312 12 19MB
English Pages 334 [339] Year 1999
F
cfion, election, jg and Uesion Edited by
David J. Buller Of
Function, Selection, and Design
Function, Selection, and Design
EDITED BY
David J. Buller
State University of New York Press
Published by State University of New York Press, Albany
©1999 State University of New York
All rights reserved Printed in the United States of America
No part of this book may be used or reproduced in any manner whatsoever without written permission. No part of this book may be stored in a retrieval system or transmitted in any form or by any means including electronic, electrostatic, magnetic tape, mechanical, photocopying, recording, or otherwise without the prior permission in writing of the publisher. For information, address State University of New York Press, State University Plaza, Albany, N.Y., 12246 Production by Diane Ganeles Marketing by Dana Yanulavich Library of Congress Cataloging-in-Publication Data
Function, selection, and design / David J. Buller, editor. p. cm. — (SUNY series in philosophy and biology) Includes bibliographical references and index. ISBN 0-7914-4211-X (alk. paper). — ISBN 0-7914-4212-8 (pbk.: alk. paper) 1. Biology—Philosophy. 2. Teleology. I. Buller, David J., 1959. II. Series. QH331.F895 1999 501—dc21 98-47343 CIP
10 987654321
J
Contents
vii
Preface
1
Introduction
Natural Teleology David J. Buller
Chapter 1
Functions Larry Wright
29
Chapter 2
Functional Analysis Robert Cummins
57
Chapter 3
Proper Functions Ruth Garrett Millikan
85
Chapter 4
Functions John Bigelow and Robert Pargetter
97
Chapter 5
An Ambiguity in the Notion ‘Function’ Ruth Garrett Millikan
115
Chapter 6
The Teleological Notion of ‘Function’ Karen Neander
123
Chapter 7
Functional Analysis and Proper Functions Paul E. Griffiths
143
Chapter 8
Function and Design Philip Kitcher
159
Chapter 9
Functions: Consensus Without Unity Peter Godfrey-Smith
185
v
vi
Contents
Chapter 10
A Modern History Theory of Functions Peter Godfrey-Smith
199
Chapter 11
Malfunctioning Karen Neander
221
Chapter 12
The Misuse of Sober’s Selection for/Selection of Distinction Richard Goode and Paul E. Griffiths
233
Biological Function, Adaptation, and Natural Design Colin Allen and Marc Bekoff
243
Chapter 13
Chapter 14
A Taxonomy of Functions Denis Walsh and Andre Ariew
Chapter 15
Etiological Theories of Function: A Geographical Survey David J. Buller
257
281
References
307
Contributors
319
Index
321
1
Preface
The principal focus of this volume is the so-called etiological theory of functions, pioneered by Wright and institutionalized by Millikan, Neander, and Godfrey-Smith. This is not to say, however, that the etiological theory is the sole focus of the volume; indeed, several chapters offer criticisms of and alternatives to the etiologi cal theory. But I have found that even the critiques share certain common commitments with the etiological theory. And it is these common commitments that unite the essays in this volume, as much as the cross-referenced debates running through the essays. One of the many interesting things about the recent philo sophical literature on biological functions is the way in which dis cussion has evolved. Consequently, I have arranged the essays ir chronological, rather than thematic, order, so that the reader cai get a vivid sense of the recent evolution of debate. In addition, j have provided a historical introduction to the volume that is pitched at the level of an advanced student without any prior acquaintance with the philosophy of biology. It is my hope that this introduction will provide orientation to the philosophical problems concerning teleological concepts in biology, an overview of the his tory of efforts to solve those problems in the period immediately preceding the essays reprinted here, a sense of the significance of the etiological theory, and a grasp of the principal issues that con stitute the foci of the debates running through these essays. (While this volume thus focuses on the philosophical literature on biologi cal functions, Allen, Bekoff, and Lauder (1998) provide a collection of interleaved philosophical and biological essays on the topic.) This volume was made possible only through the cooperation and efforts of many. First, I would like to thank all the authors for their kindness and generosity in allowing me to reprint their arti cles. In addition, several copyright administrators and editors at
vii
viii
Preface
various presses and journals were particularly accommodating and easy to work with in the process of obtaining permissions to reprint the essays that follow. In particular, for their kind assistance I would like to thank Sharon Boyle of the University of Calgary Press, Melissa Connerton of Blackwell Publishers (US), Claire Lloyd of Blackwell Publishers (UK), Joolz Longley of Oxford Uni versity Press (UK), Pamela Quick of The MIT Press, John Smylie of The Journal of Philosophy, Berendina van Straalen of Kluwer Academic Publishers, Howard Wettstein of Midwest Studies in Phi losophy, and Robert Young of the Australasian Journal of Philoso phy. For their very generous financial support in making the vol ume possible I would like to thank the Chair of the Department of Philosophy, the Dean of the College of Liberal Arts and Sciences, the Dean of the Graduate School, and the Graduate Council on Research and Artistry at Northern Illinois University. I would also like to thank David Hull for his advice, guidance, encouragement, and support of the project. In addition, I would like to thank Karen M. Blaser of the Word Processing Department at Northern Illinois University for her superb and timely work in assisting in the prepa ration of the manuscript, and James Hudson for his comments on the introduction. And last, but far from least, I would like to thank my wife, Heide Fehrenbach, for her editorial comments on the introduction—and so much else.
.J
INTRODUCTION
Natural Teleology David J. Buller
I. Introduction. Within the past decade a near-consensus has emerged among philosophers concerning how to understand teleological concepts in biology. This is not to say, of course, that there is complete agree ment; but broad agreement about certain fundamental commit ments can be discerned in the recent literature, and the essays reprinted in chapters 3-15 of this volume exemplify this consensus. Against the background of this core agreement, several issues stand out as the foci of recent debates, and the essays in this vol ume have also been selected to exemplify these debates. In this introduction, I want to identify both the core consensus regarding a theory of biological teleology and the dimensions of debate over the peripheral details of such a theory. But first some stage setting is in order.
II. The Philosophical Problem Since all theories, philosophical or otherwise, are designed to solve certain problems, it is worth being clear about the problem the essays in this volume address before examining their proposed solutions. Broadly speaking, there are two sources of the problem of biological teleology, one in the philosophy of science and one in metaphysics. I will take these in turn.
1. Philosophy of Science. Philosophy of science developed in this century as an offshoot of epistemology concerned specifically
1
4
David J. Buller
orchids, the efficient causes of the release of the pheromone are the biochemical processes that trigger its release. From the perspective of Aristotelian metaphysics, however, efficient causation does not provide a complete explanation of why any given orchid releases a mimetic sex pheromone. A complete explanation of why a thing is requires citing its final cause, “that for the sake of which” it exists (as Aristotle repeatedly put it in the Physics and Metaphysics}; and the release of the mimetic sex pheromone is “for the sake of’ polli nation, since it tricks a male thynnine wasp into thinking that in landing on the orchid it is landing on a female wasp that is ready to mate, and this results in the orchid’s pollen being dispersed by the male wasp. A complete explanation of why orchids release the mimetic sex pheromone, according to Aristotelian metaphysics, involves the claim that they release the pheromone in order to dis perse their pollen. It is thus final causation that is teleological, since the final cause of a thing is its purpose or goal, “that for the sake of which” it exists. Aristotelian metaphysics did not restrict the concept of final causation to biological phenomena. Indeed, it was applied freely to physical phenomena as well; according to Aristotelian metaphysics, all motion and change in nature was teleological, occurring “for the sake of some end. This was the most immediate source of problems for Aristotelian metaphysics with the rise of the scientific revolu tion. For the “corpuscularian” or “mechanical philosophy” sought to explain all physical phenomena as the product of particles of mat ter (called “corpuscles”) in motion, and to explain the motion of a particle of matter as the product of the direct actions of other par ticles on it (Matthews, 1989). In addition, according to the mechan ical philosophy, final causes do not do any genuine explanatory work over and above the work already performed by explanations of motion solely in terms of particles in motion acting on one another. The physical universe thus came to be seen as purely mechanical and all change within it as wholly explicable in terms of antecedent events. The image was that of a clock, whose spring unwinds and turns the gears that drive the entire mechanism inex orably forward in time. And once physical phenomena became con ceived in this way, the door was open for philosophers such as Hobbes to extend the same mechanical conception of causation not only to biological phenomena, but to mental phenomena as well (Burtt, 1964). The mechanical philosophy thus viewed all causation as efficient causation working forward in time. In this world view, pollination is merely an effect, rather than a cause, of an orchid’s
ii
Natural Teleology
5
