/11_Von Bar, Villa Vigoni 12 05 2017 GB.3d from 30.10.2020 11:36:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

K. Personalization of Tort Law?

There is no need to “personalize” tort law. Rather tort law, in conjunction with the 11 law against unjustified enrichments and property law, should develop more effective methods to confront the dreadful collection and use of data. It should strengthen again and invoke one of his most important legally protected interests: the right to protection of personal privacy. This is a crucial element for the preservation of the rule of law, perhaps even for the future of the western civilisation. Issues of damage and damages are much too trivial to sacrifice on the altar of technology what is really essential. Big data and tort law cannot be friends.

von Bar

239

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/11_Von Bar, Villa Vigoni 12 05 2017 GB.3d from 30.10.2020 11:36:57 3B2 9.1.580; Page size: 160.00mm  240.00mm

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

PART 4 TECHNOLOGICAL AND BEHAVIORAL PERSPECTIVES L. Personalized Law and the Behavioral Sciences Interdisciplinary work is not novel per se in legal analysis. What has arguably 1 changed significantly over the past two decades, however, are the disciplines that interdisciplinary legal scholarship primarily draws on. One key field that legal academics as well as regulators increasingly turn to is behavioral economics.1 The resulting legal applications in the field of behavioral law and economics have provoked a vivid debate in the past years.2 This chapter argues that one crucial, but often overlooked, problem in the operationalization of behavioral insights for lawmaking is actor heterogeneity, i.e., the fact that different people act according to different heuristics and are subject to different biases at varying degrees. Therefore, behavioral law and economics arguably stands much to gain from 2 personalizing behavioral interventions so as to match regulatory strategies with the concrete needs, capacities, and vulnerabilities of the addressees.3 Recent advances in the quantification of bias, particularly when paired with machine learning techniques, have swung the door wide open toward a greater behavioral personalization of legal content. However, the law simultaneously needs to be conscious of the inherent limits of personalized behavioral law and economics. The mere technological possibility of personalizing behavioral interventions ought not to preclude a broad legal and societal debate about its normative desirability. The more modest ambition of this chapter is to map out the potential advantages of behavioral personalization; to point to different drawbacks; and to discuss good governance mechanisms, in order to spur and facilitate such a debate. The chapter proceeds in six steps along this path. First, it presents a very brief 3 introduction to behavioral law and economics for those less familiar with this field of legal research. Second, it highlights actor heterogeneity, and the resulting knowledge problem about the degree of rationality of regulatees, as one of the key challenges for contemporary behavioral law and economics. Third, it shows, along different examples, how personalization can solve the problem of actor heterogeneity. Fourth, it points to some limits and drawbacks of personalization. Fifth, it introduces some core elements of a good governance of personalized behavioral law and economics that could potentially rein in its disadvantages. The final section concludes.

1 See, e.g., Jolls/Sunstein/Thaler, A Behavioral Approach to Law and Economics, (1998) 50 Stan. L. Rev. (Stanford Law Review) 1471; Korobkin/Ulen, Law and Behavioral Science: Removing the Rationality Assumption from Law and Economics, (2000) 88 California Law Review 1051; more recently Hacker, Verhaltensökonomik und Normativität, Mohr Siebeck 2017; OECD, Behavioural Insights and Policy: Lessons from around the World, Report, 2017, https://people.kth.se/~gryne/papers/OECD_2017.pdf; Lourenco et al., Behavioural Insights Applied to Policy. European Report 2016, European Commission 2016, http://publications.jrc.ec.europa.eu/repository/bitstream/JRC100146/kjna27726enn_new.pdf. 2 See, e.g., Sunstein, Why Nudge, Yale University Press 2014; Rebonato, Taking Liberties, Palgrave Macmillan 2012. 3 See also Hacker, Personalizing EU Private Law. From Disclosures to Nudges and Mandates, (2017) 25 European Review of Private Law 651, available at http://ssrn.com/abstract=2914393.

Hacker

241

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

I. A very short introduction to behavioral law and economics 4

To grasp the potential of the personalization of behavioral law and economics, it is first necessary to briefly survey the state-of-the-art of contemporary behavioral law and economics. Therefore, this first section provides a short primer on behavioral economics and discusses some illustrative legal applications of behavioral economics.

1. Behavioral economics Behavioral economics was born out of the confluence of cognitive psychology with empirical, economic decision theory.4 It gained importance as the key empirical challenge to the hitherto dominant theory of human decision making in economic contexts: expected utility theory (EUT). This theory, prepared by the marginalist revolution in economics at the end of the 19th century, and formalized by von Neumann and Morgenstern in the 1940s, posits that human agents choose between different alternatives such that they maximize their own utility.5 Since the future evolution of the states of the world is not always certain, they have to factor in the probabilities of different possible developments, and have to estimate the potential gains and losses under each alternative. Therefore, homo economicus essentially has two distinguishing characteristics that, simultaneously, denote rationality:6 the correct calculation of probabilistic (i.e., expected) utilities based on the information available; and the choice of an alternative that maximizes expected utility. At least, EUT posits, decision makers act as if they were fulfilling these two criteria. 6 Behavioral economics has powerfully challenged this hypothesis. First, in the 1950s already, Herbert Simon introduced the concept of bounded rationality. He argued that real decision makers often do not maximize, but limit themselves to reach an aspiration level (satisficing).7 This term was taken up, and adapted, by the heuristics and biases literature from the 1970s on, primarily developed by Daniel Kahneman and Amos Tversky.8 This literature chiefly challenged the other characteristic of homo economicus, its capability for mathematically correct calculation of outcomes. More specifically, empirical studies demonstrated systematic deviations from the axioms and predictions of EUT that can be grouped into four main categories: bounded rationality (in the modern sense); bounded willpower; bounded self-interest; and cognitive capacity limits.9 7 In the modern sense, bounded rationality means that human decision making is fraught with biases, and driven by heuristics, that prevent the correct calculation of expected utilities.10 For example, agents are supposed to be overoptimistic, overestimat5

4 For a more detailed account of behavioral economics see, e.g., Camerer/Loewenstein/Rabin, Advances in Behavioral Economics, Princeton University Press 2004; for an introduction into the development of behavioral economics, see, e.g., Hacker, supra (fn. 1), at §§ 3–4. 5 von Neumann/Morgenstern, Theory of Games and Economic Behavior, Princeton University Press 1944; see, more generally, Shoemaker, The Expected Utility Model: Its Variants, Purposes, Evidence and Limitations, (1982) 20 J. Econ. Literature 529. 6 Cf Sen, Rational Behavior, in: Durlauf/Blume (eds.), The New Palgrave Dictionary of Economics, 2nd edn., Palgrave Macmillan 2008. 7 Simon, A Behavioral Model of Rational Choice, (1955) 69 The Quarterly Journal of Economics 99. 8 See, e.g., Kahnemann/Slovic/Tversky (eds.) Judgment under Uncertainty: Heuristics and Biases, Cambridge University Press 1982. 9 See Hacker, supra (fn. 1), at § 4. 10 Cf Thaler, Doing Economics Without Homo Economicus, in: Medema/Samuels (eds.), Foundations of Research in Economics: How Do Economists Do Economics?, Edward Elgar 1996, 227, 230.

242

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

ing their own capacities and underestimating the likelihood of negative outcomes; they are overconfident, thinking that their judgment is more precise than it really is; and they are subject to confirmation bias, processing information in a selective way that fits with their own, pre-conceived opinion.11 (Cumulative) prospect theory attempts to unify some strands of biases by showing that, more generally, people evaluate outcomes relative to their point of reference (and not in absolute terms, as EUT posits); that they exhibit not only diminishing sensitivity, but also loss aversion; and that probabilities are not treated as mathematical objects, but rather subjectively transformed.12 Bounded willpower, in turn, testifies to the difficulty of following up on one’s own plans.13 People tend to attach more importance to instantaneous gratification than to long-term wellbeing: they often exhibit a present bias that can be understood as impatience or impulsiveness. Bounded self-interest posits that agents not only care about their own utility, but also about those of others, and the enforcement of community norms such as fairness.14 Finally, cognitive capacity limits highlight that there is an absolute limit to the amount of information people can cognitively process at one time, and this limit will often lie at less than 10 items, sometimes even significantly lower.15 All in all, research in behavioral economics, which may more appropriately have been 8 called cognitive economics, unearthed a vast number of biases, and more general limitations on mathematically and logically correct human decision making, that lead to systematic deviations from EUT. Particularly, the processing of information is marked by significant quantitative and qualitative bounds, which is of prime importance for regulatory strategies such as the disclosure paradigm prevalent in much of US and EU consumer law.

2. Legal applications Against the background of these behavioral findings, it does not come as a surprise 9 that a number of legal interventions take these behavioral insights into account. Some seek to actively harness biases in order to facilitate the attainment of certain regulatory goals (public nudging),16 while others have been designed to counteract biases in order to stimulate more adequate decision making.17 Broadly, these legal applications can be divided into three main groups: the revision of i) disclosures; ii) default rules; and iii) mandatory law. For instance, if consumers are overly optimistic, disclosures should highlight more 10 prominently the risks of products. And if cognitive capacity limits tend to lead to information overload, it seems sensible to make disclosures more concise, to reduce 11 See, e.g., Weinstein, Unrealistic optimism about future life events, (1980) 39 Journal of Personality and Social Psychology 806; Nickerson, Confirmation bias: A ubiquitous phenomenon in many guises, (1998) 2 Review of General Psychology 175; see also Hacker, supra (fn. 1), at 81–91 for more recent studies. 12 Kahneman/Tversky, Prospect Theory: An Analysis of Decision under Risk, (1979) 47 Econometrica 263; Tversky/Kahneman, Advances in Prospect Theory: Cumulative Representation of Uncertainty, (1992) 5 Journal of Risk and Uncertainty 297. 13 Laibson, Golden Eggs and Hyperbolic Discounting, (1997) 112 Quarterly Journal of Economics 443. 14 Güth/Kocher, More than thirty years of ultimatum bargaining experiments: Motives, variations, and a survey of the recent literature, (2013) 108 Journal of Economic Behavior & Organization 396. 15 Cowan, The magical number 4 in short-term memory: A reconsideration of mental storage capacity, (2000) 24 Behavioral and Brain Sciences 87. 16 Thaler/Sunstein, Nudge, Yale University Press 2008; for the terminology, see Alemanno/Sibony, The Emergence of Behavioural Policy-Making: A European Perspective’ in id. (eds.), Nudge and the Law, Hart 2015, 1, 11. 17 Hacker, Nudge 2.0 – The Future of Behavioural Analysis of Law, in Europe and Beyond, (2016) 24 European Review of Private Law 297, 303–304.

Hacker

243

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

complexity, and to introduce different layers that contain information of differing levels of detail. For example, in the EU, short key investor documents were introduced that are meant to provide better guidance to retail investors about the crucial characteristics of financial products.18 In the US, mortgage disclosures were redesigned to reduce information overload and to enhance the salience of important items.19 11 Default rules were designed, for example, to leverage status quo bias (the tendency to stick with the default option) in order to raise participation in retirement saving programs in the US.20 In the EU, conversely, Art. 22 of the Consumer Rights Directive21 prohibits pre-ticking boxes in online sales in order to prevent status quo bias and limited attention from making consumers “assent” to features they do not actively want. Finally, in the realm of mandatory law, it was argued that negligence standards in product liability should be heightened to better reflect the underestimation of product risk by consumers,22 and that product liability more generally should take suboptimal risk evaluations by consumers into account.23

II. The knowledge problem in behavioral law and economics 12

Many of these policy interventions are based on the assumption that the average agent, instead of acting rationally, will exhibit a specific bias. However, it is clear that not every person acts in the same, biased way. For example, the data shows that, in some situations, some actors are under- rather than overconfident; similarly, some people tend to be pessimistic, rather than overly optimistic. This finding significantly complicates drawing policy conclusions from behavioral insights,24 and provoked a critique of behavioral law and economics that postulated a novel “knowledge problem”.25 Regulators, the critique holds, are not in a position to know if an individual is afflicted at all by a certain bias, and if yes, to what extent. This critique presents a particularly serious challenge to behavioral law and economics as it starts from the very methodological premises that behavioral law and economics is based on. As we shall see, personalized law can offer a novel solution to this type of knowledge problem.

1. Uncertainty about the true rationality of market actors 13

Proponents of the knowledge problem rightly point to the limits of internal and external validity of empirical studies26 to caution about the direct use of empirical 18 Art. 6 of the Regulation (EU) No 1286/2014 of the European Parliament and of the Council of 26 November 2014 on key information documents for packaged retail and insurance-based investment products (PRIIPs), OJ 2014 L 352/1; on this see, e.g., Hacker, supra (fn. 1), at 454–457; 751–752. 19 Kleimann Communication Group, Know Before You Owe: Evolution of the Integrated TILA-RESPA Disclosures, (July 2012), http://files.consumerfinance.gov/f/201207_cfpb_report_tila-respa-testing.pdf. 20 Thaler/Benartzi, Save More Tomorrow™: Using Behavioral Economics to Increase Employee Saving, (2004) 112 (S1) Journal of Political Economy S164. 21 Directive 2011/83/EU of the European Parliament and of the Council of 25 October 2011 on consumer rights, OJ 2011 L 304/64. 22 Latin, “Good” Warnings, Bad Products and Cognitive Limitations, (1994) 41 UCLA Law Review 1193. 23 Hacker, The Behavioral Divide. A Critique of the Differential Implementation of Behavioral Law and Economics in the US and the EU, (2015) 11 European Review of Contract Law 299, 327–343. 24 ibid. 305–307; id. supra (fn. 1), at § 5. 25 Schwartz, Regulating for Rationality, 67 Stanford Law Review 2015, p 1373; Rizzo/Whitman, The Knowledge Problem of New Paternalism, (2009) Brigham Young U. L. Rev. 905. 26 See, for an introduction to these limits, Lawless/Robbennolt/Ulen, Empirical Methods in Law, Aspen 2010, 36–40.

244

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:15 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

results for policy making. Recent studies have shown, for example, that the endowment effect, is likely mitigated in market contexts.27 Other behavioral effects and biases that were found in the lab could be reproduced in the field;28 even in these cases, however, the findings cannot be generalized to the entire population due to limited external validity. Moreover, for a number of regulatory problems, suitable empirical studies are lacking altogether.29 And finally, different biases may cut in different directions.30 For example, while overoptimism would lead people to underestimate the risk of an earthquake, the overweighting of small probabilities predicted by prospect theory causes them to overestimate the risk. The estimation is further complicated by the different mental availability of earthquake damage for different people: if examples of earthquakes come more readily to mind, people will find their incidence more likely.31 Therefore, regulators today are indeed not necessarily well-positioned to determine 14 what kind of behavior to expect in a pool of regulatees in the face of limited validity, lacking empirical studies, and unclear interactions between multi-directional biases. One possible way forward is to frame regulation, and such a situation, as decision making under uncertainty over the degree of rationality of regulatees.32 Another solution lies in seeking to gather information about degrees of bias on an individual level; this is what recent empirical research, combined with advances in machine learning technology, increasingly enables. This chapter pursues this latter route.

2. Psychometrics and the quantification of bias Behavioral economics has always been founded on empirical research, be it in the lab 15 or, increasingly, in the field.33 More recently, however, it has undergone what may be termed a “quantitative turn”. Researchers have begun to develop precise metrics to evaluate just how much of a particular bias an individual exhibits.34 Perhaps most prominently, Keith Stanovich, Richard West and Maggie Toplak have developed a “Comprehensive Assessment of Rational Thinking”.35 Overall, this test is meant to capture what they call a “rationality quotient”;36 the various subtests the entire test is made of measure a variety of different biases,37 from overoptimism, overconfidence and 27 Arlen/Tontrup, Does the Endowment Effect Justify Legal Intervention? The Debiasing Effect of Institutions, (2015) 44 The Journal of Legal Studies 143; Tontrup, Does the Endowment Effect Prevail When Traders Act Strategically?, (1 July 2017), NYU Law and Economics Research Paper No. 17–18, https://ssrn.com/abstract=2997585. 28 For an overview, see, e.g., DellaVigna, Psychology and Economics. Evidence from the Field, (2009) 47 J. Econ. Lit. 315. 29 Engel, Verhaltenswissenschaftliche Analyse: Eine Gebrauchsanleitung für Juristen, in id. et al. (eds.), Recht und Verhalten, Mohr Siebeck 2007, 363, 373–374. 30 Cf Klaff, Debiasing and Bidirectional Bias: Cognitive Failure in Mandatory Employment Arbitration, (2010) 15 Harv. Negot. L. Rev. 1, 25; Arlen, Comment: The Future of Behavioral Economic Analysis of Law, (1998) 51 Vand. L. Rev. 1765, 1769. 31 Tversky/Kahneman, Availability: A heuristic for Judging Frequency and Probability, (1973) 5 Cognitive Psychology 207. 32 See, e.g., Hacker, Overcoming the Knowledge Problem in Behavioral Law and Economics: Uncertainty, Decision Theory, and Autonomy, (17 July 2015), available at https://ssrn.com/abstract=2632022; id., supra (fn. 1) § 5.A. 33 See, e.g., DellaVigna, supra (fn. 28). 34 For the construction of an overconfidence metric (δ), see Grubb/Osborne, Cellular Service Demand: Biased Beliefs, Learning, and Bill Shock, (2015) 105 American Economic Review 234, 252–253; Stanovich/ West/Toplak, The Rationality Quotient (RQ): Toward a test of rational thinking, MIT Press 2016. 35 Stanovich/West/Toplak, The Rationality Quotient (RQ): Toward a test of rational thinking, MIT Press 2016, 162 et seq. 36 Ibid. 315. 37 Ibid. part II.

Hacker

245

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

confirmation bias to bounded willpower, for example. The tests seem to be reliable38 and have been rolled out in a number of empirical studies.39 The development of the CART is crucial because it leads the way from an aggregate understanding of human rationality to a more individualistic concept that accommodates vast degrees of heterogeneity. While scholarship in behavioral economics, and particularly its legal reception, have often focused on the modal behavior fund in empirical studies, resulting in claims that “people in general tend to… [e.g., be overly optimistic]”, the quantification of bias leads to more refined understanding that can inform us about the many shades of rationality that human agents exhibit, and their individual distribution within a given group. 16 Such insights can arguably be fruitfully paired with machine learning algorithms to predict degrees of bias, and even scores of total rationality (the rationality quotient), of potential regulatees. In another context, the validity of this approach has already been demonstrated. Contemporary research in psychometrics is increasingly using machine learning techniques to predict psychological traits of human agents using their digital footprint.40 In one study, for example, the authors followed four steps to train an algorithm to this end.41 First, they elicited scores of personality traits, using the Big Five traits of personality psychology, from participants, as well as their partners and friends through questionnaires. Second, they matched these traits with the Facebook profiles, and more particularly the distribution of likes over different pages, of participants to establish correlations between psychological traits and digital likes. Third, they trained a machine learning algorithm on a fraction of the collected data, using the digital footprint (the likes) as inputs and the corresponding traits from the questionnaires as labelled outputs. Fourth, on the remaining part of the data, they had the algorithm predict the personality traits using only the digital footprint, and compared it with the elicited questionnaire data. It turned out that, if the digital footprint was rich enough, the algorithm’s predictive power outperformed personality trait predictions of friends, and in some cases even partners, of the persons in question.42 17 This shows that digitally mediated out-of-sample prediction of personality traits is already possible, in many cases, based on the information that people disclose on the Internet, or elsewhere in the digital economy. The same approach could be used to predict specific biases or even entire rationality quotients by replacing the personality tests, in the first step, with the CART (or one of its subtests). The machine learning algorithm could then be trained to predict degrees of bias, or degrees of rationality, of regulatees based on their digital profile. This is precisely what opens the door to the personalization of behavioral interventions: interventions could be made contingent on certain levels of bias.

3. Personalized law as a solution to the knowledge problem 18

Valid predictions of individuals’ degrees of specific biases, or of rationality quotients, hold the potential to solve the knowledge problem that plagues contemporary behavioral law and economics. At the moment, it is difficult to design policy proposals 38

Ibid. 222–223; 236–237. Ibid. part III. 40 Kosinski et al., Mining big data to extract patterns and predict real-life outcomes, (2016) 21 Psychological Methods 493; Kosinski/Stillwell/Graepel, Private traits and attributes are predictable from digital records of human behavior, (2013) 110 Proceedings of the National Academy of Sciences 5802, 5804. 41 Youyou/Kosinski/Stillwell, Computer-based personality judgments are more accurate than those made by humans, (2015) 112 Proceedings of the National Academy of Sciences 1036, 1037. 42 Ibid. 1038. 39

246

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

if it is unclear whether the addressees are, for example, over- or underoptimistic. If most regulatees fall into the former category, the risk of specific products could be more saliently disclosed to counter underestimation of risk by potential product users. If, on the other hand, most regulatees qualify as underoptimistic, the strategy would create over-deterrence. More generally, any uniformly applicable regulatory strategy will fall short of accommodating behavioral heterogeneity: it will address biases that point into one direction, but be ineffective for, or even exacerbate, biases that point into the other direction. Personalization would make behavioral interventions contingent on the direction 19 and level of bias that individuals exhibit. Hence, policy responses can be tailored to the individual vulnerabilities of addressees without imposing externalities on those that behave differently. This, at least, is the promise of personalized behavioral lawmaking. It goes without saying, however, that the acquisition of the data necessary for personalization entail significant costs, too, that will be investigated in Part IV of the chapter.