releasing its mimetic sex pheromone, since the pollination occurs after the release of the pheromone. As a consequence of the rise of science and the mechanical phi losophy, Aristotelian goal-directed final causation came to be seen as suffering from one or both of two principal difficulties. First, it appeared to put the cart before the horse—explaining a cause in terms of its effects—and thus to require “backward" causation. This conflicted with the mechanical philosophy’s conception of all change as explicable in terms of the motions of particles and of all of a par ticle’s motions as explicable in terms of the direct actions on it of other particles in motion. For the dispersal of pollen from an orchid cannot set in motion any actions of particles that would initiate the release of the mimetic pheromone, since it occurs after the release of the pheromone. Thus, pollination can in no way explain the pheromone’s release. So goal-directed final causation seems physi cally impossible and, hence, not explanatory. There is, however, one clear domain in which goal-directed causation seems both possible and explanatory. Intelligent beings represent to themselves the goals that they pursue in their deliber ations and actions, and these representations of goals are among the efficient causes of their deliberations and actions. For example we can explain my going to the freezer and retrieving the ice crear in terms of my desire to eat ice cream (and my belief that ice crear. is in the freezer). In offering such an explanation, we are not claim ing that the event of my eating ice cream caused the earlier event of my retrieving the ice cream from the freezer. Rather, we are claiming that my desire to eat ice cream was a cause of my retriev ing the ice cream; and this desire was temporally prior to the event that it is invoked to explain (my retrieving the ice cream). Of course, what makes my desire the desire to eat ice cream, rather than some other desire, is the fact that it involves a representation of my eating ice cream, rather than a representation of something else. Such explanations of the deliberations and actions of intelli gent beings, then, are paradigms of explanation in terms of goaldirected causation; but they do not actually require backward cau sation, since they are always formulated in terms of (antecedent) efficient causes that represent some goal. If this form of goal-directed causation is extended to the full range of cases to which Aristotelian metaphysics applied the con cept of final causation, however, a second problem arises. For this would require that sweat glands somehow represent the goal of low ering body temperature; and this, in turn, would require that sweat
6
David J. Buller
glands be “intelligent” in some sense. Similarly, it would require that the thymus somehow “know” that it must manufacture lym phocytes and that the orchid somehow have “figured out” what it must do in order to deceive male thynnine wasps into dispersing its pollen. In short, treating all final causation as goal directed in the way that intelligent behavior is appears to require that intelligence pervade the physical universe, being present in the most unlikely plants and organs of organisms. Thus, once the mechanical philosophy took hold and efficient causation was seen as sufficient to explain all change in nature, final causation appeared to be caught on the horns of a dilemma: either it involves the unacceptable postulation of backward causa tion or it involves a grossly implausible panpsychism. In short, there appeared to be no room at all for teleology, or purpose, within the scientific world view. And ever since the rise of this scientific world view, the metaphysical problem of teleology has been that of explaining whether, and if so how, there can be goal-directed pro cesses in a universe governed solely by efficient causation. In spite of the difficulties associated with teleology, however, biologists continued to use the teleological concept of function in describing the characteristics of organisms, finding the organiza tion of organisms and the operation of their parts virtually incom prehensible in strictly non-teleological terms. As Darwin wrote in a letter to A. de Candolle, it is “difficult for any one who tries to make out the use of a structure to avoid the word purpose” (quoted in Ghiselin, 1997, p. 63). This poses the following problem, which derives from the two sources just discussed: How can the biological concept of function, which is prima facie infected with final causa tion, be analyzed so as to make it compatible with a scientific world view that countenances only efficient causation? The problem is by no means easy. In the first place, the bio logical concept of function discriminates among the effects of an item: pumping blood is a function of the heart, but making noise is not; making noise is merely an “accidental” effect of the heart’s pumping. So any solution to the problem of biological teleology must explain why the concept of function discriminates between functional and accidental effects in this way. Second, the biological concept of function does seem to imply that the effects that it is the function of an item to produce in some sense explain the existence of that thing: saying that the function of sweat is to maintain con stant temperature implies that the need for constant body temper ature in some way explains why mammals sweat. This actually
"J
Natural Teleology
7
works together with the first point. The reason why pumping blood is a function of the heart, whereas making noise is not, is that its pumping blood in some sense explains why an organism has a heart, whereas its making noise does not. So any solution to the problem needs to account for how the biological concept of function can be explanatory in this way. But, third, in providing this account, the biological concept of function must be shown not to require the postulation of backward causation or panpsychism, or else it cannot be a proper scientific concept. In short, then, the prob lem of biological teleology is to explain how, for example, the claim that the function of kidneys is to remove metabolic wastes from the blood—that kidneys exist in order to do that—can in any way explain why animals have kidneys, when explanations must cite only efficient causes.
III. Recent Prehistory: The “State of the Art” in the 1960s
This was the problem that both Nagel (1961) and Hempel (1965) tried to solve within the framework of the deductive-nomological model of explanation (the classic formulation of which is to be found in Hempel & Oppenheim, 1948). According to this model, all explanations conform to the same general logical form: they all explain some phenomenon by deducing its existence or occurrence from a set of premises that include lawlike (nomological) regulari ties. When this logical model of explanation is conjoined with the assumption that all causation is efficient causation, the problem of analyzing the biologist’s concept of function in statements such as “the function of X is to Y” takes the following form: How can the existence of Xs be deduced from lawlike statements that include the fact that Ys are the effect of Xs? For example, how can the presence of kidneys in humans be deduced from lawlike statements that include the fact that kidneys have the effect of removing metabolic wastes from the blood? Both Hempel and Nagel started from the idea that a state ment such as (1) The function of the heart is to circulate the blood
is an elliptical, or shorthand, explanation of the existence of hearts. The task, then, is to make the explanation fully explicit and show how it conforms to the deductive-nomological model.
8
David J. Buller
Doing so requires noting several things. First, (1) is implicitly rel ativized to organisms with hearts, in particular vertebrates; that is, (1) is a claim about the function of the heart in the kinds of organism that possess hearts. Second, the heart can perform this function in some organism only in virtue of its having a particu lar anatomical organization; it is only in the context of the struc ture of the circulatory system as a whole, for example, that hearts can function to circulate the blood. Third, to put it crudely, the heart only performs this function in live vertebrates, and verte brates depend for life on particular chemical properties of their external environments. When we make these three things explicit, (1) becomes (2) The function of the heart in vertebrates with a particu lar anatomical organization and in a particular envi ronment is to enable them to circulate their blood.
But (2) is merely an instance of the following general schema, as Nagel pointed out:
(3) “The function of A in a system S with organization C is to enable S in the environment E to engage in process P” (Nagel, 1961, p. 403; cf. Hempel, 1965, p. 306).
This, then, was taken to be the fully explicit form of statements that describe the function of some trait of an organism. This just leaves the problem of how to convert a statement with the form of (3) into a fully explicit deductive-nomological explanation of the existence of the entity A in the system S. Since Hempel and Nagel offered very similar solutions to this problem, I will focus on Nagel’s solution and discuss Hempel’s in relation to it. Nagel proposed that the informational content of (3) can be formu lated as the following deductive argument, which, according to the deductive-nomological model of explanation, constitutes an expla nation of the existence of A in system S (1961, p. 403):
(a) “Every system S with organization C and in environ ment E engages in process P,” (b) “if S with organization C and in environment E does not have A, then S does not engage in P" (c) “hence, S with organization C must have A.’
Natural Teleology
9
Here (a) provides a law to the effect that every S (e.g. vertebrate) does P (circulate blood), and (b) states that having A (a heart) is a necessary condition for doing P. From these two premises it deduc tively follows that S must have A. Consequently, if we make the substitutions for the variables as just noted, we appear to have an explanation of the existence of hearts in vertebrates that appeals to the fact that hearts have the effect of circulating blood. And Hempel and Nagel both discriminated functional effects from accidental effects by requiring that process P in the above schema be neces sary for maintaining the system S in proper working order. So cir culating blood is in turn a necessary condition for the survival of vertebrates, whereas the noise made by the circulation of the blood is not. The problem with this explanation is that (b) is false. Cummins (chapter 2) discusses the difficulty at length, and I will not repeat his arguments here. For present purposes it is sufficient merely to note that artificial pumps could circulate the blood in vertebrates; so the heart is not necessary for circulating blood. Nagel was aware of this problem, and responded by saying that function statements in biology “are not explorations of merely logical possibilities, but deal with the actual functions of definite components in concretely given living systems” (1961, p. 404). That is, Nagel wanted to construe the “necessary” narrowly, as applying only to the available biological^ possible options. But there are two problems with Nagel’s response. First, prosthetic organs are not merely “logical possibilities,” but have replaced, and performed the functions of, “definite compo nents” in some actual “concretely given living systems.” Second, if we restrict our examples to organs (hearts, kidneys, livers, and so on), it may appear plausible that they are the only available biolog ical options for producing the effects that they have, and are in that sense necessary for those effects. But if we consider traits of organ isms generally, the plausibility vanishes. For it would be highly implausible to think that cryptic coloration is the only way that chameleons could avoid predation, that eggshell removal is the only way that black-headed gulls can protect their fledglings from preda tion, or that stotting is the only way that Thomson’s gazelles can communicate to cheetahs that they have been noticed. Surely, in each of these cases, there is a range of biologically possible ways of achieving the same effect, and in general the evolutionary process continually produces diverse solutions to adaptive problems. So con struing “the function of A is to produce process P" as entailing that A is necessary for P seems far too strong.
5
10
David J. Buller
This led Hempel (1965, pp. 310-312) to consider a variant of (b), which loosens the requirement that there be only one way of producing process P. Hempel’s suggestion amounts to letting A be merely one element of a class of items that are jointly necessary for producing P. Thus, if P is the process of circulating blood, Hempel’s proposal is to allow A to be one element of a class of items, I, that are necessary for circulating blood; so I would include the heart (A), artificial pumps, etc. This leads to the following revised version of (b): (b') if S with organization C and in environment E does not have one element of class I (of which A is a member), then S does not engage in P. While this replaces (b) with a true premise, as Hempel pointed out (p. 312) the resulting schema fails to explain why S has A, since (c) is not deducible from (a) and (b‘). The most that we could deduce from (a) and (b’) would be that S has at least one of the elements of the class Z; we cannot deduce that the element of I that S has is A. To make this concrete, if all vertebrates circulate blood (as per (a)), and vertebrates can circulate blood only if they have one element of the class of items consisting of hearts, artificial pumps, and so on (as per (b')), it only follows that a given vertebrate must have a heart or an artificial pump or one of the other items that is neces sary for pumping blood. But this means that (a) and (b') fail to explain why any given vertebrate has a heart; for, according to the deductive-nomological model of explanation, (a) and (b') explain why some vertebrate has a heart only if they jointly entail that it has a heart rather than one of the other items that could pump blood. So, although (b') provides a true premise for the explanation, it renders the deduction of (c) invalid. To make the deduction valid, (c) would have to be replaced with
(c ) hence, S with organization C must have one element of class I (of which A is a member).
J
But then we no longer have an explanation of why S has A; we have only an explanation of why S has something that produces process P. This is a dilemma. Either a statement such as “the function of the heart in vertebrates is to circulate blood” gets analyzed as a deductively valid explanation with a false premise or it gets ana-
Natural Teleology
11
lyzed as a deductively valid argument with true premises that fails to explain why vertebrates have hearts. Strangely, Hempel grasped the dilemma by both horns and said that the only case in which we have a genuine functional explanation is one in which there actu ally is one and only one element of the class I (pp. 313-314). In that event, that element is indeed necessary for producing process P and Nagel’s analysis holds. In every other case, a statement of the form “the function of X is to Y” fails to explain the existence of Xs by cit ing the fact that they produce effects of type Y. But this is just to say that Hempel and Nagel failed to solve the problem of explicat ing the teleological content of the biological concept of function. For by their analyses, the functions of traits or organs rarely if ever explain the existence of those traits or organs. And this is just to say that, by their analyses, the concept of function performs no gen uine explanatory work.
IV. Wright and Cummins
The analyses of Hempel and Nagel were typical of philosophi cal analyses of the concept of function until the mid-1970s, when two articles appeared that proved decisive in reshaping efforts to understand the concept of function. Those articles were Larry Wright’s 1973 article “Functions” (chapter 1) and Robert Cummins’ 1975 article “Functional Analysis” (chapter 2), and they are reprinted here not because they are part of the recent near-consensus, but because they have exerted such a strong influence on the work that has formed that near-consensus. Wright and Cummins both reject the idea, common to Hempel and Nagel, that function statements are elliptical deductive-nomological explanations of the existence of a functional item. In addi tion, neither Wright nor Cummins is concerned solely with expli cating the concept of function as it is used in biology ; they both offer theories that are intended to apply to all instances of function, whether biological, mechanical, institutional, or artifactual. Apart from these two commonalities, however, their theories are radically different. According to Wright (1973, p. 161), a statement of the form “the function of X is I7” means simply (a) X is there because it does Y,
(b) Y is a consequence (or result) of JCs being there.
12
David J. Buller
In this schema, (b) exhibits the fact that Y is an effect (not a cause) of A', and (a) exhibits the teleological explanation of the existence of A", since the “because” in (a) “is to be taken in its ordinary, conver sational, causal-explanatory sense” (1973, p. 157). This is where Wright offers a novel twist. Instead of construing Y itself as a cause of A’ (which would reintroduce the problem of backward causation that appeared to plague Aristotelian metaphysics), Wright’s analy sis takes the fact that Y is an effect of X to be among the (antecedent) efficient causes of X, and thus provides an efficient causal explanation of the existence of A” in terms of its producing Y. Wright calls this an “etiological” theory of teleology, since it ana lyzes statements that ascribe a function to X as explanations of the existence of X strictly in terms of its (antecedent) efficient causes— its etiology. The causal explanation of A' that substantiates (a) is some thing that Wright thinks can vary from context to context, so he doesn’t build it into the analysis. If A is a fuel injection system, then the causal explanation of why it is in a car will be formulated in terms of the design of the car engine, and the explanation may make detailed reference to how alternatives to the fuel injection system may have been tried unsuccessfully or less successfully in the design of the engine. If X is the rock holding open my office door, then the explanation of why it is at the foot of my door will simply be in terms of my intention that my door be held open. And ifX is the heart, then the causal explanation of why it is in a ver tebrate will be formulated in terms of natural selection. This last-mentioned biological case is the one that is most interesting for the purposes of this volume. Consider how natural selection provides an explanation of why humans, for example, have hearts. The heart is a complex organ and all complex traits are the product of accumulated modifications to antecedently exist ing structures. These modifications to existing structures occur randomly as a result of genetic mutation or recombination. When they occur, there is variation in a population of organisms (if there wasn’t already) with respect to some trait. If one of the variants of the trait provides its possessor(s) with an advantage in the compe tition for survival and reproduction, then that variant will become better represented in the population in subsequent generations. When this occurs, that variant of the trait has increased the rela tive fitness of its possessor(s) and there has been “selection for” that variant (see Sober, 1984c, pp. 98-102). That variant can then pro vide the basis for further modification. Thus, humans have hearts
Natural Teleology
13
because hearts were the product of randomly generated modifica tions to preexisting structures that were preserved or maintained by natural selection due to their providing their possessors with a competitive edge (see Sober, 1984c, pp. 147-155). So natural selec tion explains the presence of a trait by explaining how it was pre served after being randomly generated. Boorse (1976) constructed a counterexample to Wright’s anal ysis that played on a similarity with explaining why a trait “is there” in terms of natural selection. Boorse (p. 72) presented the fol lowing scenario:
Suppose that a scientist builds a laser which is connected by a rubber hose to a source of gaseous chlorine. After turn ing on the machine he notices a break in the hose, but before he can correct it he inhales the escaping gas and falls unconscious. ... The release of the gas [Y] is a result of the break in the hose [X]; and the break is there—that is, as in natural selection, it continues to be there—because it releases the gas. If it did not do so, the scientist would cor rect it. Here Y (the release of the gas) is clearly a consequence of X’s (the break’s) being there; so condition (b) of Wright’s analysis is satis fied. In addition, the fact that X produces Y helps to explain why X is there, since, if it were not for the release of the gas, the scientist would fix the hose and the break would not be there. This, Boorse argues, is like the case of natural selection in that the break in the hose originates as a result of some random event, and then is pre served (as a result of the scientist’s inability to fix it). This means that condition (a) of Wright’s analysis is also satisfied. But we surely wouldn’t want to say that the function of the break in the hose is to release the gas. So Wright’s analysis appears not to be fully successful. Note that one thing that Wright’s analysis has in common with Nagel’s and Hempel’s is a commitment to the idea that a func tion statement is an implicit explanation of the presence of the functional item. And one could see this commitment as the source of difficulties for all three analyses. In the case of Nagel and Hempel, the analyses failed to achieve their objective of providing an explanation of the existence of the functional item; and in the case of Wright, the analysis transforms an explanation of the exis tence of an item into a function of that item in some cases where it
14
I
David J. Buller
shouldn’t. Identifying this as the common problem of all three anal yses could lead one to reject entirely the idea that a function state ment is an implicit explanation of the presence of the functionally characterized item. And this is what Cummins does. According to Cummins, function statements are implicit explanations of a unique sort; they do not explain the existence of the functional item, but rather its contribution to an activity or capacity of a system that contains that item, where that contribu tion emerges through a functional analysis of a capacity of that system. A functional analysis of a capacity C of some system S pro ceeds by analyzing C into the capacities of simpler components of S in such a way that C emerges as the “programmed manifesta tion” of the exercise of the capacities of those simpler components (where the latter capacities may themselves admit of functional analysis, until the analysis terminates in the capacities of nondecomposable structural components of S). Given this conception of functional analysis, the statement “The function ofX is to Y” can be understood as stating that X is a component of some system S and Xs doing V features in a functional analysis of some capacity CofS. To illustrate Cummins’ analysis, consider respiration. To explain how the respiratory system (S) exhibits the capacity to exchange oxygen and carbon dioxide (C), we would specify the com ponents of the respiratory system and the capacities of those com ponents. From this it would emerge that the contraction of the diaphragm causes the expansion of the cavities containing the lungs, which in turn causes the lungs to expand and their internal pressure to drop. This drop in pressure causes outside air to fill the lungs and then oxygen and carbon dioxide are exchanged directly across the gas-permeable walls of the capillaries that cover the internal surface of the lungs. Thus, it is the function of diaphragm contraction (X) to produce the expansion of the cavities containing the lungs (Y), since the contracting diaphragm’s producing that expansion features in a functional analysis of the capacity to exchange oxygen and carbon dioxide (C) of the respiratory system (S) that contains the diaphragm. In sum, then, Cummins’ analysis of the concept of function makes the function of an item merely its causal contribution to a complex process. While this certainly succeeds in avoiding appeals to anything other than efficient causation, it does so at the cost of emptying the concept of function of all its teleological content. In addition, Cummins’ analysis seems unable to account for the ways
1
Natural Teleology
15
in which the concept of function is used in biology. Recall some of the obvious facts with which we began—for example, that biologists agree that pumping blood is the function of the heart, but that mak ing noise is not. As Cummins admits (see section III.4), his analy sis does not rule out saying that the function of the heart is to make noise. For we could take the mammalian circulatory system as S and its capacity to make “circulatory noise” as C, in which case a functional analysis would reveal that the heart contributes to C by making a thumping sound. But no biologist would reason in this way. Thus, if we are looking for the theoretical principles underly ing the biological concept of function, which discriminate between the heart’s pumping blood and its making noise, Cummins’ analy sis will not reveal them to us.