III. Examples of personalized behavioral law The following section presents examples that spell out the potential of personalized 20 behavioral lawmaking, ranging from disclosures via default rules to mandatory law, before the limits and perils of this type of regulatory intervention are addressed in the next section.

1. Disclosures Disclosures are perhaps the most important regulatory strategy in contemporary 21 private law, both in the US and the EU.43 They abound wherever one party is supposed to have superior information about a relevant feature of a transaction. However, simply conveying information to individuals via disclosures is increasingly viewed as an ineffective way of overcoming information asymmetry.44 People are often not motivated to look at disclosures at all;45 if they do, information processing may be affected by biases; and finally, dense or complex disclosures can quickly lead to information overload, particularly among those less familiar with the covered subjects.46 a) General idea. Personalization of disclosures starts from the idea that different 22 pieces of information are of different relevance for different people. Therefore, personalization would attempt to portray different items on a disclosure in a way that matches the informational need of the addressee. This can be done, for example, by modifying font size and color of textual disclosures; or by distributing pieces of information on different layers of disclosures. Both strategies change the salience of the respective items. If disclosure is to have a future in which it is made relevant for individual addressees, its content must be restricted to what is absolutely necessary and most likely to attract the attention of the recipients. Personalization can adapt this content to individual needs. 43

This part draws on Hacker, supra (fn. 3), at 651, 666–670. Ben‐Shahar/Schneider More Than You Wanted to Know: The Failure of Mandated Disclosure, Princeton University Press 2014; Hacker, supra (fn. 1), at § 9. 45 Ben‐Shahar/Chilton, Simplification of Privacy Disclosures: An Experimental Test, (2016) 45 The Journal of Legal Studies S41. 46 Eppler/Mengis, The concept of information overload: A review of literature from organization science, accounting, marketing, MIS, and related disciplines, (2004) 20 The Information Society 325. 44

Hacker

247

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives 23

b) Concrete examples. Concrete examples of personalized disclosures can start from two recently enacted European directives that contain some of the most important disclosure obligations for market transactions: the Consumer Rights Directive (CRD)47 and MiFID II.48

aa) Consumer disclosures: personalizing the Consumer Rights Directive. The CRD of 2011 expanded information duties vis-à-vis consumers both in content and scope. According to its Article 6(1), for example, 20 items, which often include several subitems, need to be disclosed to consumers for distance or off-premises contracts. This list alone is likely to provoke information overload with a significant number of readers49 – if it is read at all. Moreover, many products that we acquire online have a multi-dimensional price:50 they are made up of different pricing components for different features, shipping varieties, or use formats. This adds another layer of complexity to the evaluation and comparison of products that needs to be disclosed to consumers (see, e.g., Art. 6(e) CRD). 25 With personalization, two strategies can be pursued that could make consumer disclosures more meaningful again. First, disclosures could be split into two (or potentially more) layers of different complexity.51 The first layer would only contain a few indispensable items. Key investor documents such as those provided for in Article 6 of the PRIIP Regulation52 already make an attempt to simplify disclosures in this vein. A second (hyperlinked or annexed) layer would then provide more detailed information for those seeking it. However, recent empirical evidence suggests that merely simplifying disclosures by uniformly reducing information load will not succeed in making consumers read them.53 Personalized disclosures could present a way out of this impasse by making disclosures individually meaningful. One crucial question in this endeavour is: just how much information may the first layer contain to avoid information overload and raise motivation sufficiently for people to read the disclosures? With personalization, the amount of information disclosed on a key consumer information document could be made dependent on the predicted rationality quotient of the consumer. In this way, more rational consumers do not receive too little, and less rational consumers not too much information. 26 A second strategy consists in varying not only the amount, but also the content of the information consumers receive. On every level of consumer disclosures – be it a key consumer information document or a more complex, secondary disclosure – different items can be given varying degrees of salience depending on which items are likely to be most relevant for the respective recipient. This is particularly important for multidimensional pricing. For some products, prices will vary depending on the level of use of the product. For example, cell phone companies often charge a basic fee for a certain data package (for example, 10 € for 2 MB per month), but then charge higher overage rates for

24

47

See supra (fn. 21). Directive 2014/65/EU of the European Parliament and of the Council of 15 May 2014 on markets in financial instruments, OJ 2014 L 173/349. 49 Cf Cowan, The Magical Number 4 in Short-Term Memory: A Reconsideration of Mental Storage Capacity, (2000) 24 Behavioral and Brain Sciences 87. 50 Bar‐Gill, Price caps in multiprice markets, (2015) 44 The Journal of Legal Studies 453. 51 See Hacker, supra (fn. 1), at § 11 A.III.; Grundmann/Hacker, Conflicts of Interest – Theory and the Regime of MiFID I and II, in: Busch/Ferrarini (eds.), Regulation of the EU Financial Markets: MiFID II and MiFIR, OUP 2017, 165, 189–191, 200 et seq. 52 Regulation (EU) No 1286/2014 of the European Parliament and of the Council of 26 November 2014 on key information documents for packaged retail and insurance-based investment products (PRIIPs), OJ 2014 L 352/1. 53 Ben‐Shahar/Chilton, supra (fn. 45). 48

248

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

additional data.54 Hence, the basic fee is most relevant for those consumers staying within the bounds of their expected consumption, while the overage fee is highly relevant for those that overconsume. The salience of the respective price dimensions could be varied depending on how likely it is that consumers are able to accurately predict their future consumption behavior. This can be informed not only by previous call histories (which are monitored by cell phone companies, anyways), but also by the degree of willpower that consumers exhibit. Those with high willpower (as measured by the β factor, for example)55 will thus prominently see the basic fee most prominently, while the overage fee would be highlighted for those with low willpower. Similar strategies can be applied for any other pricing dimensions that depend on predicted use, such as teaser rate tariffs for subscriptions (for example to digital services, newspapers, health club plans or credit cards) that are followed by a period of higher use fees. Furthermore, for those consumers that score high on procrastination,56 the expiration of the withdrawal period could be included in the key consumer information document, while it could be left out for those likely to suffer less trouble in keeping deadlines. bb) Financial disclosures: personalizing MiFID II. A second category of disclosures 27 that could be personalized along behavioral lines are financial disclosures, for example those owed by financial intermediaries under the MiFID II Directive (Art. 23(2), 24). Such disclosures are of particular interest as personal (or, in the case of corporations, corporation-specific) information is already used to differentiate between different actors in securities regulation. Professional investors are trusted to a larger extent together to gather their own information and to make diligent investment decisions; hence, informational duties are reduced vis-à-vis professional investors.57 However, the criteria used to determine whether an investor qualifies as professional 28 are only crude proxies for financial sophistication: according to Annex II of MiFID II, they include not only institutions offering financial services, but also particularly large corporations; furthermore, wealthy individuals that regularly trade securities can request to be classified as professional investors. Overconfidence suggests that a number of these requests will be made despite the inability of the requesting agent to adequately assess the risk of securities;58 and Recital 104 of MiFID II more generally acknowledges that the “financial crisis has shown limits in the ability of non-retail clients to appreciate the risk of their investments”. Therefore, it seems promising to separate different types of investors not based on the criteria enshrined in MiFID II, but based on behavioral factors that measure more directly the (in)ability to appreciate risk. For example, those investors with marked overoptimism and overconfidence could be confronted with a more salient portrayal of the risk profiles of the securities they are attempting to buy.59 Empirical evidence also suggests that overconfident investors tend to lose money by trading too often;60 therefore, they could be warned if their trading frequency surpasses a certain threshold. 54

Bar-Gill, Seduction by Contract, OUP 2012, ch. 5. See Laibson, supra (fn. 13), at 478. See Tuckman, The development and concurrent validity of the procrastination scale, (1991) 51 Educational and Psychological Measurement 473; Choi/Moran, Why not procrastinate? Development and validation of a new active procrastination scale, (2009) 149 The Journal of Social Psychology 195. 57 Recital 86, Art. 24(4)(b), 30 MiFID II. 58 Cf Malmendier/Tate, CEO overconfidence and corporate investment, (2005) 60 The Journal of Finance 2661. 59 One key problem in this context is the measurement, and prediction, of “rationality” of corporate agents; empirical evidence is currently still lacking in this domain. Until this problem is solved, the current, size-based criteria may be used as for lack of a better alternative with corporate actors. 60 Odean, Do investors trade too much?, (1999) 89 American Economic Review 1279. 55 56

Hacker

249

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives 29

All in all, these examples show that there are a number of ways in which personalization can make disclosures, vis-à-vis consumers, investors or other persons, more meaningful again.

2. Default rules 30

Besides disclosure mandates, default rules are a second, important category of legal rules. They determine legal outcomes in the absence of a negotiated agreement, and hence crucially influence bargaining positions. Default rules are a key tool of behavioral legal interventions61 since, under certain conditions,62 people tend to stick to preconceived options (status quo bias).63

a) General idea. Personalized default rules can seek to vary two parameters: the content of default rules, and the altering rules determining how people can contract around the default. As Ian Ayres has suggested,64 both are equally important for the outcome of the bargaining process: the content of default rules determines the initial bargaining positions, and the altering rules specify how the parties can change these positions. With respect to the contents of default rules, Porat and Strahilevitz have suggested that a number of rules could be personalized according to certain individual traits or personality patterns.65 For example, they argued that inheritance rules in case of intestacy should be different for women and men as preferences significantly differ between the sexes for the division of the estate amongst, e.g., their partner and children.66 Furthermore, they suggested that, in consumer contracts, the default place of delivery could vary between consumers with and without disabilities: it could be at the buyer’s residence in the former and at the seller’s place of business in the latter case.67 These examples are thought-provoking suggestions for personalizing default rules; however, they do not employ behavioral metrics in the sense of behavioral economics. 32 More generally, personalized law, in private law contexts, lends itself particularly to consumer or employment contracts where one party is an individual and the other party is a professional or a corporation. In these cases, legal provisions can be tailored to the traits of the consumer or employee only. If cognitive traits of both parties have to be taken into account, personalization becomes much more complicated – a regime of “conflict of personalized laws” would have to be developed that determines what happens if the personality traits of the two parties suggest incompatible rules. This is why consumer or employment law are ideal candidates for personalization. However, under EU law, many of the most important rules of consumer and employment law are mandatory, and not default rules as in the US. However, a number of default rules exist even in EU consumer law. This chapter focuses on the personalization of their altering rules. Additional procedural safeguards could be established for those less able to fend for themselves. This would make the default rule stickier for those individuals who are deemed to benefit, rather than be harmed, by the default setting. Conversely, contracting around default rules could be made selectively easier for more rational parties.

31

See Hacker, supra (fn. 1), at § 11 C.I. See Willis, When Nudges Fail: Slippery Defaults, (2013) 80 University of Chicago Law Review 1155. 63 Samuelson/Zeckhauser, Status quo bias in decision making, (1988), 1 Journal of Risk and Uncertainty 7. 64 Ayres, Regulating Opt-Out: An Economic Theory of Altering Rules, (2012) 121 Yale Law Journal 2032. 65 Porat/Strahilevitz, supra Part 1.A. 66 ibid. 67 ibid. 61 62

250

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:16 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

b) Concrete examples. As Lauren Willis has meticulously documented in a study on defaults that were supposed to reduce fees for consumers resulting from account overdraft, default rules are unlikely to effectively steer the behavior of vulnerable parties if counterparties have better access to them than regulators, vulnerable parties do not have strong preferences, the decision environment is complex, and counterparties have an incentive to persuade the vulnerable party to deviate from the default.68 In the US, when overdraft protection was enacted by default rules that sought to ensure that consumers only opt into overdraft schemes if they are beneficial for them, banks were able to push the great majority of those consumers that generated most overdraft fees successfully back into overdraft eligibility.69 In many cases, they lobbied them until they signed a simple waiver that reinstituted overdrafting possibilities and fees.70 Empirical studies show, however, that consumers with bounded willpower are particularly likely to miss payment deadlines71 and to, generally, act more impulsively.72 This suggests that these consumers are particularly likely to accumulate overdraft fees and high borrowing costs associated with account overdrafting. Therefore, personalized altering rules could mandate that, in order to contract around a no-overdraft default, consumers scoring high on metrics of bounded willpower would need to undergo additional procedures.73 For example, one could institute a risk awareness test these consumers have to pass in order to qualify for overdraft; the test would ensure that consumers familiarize themselves with the fees and interest rates applicable to overdraft, and with alternatives for short-term borrowing. Alternatively, one could compel these consumers to conduct a counselling session with a consumer or borrower protection agency before opting into overdraft. The German legislator has recently enacted provisions that point into a similar direction.74 Under the new § 504a of the German Civil Code (BGB), lenders have to offer consumer borrowers a counselling session if borrowers use an overdraft facility continuously for six months at a level that exceeds 75 % of the maximum overdraft amount. Borrowers, however, may decline such counselling; also, the obligation is only triggered after fees, and interest rates, have already piled up over half a year (and, potentially, much longer if the overdraft amount was below the 75 % threshold). Personalization, by contrast, would use a behavioral metric in order to identify ex ante consumers that are likely to incur significant costs of overdrafting, instead of singling out consumers ex post, after debt and costs have accumulated. Similar procedures could be applied to other risky types of borrowing for consumers with bounded willpower, for example for payday loans, variable interest loans without interest rate caps etc. This would enable these consumers to access short-term, relatively high-interest credit, but would simultaneously mitigate the risk of these consumers being pushed into these schemes by lenders hoping to profit from their mistakes. 68

Willis, supra (fn. 62), at 1200–1211. ibid., 1185–1200. ibid. 71 Meier/Sprenger, Time discounting predicts creditworthiness, (2012) 23 Psychological Science 56; see also Chabris et al., Individual laboratory-measured discount rates predict field behavior, (2008) 37 Journal of Risk and Uncertainty 237, 256 on payment of credit card bills. 72 Reimers et al., Associations between a one-shot delay discounting measure and age, income, education and real-world impulsive behavior, (2009) 47 Personality and Individual Differences 973; de Wit et al., IQ and nonplanning impulsivity are independently associated with delay discounting in middleaged adults, (2007) 42 Personality and Individual Differences 111. 73 See also Hacker, supra (fn. 43), at 670–672. 74 I am grateful to Christoph Busch for bringing this example to my attention. 69 70

Hacker

251

33

34

35

36

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

3. Mandatory law 37

One final area of law that may benefit from behaviorally informed personalization is mandatory law. By this, I understand law that substantively prescribes or outlaws certain types of behavior, or the substance of contracts, rather than compelling disclosure.

38

a) General idea. Much of consumer law is mandatory in the EU; therefore, a broader approach to personalization must encompass such binding rules. Again, the motivation would be to specifically tailor protective regimes to the demonstrated vulnerabilities of addressees. For example, in related work I have suggested to personalize usury thresholds to reflect differential repayment probabilities that could be deduced from different degrees of willpower.75

39

b) Concrete examples. Perhaps the most direct way in which vulnerable consumers may benefit from personalization is to tailor protections against exploitative contracting practices. There is a growing economic literature that shows that the informational asymmetries brought about by advanced pattern recognition in machine learning facilitate exploitative contracts (also called adverse targeting).76 These are contracts in which one party benefits from mistakes of the other party;77 the offer of teaser rate contracts to consumers with bounded willpower and bounded ability to forecast their future spending and consumption behavior is a case in point. Such practices are not merely a theoretical concern; they have been demonstrated in a number of field experiments.78 Recently, internal documents furthermore suggested that Facebook enabled advertisers to strategically target psychologically instable adolescents.79 There need not be intentional malice behind exploitative contracting: to the extent that the content of contractual offers is automated with the use of machine learning,80 the goal of maximizing the profit of the user of machine learning may lead algorithms to select offers that exploit mistakes simply as a result of their adaptive learning capabilities. Under the EU framework, material standards of fairness are safeguarded, inter alia, by the provisions of the Unfair Commercial Practices Directive (UCPD); its Recital 18 stresses, in its second sentence, that the provisions aim to prevent the exploitation of vulnerable consumers. The following 75

Hacker, supra (fn. 3), at 651, 673 et seq. Heidhues/Kőszegi, Naivete-based discrimination, (2017) 132 The Quarterly Journal of Economics 1019; Kőszegi, Behavioral Contract Theory, (2014) 52 Journal of Economic Literature 1075, 1104–1110; Grubb, Overconfident Consumers in the Marketplace, (2015) 29 Journal of Economic Perspectives 9; Heidhues/Kőszegi, Exploitative Innovation, (2016) 8 American Economic Journal: Microeconomics 1; see also, more generally, Royal Society, Machine Learning, London 2017, 20 and 115; Bar-Gill, supra (fn. 54), at 14–18; Akerlof/Shiller, Phishing for Phools, Princeton University Press 2015. 77 Mik, The Erosion of Autonomy in Online Consumer Transactions, (2016) 8 Law, Innovation and Technology 1, 32–36; Hacker, Personal data, exploitative contracts, and algorithmic fairness: autonomous vehicles meet the internet of things, (2017) 7 Int’l Data Privacy L. 266, 266–267. 78 See, e.g., Hawkins, Using Advertisements to Diagnose Behavioral Market Failure in Payday Lending Markets, (2016) 51 Wake Forest Law Review 57; Shui/Ausubel, Time Inconsistency in the Credit Card Market, Working Paper (2005), http://web.natur.cuni.cz/~houdek3/papers/economics_psychology/Shui% 20Ausubel%202006.pdf; DellaVigna/Malmendier, Paying Not to Go to the Gym, (2006) 96 American Economic Review 694; and references in fn. 76. 79 Whigham, Leaked document reveals Facebook conducted research to target emotionally vulnerable and insecure youth, news.com.au (1 May 2017), http://www.news.com.au/technology/online/social/ leaked-document-reveals-facebook-conducted-research-to-target-emotionally-vulnerable-and-insecureyouth/news-story/d256f850be6b1c8a21aec6e32dae16fd. 80 See Grundmann/Hacker, Digital Technology as a Challenge to European Contract Law, (2017) 13 European Review of Contract Law 255, 277 et seq. 76

252

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

section therefore investigates to what extent these protections need to and could be personalized to counter exploitative contracting. aa) Personalizing unfair commercial practices under the UCPD. The UCPD 40 contains a number of provisions that prohibit different types of unfair commercial practices. Apart from the black list in Annex I, it contains a general clause in Art. 5(2) UCPD, provisions on misleading actions (Art. 6 UCPD) and omissions (Art. 7 UCPD) as well as aggressive commercial practices (Art. 8 UCPD). As the following analysis shows, all of these provisions stand to benefit from greater personalization. (i) Art. 5(2), 6(1) and 8(1) UCPD: Material distortion, misleading and aggressive 41 practices. Art. 5(2) UCPD specifies that commercial practices are unfair if two criteria are cumulatively fulfilled: they need to breach professional diligence; and materially distort economically relevant consumer decisions. Professional diligence is defined in Art. 2(h) UCPD as acts that, inter alia, abide by “honest market practice and/or the general principle of good faith in the trader’s field of activity”. This prong therefore necessitates a normative, not an empirical analysis.81 Arguably, such standards should be considered violated when traders seek to leverage unilateral access to data processing by benefiting from expected mistakes that become apparent in the data patterns;82 this should particularly hold if the data suggests that the advertised offer conveys negative expected value to the consumer. The second criterion, however, is more restrictive; the practice is only considered unfair 42 if it “is likely to materially distort the economic behavior with regard to the product of the average consumer whom it reaches or to whom it is addressed, or of the average member of the group when a commercial practice is directed to a particular group of consumers”. Art. 2(e) UCPD defines material distortion as a commercial practice that “appreciably impair[s] the consumer’s ability to make an informed decision, thereby causing the consumer to take a transactional decision that he would not have taken otherwise”. The wording of the definition is very broad and hence in need of restrictive criteria.83 The focus on informational autonomy84 suggests that material distortion could be understood as any change in the decision environment that appreciably and negatively impacts the consumer’s ability to make a rational decision without giving the consumer an adequate possibility to reflect on the influence.85 This, in turn, implies that practices seeking to benefit from cognitive or volitional vulnerabilities qualify as material distortion precisely because they benefit from mistakes that are, among members of this group, usually not noticed and corrected.86 81 Sibony, Can EU Consumer Law Benefit from Behavioural Insights? An Analysis of the Unfair Practices Directive, (2014) 6 European Review of Private Law 901, 909, 920; see also Willett, Fairness and Consumer Decision Making under the Unfair Commercial Practices Directive, (2010) 33 Journal of Consumer Policy 247, 268. 82 Hacker, supra (fn. 77), at 266, 279 et seq. 83 Tor, Some Challenges Facing a Behaviorally-Informed Approach to the Directive on Unfair Commercial Practices, in: Tóth (ed.), Unfair Commercial Practices: The Long Road to Harmonized Law Enforcement, Pázmány Press 2013, 9, 15. 84 Micklitz, The General Clause on Unfair Practices, in: Howells/Micklitz/Wilhelmsson (eds.), European Fair Trading Law, Ashgate 2006, 83, 102, 104 et seq. 85 Hacker, Nudging and Autonomy. A Philosophical and Legal Appraisal, in: Micklitz/Esposito/Sibony (eds.), Handbook of Consumer Research Methods, Edward Elgar 2018, 77: for the connection between the informed and a rational decision, see Köhler, in: Köhler/Bornkamm (eds.), UWG, 38th edn. 2020, § 3 para 3.25–3.26; and, more generally, Hacker, Verhaltensökonomik und Normativität, Mohr Siebeck 2017, § 9. 86 Cf also Tor, Some Challenges Facing a Behaviorally-Informed Approach to the Directive on Unfair Commercial Practices, in: Tóth, supra (fn. 83), at 9, 15.