V. Millikan.
In 1984 Millikan’s Language, Thought, and Other Biological Categories (excerpts from which are reprinted as chapter 3) made significant headway on the problem of teleology. The details of Mil likan’s theory are complex and the reader will encounter them in chapter 3; so I will provide only a brief paraphrase here. Millikan’s theory is intended to account for a very wide range of functions, from those of traits and organs of organisms to the functions of thoughts, words, artifacts, and cultural products; but it is clearly inspired by the theory of evolution by natural selection and intended to apply paradigmatically in the biological context. And, since the current focus is biological functions (and since her treat ment of artifact functions is too complex to be discussed here and is not reprinted in chapter 3), I will focus only on Millikan’s theory of biological functions and explain it by reference to a prominent way of describing the evolutionary process. While Millikan does not explain her theory in this way, my hope is that tying in Millikan’s theory to broader issues in evolutionary biology will clarify how teleology can emerge from a natural process governed solely by effi cient causation, and thereby clarify Millikan’s solution to the philo sophical problem of teleology. As Hull (1989) has characterized it, the entities that function in the process of evolution by natural selection are replicators and interactors. A replicator is “an entity that passes on its structure largely intact in successive replications” and that replicates itself in accordance with causal laws of nature (p. 96). Genes, for example,
16
David J. Buller
are replicators, since they replicate themselves by directly copying or reproducing their structure. An interactor, on the other hand, is “an entity that interacts as a cohesive whole with its environment in such a way that this interaction causes replication to be differ ential" (p. 96). (Dawkins (1989) uses the term “vehicle,” instead of “interactor,” to express roughly the same idea.) A replicator is also an interactor, since it interacts with its environment as a cohesive whole at the very least “to the extent necessary to replicate itself” (Hull, 1989, p. 96). But more paradigmatic instances of interactors are the organisms that are built by genes; for organisms interact with their environments as cohesive wholes. An organism’s inter action with its environment, of course, largely consists of competi tion with other organisms for survival and reproduction, a compe tition in which different organisms meet with differing degrees of success due to differences in their characteristics. As a result of the differential success of organisms (interactors) in reproducing, there is differential replication of the genes (replicators) that built those organisms. Thus, Hull characterizes natural selection as “a process in which the differential extinction and proliferation of interactors cause the differential perpetuation of the relevant replicators” (p. 96). As a result of this process evolution occurs—that is, there are changes across generations in the relative frequencies of types of replicator and, consequently, changes in the relative frequencies of the characteristics of interactors that are developmentally con structed bjr those replicators. Millikan’s theory is designed to show how functions can emerge within such a process. According to Millikan, an item has a function only as a member of what she calls a “reproductively established family,” where Millikan distinguishes between “firstorder” and “higher-order” reproductively established families. First-order reproductively established families consist of replica tors, which directly reproduce their structures when making copies of themselves in accordance with causal laws of nature. Thus, each temporal sequence of replicators that are related by descent (through copying) forms a first-order reproductively established family. The copies of the gene for blue eyes, for example, are mem bers of a first-order reproductively established family. Replicators, however, do not just cause their own replication; they also cause the properties of the interactors that contain them (that is, differ ences in the properties of interactors are positively correlated with differences in the types of replicator that built them). Thus, as interactors reproduce, not only are the replicators they contain
1
Natural Teleology
17
reproduced, but the properties of interactors caused by those repli cators are reproduced as well. In such cases, the replicators are directly reproduced through copying, whereas the properties of interactors are indirectly reproduced via the reproduction of the replicators that cause those properties. Such indirectly reproduced properties of interactors form higher-order reproductively estab lished families, examples of which are traits such as eye color and blood type and organs such as livers, hearts, and lungs. Members of a higher-order reproductively established family are thus related by descent also, via the direct relations of descent of the members of the first-order reproductively established family that produce the members of the higher-order family. So, in both cases, we can say that the ancestors of a particular member of a reproductively estab lished family are those temporally prior family members to which it is related by a continuous chain of (direct or indirect) reproduc tion. According to Millikan, it is the function of a member X of a reproductively established family to do Y just in case ancestors ofX did Y and their doing Y causally contributed to their family’s having greater reproductive success than competing reproductively estab lished families and, hence, causally contributed (eventually) to the production of X. In other words, the function of X is to do what its ancestors were “selected for” doing. This definition of “function” finds its initial application at the level of replicators, or first-order reproductively established families. For in their interactions with their environments some replicators do things that cause them to be more successful than others in replicating themselves. When there is such a history of competitive success within a family of replica tors, we can pick some particular replicator X and see that its ances tors successfully replicated by doing Y; according to Millikan’s the ory, it is thus the function of X to do Y, since doing Y causally contributed to the (direct) reproduction of members of Xs family. But some functions of a replicator will turn out to be developmental, whereby a replicator has the function of causally contributing to the production of an interactor property. In other words, members of some first-order family will have the function of producing members of some higher-order family. The members of that higher-order fam ily of interactor properties may also contribute to the reproductive success of the interactors bearing those properties by having some particular effect. When they do, we can pick some particular inter actor property X and see that its ancestors contributed to the repro ductive success of the interactors bearing those ancestral properties
18
David J. Buller
by doing Y; it is thus the function ofA” to do Y, since doing Y causally contributed to the (indirect) reproduction of member’s ofX’s family. In this way, Millikan’s theory of functions applies to members of all higher-order reproductively established families as well. In short, according to Millikan, it is the function of X to do Yjust in case doing Y caused the proliferation of ancestors of X (through either direct or indirect reproduction). It is worth noting how well this addresses the philosophical problem of teleology. First, it shows how “the function of X is to do Y” can causally explain the existence of X without invoking back ward causation. For the fundamental idea of Millikan’s theory is that it is not the fact that X itself produces Y that explains X’s exis tence, but the fact that ancestors of A' produced the effect Y and that Y was among the causes of the existence of X (via the processes of replication and/or development that eventually produced X). It is not a current pollination that caused this orchid to release its mimetic pheromone; rather, this orchid causally originated as a (higher-order) reproduction of ancestral orchids that released mimetic pheromones and one of those ancestral releases of mimetic pheromones succeeded in pollination that causally produced this particular orchid. That is why pollination is the function of this par ticular orchid’s release of its mimetic pheromone. Millikan’s theory is thus clearly etiological, since it analyzes a statement that ascribes to X the function of producing Y as an explanation of the existence of X strictly in terms of X’s antecedent (efficient) causes. But Millikan’s etiological theory differs from Wright’s. For, while Wright analyzes “the function of Xis to do Y” as implying that X “is there” because it does Y, Millikan analyzes it as implying that X “is there” because its ancestors did Y; while Wright’s analysis appears to focus on those instances of Y that are current effects ofX, Millikan’s analysis focuses on those instances of Y that were among the causes ofX. So Millikan’s etiological the ory takes the causal explanation of why a functional item “is there” to refer strictly to the causal history of the item, not to what it causes. Second, although clearly in the spirit of Wright’s theory, Mil likan s theory avoids Boorse’s counterexample. For an item has the function of producing some effect only if that item originated as a copy or reproduction of earlier items that had the same effect. This avoids Boorse’s counterexample, since the break in the hose did not originate as a copy of earlier breaks in hoses that had the effect of knocking scientists unconscious.