Hacker

253

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

Moreover, exploitative contracting could fall under the prohibition of aggressive commercial practices in Art. 5(4)(b), 8(1) UCPD,87 which takes precedence over the general clause of Art. 5(2) UCPD.88 Under Art. 8(1) UCPD, commercial practices are considered aggressive if, by the use of harassment, coercion or undue influence, they are “likely to significantly impair the average consumer’s freedom of choice or conduct with regard to the product and thereby causes him or is likely to cause him to take a transactional decision that he would not have taken otherwise”. Pursuant to Art. 9 UCPD, the determination of harassment, coercion and undue influence must take account of its timing, location, nature and persistence (lit. a), as well as any “exploitation by the trader of any specific misfortune or circumstance of such gravity as to impair the consumer’s judgement” (lit. c). As Natalie Helberger has convincingly argued, targeted advertising may constitute harassment if it reaches the consumer in her private sphere, for example through personal digital assistants.89 Undue influence, however, is even more likely to be exerted in the exploitation of cognitive biases. Art. 2(j) UCPD defines undue influence as “exploiting a position of power in relation to the consumer so as to apply pressure, even without using or threatening to use physical force, in a way which significantly limits the consumer’s ability to make an informed decision”. Clearly, this is intended to capture illegitimate forms of psychological persuasion techniques.90 In the case of exploitative contracting, the position of power derives from the informational advantage that algorithmic data processing confers on the trader.91 While it is not easy to draw the line between due and undue influence, the specific targeting of consumers with weak cognitive capabilities precisely uses psychological insights to an extent that makes a rational decision of the consumer unlikely, particularly if it is timed or presented in circumstances so as to maximize the likelihood that the consumer will accept the offer without significant reflection.92 It remains debatable, however, if this amounts to 'pressure' in the sense of Art. 2(j) UCPD. At any rate, to exonerate the consumer from the burden of proof,93 a finding of undue algorithmic influence should also lead to a presumption that the undue influence was likely to cause a transaction the consumer would not have otherwise made.94 44 Finally, exploitative contracting may constitute a misleading action under Art. 7(1) UCPD if the targeting “is likely to deceive the average consumer, even if the information is factually correct” in relation to a number of elements key to an informed decision, such as the benefits and risks of the product (lit. b) or its price (lit. d). As research in behavioral economics has shown, boundedly rational consumers are prone to miscalculate risks for a variety of reasons,95 a fact even acknowledged by the Commission 43

87 Cf European Commission, Staff Working Document on Guidance on the Implementation/Application of Directive 2005/29/EC on Unfair Commercial Practices”, SWD (2016)163 final, 135. 88 Cf CJEU, judgment of 19 September 2013 in CHS Tour Services GmbH, C‐435/11, EU:C:2013:574. 89 Helberger, Profiling and Targeting Consumers in the Internet of Things, in: Schulze/Staudenmayer (eds.), Digital Revolution: Challenges for Contract Law in Practice, Hart/Nomos 2016, 135, 156. 90 Howells, Aggressive Commercial Practices, in: Howells/Micklitz/Wilhelmsson (eds.), supra (fn. 84), at 167, 188. 91 See also, arguing that a position of power may derive from intellectual domination, ibid.; Köhler/ Lettl, Das geltende europäische Lauterkeitsrecht, der Vorschlag für eine EG–Richtlinie über unlautere Geschäftpraktiken und die UWG–Reform, WRP 2003, 1019, 1046; see further Willett, supra (fn. 81), at 247, 260 (“greater market understanding”); Köhler/Lettl, Das geltende europäische Lauterkeitsrecht, der Vorschlag für eine EG–Richtlinie über unlautere Geschäftpraktiken und die UWG–Reform, WRP 2003, 1019, 1046. 92 Cf also Helberger, supra (fn. 89), at 135, 157. 93 Ibid., at 135, 159. 94 Cf also Micklitz, supra (fn. 84), at 83, 110; Sibony, supra (fn. 81), at 901, 939 et seq.; Hacker, supra (fn. 1), at 673 et seq. 95 Weinstein, supra (fn. 11); Tversky/Kahneman, Rational choice and the framing of decisions, (1986) 59 Journal of Business S251; Hacker, supra (fn. 1), at § 4.

254

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

guidance on the UCPD;96 and teaser rate contracts, for example, may be designed so as to induce misperceptions about the long-term price of a product.97 For misleading omissions, timing is again acknowledged as a key factor for the analysis, Art. 7(2) UCPD. Hence, targeting consumers by providing information likely to induce biases that lead to miscalculations of key product characteristics or its price should be construed as deception;98 again, the presumption should be that the consumer was likely not to make the transaction without it. The problem with all of these different prohibitions, as has often been noted,99 45 consists in the reference consumer understood to be “reasonably well-informed and reasonably observant and circumspect”.100 In contrast to a boundedly rational consumer, such an average consumer, arguably, would not be likely to make the kind of mistakes exploitative contracting targets.101 The average consumer is the reference for the determination of misleading actions or omissions (Art. 6(1) and 7(1) UCPD, respectively) and of aggressive commercial practices (Art. 8(1) UCPD), as well as for the material distortion of Art. 5(2) UCPD.102 For all of the considered prohibitions,103 this yardstick changes in two situations: first, 46 still according to Art. 5(2) UCPD, when contracting practices are directed toward a particular group, the reference individual is an average member of that group. This group must be distinguishable by certain general characteristics or traits; arguably, specific biases or thresholds of bounded willpower could constitute such traits. However, the practice must be intentionally directed at a certain group.104 The key problem, in our context, will be to prove this intention, particularly if one does not have access to the data and algorithms of the company. Second, Art. 5(3) UCPD renders this yardstick even more granular for practices that exclusively negatively affect (without necessarily being directed toward) specific vulnerable groups. It reads: “Commercial practices which are likely to materially distort the economic behavior only of a clearly identifiable group of consumers who are particularly vulnerable to the practice or the underlying product because of their mental or physical infirmity, age or credulity in a way which the trader could reasonably be expected to foresee, shall be assessed from the perspective of the average member of that group.” Children are a key example of such a vulnerable class.105 Art. 5(3) UCPD therefore is a clear attempt at greater legal granularity. However, for all its good intentions, Art. 5(3) UCPD is subject to four significant limitations, that partially also affect the particular group standard of Art. 5(2) UCPD. (ii) Limitations of the current framework. First, while the wording of Art. 5(3) 47 UCPD is very open (mental infirmity; credulity), potentially encompassing any kind of 96 European Commission, Staff Working Document, Guidance on the Implementation/Application of Directive 2005/29/EE on Unfair Commercial Practices, SEC(2009) 1666 final, 31; see also European Commission, supra (fn. 87) 40, 52 et seq. 97 Bar-Gill, supra (fn. 54), at 92–95; 113–115. 98 Cf Sibony, supra (fn. 81), at 901, 922–926. 99 See, e.g., Sibony, supra (fn. 81), at 901, 903, 909; Micklitz, supra (fn. 84), at 83, 111; Tor, supra (fn. 83), at 9, 16 et seq.; Helberger, supra (fn. 89), at 135, 159 et seq. 100 See, e.g., 18th Recital of the UCPD; European Commission (n. 96) 24. 101 Hacker, supra (fn. 23), at 299, 311 et seq.; see also Willett, supra (fn. 81), at 247, 269 et seq.; but see now also European Commission, supra (fn. 87), at 40 et seq. 102 Willett, supra (fn. 81), at 247, 268. 103 This is not directly implied by the wording of Art. 7 and 8 UCPD; see, however, in a similar vein Wilhelmsson, Misleading Practices, in: Howells/Micklitz/Wilhelmsson (eds.), European Fair Trading Law, Ashgate 2006, 123, 133 et seq.; Alexander, Grundfragen des neuen § 3 UWG, WRP 2016, 411, 418 et seq. To the very least, the specifications become relevant via the general clause of Art. 5(2). 104 Alexander, supra (fn. 103), at 411, 415 para 34. 105 Durovic, European Law on Unfair Commercial Practices and Contract Law, Hart 2016, 43.

Hacker

255

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:17 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

behavior that significantly deviates from economic rationality,106 scholars are divided whether it covers only exceptional cases107 or should be construed broadly, as a potentially pervasive category.108 In the absence of guidance from the CJEU, one could restrict mental infirmity, for example, to serious medical conditions, such as blindness, deafness, trisomy 21, or dementia.109 In the alternative, one could include less serious deviations from economic rationality, such as the propensity to gamble110 or cognitive biases. Similarly, credulity may refer to a limitation of judgment that renders rational decision making impossible,111 likely because of an uncritical attitude and a propensity to affectionate decision making.112 Here again, cognitive and volitional imperfections that are adversely targeted can be construed to qualify as credulity, but only under an expansive notion. While the Commission suggests that Art. 5(3) covers “a wide range of situations”,113 and even mentions “behavioral characteristics”,114 the CJEU tends to interpret exceptions restrictively. Therefore, it remains highly dubitable whether cognitive biases fall under these categories. 48 Second, and potentially even more importantly: the practice must exclusively affect the vulnerable group.115 Hence, if some members of a non-vulnerable group are also affected (more precisely: if their economic behavior is also materially distorted), the vulnerable group is not the reference anymore.116 This is an important limitation in the practice of modern algorithmic targeting. Models operate on a probabilistic basis;117 but statistical correlations are never perfect. Therefore, targeting in practice is never restricted exclusively to the vulnerable group; there are always “mismatches”, members of other groups which were incorrectly included in the targeted group. Most likely, the judgments of some of these mismatches will also be distorted, meaning that the targeting does not exclusively affect the vulnerable group. 49 However, it should be sufficient that only a vulnerable group is meant to be targeted, even if the probabilistic framework of the model leads to some non-vulnerable persons being inadvertently addressed, and potentially negatively affected, too. This is reinforced by Recital 19 UCPD, which itself adopts a probabilistic perspective by requiring that the “economic behavior only of such consumers is likely to be distorted” [emphasis added]. Similarly, the Commission guidance document stresses that the provision aims to cover even practices “which reach the majority of consumers, but in reality are devised to exploit the weaknesses of certain specific consumer groups”.118 50 Third, the trader must be reasonably expected to foresee that the affected group is particularly vulnerable to the specific offer. This criterion is problematic when models 106

Durovic, supra (fn. 105), at 45. Micklitz, supra (fn. 84), at 83, 112 et seq.; Friant-Perrot, The Vulnerable Consumer in the UCPD and Other Provisions of EU Law, in: van Boom et al. (eds.), The European Unfair Commercial Practices Directive, Ashgate, 2014, 89, 100; Durovic, supra (fn. 105), at 191. 108 Köhler, supra (fn. 85), at § 3 para 5.18. 109 In this vein European Commission, supra (fn. 87), at 43 et seq.; Sosnitza, in: Ohly/Sosnitza (eds.), Gesetz gegen den unlauteren Wettbewerb, 7th edn. 2016, § 4a para 161; Micklitz, supra (fn. 84), at 83, 114. 110 Köhler, supra (fn. 85), at § 3 para 5.20. 111 Sosnitza, supra (fn. 109), at para 177. 112 Cf also European Commission, supra (fn. 96), at 29; Köhler, supra (fn. 85), at § 3 para 5.25. 113 European Commission, supra (fn. 96), at 28. 114 European Commission, supra (fn. 87), at 43; see also European Commission, Consumer vulnerability across key markets in the European Union, Final Report (2016) 196 et seq. 115 See, e.g., Bundesgerichtshof (BGH), Judgment of 12 December 2013, I ZR 192/12, Goldbärenbarren, para 16. 116 Durovic, supra (fn. 105), at 44. 117 See, e.g., Witten et al., Data Mining. Practical Machine Learning Tools and Techniques, Morgan Kaufmann 2016, 96 et seq. 118 European Commission, supra (fn. 96), at 30. 107

256

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

are used, in machine learning, which deliver results that are ex ante unforeseeable even for the developer of the model.119 However, such inscrutability should eventually burden the trader, not the consumer. The foreseeability requirement introduces a proportionality element into the analysis to protect traders against excessive monitoring and adaption duties.120 If, however, traders use algorithmic models for their own benefit, it is quite appropriate to demand that they understand the model to an extent that they can exclude exploitative behavior. If they choose to implement a model that does not allow for this, they do this at their own risk, as they also reap the rewards of targeting by machine learning. Foreseeability, then, must be understood as a general capacity to understand that the model could target vulnerable consumers in the way it does, even if the specific traits the model extrapolates and bases its decision on were not ex ante foreseeable, or ex post retrievable. It is the very choice of a self-learning model that, in the presence of documented cases of bias and exploitation, renders the specific effect sufficiently foreseeable in the abstract. Therefore, with some goodwill and an expansive understanding of the provisions – 51 but only with this –, the first three limitations can arguably be overcome. However, fourth, and most importantly, both under Art. 5(2) and Art. 5(3) UCPD, some group must exist from which the reference standard is derived. This likely implies that targeting at the level of individual persons (first-degree discrimination) is not caught. While it is logically possible to define a set with only one element, and to identify that member as its average member, the use of the word “group” clearly suggests a multitude of persons. The key example the Commission mentions for a specific group in the sense of Art. 5(2) UCPD, but also for a vulnerable group under Art. 5(3) UCPD, is the group of teenagers121 – still a very large and heterogenous pool. More importantly, the provisions embody the trade-off between consumer protection and the interests of the trader not to adapt its behavior to every single consumer.122 Therefore, as targeting becomes ever more granular, it becomes ever less likely to fall under the group standard of the cited provisions.123 This is precisely where a personalized regime would make a difference. bb) The benefits of personalization. Personalization, in this context, should take the 52 form of company-based personalization:124 if a trader has access to certain data and uses it to its benefit to make more granular distinctions in its advertisement or offer campaigns, it should equally be bound by more demanding legal standards under Art. 5–8 UCPD. More specifically, the reference group for the behavioral standard should be chosen at exactly the level at which the company targets consumers with the practice in question, even if this includes analysis of material distortion on the individual 119 See, e.g., Lou/Caruana/Gehrke, Intelligible models for classification and regression, (2012) Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining 150; Bornstein, Is Artificial Intelligence Permanently Inscrutable?, Nautilus (1 September 2016), http://nautil. us/issue/40/Learning/is-artificial-intelligence-permanently-inscrutable; this is different only for interpretable machine learning, see Lipton, The mythos of model interpretability, arXiv preprint arXiv:1606.03490 (2017); Vellido/Martín-Guerrero/Lisboa, Making machine learning models interpretable, (2012) 12 ESANN 163 (2012); Rudin, Algorithms for Interpretable Machine Learning, (2014) 20th ACM SIGKDD Conf. on Knowledge Discovery & Data Mining 1519. 120 European Commission, supra (fn. 96), at 30; European Commission, supra (fn. 87), at 46; Micklitz, supra (fn. 84), at 83, 115 et seq. 121 European Commission, supra (fn. 96), at 27–29; European Commission, supra (fn. 87), at 45. 122 European Commission, supra (fn. 96), at 24; Köhler, supra (fn. 85), at § 3 para 5.2. 123 Cf also Ebers, Beeinflussung und Manipulation von Kunden durch Behavioral Microtargeting, (2018) MultiMedia und Recht 423, 425. 124 See, on the distinction between company- and government-based personalization Hacker, supra (fn. 3), at 651, 655 et seq.

Hacker

257

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

level. Thus, if the algorithmic model a trader uses, intentionally or unintentionally, singles out an input combination of traits that may even be unique to a certain individual, but that leads to an output which puts that individual at a material disadvantage in the decision-making situation and enables exploitative contracting, this should be deemed unfair under Art. 5 UCPD.125 The reason for sparing the trader the burden of certifying fairness at the individual level does not apply in these situations anymore. It was based on the assumption that it would be too onerous for the trader to gather and evaluate information at the individual level, the reference being large advertisement campaigns in the pre-personalized era. However, if analysis at the individual level is precisely what targeting increasingly engages in, the legal standard should adapt to the increasing granularity of commercial practices, too. 53 In practice, this would mean that certain types of contracts that generate negative expected value under a certain combination of traits (for example, high optimism bias; or very bounded willpower) may not be offered to individuals for which the data suggests this precise combination of traits. The trader must, as part of his professional diligence,126 implement features that reasonably prevent exploitation even at an individual level if, and only if, the trader also targets consumers with distinctions on the individual level. As mentioned, if the model is inscrutable as to the decision weights and traits it uses, this should not exonerate the trader if it can be shown, ex post, that a certain targeted individual did possess the required combination of traits. Of course, this may necessitate types of algorithmic auditing, in order to assess whether targeted individuals did manifest certain traits; we shall return to this question in Part V of the chapter.

IV. The limits of personalized behavioral law 54

It is important to note that there are quite obviously not only benefits, but also significant costs and risks associated with personalization. One of the key challenges for personalized law is safeguarding the privacy of the affected individuals; I have dealt at length with these issues in related publications.127 Suffice to say that concerns over privacy and data abuse are obviously a key risk of government-based personalization, i.e., when government bodies themselves conduct the profiling. The next section of the chapter, focusing on good governance aspects, will take up the issue of privacy in detail. In this section, I would like to touch on two different limits of personalization: the strength of empirical correlations that may be too weak to support personalization; and algorithmic bias that may lead to discriminatory content of personalized law.

1. The strength of empirical correlations 55

Personalized law will only present an advance if the addressees of regulation are correctly categorized, so that the personalized legal provision tightly matches behavioral reality. To the extent that personalized law is based on probabilistic Big Data techni125 Under the additional criteria outlined above, it could even constitute a misleading or even aggressive commercial practice. 126 On the link between Art. 5(3) UCPD and professional diligence, see Alexander, supra (fn. 103), at 411, 419 para 80. 127 Hacker, supra (fn. 3), at 651, 664 et seq.; id., The Ambivalence of Algorithms. Gauging the Legitimacy of Personalized Law, in: Bakhoum et al. (eds.) Personal Data in Competition, Consumer Protection and IP Law – Towards a Holistic Approach?, Springer, 2018, under 4.1.

258

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

ques, matching will never be perfect, of course. Someone classified as particularly optimistically biased might, in reality, be perfectly calibrated. However, one must demand that personalized law should provide, at least, a closer match to reality than non-personalized law. More specifically, the misclassification rate (defined, for example as the product of the extent with the number of instances of a prediction error)128 should be smaller for personalized than for traditional law. Of course, it will be difficult to calculate the misclassification rate in reality; therefore, 56 the models used to personalize law should be thoroughly scrutinized for their internal and external validity.129 More precisely, for government-based personalization one should at a minimum demand that the empirical correlations (between data patterns and behavioral traits on which legally relevant classifications are based) surpass a certain threshold. Depending on the intrusiveness of the provision, this threshold may vary from norm to norm. It should be larger for default rules than for disclosures, and larger still for mandatory rules. For example, one could demand that the sample Pearson’s correlation coefficient r, a frequently used measure in empirical studies,130 be greater than 0.7 for disclosures, greater than 0.8 for default rules, and greater than 0.9 for mandatory law. This would mean that variation in the data can explain 49 %, 64 %, and 81 % of the behavioral variation, respectively. However, current empirical studies using Big Data techniques to infer behavioral 57 attributes often fall short of such thresholds. For example, the above-mentioned seminal study in which personality traits were derived from Facebook likes only attained an r of 0.56 for an average individual with 227 likes, and a maximum of r = 0.66 only in the top group of persons with more than 500 likes.131 In a preceding study on Facebook likes, the r values were (with the exception of age: r = 0.75) even lower.132 It can be expected that technological progress, for example more powerful machine learning methods, will lead to a steady rise of r over time. However, recent findings also suggest that users can easily “cloak” a minimal amount of highly predictive data points to prevent meaningful inference; for example, removing “Lady Gaga” from one’s Facebook likes significantly reduces the probability of being identified as gay.133 The effect of such obfuscation strongly depends on the scoring technique used;134 nevertheless, it shows that datadriven scoring is far from perfect, and that scores are subject to manipulation and strategic behavior by those scored. Therefore, utmost care must be dedicated to the precise calculation of the correlation strength in the models used for personalized law.

2. Algorithmic bias and discrimination A second phenomenon is increasingly documented in theoretical and empirical 58 studies that demonstrates the potential failure of algorithm-driven decision making: 128 For non-personalized law, the misclassification rate must be inferred from an implicit behavioral benchmark inherent in a provision, and the extent and number of instances addressees deviate from it; see also Hacker, supra (fn. 3), at 651, 676. 129 For the concepts of internal and external validity, see Lawless et al., Empirical Methods in Law, Wolters Kluwer 2010, 36 et seq. 130 Rodgers/Nicewander, Thirteen ways to look at the correlation coefficient, (1988) 42 The American Statistician 59, 61; cf also Kritzer, The (Nearly) Forgotten Early Empirical Legal Research, in: Cane/ Kritzer, The Oxford Handbook of Empirical Legal Research, OUP 2010, 875, 884; Lawless et al., supra (fn. 129), at 254. 131 Youyou/Kosinski/Stillwell, supra (fn. 41), at 1036, 1037. 132 Kosinski/Stillwell/Graepel, supra (fn. 40), at 5802, 5804. 133 Chen et al., Enhancing Transparency and Control When Drawing Data-Driven Inferences About Individuals, (2017) 5 Big Data 197, 201–204. 134 Ibid., at 197, 206–209.

Hacker

259

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

discrimination by algorithms.135 The application of machine learning techniques can perpetuate and even exacerbate human bias. This is due to a number of reasons.136 First, and perhaps most often, discriminatory algorithmic processes stem from imperfect training data for machine learning algorithms (so-called sample bias).137 For instance, the face recognition algorithm used by Google inadvertently labelled black people as gorillas as a result of training data that comprised too few black persons.138 The training data can also be naturally distorted as a result of discriminatory social conditions, as a recent study on bias in automated translation techniques has shown for the semantics of natural language.139 Second, in supervised learning, data has to be labelled by human coders to feed and train the algorithm. Here again, incorrect labeling may lead to bias.140 This was documented in job selection algorithms, for example, when previously successful candidates were treated as examples for positive future job performance, but these candidates were overwhelmingly white and male due to the social circumstances at the time of their hiring.141 Third, unequal base rates between different groups necessarily lead to the violation of fairness norms in algorithmic processes. Most prominently, this has been shown in the workings of the COMPAS algorithm used to predict the recidivism rate in criminal justice proceedings in a number of US states; this algorithm discriminates against black persons according to one important metric.142 59 The use of machine learning techniques for lawmaking purposes therefore not only holds the potential of more specifically tailoring legal protection to particularly vulnerable individuals; it also threatens to make marginalized groups and individuals even 135 Calders/Žliobaitė, Why Unbiased Computational Processes Can Lead to Discriminative Decision Procedures’, in: Custers et al. eds., Discrimination and Privacy in the Information Society, Springer 2013, 43; Romei/Ruggieri, A multidisciplinary survey on discrimination analysis, (2014) 29 Knowledge Engineering Rev. 582; Kamiran/Žliobaitė/Calders, Quantifying explainable discrimination and removing illegal discrimination in automated decision making, (2013) 35 Knowledge & Information Systems 613; Žliobaitė, A Survey on Measuring Indirect Discrimination in Machine Learning, Association of Computer Machinery, (2015), https://arxiv.org/abs/1511.00148; Lepri et al., The Tyranny of Data? The Bright and Dark Sides of Data-Driven Decision-Making for Social Good, in: Cerquitelli/Quercia/Pasquale (eds.), Transparent Data Mining for Big and Small Data, Springer 2017, 3; Žliobaitė, Measuring discrimination in algorithmic decision making, (2017) 31 Data Mining & Knowledge Discovery 1. 136 See Hacker, Teaching Fairness to Artificial Intelligence: Existing and Novel Strategies Against Algorithmic Discrimination Under EU Law, (2018) 55 Common Market Law Review 1143, 1146–1150; Barocas/Selbst, Big Data’s Disparate Impact, (2016) 104 Cal. L. Rev. 671, 677–693. 137 Calders/Žliobaitė, supra (fn. 135), at 51. 138 Barr, Google Mistakenly Tags Black People as ‘Gorillas,’ Showing Limits of Algorithms, Wall Street Journal (1 July 2015), https://blogs.wsj.com/digits/2015/07/01/google-mistakenly-tags-black-people-asgorillas-showing-limits-of-algorithms/. 139 Caliskan et al., Semantics derived automatically from language corpora contain human-like biases, (2017) 356 Science 183. 140 Calders/Žliobaitė, supra (fn. 135), at 50–51. 141 Lowry/Macpherson, A Blot on the Profession, (1988) 296 British Medical J. 657; see also Reuters, Amazon ditched AI recruiting tool that favored men for technical jobs, The Guardian (11 October 2018), https://www.theguardian.com/technology/2018/oct/10/amazon-hiring-ai-gender-bias-recruiting-engine. 142 Errors are very differently distributed between races in COMPAS: risk predictions for white defendants are often too low (false negative), while those for black defendants are often too high (false positive). See Larson et al., How We Analyzed the COMPAS Recidivism Algorithm, Pro Publica (23 May 2016), https://www.propublica.org/article/how-we-analyzed-the-compas-recidivism-algorithm. Mathematically, it is impossible to equalize both the error rate (so-called balance of positive and negative class) and predictive accuracy across the two groups if they have an unequal base rate of recidivism, see Kleinberg/ Mullainathan/Raghavan, Inherent trade-offs in the fair determination of risk scores, Working Paper (2016), https://arxiv.org/abs/1609.05807, at 5–6; Chouldechova, Fair prediction with disparate impact: A study of bias in recidivism prediction instruments, (2017) 5(2) Big Data, 153; see also, on the predictive parity of COMPAS, Dieterich/Mendoza/Brennan, COMPAS Risk Scales: Demonstrating Accuracy Equity and Predictive Parity, Technical Report, Northpointe (July 2016), http://www.northpointeinc.com/northpointe-analysis.

260

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

worse off if algorithmic bias slips into the decision-making process. An algorithmic hiring tool developed by the Austrian unemployment agency powerfully shows this: it discriminates against female and elder jobseekers.143 This presents a powerful challenge to varieties of personalized law that rely on machine learning to render greater legal granularity effective. This is not to say that machine learning cannot play any role in the future of lawmaking and adjudication. In a recent empirical study, an algorithm was successfully trained to make bail decisions in ways less prone to predictive error than the same decisions made by human judges.144 However, the literature on algorithmic bias shows that such results are far from self-evident. If this problem is not tackled, personalized law cannot be implemented in societies that adhere to principles of equal protection and non-discrimination.

V. Good governance of personalized behavioral law These challenges point to the crucial importance of installing good governance mechan- 60 isms in any variety of personalized (behavioral) law. The remainder of this article discusses two specific varieties: privacy respecting metrics, and algorithmic auditing techniques.

1. Privacy respecting metrics A number of features can be implemented that mitigate privacy concerns. First, 61 narrow, specific metrics should be used that have high predictive value for the desired traits, but that do not correlate well with other behavioral characteristics.145 In this way, the information content on each scored individual is minimized. Second, any scores that are communicated to third parties as a part of the implementation of personalized law need to be protected, to a maximum amount, against “reverse engineering”, i.e., against the possibility of deriving specific vulnerabilities from the scores. This can be done, for example, by “mixing” several types of vulnerabilities together into one single combined score. Third, similar protection needs to be afforded against hacks of the database on which personal data is stored. This necessitates, for example, advanced encryption techniques. Fourth, and perhaps most importantly, company-based personalization does not require state authorities to collect and process any personal data; rather, personalization is based on data collected by companies to the extent that these companies do lawfully process personal data. The personalization of the treatment of unfair commercial practices was a case in point: only to the extent that companies do engage in granular targeting should they be obliged to implement measures mitigating the exploitation of those groups or individuals they target. Company-based personalization, therefore, does not lead to greater privacy and data protection concerns than the status quo (which does not imply, of course, that stronger privacy and data protection norms should not be discussed to change the status quo, too).146 Finally, users could be 143 Dornis, Arbeit aus dem Automaten, Süddeutsche Zeitung (22 October 2018), https://www.sueddeutsche.de/digital/digitalisierung-arbeitslosigkeit-jobcenter-1.4178635. 144 Kleinberg et al., Human Decisions and Machine Predictions, (2018) 133 The Quarterly Journal of Economics, 237; see also Berg, Forecasting Domestic Violence: A Machine Learning Approach to Help Inform Arraignment Decisions, (2016) 13 J. Emp. Leg. Stud. 94; for a classical study on human error in the courtroom, see, e.g., Danziger/Levav/Avnaim-Pesso, Extraneous factors in judicial decisions, (2011) 108 PNAS 6889. 145 Hacker, supra (fn. 3), at 651, 660 et seq., 665. 146 See, e.g., Hacker/Petkova, Reining in the Big Promise of Big Data. Transparency, Inequality, and New Regulatory Frontiers, 15 Northwestern Journal of Technology and Intellectual Property, (2017); Hacker, supra (fn. 77), at 266, 280 et seq.

Hacker

261

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

given a choice between opting into a personalized behavioral regime and staying with a non-personalized legal system.147 This could be deployed particularly in the case of government-based personalization so that those fearing excessive privacy risks may prevent personalization from applying to them.

2. Oversight and algorithmic auditing Countering privacy concerns is one key dimension of a successful implementation of personalized law. The second component of good governance mechanisms in this field certainly relates to robust oversight and auditing mechanisms.148 They should be designed so as to minimize (i) the risk of the abuse of data collected for personalization purposes, be it by government agencies or private companies, and (ii) the risk of discrimination. Data protection impact assessments are a first step toward the former goal.149 On the latter point, a proliferating literature in computer science formalizes algorithmic fairness measures that can be used to detect and correct for algorithmic bias.150 Such techniques could be mandated, and their implementation verified, in the course of algorithmic audits.151 In this way, core societal values, such as fairness and non-discrimination, may be implemented at the very code level.152 A regulator wishing to implement or oversee algorithmic fairness procedures faces, however, a non-trivial selection task as different fairness measures are being discussed in the computer science literature.153 They range from the reconfiguration of input data154 via the control of the algorithmic process155 to the transformation of the algorithmic output.156 Here, a sector specific approach is necessary in which measures fitting the precise regulatory environment (be it the personalization of disclosure or the use of scoring for criminal justice purposes) are broadly discussed and chosen. In many cases, this involves a trade-off between individual and group fairness that can, however, also be operationalized on a mathematical and technical level.157 63 Such procedures require frequent and thorough vetting of the very techniques that are used for algorithmic scoring. However, we should not hesitate to integrate such 62

147 Busch, The Future of Pre-Contractual Information Duties: From Behavioural Insights to Big Data, in: Twigg‐Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Edward Elgar 2016, 221, 237 et seq.; Hacker, supra (fn. 3), at 651, 663 n. 61. 148 Cf also Citron/Pasquale, The Scored Society: Due Process for Automated Predictions, (2014) 89 Washington Law Review 1, 20 et seq. 149 Art. 35 GDPR. 150 For an overview, see, e.g., Žliobaitė, Fairness-aware machine learning: a perspective, Working Paper (2 August 2017), https://arxiv.org/abs/1708.00754; Lepri et al., Fair, Transparent, and Accountable Algorithmic Decision-Making Processes, Phil. & Tech. 1 (2017). 151 Cf Tutt, An FDA for Algorithms, (2017) 69 Admin. L. Rev. 83; Sachverständigenrat für Verbraucherfragen, Verbraucherrecht 2.0 – Verbraucher in der digitalen Welt (Berlin: Sachverständigenrat für Verbraucherfragen beim Bundesministerium der Justiz und für Verbraucherschutz, 2016), 74–76. 152 For legal applications of algorithmic fairness, see Kroll et al., Accountable Algorithms, (2017) 165 U. Pa. L. Rev. 633, 685–690; Hacker (n 136) 1170–1183. 153 See Zehlike/Hacker/Wiedemann, Matching code and law: achieving algorithmic fairness with optimal transport, (2020) 34 Data Mining and Knowledge Discovery, 163, 184–190; Dunkelau/Leuschel, Fairness-Aware Machine Learning, Working Paper (2019); Friedler et al., On the (im)possibility of fairness, Working Paper (2016), arxiv.org/abs/1609.07236, at 12 et seq.; Berk et al., Fairness in Criminal Justice Risk Assessments: The State of the Art, (2018) Sociological Methods & Research, Article 0049124118782533, at 12–15. 154 See, e.g., Zemel et al., Learning Fair Representations, (2013) 28 Proceedings 30th International Conference on Machine Learning 325. 155 Dwork et al., Fairness Through Awareness, (2012) Proceedings 3rd Innovations in Theoretical Computer Sciences Conference 214; Friedler et al., supra (fn. 153). 156 Zehlike/Hacker/Wiedemann, supra (fn. 153). 157 Zehlike/Hacker/Wiedemann, supra (fn. 153).

262

Hacker

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/12_Hacker Personalization and Behavioural Sciences 181025.3d from 30.10.2020 11:37:18 3B2 9.1.580; Page size: 160.00mm  240.00mm

L. Personalized Law and the Behavioral Sciences

oversight mechanisms into both the digital economy and digital government. After all, in other core sectors of society – be it the safeguarding of tax revenues or of key competitive processes – audits and investigations by independent parties form part and parcel of the regulatory framework. As our societies increasingly shift into the algorithmic mode, algorithmic audits should become an established feature to safeguard the purpose limitation and fairness of the algorithmic processes that govern our lives.

VI. Conclusion This chapter makes three contributions at the intersection of personalized law and 64 the behavioral sciences. First, by discussing a number of examples ranging from disclosures via default rules to mandatory law, it shows how personalization can be used to tailor legal protections to those particularly vulnerable in cognitive and volitional ways. Thus, the knowledge problem in behavioral law and economics can be overcome. For example, disclosures can be designed so as to render aspects salient that are of particular importance for the disclosee; default rules can render legal protections stickier for those prone to contract out of them against their own interest; and companies can be forced to observe fairness norms, for example under the Unfair Commercial Practices Directive, at a more granular level, matching individualized targeting of consumers. Second, however, personalized law faces a number of important challenges. This is 65 quite evident for privacy concerns. This chapter furthermore discusses the necessary strength of empirical correlations (which is often lacking at the moment); and the risk of algorithmic bias against marginalized or digitally underrepresented groups. Discrimination by algorithms is increasingly documented in applications of machine learning and constitutes a limiting factor to the expansion of machine learning into lawmaking. Therefore, third, the chapter introduces a number of governance mechanisms to 66 mitigate these risks. For example, it advocates the use of privacy sensitive metrics and other measures to minimize privacy risks, such as company-based personalization. Additionally, it suggests strict processes of algorithmic auditing designed to ensure the validity of inferences made from data, and the freedom of algorithmic processes from discriminatory outcomes. In this endeavor, governance mechanisms may draw on the vast literature of algorithmic fairness currently developing in computer science. The intersection between behavioral economics, computer science and regulation therefore promises to be a particularly fruitful field for future research.

Hacker

263

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:32 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contract”, “Granular Norms” and Non-Discrimination* An elderly university professor used to dress with a grey suit, a white shirt, a Jermyn Street tie and classical Italian shoes, still shudders when he remembers those past ages when suddenly, like in a biblical curse, from nowhere appeared sharp-pointed shoes, tight jeans-like trousers and large, Al Capone style, ties. 2 Fortunately, those times have passed and less boisterous (and more elegant) fashion has returned and is easily available for those who still believe, two centuries and half later, that “le style c’est l’homme”. 3 Experience, therefore, suggests reacting with calm when equivalent monstrosities are dragged in the legal arena by fashionable authors, and apparently one is not à la page if one does not exalt the novelty brought by “granular norms” and “smart contracts”, and the rise of a new era in legal studies. 4 It is ironical that an anti-formalist and legal-realist author is forced to point out that the term “smart contracts” is, from a legal point of view, sheer nonsense and that the term “granular norms” requires a great deal of delving before one can properly define it; and that it would be desirable if those – the many – scholars who believe that the law has a sense could contribute in avoiding it becoming senseless.1 1

I. Only words The law does not exist in some physical form. The law is a linguistical convention by which a community gives to words and expressions a common meaning. 6 We cannot understand and perform a command which is expressed in a language which is unknown to us. In a trans-Atlantic classroom there will be continuous misunderstandings when the term “property” is used. 7 Sometimes – typically in traffic signs or on a football pitch – rules are expressed through signs or sounds, but again their semiotics must be shared by those who are driving or by the 22 players. 8 For this reason, it is essential that the community, both in a broad sense (those to whom the legal terms are addressed) and in a more restricted sense (legal scholars), agree on the meaning of the legal terms. 5

* A disclaimer is necessary. This paper was thought of and presented at a seminar held in March 2017 and subsequently adapted for publishing. In the meantime lots of ink has run on the topic. Legal writings rarely are like grand-crus which improve over the years in tightly sealed barriques. The reader will be the best judge. At any rate one can by now consider an acquired wisdom that “smart contracts” are nor “contracts” nor “smart”. However, many issues still remain open, which justify presenting this article. 1 One can start by following the directions contained in Grundmann/Hacker, The Digital Dimension as a Challenge to European Contract Law. The Architecture, in: Grundmann (ed.), European Contract Law in the Digital Age, Intersentia 2018, at 3; and in Schulze/Staudenmayer, Digital Revolution – Challenges for Contract Law, in id. (eds.), Digital Revolution: Challenges for Contract Law in Practice, Hart-Nomos 2016, at 19 et seq.

264

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:33 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

Contrary to other social sciences, the law cannot suffer vague expressions or, even worse, similitudes. A contract is a contract. A norm is a norm. Not something that resembles them. If a legal object is different it has a different name, and there can only be confusion if one calls differently the same legal objects, or if different legal objects are given the same name.2 Unless one chooses to follow Humpty Dumpty’s inescapable logic: “When I use a word it means just what I choose it to mean – Neither more nor less”. a) Trying to make sense out of Humpty Dumpty nonsense, one can start by destructuring so-called “smart contracts”. Setting aside the adjective (we shall see later who is “smart” and who is “dumb”) let us focus on the noun. Here are some of the definitions currently given in what appear to be the most downloaded and cited papers:3 “A smart contract is a software, which computer code binds two, or a multitude, of parties in view of the execution of predefined effects, and that is stored on a distributed ledger.”4 “Smart contracts are algorithmic, self-executing and self-enforcing computer programs.”5 “Agreements existing in the form of software code implemented on the Blockchain platform, which ensures autonomy and self-executive nature of Smart contract terms based on predetermined set of factors.”6 “Smart contracts” are decentralized agreements built in computer code and stored on a blockchain.”7 “Smart contracts are technically defined as an event-driven programs, with state, that run on a distributed, decentralized, shared and replicated ledger (blockchain) and that can take custody over and transfer assets on the ledger. This new invention enables declarations of will to be expressed as self-executing computer code. The fact that smart contracts can transfer assets without the need for judicial system creates many questions about their place in the civil law.”8 “These are in essence programs that perform part of the contractual obligations, and may contain and execute contractual conditions, as well as invoke physical remedies (such as providing or withholding access to a room, interrupting the starter of a car).”9 2

See Sartor et al. (eds.), Approaches to Legal Ontologies, Springer 2011. Please note that not all the authors cited in the following footnotes adhere enthusiastically to the “smart contracts” credo. 4 Jaccard, Smart Contracts and the Role of Law, in Jusletter IT, 23.11.2017 (available on SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3099885). 5 Lauslahti/Mattila/Seppala, Smart Contracts. How will Blockchain Technology Affect Contractual Practices?, in ETLA Reports 9.1.2017 (available online at https://www.etla.fi/wp-content/uploads/ETLARaportit-Reports-68.pdf). 6 Savelyev, Contract Law 2.0: “Smart” Contracts As the Beginning of the End of Classic Contract Law, Higher School of Economics, Moscow, Working Papers 2016 (available online at https://wp.hse.ru/data/ 2016/12/14/1111743800/71LAW2016.pdf). 7 Sklaroff, Smart Contracts and the Cost of Inflexibility, 166 U. Pa. L. Rev. 263 (2017). 8 Szczerbowski, Place of smart contracts in civil law. A few comments on form and interpretation (available at SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3095933). Incidentally, one might observe that the statement according to which a judicial system is essential for the transfer of assets is, to put it mildly, a misconception. Clearly any legal provision, in any field, lives in an environment where there are means (not necessarily judicial) of enforcement: a contract of pledge which allows self-relief is not a “smart contract”. 9 Tjong Tjin Tai, Force Majeure and Excuses in Smart Contracts, Tilburg Private Law Working Paper Series, No. 10/2018 (available at SSRN https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3183637). 3

Zeno-Zencovich

265

9

10

11 12

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:33 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

“‘Smart contracts’ [are] a new digital innovation that leverages the blockchain (the technology underlying Bitcoin) to encode obligations so they execute automatically when certain triggering conditions are met.”10 “Smart contracts are self-executing digital transactions using decentralized cryptographic mechanisms for enforcement.”11 13 14