1
Natural Teleology
19
Third, Millikan’s theory is able to successfully distinguish functional effects from accidental effects. For example, the reason that it is the function of my heart to pump blood, but not to make noise, is that the hearts of my ancestors contributed to their repro ductive success—and thus to the reproduction of hearts—by pump ing blood, not by making noise. Fourth, Millikan’s theory explains how teleology can emerge within a universe governed solely by efficient causation. For in the beginning the only interactors were probably replicators differen tially replicating themselves in “the primeval soup” (see Dawkins, 1989, chap. 2), where their differential replication was due to dif ferences in what they did in interacting with their environments. The first functions emerged, at this point in the universe, as func tions of members of first-order reproductively established families. Eventually, however, some replicator interactions resulted in types of replicator bonding with one another to form more complex inter actors consisting of “teams” of replicators. The properties of these complex teams of interactors biased interactions in their favor and, as a result, they began to enjoy greater differential success in repli cation. This eventually led to increased team size and the formatio' of protein walls for protection, and the first cells thus appearet Multicellular interactors eventually followed and finally the com plex organisms that we take to be paradigmatic interactors. And the emergence of these higher levels of complexity were accompa nied by the emergence of the functions of members of higher-order reproductively established families. While this provides a plausible scenario about the origins of complex organisms and the functions of their traits and organs, now’here does it make reference to any thing but efficient causation and the differential perpetuation of replicators (which is a function of differential success in interaction with the environment).
VI. The Core Consensus and the Peripheral Disagreements
Millikan’s approach to defining the function of an item in terms of its evolutionary, causal history succeeded in setting the agenda for subsequent discussions of the biological concept of func tion. Indeed, her approach so successfully set the agenda that the term “etiological theory” has come to refer in the literature to theo ries that define the function of an item in terms of its evolutionary history, in spite of the fact that Wright authored the (broadly) etio-
20
David J. Buller
logical approach to understanding teleology. Goode and Griffiths (chapter 12), for example, reserve the term “etiological” for Millikanesque approaches and refer to Wright’s theory as “proto-etio logical." While this terminological convention does some violence to the history of philosophical work on functions, for the sake of expe diency I will follow it, since the evolutionary approach of Millikan has exerted the strongest influence on recent work. In fact, of cur rent theories, the (evolutionary) etiological theory has the largest number of adherents, with versions of it defended in this volume by Millikan (chapters 3 and 5), Neander (chapters 6 and 11), Griffiths (chapter 7), Godfrey-Smith (chapters 9 and 10), Goode and Griffiths (chapter 12), Allen and Bekoff (chapter 13), and Buller (chapter 15). Alternatives to the etiological theory are defended by Bigelow and Pargetter (chapter 4), Kitcher (chapter 8), and Walsh and Ariew (chapter 14). Despite the disagreement, however, there is a common core of agreement that unites etiological theorists with the dissenters just mentioned, and this core of agreement represents as great a con sensus as has been achieved in philosophy. For, in one way or another, all the authors agree that the biological concept of function is to be analyzed in terms of the theory of evolution by natural selection. That is, all agree that a trait or organ has a function in virtue of its role in a selection process—either in virtue of its role in a selection process that a lineage bearing that trait or organ actually has undergone, or in virtue of a selection process it is cur rently undergoing or is set to undergo. In this sense, there is con sensus that the theory of evolution by natural selection can provide an analysis of the teleological concept of function strictly in terms of processes involving only efficient causation (although see Man ning (1997) for an argument that this consensus viewpoint is mis taken). Agreement regarding this fundamental idea, however, still leaves a great deal of room for disagreement concerning the details of a theory of functions. In what remains of this introduction, I will sketch several foci of the debates that run through the essays that follow. 1. Looking Forward Versus Looking Back. Defining the func tion of a trait as its role in a selection process still leaves the fol lowing question unanswered: At what point in evolutionary history lies the selection process relevant to characterizing a trait’s func tion? According to etiological theorists, the function of a trait is the past contribution that the trait made to the fitness of its bearers,
Natural Teleology
21
and most etiological theorists agree with Godfrey-Smith (chapter 10) that the relevant past is the recent evolutionary past. Bigelow and Pargetter (chapter 4), however, argue that, since the concept of fitness plays a role in the definition of “function,” the concept of function should reflect the best available analysis that we have of the concept of fitness. According to that analysis, the fit ness of an organism is not to be measured in terms of actual repro ductive output, but in terms of potential reproductive output, which in turn is to be measured in terms of an organism’s propensity to survive and reproduce in its environment (Mills & Beatty, 1979). Thus, Bigelow and Pargetter argue, the function of a trait should similarly be defined in terms of its propensity to contribute to fit ness in the immediate possible selective future of the organisms with that trait. Whereas the etiological theory is “backward look ing,” defining function in terms of past selection, they offer a propensity theory that is “forward looking,” defining function in terms of future selection. Millikan (chapter 5) and Neander (chap ter 6) provide critiques of the propensity theory from the standpoint of the etiological theory. In a somewhat different spirit, Walsh and Ariew (chapter 14) argue that fitness is to be defined in terms of a “selective regime,’ which is a set of environmental demands placed on organisms. In defining “fitness,” we can focus on a past selective regime, a current regime, or an immediate possible future regime, and there is no definitive reason for privileging one of these possible regimes over the others. This moves them to defend a “relational theory” of func tions according to which the function of a trait is its contribution to fitness relative to a selective regime (see Walsh (1996) for a more detailed defense of the relational theory). If we look “backward” to a past selective regime we get the etiological conception of function, whereas if we look “forward” we get the propensity conception. Both are legitimate, according to the relational theory, but neither is to be preferred in any absolute sense over the other.
2. Pluralism Versus Unification. Millikan (chapter 5) argues that the propensity theory’ of Bigelow and Pargetter confuses two distinct concepts of function that are at work in biology; and, although this argument is directed at Bigelow and Pargetter, it can be generalized to the relational theory of Walsh and Ariew as well. According to Millikan, one concept of function is the historical con cept that is captured by her etiological theory, according to which a function of an item is the contribution its ancestors made to the fit-
22
David J. Buller
ness of its bearers; the other concept is that captured by Cummins’ theory of functions, according to which the function of an item is its causal contribution to a complex process. Millikan thinks that the propensity theory, in effect, confuses the historical concept of func tion with a current propensity to contribute to fitness, which is actually the Cummins-concept of functions. Godfrey-Smith (chap ter 9) offers a similar argument. He claims that the historical con cept of function is at work in subdisciplines of biology such as evo lutionary theory and behavioral ecology' and helps to explain the presence of a trait in its bearers, whereas the Cummins-concept of function is at work in fields such as physiology' and helps only to explain how a trait or organ contributes to maintaining certain fea tures of organisms. Allen and Bekoff (chapter 13) also endorse this idea that there are at least two distinct concepts of function at work in biology. Opposed to these “pluralists” regarding the concept of function in biology are the “Unitarians,” who claim that the plural senses of “function” are merely an artifact of the failure of certain versions of the etiological theory’ to explicate adequately the unified concept of function at work in all areas of biology. Griffiths (chapter 7) and Buller (chapter 15) defend syntheses of the etiological theory with Cummins’ theory, and argue that their syntheses provide a unified account of the concept of function in biology. And Kitcher (chapter 8) defends a theory according to which the function of an item is what it is designed to do, arguing that the concept of design is capa ble of unifying the uses of the concept of function in all areas from behavioral ecology to physiology. (Godfrey-Smith (chapter 9) pro vides a detailed critique of Kitcher’s theory from the standpoint of the etiological theory.) While rejecting Kitcher’s design theory of functions, Walsh and Ariew (chapter 14) agree that there is a uni fied concept of function in biology and that it is captured by the relational theory. 3. Selection for Traits Versus Selection of Bearers. In Mil likan’s version of the etiological theory, a trait X has a function only if there was selection for X at some point in evolutionary history. This requires that, at some point in evolutionary history, there was variation in a population with respect to the possession of X and that organisms with X enjoyed greater reproductive success than organisms without X at least partly because of their having X. Thus, Millikan’s etiological theory entails that no trait that hap pened to be universal in a population or that didn’t causally con-
Natural Teleology
23
tribute to making its bearers better adapted than organisms not bearing that trait counts as having a function, regardless of whether the trait was actually useful to its bearers. And most etio logical theorists have followed Millikan in taking these strong requirements to be necessary for a trait’s having a function; this view, for example, is to be found in Neander (chapters 6 and 11), Griffiths (chapter 7), and Godfrey-Smith (chapters 9 and 10). This way of defining a trait’s function requires that there has been a history of selection acting directly on a trait to favor it over its rival variants. But this is not the only way in which selection can play a role in defining a trait’s function. For different reasons, both Kitcher (chapter 8) and Buller (chapter 15) argue that a trait has a function provided that it has contributed to the success of its bear ers in a selection process, regardless of whether there has been direct selection for that trait over any rival variants (indeed, regardless of whether there have been rival variants). Thus, whereas Millikan, Neander, Griffiths, and Godfrey-Smith define a trait’s function in terms of that trait’s having undergone a selection process, Kitcher and Buller define a trait’s function in terms of its bearers having undergone a selection process and that trait’s hav ing been useful to its bearers in that process. 4. Levels of Functional Description. Every’ functional trait pro duces numerous effects ranging from the proximal to the very remote, and to every one of which there corresponds a possible func tional description of the trait. For example, chameleons possess specialized skin cells called “chromatophores,” which contain gran ules of pigment, and a chameleon’s famous ability to change color is autonomically brought about by the aggregation or dispersion of the pigment granules within these chromatophores (Cott, 1966). This rearrangement of pigment granules has the effect of making the chameleon’s skin match the coloring of its arboreal surroundings. This, in turn, has the effect of camouflaging the chameleon, which in turn effects the avoidance of predation from the large birds and large lizards that typically prey on chameleons. Successful preda tion avoidance then has the effect of contributing to the fitness of chameleons. What is the function of chromatophores? Is it merely their function to rearrange pigment granules? Or is it to enhance the chameleon’s fitness? Or is there a function of chromatophores corresponding to every effect in the series linking pigment rear rangement with fitness enhancement? Millikan (chapter 3) argues that each effect in such a series
24
David J. Buller
2 to ■ > a correct description of the function of the item corresponds that produces the series, since each effect in the series plays a role in explaining why members of that item’s reproductively estab lished family were successful under selection. So functional items rarely have unique functional descriptions, but rather as many correct functional descriptions as there are effects in the series that contributed to the successful reproduction of their family members. Millikan calls such series of effects the “serial func tions” of an item. Neander (chapter 11) argues, in contrast, that a functional item has a “privileged” level of description, which corresponds to the level at which the item can be described as “malfunctioning.” If someone’s carotid artery were opened, there would be a relatively rapid failure of their heart’s ability to circulate their blood, since they would soon lose their body’s blood supply through the opened artery. In such a case, however, we would not say that their heart had malfunctioned; for their heart continued to beat regularly, pumping their blood out through the opened artery. We would only say that someone’s heart malfunctioned when it failed to produce the characteristic pumping action. Thus, Neander argues, “the function of the heart is to pump blood” is the “privileged” descrip tion of the heart’s function, not “the function of the heart is to cir culate the blood,” since that is the description that corresponds to a description of the heart’s malfunctioning. (In the context of a dis cussion of the functions of behavior control mechanisms, Buller (1997) defends a view very similar to Neander’s.) Goode and Griffiths (chapter 12) respond to Neander by argu ing that there are many levels at which selection processes can be described, ranging from a detailed description of the benefits of an organism’s neuroanatomy to more general descriptions that con sider the organism embedded in ever wider ecological systems in its environment. These different levels, they argue, are complemen tary, not mutually exclusive. And, since the function of a trait is its role in a selection process, there should be as many legitimate and complementary descriptions of a trait’s functions as there are lev els at which the selection of that trait can be described.