15

16 17

18 19

From these descriptions it would appear that there is fundamental confusion between a contract and its performance. This is even more surprising when this muddle comes from authors who one supposes have studied private law in a continental European university and who, in their first year, have – must have – learnt that delivery of the goods and payment of the price are not the contract, but unilateral legal acts which ensue from the contract.12 Even if one does not share – as the author of these notes does – the opinion that “Bitcoins” and similar crypto-currencies are a gigantic fraud, they are issued and circulate under some kind of agreement.13 The contract is not in the crypto-currency or in the block-chain technology underlying it, but in the conditions under which they are issued or exchanged. Fascination for the digital dernier-cri obliterates knowledge of past experiences. Over a century ago, when the ancestors of e-contracts were put into circulation, scholars asked themselves whether automated vending machines gave rise to a contract. One put a coin in a slot, pressed a button and pop! out came the chosen product. Where was the meeting of the minds? The consensus in idem? Offer and acceptance? What if the coinage was insufficient, or if the item was different from that requested? It did not take much brainwork to put the new, automated, wine into the century-old jars of contractual theory.14 And frankly speaking it is difficult to see what is presented as “revolutionary” in “smart contracts” when one announces grandly that they possess a guaranteed enforcement. For at least nearly half a century we have been using metro tickets with a magnetic stripe: no entry to and no exit from the metro station with an already used 10 Verstraete, The Stakes of Smart Contracts’, in Arizona Legal Studies, Univ. Arizona, Discussion Paper No. 18–20 (May 2018) (available at SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3178393). 11 Werbach, Nicolas Cornell, Contracts Ex Machina, 67 Duke L.J. 313 (2017). 12 Embedded in our legal culture is the principle stated in Digest L, 16, de verborum significatione, 176 (the title of the chapter is very significant) “solvere dicimus eum qui fecit, quod facere promisit”. Even less understandable, from a continental European perspective, is the juxtaposition of “contract” with its enforcement: Koulu, Blockchains and Online Dispute Resolution: Smart Contracts as an Alternative to Enforcement, in 13 Scripted 40 (2016) (available online at https://script-ed.org/article/blockchains-andonline-dispute-resolution-smart-contracts-as-an-alternative-to-enforcement/). In a common law environment such distinctions appear to be immaterial: “Smart contracts are models of legal efficiency, reducing the need for a complex court system to enforce transactions because the contracts themselves are selfenforcing” (McKinney/Landy/Wilka, Smart Contracts, Blockchain, and the Next Frontier of Transactional Law, 13 Wash. J.L. Tech. & Arts 313 (2018). 13 “Some kind”: as this currency can be created and can circulate only on the basis of an agreement I have vainly searched for indications on the bitcoin.org website. Not surprisingly – as these cryptocurrencies can exist only in the shadow of the law – one finds only a series of usual disclaimers [https://bitcoin. org/en/legal] and a revealing page [https://bitcoin.org/en/scams] titled “Avoid Scams” (de te fabula narratur?). Unless one dumps any contractual theory and imagines that the system is legally run in accordance with the article by Nakamoto, Bitcoin: A Peer-to-Peer Electronic Cash System [https://bitcoin. org/bitcoin.pdf] which purportedly presents the functioning of the system. 14 In Germany see, even before the BGB, Auwers, Der Rechtsschutz der automatischen Wage nach gemeinem Recht, Kästner 1891 (followed by the works of Günther, Schels, Schiller, Ertel and Neumond). In Italy see Cicu, Gli automi nel diritto privato, in Il Filangieri 1901, 561 and Scialoja, L’offerta a persona indeterminata ed il contratto concluso mediante automatico, Lapi 1902. For an appraisal see Gambino, L’accordo telematico, Giuffré 1997.

266

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:33 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

ticket. The ticket, in itself, is not the contract but is the document that provides evidence that the holder has entered an agreement with the carrier. All legal activities (thousands, and not only contractual) on a telecommunication network (from buying an airline ticket, withdrawing cash from an ATM, filling a form for one’s children’s school, enlisting in an unemployment roll, presenting a tax form, etc.) have an underlying software that obliges compliance.15 One would hope that by now Lawrence Lessig’s tenet – code is law16 – is sufficiently interiorized, without the need to invent again – as one says in Italy – the umbrella or the water-boiler. Once one has disposed of the idea that the definitions provided above have anything to do with a “contract” (not a wannabee contract; not a make-believe contract; not a “contractish” legal act; not an imitation contract; not a pseudo-contract), one must take into account a much more appropriate notion of “smart contract”, where the accent goes on the adjective. “Smart” is referred to the fact that the contract – here in its proper sense – is shaped, in its terms and conditions, by one of the parties on the basis of the available big data. This aspect will be examined in a following paragraph. b) Destructuring “granular norms” is more complex. In this case the adjective is sufficiently identifiable and clear. What is widely debated is the noun. It is difficult to find a term that is open to so many, contrasting, definitions as “norm”. Does it mean a law, or some piece of general provision issued by a public authority empowered to produce it? Does it include provisions formed by private parties and binding for a non-determined number of other private parties? Are contractual terms and conditions to be considered as “norms” (the Code Napolèon taught us that “les conventions légalement formées tiennent lieu de loi à ceux qui les ont faites”). And what about “social norms”? Technical rules? Standards? An enormous amount of brainpower and of ingenuity have been put – mostly by philosophers and by general theorists – in the attempt to set out a shared view. The results are not encouraging, for every new analysis ends up by providing one further definition.17 For the limited purposed of this paper one will adopt a broad notion. The choice is somehow obliged. If one were to choose a quasi-Kelsenian approach (norms are general commands which prohibit, enable or promote, issued by a public authority entrusted with its enforcement, in accordance with a clearly defined hierarchy) this would be incompatible with the adjective “granular” which supposes individualized rules, one different from the other, tailored on the specific features of one (or both) of the parties. The institutional approach can, however, be recovered imagining that “granular norms” are set (rectius, should be set) by a more general rule that enables such individualization. 15 To bring things back to a more orderly system the suggestion is that of focusing on the so-called “software agents”. For a comprehensive view of the issues see Sartor, Cognitive automata and the law: electronic contracting and the intentionality of software agents, in 17 Artificial Intelligence and Law 253 (2009); and Id., Agents in Cyberlaw, in: Cevenini (ed.), The law of electronic agents, Gedit, Bologna 2004. Incidentally the “software agents” theory allows us to tackle the legal issues related to robots. 16 Lessig, Code and other laws of cyberspace: Version 2.0, Basic Books, 2006. 17 See Hydén/Svensson, The Concept of Norms in Sociology of Law, in 53 Scandinavian Studies in Law 15 (2008) who (at p. 25) distinguish, from their view-point, between “Legal”, “Social”, “Technical”, “Economical”, “ Bureaucratic”; Carbonnier, Il y a plus d’une definition dans la maison du droit, in 11 Droits 5 (1990) . From a legal realist perspective “The ‘nature of the law’ is the main problem of jurisprudence” (Ross, On Law and Justice, University California Press 1959, at 11. The inevitable conclusion seems to have been reached over two centuries ago: “Jurists are still without a complete definition of the idea of right” (Kant, Critique of pure reason. Transcendental doctrine of method, ch. 1, § 1 (1781).

Zeno-Zencovich

267

20

21

22

23

24

25

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:33 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

The problem with “granular norms” is different, therefore, from the one faces with “smart contracts”, which in their original meaning are simply sheer nonsense (from a legal point of view, obviously). “Granular norms” impose a choice which is not only theoretical (what do we mean by “norm”) but also deontic: what is the purpose of a norm? Is any “norm” a norm? Should “granular norms” be admissible in the Western legal tradition? The answers to these questions are complex, especially because, as we shall see, there does not appear to be a clear-cut boundary between “granular norms” and “non-granular norms”. 27 At any rate, like for any serious analysis of legal situations, words – correct and shared words – are essential. Even more so when the legal discourse is invaded by juvenile-digital jargon. The need to call things by their proper name is common to many other areas of human knowledge, especially hard and bio-sciences. One just has to imagine what would happen if one started to call parts, functions or disfunctions of the human body with terms created by people with no medical training and, especially, no idea of physiology and pathology. 26

II. How “smart” can contracts be? Let us now turn to the second, and more reliable, notion of “smart contracts”, contracts that through big data have been tailored on the specific features and the (supposed) needs of the parties. 29 The advantages of such contracts have been well set out by professors Ben‐Shahar and Porat:18 28

“Instead of one-size-fits-all protective mandates, the law would tailor the protection to the personal attributes of each protected party. Similar to the method through which other services like insurance, education, medicine, and marketing are personalized – firms using Big Data to tailor their product to the predicted personal needs of each client – the service of legal protection could be personalized to correspond to the predicted protective needs of contracting parties. We argue that, if done properly, personalization could increase the benefits and reduce the unintended costs of mandatory law. Protective needs would be better addressed, and more consumers would be served”.19 From a theoretical point of view there can be no objection to personalized contracts: when we go to the restaurant everybody – even those who sit around the same table – choose different dishes according to his or her taste. And this is exactly the difference with so-called “fast-foods” where each item (burger, drink, chips) is standardized. 31 The challenge brought by “smart contracts” is deeper and goes to the roots of the debate that quite aptly has been labelled as “The rise and fall of freedom of contract”.20 One assumes that the cultivated reader is familiar with the reasons that brought to the 30

18 Ben‐Shahar/Porat, Personalizing Mandatory Rules in Contract Law, (available at SSRN https:// papers.ssrn.com/sol3/papers.cfm?abstract_id=3184095). 19 A similar approach can be found in Porat/Strahilevitz, supra Part 1.A (“Under a personalized approach to default rules, individuals are assigned default terms in contracts or wills that are tailored to their own personalities, characteristics, and past behaviors. Similarly, disclosures by firms or the state can be tailored so that only information likely to be relevant to an individual is disclosed and information likely to be irrelevant to her is omitted.”). However, one should note that quite often one tends to forget one of the basic distinctions between the continental European approach and the US one. In the former, “default rules” for contracts are historically sedimented in the Civil Code and in statute law. In the latter, the role of legislation is much less cogent and rules are forged on a case-by-case basis by the courts. 20 The reference is obviously to Patrick Atiyah’s classical book (OUP 1985).

268

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

genesis of “consumer contracts”, as a response to bargaining inequality. The reasons of such inequality have been – extensively and for well over a century – thoroughly investigated and include not only the economic size of the parties of the contract, but also their informational unbalance. This element is well described in the standard definition of consumer in all EU directives and regulation: “‘Consumer’ means any natural person who, in a contract with a professional, is acting for purposes which are outside his trade, business, craft or profession”. A professional, in his or her field of business, possesses a specific knowledge that enables to bargain at arm’s length, and even with a considerable advantage over the other party (the typical example is that of the computer consultant that can bamboozle any firm that does not have in-house expertise). In the big data environment this informational unbalance increases covering another dimension21. Until now the business knew everything of its products, the structure of their costs, their functionality, their possible limits and defects. The answer to this has been that of imposing an overload of information of all kinds from the business to the consumer. A practice that, on the whole, is considered quite unsuccessful.22 Now the business, though profiling techniques, knows, (nearly) everything of the consumer: his or her income, family group, educational background, profession, spending habits and preferences, previous purchases, whereabouts, credit reliability.23 This practice, provided it complies with data protection regulations, is perfectly legitimate and if not followed would raise serious doubts on the ability in running a business. It conforms with the golden rule of any commercial activity, since it began: know your client.24 It is therefore natural that businesses should segment their clients on the basis of algorithms and predictive analytics. It appears that this is already a common practice especially for those businesses that operate on the Internet. The selection of the clients touches, initially, two basic elements: what is offered (i.e. the nature of the goods and services, tailored to the supposed preferences of the potential client), and the price requested. The informational unbalance is structural: it is pointless for the consumer to profile the business. The most he or she can do is compare the ratings on the service or the product, hoping that they are reliable and not rigged25. One can imagine also that the business is able to know, in real time, through the cookies it has installed, if the consumer is comparing prices and products, if he or she is considering other providers

21 In general, see Wendehorst, Consumer Contracts and the Internet of Things, in Schulze/Staudenmayer (eds.), Digital Revolution: Challenges for Contract Law in Practice, cit., 189 et seq. 22 The literature on the point is vast. One can refer, for the most reliable presentation, to Bar‐Gill, Consumer Transactions, in Zamir/Teichman (eds.), The Oxford Handbook of Behavioral Law and Economics, OUP 2014, 465 et seq. 23 See Sein, What Rules Should Apply to Smart Consumer Goods: Goods with Embedded Digital Content in the Borderland between the Digital Content Directive and Normal Contract Law, 8 JIPITec 96 (2017). 24 Helberger, Profiling and Targeting Consumers in the Internet of Things – A New Challenge for Consumer Law, in Schulze/Staudenmayer (eds.), Digital Revolution: Challenges for Contract Law in Practice, cit., 135 et seq. 25 This was the conclusion of the Italian Competition and Consumer Authority in the TripAdvisor case (proceeding PS9345 of 22 December 2014, available at http://www.agcm.it/ trasp-statistiche/doc_download/4619-ps9345scorrsanz-omi.htm). The decision was, however, quashed by the Rome Administrative Tribunal (TAR Lazio), 13 July 2015, n. 9355, in Diritto informazione informatica 2015, 494. The case is analysed in detail by De Franceschi, The Adequacy of Italian Law for the Platform Economy, (2016) 1 EuCML 56.

Zeno-Zencovich

269

32

33

34

35

36

37

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

38

39 40

41 42

43

44 45 46

47

48

of services.26 Therefore, every rational cost reducing move of the consumer can be automatically countered.27 As one can see, “smart contracts” are an efficient practice to extract the maximum value from the consumer. The business is on the “smart” side, the consumer on the “dumb” one.28 Can this conclusion be avoided through existing legal instruments?29 The first path could be that of considering price differentiation as an unfair commercial practice. This would happen when the same product or service is sold at different prices. One must be aware, however, that the price can depend on external circumstances: typically, how much in advance is a travel ticket or a hotel room bought. To avoid the accusation of unfairness it would be sufficient to add to the more expensive product or service a small bonus which purportedly is the consideration for the price increase. Furthermore, price differentiation is one of the most complex legal issues, quite commonly tackled by competition law and in regulated markets.30 In the first place we find geographical segmentation in which prices vary, upwards, in accordance with transport costs and, downwards, in accordance to consumption rate or power of expenditure. Nobody complains – from an EU common market point of view – if exactly the same toothpaste costs more in a Luxembourg supermarket that in a Bulgarian one. Other quite common forms of price differentiation are on the basis of age: children or old-age pensioners may receive a discount not only in public services, but also in entirely privately-run business, such as cinemas or supermarkets, the assumption being that they have a limited purchasing power. One could therefore argue that “smart contracts” are simply an evolution of firm and well accepted business practices. This, however, is not always true. Price discrimination is not allowed in most public services, especially if provided by public utilities (energy, water, telecommunications). In the EU, article 102 of the TFEU prohibits price discrimination by firms which are in a dominant position or that hold a significant market share. And, obviously, article 101 prohibits concerted practices between businesses aimed at price discrimination. This, implies, a contrario, that firms that do not fall in such condition are allowed to discriminate among their clients. They will be – or should be – sanctioned only by competition on the price. This brings us further in the analysis of the notion of contractual capacity. 26

Langhanke/Schmidt-Kessel, Consumer Data as Consideration, (2015) EuCML 218 et seq. As for privacy concerns: cf Wendehorst, Of Elephants in the Room and Paper Tigers: How to Reconcile Data Protection and the Data Economy, in: Schulze/Staudenmayer/Lohsse (eds.), Trading Data in the Digital Economy: Legal Concepts and Tools, Hart-Nomos 2017, 327 et seq. 28 This tendency in computer contracts was quite clear over 30 years ago: with permission, one can refer to Zeno-Zencovich, Sul rilievo pratico e sistematico della categoria del c.d. contratti di informatica, in: Alpa/Zeno-Zencovich (eds.), I contratti di informatica, Giuffré 1987, 31 (at p. 39: “It is not unrealistic to suggest that computer contracts are a sort of ‘laboratory’ for contractual models more profitable for businesses. The balance one finds in computer contracts appears to be a reply to the pervasive regulation (legislative, administrative, judicial) of contracts promoted by the instances of consumerism.” – translation supplied). 29 On similar issues, though from a different perspectives, Janal, Fishing for an Agreement: Data Access and the Notion of Contract, in: Schulze/Staudenmayer/Lohsse (eds.), Trading Data in the Digital Economy: Legal Concepts and Tools, cit., 273 et seq., 280 et seq.; Schulze, Supply of Digital Content. A New Challenge for European Contract Law, in De Franceschi (ed.), European Contract Law and the Digital Single Market. The Implications of the Digital Revolution, Intersentia 2016, 127 et seq. 30 One can refer to the extremely detailed article by Maggiolino, Big Data e prezzi personalizzati, (2016) Concorrenza e Mercato 23, 95 and to the vast literature (both legal and economic) cited therein. 27

270

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

III. Creditworthiness “Creditworthiness” is a concept that, under different words, has been around for millennia. Any sound business judgment is based on it when payment is differed in time, or when money is lent. Again, one can assume that the cultivated reader is familiar with its use in capitalist societies for at least the last two centuries, in company management, in financial institutions, in contracts. We shall focus here on the meaning – the legal meaning – of the term in the last 10 years, i.e. after the dramatic 2008 financial crisis that struck the US and involved also Europe. Extremely succinctly, the macro-prudential measures adopted by the Basel Committee on Banking Supervision and that commonly go under the shorthand of “Basel III”,31 set a number of minimum requirements that apply to most banking institutions in order to ensure their resilience in case of unexpected financial crisis. This entails strong – even stronger than in the past – powers of regulation, supervision and risk management in financial activities. These regulations clearly do not affect only the relations between central banks and supervised institution, but, downstream, the relations with all those who entertain credit relations with the latter.32 Expressed in very simple terms, banks – most European banks – collect money from savers and depositors and sell credit to all those – individuals, families, firms – that are in need of liquidity. Lending has always been based on two essential elements: guarantees (typically pledges, mortgages, liens) and creditworthiness. After the 2008 crisis they both have been strengthened bringing to the present “credit crunch”. While credit-rating for businesses has always been used, generally under some kind of legislative framework and public supervision, the aspect we are interested in here is consumer creditworthiness.33 While firms resort to credit for the ordinary or extra-ordinary course of their business, individuals and families ask credit in order to purchase goods and services: first of all, a house, but also automobiles, household appliances, educational and medical services, and more frivolous expenses (cruises, jewelry, hi-tech apparatuses). Therefore, there is a direct relation between creditworthiness and contractual capacity. Those who have a high score on the first, can enter into contracts with delayed forms of payment. Those who have a low rating cannot.34 One could object that lack of (or limited) creditworthiness does not deprive them of contractual capacity: it is sufficient that they save until they can dispose of the necessary cash.

31 Basel Committee on Banking Supervision, Basel III: A global regulatory framework for more resilient banks and banking systems (available online at https://www.bis.org/publ/bcbs189.pdf). 32 See in general, Grundmann/Atamer (eds.), Financial Services, Financial Crisis and General European Contract Law. Failure and Challenges of Contracting, Wolters Kluwer 2011. 33 Cf Comparato/Domurath, Financialisation and Its Implications for Private Autonomy in Consumer Credit Law, (2015) Osservatorio del diritto civile e commerciale, 280–281. 34 Or they can only by paying higher interest rates, therefore confirming that they are at risk: see Liberman/Paravisini/Pathania, High-Cost Debt and Perceived Creditworthiness: Evidence from the U.K. (available on SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2797383).

Zeno-Zencovich

271

49

50

51

52

53

54 55

56 57 58

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

In these times of consumer and demand-driven economies the objection, however, seems to echo Marie-Antoinette’s rejoinder: “Qu’ils mangent des brioches”. From a strictly legal point of view, however, what needs to be investigated is to what extent creditworthiness is a mandatory requirement, and to what extent, instead, it is left to fair business judgment.35 60 Significant indicia come from the EU Directive 2008/48 on “Credit agreements for consumers”. The recitals are quite explicit: “Creditors should not engage in irresponsible lending or give out credit without prior assessment of creditworthiness” and “should bear the responsibility of checking individually the creditworthiness of the consumer” (Recital 27). The same recital adds – and the wording is relevant for this paper – that “to that end, they [the creditors] should be allowed to use information provided by the consumer”. This phrase leaves open the issue of the use of information provided by third parties (typically profiling of consumers). 61 Article 8 appears to be mandatory. The heading is “Obligation to assess the creditworthiness of the consumer”. Its para. 1 states: 59

“Member States shall ensure that, before the conclusion of the credit agreement, the creditor assesses the consumer’s creditworthiness on the basis of sufficient information, where appropriate obtained from the consumer and, where necessary, on the basis of a consultation of the relevant database. Member States whose legislation requires creditors to assess the creditworthiness of consumers on the basis of a consultation of the relevant database may retain this requirement”. 62 63

64

65 66

What does “relevant database” mean? Does it include big data that allow profiling of the consumer? More recently the Commission has informed the other EU institutions, including the ECB, that “To prevent possible new NPLs in the context of consumer loans, Member States are also invited to put in place rules for the assessment of consumer affordability”.36 The Consumer Financial Services Action Plan issued by the EU Commission in 2017 states that: “While the increased availability and easier access to consumer credit create opportunities for business and result in lower costs for borrowers, there is also an increased risk of irresponsible lending and borrowing causing over-indebtedness. This risk needs to be mitigated”.37 All this is strengthened by the ECB directives on the collection of granular credit data.38 Similar provisions are contained in the 2014/17 Directive on “Credit agreements for consumers relating to residential immovable property”. Article 18 sets, in its heading, an “Obligation to assess the creditworthiness of the consumer”, and para. 1 that “Member States shall ensure that, before concluding a credit agreement, the creditor makes a thorough assessment of the consumer’s creditworthiness”. But what 35 Cf on these issues Kruithof, A Differentiated Approach to Client Protection: The Example of MiFID, in: Grundmann/Atamer (eds), Financial Services, Financial Crisis and General European Contract Law. Failure and Challenges of Contracting, cit., 116 et seq. 36 Communication from the Commission to the European Parliament, the European Council and the European Central Bank – Second Progress Report on the Reduction of Non-Performing Loans in Europe (14.3.2018 COM(2018) 133 final). 37 Communication from the Commission to the European Parliament, the Council and the European Central Bank, the European Economic and Social Committee and the Committee of the Regions – Consumer Financial Services Action Plan: Better Products, More Choice (23.3.2017 COM(2017) 139 final). 38 See e.g. the Guideline (EU) 2017/2335 of the European Central Bank of 23 November 2017 on the procedures for the collection of granular credit and credit risk data (ECB/2017/38).