5. Function and Design. Millikan (chapter 3) identifies the function of an item with what it was “designed to do,” and takes the objective of a theory of biological functions to be explaining design in terms that do not presuppose an intelligent creator. But Millikan does not articulate what might be meant by “design” or the role
Natural Teleology
25
that this plays in understanding an item’s function. Kitcher (chap ter 8), on the other hand, defends a theory of functions that also identifies an item’s function with what it was designed to do, but supplements that with an extended explanation of how natural selection is the source of biological design and how the phenomenon of design is related to that of function. Allen and Bekoff (chapter 13), however, argue that the identi fication of function with design is misguided. Function, they argue, should be seen as a necessary, but not sufficient, condition for design. While defending an etiological theory of functions, they argue that a trait is not “designed” to perform its function unless it has undergone a history of modification under natural selection for enhanced performance of its functional role, in much the way that a rock is not “designed” to be a paperweight unless it has been mod ified (e.g. flattened on one side) for better performance as a paper weight, although it can have the function of being a paperweight without being modified. This view makes design a rarer biological phenomenon than function and, if correct, would undermine efforts, like those of Kitcher, to define function in terms of design. 6. Artifact Functions. The analyses of the concept of function offered by Wright and Cummins are intended to apply equally to biological functions and artifact functions. If these analyses are rejected, however, the question arises as to whether a theory of bio logical functions can also extend to artifacts or whether a distinc account is required for artifact functions. Even post-Wrightian etiological theorists disagree about th answer to this question. On Neander’s etiological theory (chapter 6), distinct accounts are required for biological and artifact func tions. According to Neander, the function of a biological trait is the effect for which it was selected, whereas the function of an artifact is the effect that an agent intends it to have. One major difference then is that biological items have functions only when they are reproductions of items with similar effects, whereas artifacts can have functions without being reproductions at all. In contrast, Grif fiths (chapter 7) argues that his etiological theory can be smoothly extended to artifacts as well. According to Griffiths, even the origi nal or prototypical tokens of artifacts are reproductions of a sort: They reproduce in actuality the effects that they had in the imagi nations of their creators. Further, like traits with biological func tions, artifact functions are the result of a selection process; but, whereas the selection process that determines biological functions
26
David J. Buller
is natural selection, the selection process that determines artifact functions is a process of selection among hypothetical alternative artifacts, each of which is hypothetically capable of producing the desired functional effect. Neander (chapter 6) rejects Griffiths’ account, however, on the grounds that, in the case of some artifacts, there may be only one way to achieve the desired effect; so not all artifact functions are the result of selection among hypothetical alternatives. Among the non-etiological theorists reprinted here, there are two proposals for a unified account of biological and artifact func tions. Bigelow and Pargetter (chapter 4), as we have seen, argue that a biological item has a function in virtue of having a propen sity to contribute to survival under natural selection because of producing some effect. Similarly, they contend, an artifact has a function in virtue of having a propensity to survive a process of intentional selection by conscious agents because of its producing some desired effect (or because the selecting agents believe it will produce that desired effect). Of course, if the etiological critiques of Bigelow and Pargetter’s theory7 of biological functions are corect, then their account fails to unify biological and artifactual motions by failing to provide a correct account of biological funcions. Kitcher (chapter 8), in contrast, argues simply that his equation of an item’s function with what the item was designed to do provides a seamless unification of biological and artifactual functions. The only difference between biological and artifactual functions, according to Kitcher, is that natural selection is the source of design in the biological case, and the intentions of agents are the source of design in the artifactual case. If Allen and Bekoffs (chapter 13) arguments are right, however, the equa tion of function with design is no less problematic in the artifac tual case than it is in the biological case, and Kitcher’s proposed unification is unsuccessful.
VII. Nonconclusion
While the achievement of a near consensus concerning how to understand the biological concept of function certainly represents progress in a field notorious for failing to produce any, much work still remains to be done by way of resolving the issues just dis cussed. Indeed, resolution of these issues constitutes the major part of the agenda for continued philosophical work on the biological
Natural Teleology
27
concept of function. Thus, although the essays reprinted in this vol ume do not solve all the problems regarding teleology in biology, they do, I believe, represent progress toward that goal and serve to point the direction in which future work should proceed if it is to be achieved.
1
CHAPTER 1
Functions* Larry Wright
The notion of function is not all there is to teleology, although it is sometimes treated as though it were. Function is not even the central, or paradigm, teleological concept. But it is interesting and important; and it is still not as well understood as it should be, con sidering the amount of serious scholarship devoted to it during the last decade or two. Let us hope this justifies my excursion into these murky waters. Like nearly every word in English, “function” is multilaterally ambiguous. Consider:
1. y = /W/The pressure of a gas is a function of its temper ature.
2. The Apollonaut’s banquet was a major state function. 3. I simply can’t function when I’ve got a cold. 4. The heart functions in this way . . . (something about serial muscular contractions). 5. The function of the heart is pumping blood.
6. The function of the sweep-second hand on a watch is to make seconds easier to read.
7. Letting in light is one function of the windows of a house. * Reprinted from Philosophical Review 82 (1973), pp. 139-168. Copyright 1973 Cornell University. Reprinted by kind permission of the publisher and the author.
29
30
Larry Wright
8. The wood box next to the fireplace currently functions as a dog’s sleeping quarters. It is interesting to notice that the word “function” has a spectrum of meanings even within the last six illustrations, which aie the only ones at all relevant to a teleologically oriented study. Numbers 3, 4, and 8 are substantially different from one another, but they are each, from a teleological point of view, peripheral cases by com parison with 5, 6, and 7, which are the usual paradigms. And even these latter three are individually distinct in some respects, but much less profoundly than the others. Quite obviously, making some systematic sense of the logical differentiation implicit in categorizing these cases as peripheral and paradigmatic is a major task of this paper. But a clue that we are on the right track here can be found in a symptomatic gram matical distinction present in the last six illustrations: in the peripheral cases the word “function” is itself the verb, whereas in the more central cases “function” is a noun, used with the verb “to be.” And since the controversy revolves around what the function of something is, the grammatical role of “function” in 5, 6, and 7 makes them heavy' favorites for the logical place of honor in this discussion.