272

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:34 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

is even more important is para. 5, letter a) which is expressed in mandatory terms: “Member States shall ensure that the creditor only makes the credit available to the consumer where the result of the creditworthiness assessment indicates that the obligations resulting from the credit agreement are likely to be met in the manner required under that agreement”. The use of terms such as “thorough” and “only” indicate a very clear line of conduct to financial institutions and, together with Directive 2008/48, are evidence of an acquis. Setting aside legal quibbles it appears that uniform financial practices impose advanced data analytics when assessing creditworthiness of consumers.39 There is (maybe) a duty to disclose on which elements the refusal to grant credit is made, or the more onerous conditions and collaterals required. But the bottom-line is that access to consumer credit markets can and should be restricted.40 Does this imply that contractual capacity is not equal, and that some individuals are legally prevented from entering certain contracts? Or to see things from the other side, that financial institutions may not enter into credit agreements with individuals whose creditworthiness is very low?41 Again, one should not be surprised. There are several cases in which contractual capacity is legally limited: historically merchants who went bankrupt, or individuals who have issued an uncovered cheque. However, the difference is striking: in these cases, limitations are a sanction for violation of the law and of sound business practices. In the case of creditworthiness what is sanctioned is – to put it bluntly – poverty. Which is quite incompatible with all European principles on fundamental and social rights. It is difficult to strike an appropriate balance between the competing values. On the one side prudential policies – such as credit-rating and creditworthiness – pursue paramount public interests. One should not indulge in crass demagogy – quite common in these times of ramping populism – arguing that prudential policies are meant to protect, enrich and fatten bankers and their like. Financial crisis and credit crunches hit in the first place all those firms – the majority in continental Europe – which resort to bank credit to keep their activity running. Millions of workers depend on the viability of such businesses; efficient financial institutions are essential for the well-being and development of whole regions. Financial markets are global, and it is essential to avoid domino effects in geographical areas only apparently unrelated. On the other side one should ask oneself to what extent such public interest can indent the fundamental principle of equality (or, to see it the other way, of nondiscrimination).

39 The problem is that availability of credit reports and data is quite differently regulated through the EU: see Ferretti, Credit Bureaus Between Risk-Management, Creditworthiness Assessment and Prudential Supervision, (May 2015), (available on SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_ id=2610142). 40 Yap/Ong/Husain, Using data mining to improve assessment of credit worthiness via credit scoring models, 38 Expert Systems with Applications 13274 (2011). 41 Is there a right to “creditworthiness”? See Neethling, Blacklisting of a Debtor as a Credit Risk – Infringement of a Debtor’s Rights to Creditworthiness and Earning Capacity as Personal Immaterial Property Rights, 18 SA Merc. 376 (2006). In this line one should consider the Opinion of the European Economic and Social Committee on ‘Consumer protection and appropriate treatment of over-indebtedness to prevent social exclusion’ (exploratory opinion) (in OJ C 311, 12.9.2014). Although it has not brought to general EU legislation, some Member States have moved in this direction: see the Italian Law 27.1.2012, n.3 on “Usury, extorsion and solution of crisis from over-indebtedness” which devotes Articles from 6 to 20 to the topic.

Zeno-Zencovich

273

67 68

69

70 71

72

73

74

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

A reasonable answer might lie in the principle of proportionality. Given the legitimacy of (rectius: the need for) policies that pursue financial stability, is the obligation to previously assess creditworthiness, excluding from the market a large number of citizens, proportional?42 Or can those goals be attained through a lesser sacrifice of fundamental rights? For example, could the (inevitable) risk of nonperformance be covered through mutual insurance founds financed in part also by consumer sales? Could NPLs – which are a stable component of mature financial market – be regulated in a permanent way such as to insulate banks from recurrent resolutions?43 76 There is a further clash of values and of rules that has to be pointed out: financial markets are efficient if they are transparent and therefore if all actors possess (or have access to) the same amount of information. This is the reason why insider trading and abuse of privileged information are severely punished. At the same time, however, in Europe we have raised data protection to the level of a fundamental right, on the basis of the principle of “informational self-determination”. This right is set aside when individuals are forced to disclose all the data concerning their economic life (including life styles) if they want to access the credit market. This conflict is becoming every day more obvious with the implementation of the second Payment Services Directive (2015/2366) when third party service providers ask access to financial data held by banks. 77 The quest for “smarter” solutions is justified not only by individual justice instances. It is well known that refusal to grant credit is the main fuel of shadow banking and usury, which represent a major concern for all developed economies because they undercut all the macro-prudential efforts. 78 What are the consequences of this analysis of consumer credit contracts and of creditworthiness on the general issue of “smart contracts”? In the financial sectors restrictions to contractual capacity are covered by normative provisions issued under compelling public policy reasons. When it comes to ordinary contracts one might argue that such an umbrella is lacking. Fairness – i.e. equal treatment – is an essential element in consumer contracts that cannot be circumvented through sophisticated data analytics and algorithms. This appears to be an issue that will have to be settled de lege ferenda44 or – as quite common in the EU – by incremental decisions of the Court of Justice.45 75

42 The German implementation of the GDPR (DSAnpUG-EU of 30-6-2017) attempts to rationalize the procedure in its Article 31, devoted to the “Protection of commercial transactions in the case of scoring and credit reports”: “(1) For the purpose of deciding on the creation, execution or termination of a contractual relationship with a natural person, the use of a probability value for certain future action by this person (scoring) shall be permitted only if: 1. the provisions of data protection law have been followed; 2. the data used to calculate the probability value are demonstrably essential for calculating the probability of the action on the basis of a scientifically recognized mathematic-statistical procedure”. 43 Again, one can notice the difference between the European and the US approach. While in the former one focuses more on substantive guarantees (establishing the balance between the different interest), in the latter the approach is procedural: see Citron/Pasquale, The Scored Society: Due Process for Automated Predictions, 89 Washington L. Rev. 1 (2014). 44 The proposals from the Commission to increase consumer awareness do not appear to be a substantial and adequate response to the problem: see the Communication on “Consumer Financial Services Action Plan: Better Products, More Choice” (23.3.2017 – COM (2017) 139 final). 45 For some decisions in this direction see CJEU 27.3.2014, in Case C-565/12, Credit Lyonnais and CJEU 9.11.2016 in Case C-42/15, Home Credit Slovakia.

274

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

IV. “Granular norms” Once one has set aside – for discussion’s sake – the idea that “granular norms” is an oxymoron, and one has acceded to a very broad notion of “norm”, that includes individualized commands, tailored on specific qualities of the addressee, the attempt is to understand if, when, and how such “norms” can find place in our legal systems.46 A comparatist notes immediately the cultural origin of the notion. It appears to be quite obvious that in legal systems which have grown incrementally through the stratification and consolidation of case-law – i.e. norms set case by case on the basis of the facts, the qualities of and the relationship between the parties – granular norms are simply (?) a return to the past. One destructures a general rule in its thousands, millions of occurrences and applies it casuistically. In this case however the destructuring is not done though a microscopic analysis of precedents, but though digital technologies which, analyzing data concerning the parties involved, circumstances, goals (e.g. efficiency), are able to set, ex ante, an individualized rule. This rule can be applied to contractual relations,47 but even, mostly, to cases in which there is a command issued by some public authority which requires fine-tuning according to the addressee.48 Clearly a continental European lawyer may remain somewhat astonished, and even scandalized, but this appears to be the equivalent of the surprise which struck continental European lawyers when, at the beginning of the 19th century, they discovered that English law had developed through the centuries by gentlemen who had never gone to university and graduated in law, and who were creating a system based – believe it or not – on judicial precedents. Seen from the continental European tradition norms are – must be – general provisions addressed to an indistinct recipient and aimed at regulating future actions and legal relations. This view has very sound political basis, the first being that of outlawing individual, ad personam, rules typical – one argued – of the ancien régime. Furthermore, once the law-making process has been concentrated in Parliament which is the expression of popular sovereignty, a norm can only be générale because it is the result of the volonté générale. This model clearly is replicated in all the lower norm-producing instances: norms are, must be, abstract and general, designed for the future. Therefore, around the debate on “granular norms” there is, in the first place, a cultural divide, which if not clarified fosters only misunderstandings: for a Scot a Scotsman is clearly identifiable because he wears a kilt. For a European he cannot be a man – or he is a very queer man – because men do not wear skirts; and we could go on and on with similar ludicrous arguments and discussions.

46 Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in: Mak et al. (eds.), Research Handbook on Data Science and Law, Edward Elgar 2018 (available on SSRN at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3181914). 47 Cf e.g. Busch, Implementing Personalized Law: Personalized Disclosures in Consumer Law and Privacy Law, forthcoming in University of Chicago Law Review, (available on SSRN at https://papers.ssrn. com/sol3/papers.cfm?abstract_id=3181913). 48 See Casey/Niblett, supra Part 1.C: “Advances in technology (such as big data and artificial intelligence) will give rise to this new form – the micro-directive – which will provide the benefits of both rules and standards without the costs of either. Lawmakers will be able to use predictive and communication technologies to enact complex legislative goals that are translated by machines into a vast catalog of simple commands for all possible scenarios”.

Zeno-Zencovich

275

79

80

81

82

83 84

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives 85

86

87

88

89

90

91

A comparatist however points out that legal systems and traditions cannot be compared en blanc et noir and that in the US and UK legislation and regulation are widespread and pervasive, and present the same characteristics of continental Europe. And that in the US the decisions of the Supreme Court have – to use an EU expression – both a vertical and a horizontal effect. On this side of the Atlantic it is pointless to elaborate on the law-making role of jurisprudence, already in the positivistic 19th century atmosphere. Judges do not only interpret the law, they never have been simply la bouche de la loi; they create rules ex nihilo which are applied to the facts of that specific case and become a precedent which is followed in other cases and eventually is formalized in a piece of legislation. Let us therefore assume that “granular norms” are not foreign to the common law tradition, and that there are no logical or deontic reasons that bar them from entering in the guarded citadel of the law. What appear to be objectionable – from a different perspective – are some of the arguments used to support their legitimacy, in particular that of a purported efficiency of such norms. One could point out that a significant part of hard-core law & economics, being based on purely hypothetical and unverified assumptions,49 has been greatly discredited in most fields where it has attempted to claim its rule.50 On the other hand, one should underline that the suggested notion of efficiency is – like in the case of “smart contracts” – generally one-sided, in the sense that it entails less costs for only one part of the legal relations, shifting the cost on the other side or on general interests, freedoms and liberties. Surely – as in all digital transactions – there is a considerable amount of time which is saved by all parties, including consumers. This can be considered socially “efficient”, until the moment in which some non-performance occurs and it is necessary to destructure the whole contract to analyse its actual effects. But are “granular norms” entirely alien to the continental European tradition? If one looks at the norms-producing machinery one can see that it is extremely difficult to establish, at times, a clear-cut distinction between a general provision and a personalized decision addressed to only one individual or entity. Let us take the extremely broad area of relations between a public body and an individual or an entity. The distinction between the general legislative provision concerning public procurement and the tender rules, on the one side, and the final contract awarded to the best offeror, on the other side, is quite clear. But when we enter many of the welfare-state annexed services the separation is extremely blurred. When an individual fills in his or her tax form, or his or her request for old-age pension with the assistance of a publicly provided software, are we not contemplating a form – albeit still primitive – of “granular norms”? There is a general legal framework but the amount the individual must pay (taxes i.e. duty), or receive (pension, i.e. rights) depends on the information provided or already available to the public body and the “class” in which the individual falls into. The norm is incorporated in the software which establishes in univocal way how the individual must comply: there is hardly any margin of error because the software does not allow to complete the procedure if it detects deviation from the rule. 49 “The central assumption of classical economics (…) lends itself admirably to technical and mathematical refinement. This, in turn, is tested not by its representation of the real world but by its internal logic and the theoretical and mathematical competence that is brought to bear in analysis and exposition” (Galbraith, Economics in Perspective. A Critical History, (Houghton Mifflin 1987) at p. 285) (italics added). 50 Being, presently, replaced by so-called “behavioral law and economics”: see Zamir/Teichman, supra (fn. 22).

276

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

M. “Smart Contracts”, “Granular Norms” and Non-Discrimination

We can therefore say that the legal system takes into great account differences between individuals and, far from applying a Procrustes-bed rule, adapts itself to the varying subjective and objective conditions. In all these cases it would be extremely difficult to assert that there is a discrimination. The problem therefore is in the way the selection of the various typologies is made; what is the rationale behind them; if the variations are justified; if and to what extent proportionality has been taken into account. This implies moving from the traditional area of control over written norms in the unchartered waters of what is called algorithmic accountability.51 We can assume as an acquired knowledge that algorithms – and in general any form of data collection, processing and analysis – are not “neutral” operations. There are humans and human organizations behind them, with their interests, goals, biases. Often these are quite legitimate and appropriate, but not transposable in different contexts where public policies may be significantly diverse. Control over data (and especially big data) analytics and algorithms is complex for a series of reasons.52 First of all, the professional status of those who work in the field is uncertain, there are no mandatory qualifications or shared professional ethics.53 Secondly the community of “algorithm makers” belongs nearly entirely to the private sector – which is the main consumer and contractor of their products. This community is global and constantly interconnected: solutions developed in Shanghai in a few weeks could be adapted by someone else in Hamburg or San Francisco. While this may be quite appropriate in the running of a private business which pays for its own mistakes, it raises significant doubts when such digital products should pursue public policies which, quite naturally, are totally unknown to the developer of the algorithm.54 There is therefore an obvious cleavage between lawmakers and informaticians which requires to be filled in through patient mutual understanding.55 At any rate one can realistically conclude that “granular norms” are with us and will not be easily banned.

51 Among others, Vedder/Naudts, Accountability for the use of algorithms in a big data environment, (2017) 31 International Review of Law, Computers & Technology, p. 206 et seq. 52 In the European debate, from different perspectives, see e.g. Zech, Data as tradeable commodity, in: De Franceschi (ed.), European Contract Law and the Digital Single Market, cit., 51 et seq.; Hugenholtz, Data Property in the System of Intellectual Property Law: Welcome Guest or Misfit, in: Schulze/ Staudenmayer/Lohsse (eds.), Trading Data in the Digital Economy: Legal Concepts and Tools, cit., 75 et seq. 53 For some attempts to regulate the sector see the 12.1.2017 American Association for Computing Machinery, “Statement on Algorithmic Transparency and Accountability” (available online at https://www. acm.org/binaries/content/assets/public-policy/2017_usacm_statement_algorithms.pdf); and Rosenblat/ Kneese/Boyd, Algorithmic Accountability, (available online at https:// datasociety.net/pubs/2014–0317/AlgorithmicAccountabilityPrimer.pdf); Kroll/Barocas/Felten/Reidenberg/Robinson/Yu, ‘Accountable Algorithms’, (2017) 165 U. Pa. L. Rev. 633. 54 Bass, Big Data and Government Accountability: An Agenda for the Future, 11 I/S Journal of Law and Policy for the Information Society 13 (2015). 55 For some remarks on this issue, Twigg-Flesner, Disruptive Technology – Disrupted Law? How the Digital Revolution Affects (Contract) Law, in: De Franceschi (ed.), European Contract Law and the Digital Single Market, cit., 21 et seq.

Zeno-Zencovich

277

92

93 94

95 96

97

98

99

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/13_V. Zeno-Zencovich, Smart Contracts, Granular Norms and Non Discrimination, 24.11.2019.3d from 30.10.2020 11:37:35 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

V. Non-discrimination in the age of big data 100

101

102 103

104

105

106

107 108

However, as not all evil comes to harm, granular norms help pierce the veil on a basic contradiction of our contemporary, opulent, Western societies. On the one hand we enshrine in our Constitutions and fundamental rights charters the principles of equality and of non-discrimination. They are constantly invoked and applied in legal instruments and judicial decisions. At the same time the most diverse interests group together and claim specific benefits to be granted through legal provisions: minorities, individuals who are – or believe to be – physiologically or psychologically disabled, people who live on islands, in mountainous regions, rural areas, downgraded suburbs, young job-seekers and redundant workers, immigrants or native populations, life-health-food style believers, animal lovers and nudists, regulated professions and public service providers, the list would takes pages and pages to be complete. Each of these groups claims – and often is successful in doing so – the enactment of special provisions that differentiate them from what is considered the majority. In this there is nothing really new: the middle ages and their sequel – which lasted until at least the French revolution – were a highly legally regulated society – albeit with significant problems of enforcement and adjudication – based on the multiplication of individual, collective and class privileges and statuses. Each individual was different from the others for the bundle of legal rights and obligations that were attached to him or her. The contemporary tendency – which is appropriately investigated by social and political scientists – is that differences, shunning “normality” (in its statistical meaning: within the norm)56 and considering public authorities as the dispensers of benefits related to a recognized status, to which often economic advantages are attached. “Granular norms” fit perfectly in this scenario and increase its importance. It is no longer necessary to organize pressure groups which must necessarily lobby the legislature and the other law-making institutions. Differences in income, consumption, habits, education, interests, mobility, health, age, sex, household, are all duly recorded and processed. Setting aside the even more complex issues related to so-called “predictive analytics”, the whole society is profiled and segmented shaping the norms that should apply to a certain cluster of individuals. In this context the age-old rhetoric on equality and non-discrimination makes little sense, and tends to have a humpy-dumpty meaning, as in the initial quote. The issue is clearly a political one, namely that only citizens can decide (maybe exhuming the “equal but different” notion?). The role of legal doctrine, however, ends here in pointing out an irreconcilable contradiction of our times. 56 And vice versa: the term “norm” immediately suggests that of “normal”: see Foucault, Sécurité, Territoire, Population, Paris, Gallimard – Seuil 2004, 59: “Si la loi se réfère à une norme, la loi a donc pour rôle et fonction de codifier une norme (…), j’essaie de repérer comment, en dessous, dans les marges et peutêtre même à contresens d’un système de la loi se développent des techniques de normalisation”.

278

Zeno-Zencovich

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy* I. Introduction In recent years, a small but growing body of literature on personalized law has started 1 to explore whether technological advances in data collection and data science could be used to tailor legal norms to specific individuals.1 In this perspective, the use of big data and artificial intelligence could redefine the optimal complexity of legal rules and refine their content to a hitherto unachievable level of granularity. Personalized legal rules could thus consider actor heterogeneity to a degree impersonal laws are unable to do. As a result, regulatory errors stemming from over- and under-inclusive norms based on coarse-grained typifications could be reduced. Moreover, legal categories could be designed in a more precise and granular way taking into account insights from behavioral science. Against this backdrop, this Essay aims to make two contributions: The first part of 2 the Essay explains how algorithmic personalization of legal rules could be operationalized on the basis of user profiles for tailoring disclosures, mitigating discrimination in the sharing economy and optimizing the flow of traffic in smart cities. The second part of the Essay looks into an aspect of personalized law that has so far been rather underresearched: a transition towards personalized law involves not only changes in the design of legal rules, but also necessitates modifications regarding compliance monitoring and enforcement. It is argued that personalized law can be conceptualized as a form of algorithmic regulation or governance-by-data. Considering that personalized law builds on algorithmic processes and data analytics, the implementation of personalized law requires setting up a regulatory framework for ensuring algorithmic accountability. Thus, in a broader perspective, this Essay aims to create a link between the scholarly debate on algorithmic decision-making and automated legal enforcement and the emerging debate on personalized law. The Essay proceeds as follows: Part II sets the stage by very briefly introducing the 3 broader societal context of personalized law as a form of algorithmic regulation. Part III introduces the concept of personalized law and discusses three use cases relating to digital marketplaces and smart cities. It also analyses the relationship between persona* I am grateful to the participants of the Yale Information Society Project Spring Conference on “(Im) Perfect Enforcement” (Yale Law School, April 2019) and the conference “AI Policy and Law” organized by the Chinese Academy of Social Sciences (Hangzhou, October 2019) for their very helpful comments. 1 See, e.g., Porat/Strahilevitz, supra Part 1.A; Ben‐Shahar/Porat, supra Part 1.B; Sunstein, Choosing not to Choose: Understanding the Value of Choice, Oxford 2015, 157–73; Busch, The Future of Pre-contractual Information Duties: From Behavioural Insights to Big Data, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law, Edward Elgar 2016, 221; Casey/Niblett, supra Part 1.C; Hacker, Personalizing EU Private Law: From Disclosures to Nudges and Mandates, 25 European Review of Private Law 651 (2017); Busch/De Franceschi, Granular Legal Norms: Big Data and the Personalization of Private Law, in: Mak et al. (eds.), Research Handbook in Data Science and Law, Cheltenham 2018, 408; Hacker, The Ambivalence of Algorithms: Gauging the Legitimacy of Personalized Law, in: Bakhoum et al. (eds.), Personal Data in Competition, Consumer Protection and Intellectual Property Law, Heidelberg 2018, 85; Ben‐Shahar/Porat, Personalizing Mandatory Rules in Contract Law, 86 U Chi L Rev 255 (2019); Busch, Implementing Personalized Law: Personalized Disclosures in Consumer Law and Data Privacy Law, 86 U Chi L Rev 309 (2019).

Busch

279

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

lized law and personalization techniques applied by firms. Part IV explains why personalized law can be described as a form of algorithmic regulation and argues for setting up a governance framework for ensuring accountability under a system of personalized law.

II. Big Data and the Crisis of Generalities As the digital transformation expands into more and more areas of life, new datadriven business models are being developed at a rapid pace which are based on algorithms and data analytics.2 The spectrum ranges from personalized advertising to telematics-based insurance and personalized medicine. In other areas, artificial intelligence is used for building “prediction machines”3 which are utilized for a broad range of purposes from predictive maintenance of aircraft engines to predictive policing.4 A common feature of the new technologies is that they are fuelled by the availability of granular data about individuals, objects and locations as well as significant advances in data-processing capacity. 5 These developments go hand in hand with a pervasive trend towards personalisation – some would even say “hyper-individualisation” – which, according to some observers, will result in a fundamental change in the relationship between the individual and society. The ubiquitous quantification and datafication of individuals and their social relations leads to a dissolution of collective categories, such as “citizen” and “consumer” and shifts the focus towards quantifiable differences between individuals. In the emerging “society of singularities”5 the individual is no longer considered as a representative of a certain social group defined by general criteria based on an average model,6 but rather as a singular and solitary being defined by a cloud of data points. According to some observers this shift from generality to singularity is a symptom of a more fundamental “crisis of generalities” caused by the advent of big data.7 6 Recently, research on the legal implications of big data and artificial intelligence has been gathering momentum. So far, the focus primarily lies on analysing the potential dangers for privacy and autonomy, and on elaborating a regulatory framework for the collection and processing of personal data and new data-driven business models.8 However, the digital transformation may require a more fundamental recalibration of the relationship between individuality and legal norms. Adjustments will be necessary along two axes: One the one hand, it is necessary to define limits for the novel personalization techniques that enable firms to target individuals at a level of granularity that was unimaginable only a few years ago. On the other hand, the new technologies 4

2 See, e.g., Mayer-Schönberger/Cukier, Big Data: A Revolution That Will Transform How We Live Work and Think, Boston 2013; Spencer, Privacy and Predictive Analytics in E-Commerce, 45 New England L Rev 101 (2015). 3 Agrawal/Gans/Goldfarb, Prediction Machines: The Simple Economics of Artificial Intelligence, Boston 2018. 4 See, e.g., Ferguson, Policing Predicitive Policing, 94 Wash U L Rev 1109 (2017); Selbst, Disparate Impact in Big Data Policing, 52 Ga L Rev 109 (2017). 5 Reckwitz, Die Gesellschaft der Singularitäten, Berlin 2017. An English translation of this lucid analysis of the “crisis of generalities” has been published as Reckwitz, The Society of Singularities, Cambridge 2019. 6 See also Todd Rose, The End of Average, New York 2016. 7 See Reckwitz, supra (fn. 5). 8 See, e.g., Calo, Digital Market Manipulation, 82 Geo. Wash. L. Rev. 995 (2014); Wagner/Eidenmüller, Down by Algorithms? Siphoning Rents, Exploiting Biases, and Shaping Preferences: Regulating the Dark Side of Personalized Transactions, 86 U Chi L Rev 581 (2019); Zuboff, The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power, London 2019.

280

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

can be utilized for tailoring legal rules to small segments of the population or even individuals. So far, the scholarly discourses about the two aspects of law and personalization have been somewhat detached from each other, but it seems that they are two sides of the same coin.

III. Making Laws for the Personalized Economy 1. Granular Legal Norms: The Demise of Typifications? Starting from this very brief diagnosis of social change, the proponents of persona- 7 lized law argue that the rise of the data-driven society could change the way how legal rules are being designed.9 Usually, legal norms formulate impersonal and abstract rules that are supposed to cover a large number of individual cases: To legislate means to generalize.10 A tool for generalizing commonly used by the legislator are so-called “typifications”. These are normative models that divide the infinite variations of the social world into certain categories that create meaningful order.11 Through the use of typifications, situations which are on closer inspection heterogeneous are typified as being homogeneous. A classic example for legal typification is the determination of legal capacity by reference to age in order to avoid making difficult inquiries into the actual cognitive faculty and maturity of a person.12 The rules on legal capacity do not take into account the actual maturity of judgment of an individual, but schematically define age limits. Similarly, the disclosure rules of consumer law do not take into consideration the informational needs of the individual consumer. Instead they are based on the model of the average consumer.13 The same is true for liability rules in contract and tort law where a generalized objective standard is used. The normative model which is usually applied in this context is the reasonable person test referring to the “normal, reasonable person of average competence”.14 However, the rather crude one-size-fits-all design of legal norms based on typifications suffers from a certain degree of imprecision. The underlying typifications represent only a blurred picture of reality and ignore what Oliver Wendell Holmes called the “personal equation” of the individuals.15 In mathematical terms, typifications only offer an approximate value.16 They are heuristic 9 This section draws on Busch, supra (fn. 1), in 86 Chi L Rev, at 313–14 and Busch/De Franceschi, supra (fn. 1), at 410–13. 10 See Kirchhof, Allgemeiner Gleichheitssatz, in: Isensee/Kirchhof (eds.), Handbuch des Staatsrechts der Bundesrepublik Deutschland, VIII, 3rd edn. 2010, 697, 773; see also Kelsen, Allgemeine Staatslehre, Berlin 1925, 231–32. 11 See Barber, Social Typifications and the Elusive Other: The Place of Sociology of Knowledge in Alfred Schutz’s Phenomenology, Lewisburg 1989 (on the use of typifications in sociology). 12 von Jhering, Der Geist des römischen Rechts auf den verschiedenen Stufen seiner Entwicklung, Part 1, Volume 1, Leipzig 1854, 53–4; see also Kennedy, Form and Substance in Private Law Adjudication, 89 Harvard Law Review 1685 at 1688–89 (1976). 13 CJEU, Case C-210/96 – Gut Springenheide and Rudolf Tusky [1998] ECR I-4657, para 31; see generally Klinck/Riesenhuber (eds.), Verbraucherleitbilder, Berlin 2015. 14 See, e.g., § 276(2) of the German Civil Code; Kötz/Wagner, Deliktsrecht, 13th edn., Munich 2013, at 114; see also Moran, Rethinking the Reasonable Person: An Egalitarian Reconstruction of the Objective Standard, Oxford 2003. The legislative method of typification has been expressed very clearly by Holmes, The Common Law, London 1881, at 108: “The standards of the law are standards of general application. The law takes no account of the infinite varieties of temperament, intellect, and education which make the internal character of a given act so different in different men.” 15 Holmes, supra (fn. 14), at 108. 16 One could argue that the reasonable man standard is an ‘aspirational norm’ which does not necessarily reflect reality, but provides an incentive to strive for a higher level of care. On aspirational norms, see Martin S. Flaherty, Rights, Reality and Utopia, 72 Fordham Law Review, 1789, 1791 (2003).

Busch

281

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

approximations that simplify the problems caused by the infinite complexity of the real world. However, the use of legal approximations leads to regulatory errors and inequities and cause potential losses of efficiency resulting from the over- and underinclusiveness of the normative models. The imprecisions resulting from typifications are only partially mitigated by applying group-specific standards and, in exceptional cases, typified adjustments. In hard cases, the inequities caused by coarse-grained typifications are compensated through general clauses.17 8 From an economic point of view, typifications can be conceptualized as means for reducing complexity costs.18 The more complex a legal norm is, the more difficult and thus costlier is rule drafting, ex ante compliance and ex post adjudication. Thus, in the past a higher degree of individual fairness can only be achieved at the price of less legal certainty or higher complexity costs. In this model, complexity costs are directly linked to the limited capacity of human information-processing. Thus, one could argue that the optimal complexity of legal rules – and the granularity of the entire legal system – is limited by the bounded capacity of human-information processing. From this perspective, one could conceptualize the widespread use of typifications as the answer to an information problem and a concession to the imperfections of a legal system administered by humans. If this is true, we might be heading for the “demise of typifications”. In the near future, big data, super-human information-processing capabilities and artificial intelligence could redefine the optimal complexity of legal rules and refine their content to a hitherto unachievable level of granularity. In such a scenario, personalized or “granular” legal rules could take into account actor heterogeneity to a degree impersonal laws are unable to do. As a result of granularization, regulatory errors stemming from over- and underinclusive norms based on coarse-grained typifications could be reduced. In particular, legal categories could be designed in a more precise and granular way taking into account insights from behavioural science and the available data about individual actors. For example, it has been suggested that personalized disclosures could replace standardized information in consumer law and data privacy law.19 Similarly, in a world of “quantified selves” rules on negligence could be tailored to the “personal equation” of individual actors.20 Some even envisage personalized default rules in family and inheritance law.21

2. Use Cases of Personalized Law 9

The following section will briefly discuss three use cases that illustrate how personalized law could be applied in digital marketplaces and smart cities. The first example relates to personalized disclosures, which have become a favorite topic among those interested in personalized law.22 The second example illustrates how personalization techniques could be used as a tool for mitigating discrimination in online marketplaces. Finally, the third example shows how personalized norms could be embedded into the regulatory environment of smart cities. 17 Auer, Materialisierung, Flexibilisierung, Richterfreiheit, Tübingen 2005, at 140; see also Schauer, The Convergence of Rules and Standards, 2003 N Z L Rev 303 at 308–09. 18 See, generally, Kaplow, A Model of the Optimal Complexity of Legal Rules, 11 Journal of Law, Economics & Organization 150 (1995). 19 See Busch, supra (fn. 1), in 86 Chi L Rev, at 319–24. 20 Ben‐Shahar/Porat, supra (fn. 1). 21 Porat/Strahilevitz, supra (fn. 1); see also Busch/De Franceschi, supra (fn. 1). 22 See, e.g., Porat/Strahilevitz, supra (fn. 1); Busch, in: Twigg-Felsner, supra (fn. 1), at 221; Hacker, supra (fn. 1).

282

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:47 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

a) Personalized Disclosures on Digital Marketplaces. One field of application, 10 which lends itself well to the use of personalized law, are mandatory disclosures.23 Through personalized disclosures, it could be possible to provide consumers with information that is tailored to their situations, personalities, demographic characteristics and cognitive capabilities. The provision of such behaviorally informed (personalized) information instead of standardized (impersonal) information could reduce the amount of information to be provided and, at the same time, increase the relevance of a disclosure for the individual recipient of the information. Under such a model, data on consumers’ purchasing habits and other patterns of past behavior could be used for reducing both the quantity problem of information overload and the quality problem of information mismatch that is associated with the one-fits-all approach to disclosure.24 A famous and often cited example that illustrates this use case is the retailer Target, 11 who used data mining to identify pregnant women among its customers.25 Target’s data miners observed that pregnant women were likely to buy certain nutritional supplements in their first trimester, unscented lotion in their second trimester, and hand sanitizer close to their due dates.26 Knowing that the birth of a child is a watershed moment in the customer relationship, when shopping behaviors are open to change and new brand loyalties are likely to emerge, Target used the information to send personalized advertising and coupons to the pregnant women.27 From a regulatory perspective, one could consider whether a retailer who has obtained such insights through data analytics should be obliged to use this information to provide consumers with targeted health warnings.28 For example, a customer with a high “pregnancy prediction score” could be confronted with a specific warning message if she buys alcoholic beverages or raw cheese in an online shop. Maybe this example seems a little bit creepy and overly paternalistic. This could 12 indeed be the case. Let me be clear: I am not saying that the law should require online retailers to identify pregnant customers and confront them with unwanted warnings. What I am saying is that the law could do this on the basis of data analytics. This is a regulatory option that was not available a few years ago. Therefore, it has to be decided in which cases it is appropriate to use the new type of data-driven disclosure mandates and where to draw a line. This is of course a policy question that may be subject to conflicting points of view. b) De-biasing Customer Ratings in the Sharing Economy. The second example of 13 algorithmic personalization relates to the design of online reputation systems which are a common feature in the so-called sharing economy.29 In particular, ridesharing platforms, such as Lyft and Uber, are using reputation systems that allow customers to rate 23

This section draws on Busch, supra (fn. 1), in 86 Chi L Rev, at 316–17. For a critical analysis of impersonal disclosure mandates see, e.g., Ben‐Shahar/Schneider, More Than You Wanted to Know: The Failure of Mandated Disclosure, Princeton 2014); see also Busch, in: TwiggFelsner, supra (fn. 1), at 221. 25 See Duhigg, How Companies Learn Your Secrets, NY Times Magazine, 16 February 2012, archived at http://perma.cc/8VGY-D93F. 26 Id. 27 Id. 28 See Busch, in: Twigg-Flesner, supra (fn. 1) at 234. 29 See, e.g., Tadelis, Reputation and Feedback Systems in Online Platform Markets, 8 Annual Review of Economics 321 (2016); Ranchordas, Online Reputation and the Regulation of Information Asymmetries in the Platform Economy, 5 Critical Analysis of Law 127 (2018); see also Busch, Crowdsourcing Consumer Confidence: How to Regulate Online Rating and Review Systems in the Collaborative Economy, in: De Franceschi (ed.), European Contract Law and the Digital Single Market, Cambridge 2016. 24

Busch

283

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

the work performance of drivers. After the service is rendered, customers are prompted to rate the service provider. These systems facilitate the semi-automated management of large, disaggregated and non-traditional workforces by outsourcing performance evaluations to customers.30 However, recent research indicates that rating systems can be a vehicle for workplace discrimination if customer’s biases creep into the evaluation for drivers.31 This is particularly problematic where work-related decisions by platform operators are based on customer ratings. For example, it has been reported that Uber deactivates drivers’ accounts if their average rating falls below 4.6 out of 5.32 Under antidiscrimination laws, employers are legally prohibited from making employment decisions based on protected characteristics of workers. However, reliance on potentially biased costumer ratings to make material determinations may actually lead to a disparate impact in employment outcomes.33 14 A number of regulatory interventions have been proposed for reducing the impact of biased rankings, such as raising awareness through establishing baseline statistics, reducing information available to raters or resituating the use of ratings within organizational structures.34 Another, more technological approach would involve statistical pattern recognition and automated correction of biased ratings. In such a model, the rating behavior of customers would be scrutinized in order to detect patterns that indicate bias. For example, as suggested by Rosenblat et al. ratings could be checked for significant “statistical disparities between ratings assigned by a particular rater to workers inside and outside a protected group, according to matched comparisons based on other observable attributes”.35 If evidence of bias is found, the biased ratings provided could be assigned lower weight or corrected by a personalized “de-biasing factor” in order to ensure that the biased ratings will not influence the aggregate rating of the platform worker.36 The example also shows the limits of algorithmic personalization as a regulatory instrument. Since, in practice, the implementation of such a model may be complicated by a lack of sufficient data that would allow a reliable detection of discriminatory rating patterns.37 15 Such a personalized de-biasing scheme could be implemented at various levels and does not necessarily require a public regulator to introduce such a system. Instead, one 30 It is a matter of controversy whether drivers working for ridesharing platforms can be categorized as employees. However, this question can be left open for the purposes of our analysis. See generally Prassl/ Risak, Uber, Taskrabbit and Co: Platforms as Employers? Rethinking the Legal Analysis of Crowdwork, 37 Comp Lab L Pol J 619 (2016). 31 Rosenblat/Levy/Barocas/Hwang, Discriminating Tastes: Uber’s Customer Ratings as Vehicles for Workplace Discrimination, 9 Policy and Internet 256 (2017); Ducato/Kullmann/Rocca, Customer Ratings as a Vector for Discrimination in Employment Relations? Pathways and Pitfalls for Legal Remedies, 2018, available at https://ssrn.com/abstract=3141156; see also Edelman/Luca/Svirsky, Racial Discrimination in the Sharing Economy: Evidence from a Field Experiment, 9 American Economic Journal: Applied Economics 1 (2017) (studying racial discrimination on Airbnb); see also Fisman/Luca, Fixing Discrimination in Online Marketplaces, 94 Harvard Business Review 88 (2016). 32 Prassl/Risak, supra (fn. 30), at 638. 33 Rosenblat et al., supra (fn. 31); see also Kullmann, Platform Work, Algorithmic Decision-Making, and EU Gender Equality Law, 34 International Journal of Comparative Labour Law and Industrial Relations, 1 at 11 (2018). 34 See, e.g., Dellarocas, Immunizing Online Reputation Reporting Systems Against Unfair Ratings and Discriminatory Behavior, Proceedings of the 2nd ACM Conference on Electronic Commerce. New York: ACM, 150–7 (2000) (suggesting a controlled anonymity scheme and cluster filtering); see also Rosenblat et al., supra (fn. 31), at 269–74. 35 See also Rosenblat et al., supra (fn. 31), at 269–74. 36 See also Fisman/Luca, supra (fn. 31) (suggesting that Uber could underweight ratings from those passengers, who have revealed themselves to be discriminatory, when calculating overall feedback scores). 37 See also Rosenblat et al., supra (fn. 31), at 271.

284

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

could imagine the implementation of a personalized de-biasing scheme by means private ordering.38 It would be no surprise if some platform operators themselves take the initiative to introduce personalized de-biasing as part of their corporate nondiscrimination policy. In fact, in the recent literature, platforms have been rightfully described as “the new governors”39 reflecting that many platform operators are quite active regulating communications and activities of platform users.40 Secondly, a requirement to implement a personalized de-biasing system could be included into an industry standard such as the recently adopted ISO 2018:20488 for online reviews.41 In fact, the new ISO standard already contains a provision according to which the platform operator “should use dedicated IT programs for analysis within the computer systems used to moderate content for the purpose of verifying its appropriate, relevant and impartial character by automated means”.42 While this provision does not yet explicitly require the use of any automated de-biasing system, the implementation of a personalized de-biasing mechanism would probably be in compliance with the ISO standard. c) Personalized Traffic Rules in the Smart City. Another use case of personalized 16 algorithmic regulation could be the deployment of personalized traffic rules in smart cities.43 The management and control of traffic flows through Advanced Traffic Management Systems (ATMS) is at the very core of research on data-driven urbanism. While most of the current literature focuses on the use of aggregate data of traffic streams, one could also consider the use of algorithmic personalization to implement a more granular regulatory policy. Indeed, such technologies are already being piloted in some cities. For example, in the Netherlands the city of Tilburg has recently been testing a new app called “Crosswalk” that alters crossing times at traffic lights based on the individual mobility needs of pedestrians.44 A sensor in the traffic lights constantly scans the pavement in its vicinity and if it recognizes a pedestrian with the app installed on her smartphone, it adjusts the green-light time. The app comes pre-installed with one of four time settings, depending on the user’s level of mobility, to minimize delays to other traffic. It is easy to conceive applications that apply similar technology for other applications. Dynniq, the Dutch company that has developed the “Crosswalk” app is also developing an app for cyclists called “CrossCycle” which will sense when bikes are approaching a junction and change the lights sooner. Another version of the app shall detect visually impaired pedestrians and individually activate the ticking sounds that tell 38 See, generally, Verstein, Privatizing Personalized Law, 86 U Chi L Rev 551 (2019) (discussing in which scenarios personalized lawmaking could be facilitated by non-state actors). 39 Klonick, The New Governors: The People, Rules, and Processes Governing Online Speech, 131 Harv L Rev 1598 (2018). 40 See, generally, Finck, Digital co-regulation: designing a supranational legal framework for the platform economy, 43 Eur L Rev 47 (2018); see also Busch, Self-Regulation and Regulatory Intermediation in the Platform Economy, in: Cantero Gamito/Micklitz (eds.), The Role of the EU in Transnational Legal Ordering: Standards, Contracts and Codes, Cheltenham 2020, 115–134. 41 ISO 20488:2018 (Online consumer reviews – principles and requirements for their collection, moderation and publication). 42 ISO 20488:2018, Section 6.5.3 (emphasis added). 43 There is little consensus regarding the definition of “smart city”. For an overview of the literature, see, e.g., Albino/Berardi/Dangelico, Smart Cities: Definitions, Dimensions, Performance, and Initiatives, 22 Journal of Urban Technology 3 (2015); Caragliu/Del Bo/Nijkamp, Smart Cities in Europe, 18 Journal of Urban Technology 65 (2014); see also Kichin/Lauriault/McArdle (eds.), Data and the City, New York 2018. 44 Darroch, The slow lane: Dutch app allows elderly to ‘hack’ traffic lights, The Guardian, 12 July 2017, https://www.theguardian.com/cities/2017/jul/12/dutch-app-elderly-hack-pedestrian-crossings; see also The pros and cons of placebo buttons, The Economist, 26 January 2019, https://www.economist.com/ science-and-technology/2019/01/26/the-pros-and-cons-of-placebo-buttons.