Some Rudimentary Distinctions 1. Functions v. goals. There seems to be a strong temptation to treat functions as representative of the set of central teleological concepts which cluster around goal-directedness. However, even a cursory examination of the usual sorts of examples reveals a very important distinction. Goal-directedness is a behavioral predicate. The direction is the direction of behavior. When we do speak of objects (homing torpedoes) or individuals (General MacArthur) as being goal-directed, we are speaking indirectly of their behavior. We would argue against the claim that they are goal-directed by appeal to their behavior (for example, the torpedo, or the General, did not change course at the appropriate time, and so forth). On the other hand, many things have functions (for example, chairs and windpipes) which do not behave at all, much less goal-directedly. And behavior can have a function without being goal-directed—for example, pacing the floor or blinking your eye. But even when goaldirected behavior has a function, very often its function is quite dif-
Functions
31
ferent from the achievement of its goal. For example, some fresh water plankton diurnally vary their distance below the surface. The goal of this behavior is to keep light intensity in their environment relatively constant. This can be determined by experimenting with artificial light sources. The function of this behavior, on the other hand, is keeping constant the oxygen supply, which normally varies with sunlight intensity. There are many instances to be found in the study of organisms in which the function of a certain goaldirected activity is not some further goal of that activity, as it usu ally is in human behavior, but rather some natural concomitant or consequence of the immediate goal. Other examples are food-gath ering, nest-making, and copulation. Clearly function and goaldirectedness are not congruent concepts. There is an important sense in which they are wholly distinct. In any case, the relation ship between functions and goals is a complicated and tenuous one; and becoming clearer about the nature of that relationship is one aim of this essay.
2. A function v. the function. Recent analyses of function, including all those treated here, have tended to focus on a function of something, by contrast with the function of something. This ten dency is understandable; for any analysis of this sort aims at gen erality, and “a function” would seem intrinsically more general than “the function” because it avoids one obvious restriction. This generality, however, is superficial: the notion of a function is deriv able from the notion of the function (more than one thing meets the criteria) just as easily as the reverse (only one thing meets the cri teria). Furthermore, the notion of a function is much more easily confused with certain peripheral, quasi-functional ascriptions which are examined below. In short, the discussion of this paper is concerned with a function of A' onlj- in so far as it is the sort of thing which would be the function of A' if A” had no others. Accordingly, I take the definite-article formulation as paradigmatic and will deal primarily with it, adding comments in terms of the indefinite-arti cle formulation parenthetically, where appropriate.
3. Function v. accident. Very likely the central distinction of this analysis is that between the function of something and other things it does which are not its function (or one of its functions). The function of a telephone is effecting rapid, convenient communica tion. But there are many other things telephones do: take up space on my desk, disturb me at night, absorb and reflect light, and so forth. The function of the heart is pumping blood, not producing a
32
Larry Wright
thumping noise or making wiggly lines on electrocardiograms, which are also things it does. This is sometimes put as the distinc tion between a function, and something done merely by accident. Explaining the propriety of this way of speaking—that is, making sense of the function/accident distinction—is another aim, perhaps the primary aim of the following analysis.
4. Conscious u. natural functions. The notion of accident will raise some interesting and important questions across another rudimentary distinction: the distinction between natural functions and consciously designed ones. Natural functions are the common organismic ones such as the function of the heart, mentioned above. Other examples are the function of the kidneys to remove metabolic wastes from the bloodstream, and the function of the lens of the human eye to focus an image on the retina. Consciously designed functions commonly (though not necessarily) involve artifacts, such as the telephone and the watch’s sweep hand mentioned previ ously. Other examples of this type would be the function of a door knob, a headlight dimmer switch, the circumferential grooves in a pneumatic tire tread, or a police force. Richard Sorabji has argued1 that “designed” is too strong as a description of this category, and that less elaborate conscious effort would be adequate to give some thing a function of this sort. I think he is right. I have used the stronger version only to overdraw the distinction hyperbolically. In deference to his point I will drop the term “designed” and talk of the distinction as between natural and conscious functions. Of the two, natural functions are philosophically the more problematic. Several schools of thought, for different reasons, want to deny that there are natural functions, as opposed to conscious ones. Or, what comes to the same thing, they want to deny that nat ural functions are functions in anything like the same sense that conscious functions are. Some theologians want to say that the organs of organisms get their functions through God’s conscious design, and hence these things have functions, but not natural functions as opposed to conscious ones. Some scientists, like B. F. Skinner, would deny that organs and organismic activity have func tions because there is no conscious effort or design involved. Now it seems to me that the notion of an organ having a func tion—both in everyday conversation and in biology—has no strong theological commitments. Specifically, it seems to me consistent, appropriate, and even common for an atheist to say that the func tion of the kidney is elimination of metabolic wastes. Furthermore,
■
Functions
33
it seems clear that conscious and natural functions are functions in the same sense, despite their obvious differences. Functional ascriptions of either sort have a profoundly similar ring. Compare “the function of that cover is to keep the distributor dry" with “the function of the epiglottis is to keep food out of the windpipe.” It is even more difficult to detect a difference in what is being requested: “What is the function of the human windpipe?” versus “What is the function of a car’s exhaust pipe?” Certainly no analysis should begin by supposing that the two sorts are wildly different, or that only one is really legitimate. That is a possible conclusion of an analysis, not a reasonable presupposition. Accordingly, the final major aim of this analysis will be to make sense of natural func tions, both as functions in the same sense as consciously contrived ones, and as functions independent of any theological presupposi tions—-that is, independent of conscious purpose. It follows that this analysis is committed to finding a way of stating what it is to be a function—even in the conscious cases—that does not rely on an appeal to consciousness. If no formulation of this kind can be found despite an honest search, only then should we begin to take seri ously the view that we actually mean something quite different by “function” in these two contexts.
Some Analyses of Function
The analysis of function for which I wish to argue grew out o a detailed critical examination of several recent attempts in the lit erature to produce such an analysis, and it is best understood in that context. For this reason, and because it will help clarify the aims I have sketched above, I will begin by presenting the kernel of that critical examination. The first analysis I want to consider is an early one by Morton Beckner.2 Here Beckner contends that to say something s has func tion F' in system s' is to say that There is a set of circumstances in which: F' occurs when s' has s, AND F' does not occur when s' does not have s.3 (p. 113)
For example, “the human heart has the function of circulating blood” means that there is a set of circumstances in which circula tion occurs in humans when they have a heart, and does not when
34
Larry Wright
they do not. Translated into the familiar jargon, s has function F in s' if and only if there is a set of circumstances containing s which are sufficient for the occurrence of F' and which also require s in order to be sufficient for F'. Now it is not clear whether the “requirement” here is necessity or merely non-redundancy. If it is necessity, then under the most natural interpretation of “circum stances” (environment), it is simply mistaken. There are no cir cumstances in which, for example, the heart is absolutely irre placeable: we could always pump blood in some other way. On the other hand, if the requirement here is only non-redundancy, the mistake is more subtle. In this case Beckner's formula would hold for cases in which s merely does F', but in which F‘ is not the function of s. For exam ple, the heart is a non-redundant member of a set of conditions or circumstances which are sufficient for a throbbing noise. But mak ing a throbbing noise is not a function of the heart, it is just some thing it does—accidentally. In fact, there are even dysfunctional cases which fit the formula: in some circumstances, livers are nonredundant for cirrhosis, but cirrhotic debilitation could not con ceivably be the (or a) function of the liver. So this analysis fails on the functional/accidental distinction: it includes too much. After first considering a view essentially similar to this one, John Canfield has offered a more elaborate analysis? According to Canfield: A function of I (in S) is to do C means I does C and that C is done is useful to S. For example, “(In vertebrates) a func tion of the liver is to secrete bile” means “the liver secretes bile, and that bile is secreted in vertebrates is useful to them.” (p. 290) Canfield recognizes that natural functions are the problematic ones, but he devotes his attention solely to those cases. He treats only the organs and parts of organisms studied by biology, to the exclusion of the consciously designed functions of artifacts. As a result of this emphasis, his analysis is, without modification, almost impossible to apply to conscious functions. But even with appropriate modifications, it turns out to be inadequate to the char acterization of either conscious or natural function. In the conscious cases, there is an enormous problem in iden tifying the system S, in which I is functioning, and to which it must be useful. The function of the sweep-second hand of a watch is to
1
Functions
35
make seconds easier to read. It would be most natural to say that the system in which the sweep hand is functioning—by analogy with the organismic cases—is the watch itself; but it is hard to make sense of the easier reading being useful to the mechanism. On the other hand, the best candidate for the system to which the eas ier reading is useful is the person wearing the watch; but this does not seem to make sense as the system in which the sweep hand is functioning. The crucial difficulty of Canfield’s analysis begins to appear at this point; no matter what modifications we make in his formula to avoid the problem of identifying the system S, we must retain the requirement that C be useful. This is really the major contribution of his analysis, and to abandon it is to abandon the analysis. The difficulty with this is that, for example, in the watch case, it is clearly not necessary that easily read seconds be useful to the watch-wearer—or anyone else—in order that making seconds eas ier to read be the function of the sweep hand of that wearer’s watch. My watch has a sweep-second hand, and I occasionally use it to time things to the degree of accuracy it allows: it is useful to me. Now suppose I were to lose interest in reading time to that degree of accuracy. Suppose my life changed radically so that nothing I ever did could require that sort of chronological precision. Would that mean the sweep hand on my particular watch no longer has the function of making seconds easier to read? Clearly not. If some one were to ask what the sweep hand’s function was (“What’s it do?,” “What’s it there for?”) I would still have to say it made seconds easier to read, although I might yawningly append an autobio graphical note about my utter lack of interest in that feature. Sim ilarly, the function of that button on my dashboard is to activate the windshield washer, even if all it does is make the mess on the wind shield worse, and hence is not useful at all. That would be its func tion even if I never took my car out of the garage—or broke the windshield. It is natural at this point to attempt to patch up the analysis by reducing the requirement that C be useful, to the requirement that C usually be useful. But this will not do either, because it is easy to think of cases in which we would talk of something’s having a function even though doing that thing was quite generally of no use to anybody. For example, a machine whose function was to count Pepsi Cola bottle caps at the city dump; or M.I.T.’s ultimate machine of a few years back, whose only function was to turn itself off. The source of the difficulty in all of these cases is that what the