Busch

285

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

them whether the light is red or green. In the same vein, preferential parking permits could be issued to individuals based on relevant temporary or permanent health conditions or family circumstances (for example, driving young children).45 17 A common feature of the examples cited here is that they involve tailored norms which are embedded in the digital infrastructure or the “regulatory environment”46 of smart cities and are applied in real time.47 This shows that the prospect of algorithmic personalization is closely linked to the development of Internet of Things (IoT) infrastructure. At the same time, it is important to note, that the above cited examples do not necessarily involve “technological management”48 in the sense that traffic commands are superseded by technological traffic control that would leave no autonomous choice for citizens.

3. Fighting Fire with Fire: Adverse Targeting meets Personalized Law While the use cases of personalized law discussed above are still at an experimental phase or merely hypotheticals, businesses are using big data and artificial intelligence already today for personalizing their interactions with customers. In the emerging “personalized economy”, sellers are using big data and sophisticated algorithms to tailor their commercial advertising, products, and prices with the aim of targeting individual consumers. While the welfare effects of first-degree price discrimination and the need for regulatory action against personalized pricing are a matter of controversial debate,49 there is little doubt that the law should set limits to the use of personalization tools that systematically target and exploit idiosyncratic vulnerabilities of individual consumers.50 It is less clear, however, which regulatory instruments should be used and where to draw the line between “smart” personalization techniques that are still acceptable and those that require regulatory action. 19 Consider, for example, that businesses with access to social media data can use information harvested from Facebook posts for individually targeting consumers in specific life situations. Thus, advertising for online dating services could be displayed to users who have just changed their relationship status from “in a relationship” to “single” or “it’s complicated”. While this may be considered an acceptable way of targeting consumers, big data and artificial intelligence can be utilized for more problematic practices. As Ryan Calo underlines, “trouble arises when firms start looking at the consumer behavior dataset to identify consumer vulnerabilities”.51 20 Indeed, new technologies allow businesses to exploit informational asymmetries and consumer biases in novel ways and open the door to the darker side of personalization in algorithmic marketplaces.52 Recent research shows that the psychological character18

45 See Elkin-Koren/Gal, The Chilling Effect of Governance-by-data on Innovation, 86 U Chi L Rev 401 (2019); see also Casey/Niblett, supra (fn. 1). 46 Brownsword, In the Year 2061: From Law to Technological Management, 7 Law, Innovation and Technology 1 (2015). 47 See generally Kitchin, The Realtime City? Big Data and Smart Urbanism, 79 GeoJournal 114 (2014). 48 Brownsword, supra (fn. 46). 49 See, e.g., Bar‐Gill, Algorithmic Price Discrimination When Demand Is a Function of Both Preferences and (Mis)perceptions, 86 U Chi L Rev 217 (2019); Bourreau/de Streel, The Regulation of Personalized Pricing in the Digital ERA, OECD Note, DAF/COMP/WD(2018)150. 50 Calo, supra (fn. 8), at 1010; Wagner/Eidenmüller, supra (fn. 8). 51 Calo, supra (fn. 8), at 1010. 52 Wagner/Eidenmüller, supra (fn. 8); Hacker, supra (fn. 1), at 88–92; Helberger, Profiling and Targeting Consumers in the Internet of Things: A New Challenge for Consumer Law, in: Schulze/ Staudenmayer (eds.), Digital Revolution: Challenges for Contract Law in Practice, Baden-Baden 2016, 135, 151–52.

286

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

istics of consumers can be quite accurately predicted from their “digital footprints”, such as Facebook likes, Tweets, or Instagram posts.53 For example, a recent study demonstrated that machine learning tools using color analysis, metadata components, and algorithmic face detection can make it possible to identify markers of depression in Instagram posts.54 These results could provide new avenues for early screening and detection of mental illness. However, they could also enable businesses to exploit the vulnerabilities for adverse targeting and manipulative marketing techniques. Thus, individual customers could be addressed in their weakest moments. Existing contract law rules and doctrines, such as unconscionability, undue influence, 21 misrepresentation and duress may provide some minimum protection for consumers in extreme cases.55 But they fail in cases where businesses use more subtle techniques to target psychologically instable consumers.56 Another option for regulatory action would be to address the foundation of exploitative algorithmic techniques: big data. As adverse targeting is fueled by big data, this digital manipulation could be curbed by limiting sellers’ access to information about consumers. But, as Oren Bar-Gill argues in the context of personalized pricing, attacking the big data foundation “runs the risk of throwing the baby out with the bathwater. Given the benefits that personalization provides, cutting the flow of information might be a net loss for consumers.”57 A different solution has recently been proposed by Gerhard Wagner and Horst 22 Eidenmüller, who suggest that consumers whose choices have been influenced by big data and artificial intelligence tools should be granted a special right to withdraw from the transaction.58 They argue that such a remedy would be a justifiable extension of existing withdrawal rights under European Union (EU) law, which address exogenous and endogenous distortions of preferences. Generally speaking, in the EU mandatory rights of withdrawal exist for three types of scenarios: (1) distance contracts, for example, contracts concluded by phone or on the Internet; (2) off-premises contracts, for example, contracts concluded at the consumer’s home or at her place of work; and (3) certain types of contracts that the legislator considers to be particularly complex, for example, insurance contracts and the sale of timeshares.59 The proposal by Wagner and Eidenmüller essentially means taking the existing 23 withdrawal rights, which are based on coarse-grained typifications of contracting situations or contract types that typically raise issues of impaired consumer choice as a starting point and transforming them into more tailored instruments of consumer protection. However, as a start, it is not entirely clear whether the new withdrawal right shall apply in all cases where business utilize big data and artificial intelligence tools to 53 Stillwell et al., Psychological targeting as an effective approach to digital mass persuasion, Proceedings of the National Academy of Sciences of the United States of America, 114 (48), 12714 (2017). 54 Reece/Danforth, Instagram photo reveal predictive markers of depression, 6 EPJ Data Science 15 (2017). 55 Wagner/Eidenmüller, supra (fn. 8). 56 But see Helleringer, Profiling and Targeting Consumers in the Internet of Things, at 153–60 (suggesting that unfair commercial practices law might offer more flexible solutions); see also Pałka/ Jabłonowska/Micklitz/Sartor, Before Machines Consume the Consumers, EUI Department of Law Research Paper No. 2018/12, 6. 57 Bar‐Gill, supra (fn. 49). 58 Wagner/Eidenmüller, supra (fn. 8). 59 See generally Watson, Withdrawal Rights, in: Twigg-Flesner (ed.), Research Handbook on EU Consumer and Contract Law 241, Cheltenham 2016; Twigg-Flesner/Schulze/Jonathon Watson, Protecting Rational Choice: Information and the Right of Withdrawal, in: Howells/Ramsay/Wilhelmsson (eds.), Handbook of Research on International Consumer Law, Cheltenham 2018, 111; for a critical analysis see Eidenmüller, Why Withdrawal Rights?, 7 European Review of Contract Law 1 (2011); see also Ben‐ Shahar/Posner, The Right to Withdraw in Contract Law, 40 J Legal Stud 115 (2011).

Busch

287

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

influence consumers or only in the more problematic cases where particular vulnerabilities are exploited. The first case would almost certainly translate into a massive and undesired proliferation of withdrawal rights as the use of big data and artificial intelligence becomes common practice among businesses. The second case, which appears more plausible, would probably raise enforcement problems which will be addressed in Part IV of this Essay.

IV. Governance of Algorithms for Personalized Law 24

The remainder of this Essay explains why personalized part can be conceptualized as a form of algorithmic regulation and identifies some key elements of a governance framework for personalized law regimes. It is argued that it will be necessary to take privacy seriously, preserve autonomy and ensure the quality of data and the accuracy of the statistical model on which a personalized norm is built. In addition, a transition towards personalized law will require setting up an appropriate framework of compliance monitoring and algorithmic auditing.

1. Personalized Law as Algorithmic Regulation The use cases of personalized law analyzed in Part III of this Essay show that personalized law is typically based on algorithmic processes. It can therefore be described as a form of algorithmic regulation.60 As Brauneis and Goodman explain, an algorithmic process typically involves “(1) the construction of a model to achieve some goal, based on analysis of collected historical data; (2) the coding of an algorithm that implements this model; (3) collection of data about subjects to provide inputs for the algorithm; (4) application of the prescribed algorithmic operations on the input data; and (5) outputs in the form of predictions or recommendations based on the chain of data analysis.”61 26 In the example of personalized disclosures discussed above,62 these five elements of an algorithmic process are present: (1) The personalized health warnings would be based on model that builds on correlations between particular purchasing patterns and pregnancy that have been observed in historical data. (2) On the basis of these correlations, an algorithm is coded that implements this model. In the example, the algorithm was developed by Target for the purpose of personalized advertising. The same algorithm could be repurposed for delivering personalized health warnings. (3) In order to identify customers that need a specific health warning, the purchasing history of individual customers is recorded. This data serves as input for the algorithm. (4) The algorithm is applied on the customer data; and (5) outputs a “pregnancy prediction 25

60 See generally Yeung, Algorithmic Regulation: A criticial Interrogation, 12 Regulation & Governance 505 (2018) (defining algorithmic regulation as “decisionmaking systems that regulate a domain of activity in order to manage risk or alter behavior through continual computational generation of knowledge from data emitted and directly collected (in real time on a continuous basis) from numerous dynamic components pertaining to the regulated environment in order to identify and, if necessary, automatically refine (or prompt refinement of) the system’s operations to attain a prespecified goal”); see also O’Reilly, Open Data and Algorithmic Regulation, in: Goldstein/Dyson (eds.), Beyond Transparency: Open Data and the Future of Civic Innovation, San Francisco 2013, 289. 61 Brauneis/Goodman, Algorithmic Transparency for the Smart City, 20 Yale J L & Tech 103 (2018) at 113–4; see also Zarsky, Transparent Predictions, 2013 U. Ill. L. Rev. 1503, 1517–20; Gillespie, The Relevance of Algorithms, in: Gillespie et al. (eds.), Media Technologies: Essays on Communication, Materiality, and Society, 2014, 167; Kroll et al., Accountable Algorithms, 165 U Pa L Rev 633, 640 (2017). 62 See text accompanying supra fns 23–28.

288

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:48 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

score”. On the basis of this score, it is decided whether a personalized health warning is displayed or not. While this system seems not overly complex, the implementation of personalized law 27 can even be simpler. As illustrated by the example of the “Crosswalk” app in Tilburg,63 the application of personalized law does not necessarily require any complex analysis of input data. In this example, the personalization is made possible simply by real time communication technology that enables direct interaction between traffic lights and the apps installed on the smartphones of individual pedestrians.

2. Privacy and Choice Personalized law is built on user profiling. Therefore, it is obvious that this regulatory 28 model raises privacy concerns.64 One could even ask whether the classic conflict between legal certainty and individual fairness, which personalized law purportedly is meant to solve,65 is just replaced by a new conflict between individual fairness and privacy. From a privacy perspective, within the EU a system of personalized law would have 29 to comply with Article 8(1) of the Charter of Fundamental Rights of the European Union (CFR) and Article 16(1) Treaty on the Functioning of the European Union (TFEU) which both guarantee the protection of privacy. At the level of secondary legislation, these fundamental principles are mainly implemented by the General Data Protection Regulation (GDPR). Therefore, a system of personalized law introduced in the EU would have to be in line with the principles laid down by the GDPR. First, the GDPR would require the enactment of a law sanctioning the collection and data for the purpose of personalization.66 Second, it is important to note that under the GDPR customer profiling is not prohibited as such. However, Article 22(1) GDPR gives every natural person the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.67 This provision is subject to several exceptions spelled out in Article 22(2) GDPR. In particular, Article 22(2)(c) GDPR allows measures based on profiling if the processing of data is based on the data subject’s explicit consent.68 In line with the principle volenti non fit iniuria, which is the foundation of Article 22 30 (2)(c) GDPR, some applications of personalized law could be designed as an opt-in model.69 This regulatory technique would probably be most suitable for personalized disclosures. Under an opt-in regime the consumer would have the right to choose between impersonal and personalized information. As a consequence, the degree of 63

See text accompanying supra fns 43–48. This section draws on Busch, supra (fn. 1), in 86 Chi L Rev, at 326. 65 See text accompanying supra fns 18. 66 See Article 6(3) GDPR and Recital 45 of the GDPR. 67 On the scope of Article 22 GDPR see Wachter/Mittelstadt/Floridi, Why a Right to Explanation of Automated Decision-Making Does not Exist in the General Data Protection Regulation, 7 International Data Privacy Law 76 (2017); see also Edwards/Veale, Slave to the Algorithm? Why a “Right to an Explanation” is Probably Not the Remedy You Are Looking For, 16 Duke Law and Technology Review 18 (2017). 68 For these cases, Art. 22(3) GDPR requires the data controller “shall implement suitable measures to safeguard the data subject’s rights and freedoms and legitimate interests”. This can be interpreted to the effect that the GDPR requires the implementation of discrimination-aware profiling techniques, see Veale/Edwards, Clarity, surprises, and further questions in the Article 29 Working Party draft guidance on automated decision-making and profiling, 34 Computer Law & Security Review 398, 403 (2018). 69 Busch, in: Twigg-Flesner, supra (fn. 1) at 237–8. 64

Busch

289

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

personalization of the information provided to the individual consumer would depend on her preference for privacy. This approach would reflect actor heterogeneity and take into consideration that different consumers may have different attitudes to privacy. If a consumer prefers the benefits of personalized law, she must accept customer profiling. If, however, a consumer is not willing to accept the processing of personal data for the purpose of customer profiling, she voluntarily forgoes the benefits of personalized law. In other use cases, such the automated de-biasing of customer ratings, it may be justified to apply personalization schemes irrespective of the data subject’s explicit consent.

3. Quality of Data and Models Considering that personalized law is a form of governance-by-data, the quality of any relevant training data as well as the quality of the input data on which the prescribed algorithmic operations are applied is of key importance. The use of polluted or partial data could lead to inaccurate profiling of individuals and thus distort the application of personalized norms.70 As profiling is based on statistical techniques, it is vulnerable to the problem of false positives and false negatives. As a result, an individual could be treated as part of a segment to which she does not in fact belong (false positive or type I error) or treat her as a person who does not belong to a category to which she in fact does (false negative or type II error).71 The impact of such errors is very much contextdependent. In the case of personalized health warnings, the costs of a type I error are relatively modest, if a warning message about alcoholic beverages is displayed to a customer who was wrongly identified as “probably pregnant”. By contrast, both type I and type II errors would have a distorting impact on online reputation systems in the case of automated de-biasing of customer ratings. The risk of erroneous classifications is particularly high if only little relevant data is available. Therefore, the implementation of personalized law regimes could be made conditional on a certain minimum amount of available data that allows sufficiently reliable inferences. 32 The reliability of personalized law not only depends on the quantity and quality of the data that is used for training and input, but also on the accuracy of the underlying model. The standard of accuracy (i.e. the strength of the empirical correlations) that is required may vary from case to case depending on the intrusiveness of the norm. For personalized rules concerning issues of health and safety a higher standard of accuracy (i.e. a lower misclassification rate) will be required than for information duties of lesser importance. 31

4. Compliance Monitoring and Algorithmic Auditing 33

Finally, a transition from coarse-grained typifications to highly-granular personalized law has also consequences on the level of compliance and enforcement.72 Monitoring compliance with uniform rules such as standardized information duties is, as a general rule, rather simple. Enforcement authorities such as the FTC in the United States or other market participants in countries with a system of decentralized enforcement of consumer law (for example, Germany) only have to verify whether the information provided by a trader complies with the list of disclosure items defined by the law. Compliance is even simpler if the law requires the use of certain standard forms for 70 Elkin-Koren/Gal, supra (fn. 45); see also Casey/Niblett, A Framework for the New Personalization of Law, 86 U Chi L Rev 333 (2019). 71 Hildebrandt/Koops, The Challenges of Ambient Law and Legal Protection in the Profiling Era, 73 Modern Law Review, 433–4 (2010). 72 This section draws on Busch, supra (fn. 1), in 86 U Chi L Rev, at 328–30.

290

Busch

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

N. Algorithmic Regulation and (Im)Perfect Enforcement

informing consumers, such as the Standard European Consumer Credit Information (SECCI)73 or the European Standard Information Sheet for Mortgage Credit (ESIS).74 In contrast, monitoring compliance with personalized information duties is more 34 complex. In the above-mentioned example of personalized health warnings based on a “pregnancy prediction score”,75 the question whether an online retailer has to display a personalized warning message depends on the data available to the company about the customer’s purchasing history. Similarly, the automated de-biasing of customer ratings depends on correctly identifying patterns of discrimination in the rating history. Consequently, compliance monitoring in these cases would involve testing whether the business used the data which was available and has made the right inferences on the basis of the data set. Similarly, the idea of personalized withdrawal rights as a remedy against adverse 35 targeting, although appealing at first view, raises difficult enforcement problems. While many European consumers have a general knowledge about the existence of withdrawal rights in case distance contracts and off-premises contracts, the applicable rules will be less obvious if they depend on “stealth infringements”,76 i.e. the use of exploitative marketing techniques that are specifically targeted at an individual consumer. In other words, a consumer will almost certainly be unaware that her choice has been influenced by targeted sales techniques that exploit her idiosyncratic vulnerabilities and that the law therefore grants her a special right of withdrawal. Therefore, the consumer will not be aware of her withdrawal right if a rogue trader does not provide her with the legally required information about the withdrawal right. Monitoring whether a business that utilizes algorithmic targeting techniques complies with the duty to inform the most vulnerable consumers about their personalized withdrawal rights, would require that an enforcement authority opens the seller’s “black box” and checks whether he has sensitive information about individual vulnerabilities. From a market control perspective, the implementation of personalized will con- 36 siderably increase the complexity of the “compliance landscape” and may lead to an “atomization” of market practices. Therefore, it is much more difficult and maybe even impossible for private actors such as consumer organizations to monitor whether a business complies with the applicable rules. As a consequence, effective enforcement of personalized law most probably requires some form of public enforcement. From a practical perspective, compliance monitoring would also require that the enforcement authority performs regular algorithm audits in order to ensure that the personalization algorithms perform as provided by the law (for example, use the right criteria for generating personalized disclosures). Such audits would also have to cover the data pools that are used for profiling customers in order to assess the validity of data and to ensure that the data is unbiased.

V. Conclusion Technological advances in data collection and data science could be used to tailor 37 legal norms to specific individuals. Thus, regulatory errors resulting from the over- and 73 See Annex II of Directive 2008/48/EC of the European Parliament and of the Council of 23 April 2008 on credit agreements for consumers, OJ L 133/66 (2008). 74 See Annex II of Directive 2014/17/EU of the European Parliament and of the Council of 4 February 2014 on agreements for consumers relating to residential immovable property, OJ L 60/34 (2014). 75 See supra paras 10–12. 76 Pałka et al., Before Machines Consume the Consumers, supra (fn. 56), at 5.

Busch

291

Reemers Publishing Services GmbH O:/Beck/Busch_978-3-406-74391-7/3d/14_Busch, Algorithmic Regulation and (Im)Perfect Enforcement in the Personalized Economy, 2019-11-26.3d from 30.10.2020 11:37:49 3B2 9.1.580; Page size: 160.00mm  240.00mm

Part 4. Technological and Behavioral Perspectives

under-inclusiveness of coarse-grained legal rules can be reduced and the level of regulatory precision can be increased. Personalized law could even be used as in instrument for limiting the adverse impact of personalization techniques utilized by firms for targeting particular vulnerabilities of individual consumers. However, as a form of algorithmic regulation or governance-by-data, personalized law is itself vulnerable to the limitations of algorithmic processes. The precision of personalized law very much depends on the quality and quantity of data which fuels the personalization machine and the accuracy of the underlying model. Therefore, personalized law offers not a blueprint for “perfect enforcement”77 but is subject to the imperfections that are inherent in algorithmic regulation. Therefore, it is necessary that a transition to personalized law is combined with an effective regulatory framework ensuring good governance of algorithms for personalized law. 77 Zittrain, Perfect Enforcement on Tomorrow’s Internet, in: Brownsword/Yeung (eds.), Regulating Technologies: Legal Futures, Regulatory Frames and Technological Fixes, Oxford 2008, 125.

292

Busch