36
Larry Wright
thing in question (watch, washer button, counting machine) was designed to do has been left out of the calculation. And, of course, in these cases, if something is designed to do X, then doing X is its function even if doing X is generally useless, silly, or even harmful. In fact, intention is so central here that it allows us to say the func tion of’/ is to do C, even when / cannot even do C. If the windshield washer switch comes from the factory defective, and is nevei repaired, we would still say that its function is to activate the washer system; which is to say: that is what it was designed to do. It might appear that this commits us to the view that natural and consciously contrived functions cannot possibly be the same sort of function. If conscious intent is what determines the function an artifact has got, there is no parallel in natural functions. I take this to be mistaken, and will show why later. For now it is only important to show, from this unique vantage, the nature of the most formidable obstacle to be overcome in unifying natural and conscious functions. The argument thus far has shown that meeting Canfield’s cri teria is not necessary for something to be a function. It can easily be shown that meeting them is also not sufficient. We are always hearing stories about the belt buckles of the Old West or on foreign battlefields which save their wearers’ lives by deflecting bullets. From several points of view that is a very useful thing for them to do. But that does not make bullet deflection the function—or even a function—of belt buckles. The list of such cases is endless. Arti facts do all kinds of useful things which are not their functions. Blowouts cause you to miss flights that crash. Noisy wheel bearings cause you to have the front end checked over when you are nor mally too lazy. The sweep hand of a watch might brush the dust off the numbers, and so forth. All this results from the inability of Canfield’s analysis to han dle what we took to be one of the fundamental distinctions of func tion talk: accidental versus nonaccidental. Something can do some thing useful purely by accident, but it cannot have, as its function, something it does only by accident. Something that I does by acci dent cannot be the function of I. The cases above allow us to begin to make some fairly clear sense of this notion of accident, at least for artifacts. Buckles stop bullets only by accident. Blowouts only accidentally keep us off doomed airplanes. Sweep hands only acci dentally brush dust, if they do it at all. And this brings us back to the grammatical distinction I made at the outset when I divided the list of illustrations into “central” and “peripheral” ones. When
fl
Functions
37
something does something useful by accident rather than design, as in these examples, we signal the difference by a standard sort of “let’s pretend” talk. Instead of using the verb “to be” or the verb “to have,” and saying the thing in question has such and such a func tion, or saying that is its function, we use the expression “function ing as.” We might say the belt buckle functioned as a bullet shield, or the blowout functioned as divine intervention, or the sweep hand functions as a dust brush. Canfield’s analysis does not embrace this distinction at all. So far I have shown only that Canfield’s formula fails to han dle conscious functions. This means it is incapable of showing nat ural functions to be functions in the same full-blooded sense as con scious ones, which is indeed serious; but that, it might be argued, really misses the point of his analysis. Canfield is not interested in conscious functions. He would be happy just to handle natural func tions. For the reasons set down above, however, I am looking for an analysis which will unify conscious and natural functions, and it is important to see why Canfield’s analysis cannot produce that unifi cation. Furthermore, Canfield’s analysis has difficulties in handling natural functions that closely parallel the difficulties it has with conscious functions; which is just what we should expect if the two are functions in the same sense. For example, it is absurd to say with Pangloss that the func tion of the human nose is to support eyeglasses. It is absurd to sug gest that the support of eyeglasses is even one of its functions. The function of the nose has something to do with keeping the air we breathe (and smell) warm and dry. But supporting a pincenez, just as displaying rings and warpaint, is something the human nose does, and is useful to the system having the nose: so it fits Can field’s formula. Even the heart throb, our paradigm of non-function, fits the formula: the sound made by the heart is an enormously use ful diagnostic aid, not only as to the condition of the heart, but also for certain respiratory and neurological conditions. More bizarre instances are conceivable. If surgeons began attaching cardiac pacemakers to the sixth rib of heart patients, or implanting micro phones in the wrist of C.I.A. agents, we could then say that these were useful things for the sixth rib and the wrist (respectively) to do. But that would not make pacemaker-hanging a function of the sixth rib, or microphone concealment a function of the human wrist. There seems to be the same distinction here that we saw in conscious functions. It makes perfectly good sense to say the nose functions as an eyeglass support; the heart, through its thump,
38
Larry Wright
functions as a diagnostic aid; the sixth rib functions as a pacemaker hook in the circumstances described above. This, it seems to me, is precisely the distinction we make when we say, for example, that the sweep-second hand functions as a dust brush, while denying that brushing dust is one of the sweep hand’s functions. And it is here that we can make sense of the notion of accident in the case of natural functions: it is merely fortuitous that the nose supports eyeglasses; it is happy chance that the heart throb is diagnostically significant; it would be the merest serendipity if the sixth rib were to be a particularly good pacemaker hook. It is (would be) only acci dental that (if) these things turned out to be useful in these ways. Accordingly, we have already drawn a much stronger parallel between natural functions and conscious functions than Canfield’s analysis will allow. Thus far I have ignored Canfield’s analysis of usefulness:
[In plants and animals other than man, that C is done is useful to S means] if, ceteris paribus, C were not done in S, then the probability of that S surviving or having descen dants would be smaller than the probability of an S in which C is done surviving or having descendants, (p. 292) I have ignored it because its explicit and implicit restrictions make it even more difficult to work this analysis into the unifying one I am trying to produce. Even within its restrictions (natural func tions in plants and animals other than man), however, the extended analysis fails for reasons very like the ones we have already examined. Hanging a pacemaker on the sixth rib of a cardiovascularly inept lynx would be useful to that cat in precisely Canfield’s sense of “useful”: it would make it more likely that the cat would survive and/or have descendants. Obviously the same can be said for the diagnostic value of an animal’s heart sounds. So usefulness—even in this very restricted sense—does not make the right function/accident distinction: some things do useful things which are not their functions, or even one of their functions. The third analysis I wish to examine is a more recent one by Morton Beckner.5 This analysis is particularly interesting for two reasons. First, Beckner is openly (p. 160) trying to accommodate both natural and conscious functions under one description. Sec ond, he wants to avoid saying things like (to use his examples) “A function of the heart is to make heart sounds” and “A function of the Earth is to intercept passing meteorites.” So his aims are very
Functions
39
like the ones I have argued for: to produce a unifying analysis, and one which distinguishes between functions and things done by acci dent. And since the heart sound is useful, and intercepting mete orites could be (perhaps already is), Beckner would probably agree in principle with the above criticism of Canfield. Beckner’s formulation is quite elaborate, so I will present it in eight distinct parts, clarify the individual parts, and then offer an illustration before going on to raise difficulties with them collec tively as an analysis of the concept of function. That formulation is: P has function F in S if and only if:6
1. P is a part of S (in the normal sense of “part”). 2. P contributes to F. (P's being part of S makes the occur rence of F more likely.) 3. F is an activity in or of the system S. 4. S is structured in such a way that a significant number of its parts contribute to the activities of other parts, and of the system itself.
5. The parts of S and their mutual contributions are iden tified by the same conceptual scheme which is employed in the statement that P has function F in system S.
6. A significant number of critical parts (of S) and their activities definitionally contribute to one or more activi ties of the whole system S. 7. F is or contributes to an activity A of the whole system S.7
8. A is one of those activities of S to which a significant number of critical parts and their activities definitionally contribute. Two points of clarification must be made at once. First, the notion of “the same conceptual scheme” in number 5 is obscure in some respects, and the considerable attention devoted to it by Beck ner does not help very' much. In general all one can say is that P, F, and the other parts and activities of S must be systematically related to one another. But in practice the point is easier to make. For example, if we wish to speak of removing metabolic wastes as the function of the human kidney, the relevant conceptual scheme
40
Larry Wright
contains other human organs, life, and perhaps ecology in general, but not atoms, molecular bonds, and force fields. The second point concerns the “definitional contribution” in number 6. A part (or activity) makes a definitional contribution to an activity if that con tribution is part of what we mean by the word which refers to that part (or activity). For example, part of what we mean by “heart” in a biological or medical context is “something which pumps blood”: we would allow considerable variation in structure or appearance and still call something a heart if it served that function. Beckner illustrates how all these steps work together, once again using the heart. It is true that a function of the heart is to pump blood. The heart does pump blood; the body is a complex system of parts that by definition aid in certain activities of the whole body, such as locomotion, self-maintenance, copulation; the concepts “heart” and “blood" are recognizably components of the scheme we employ in describing this complex system; and blood-pumping does contribute to activities of the whole organism to which many of its organs, tissues and other parts definitionally contribute, (p. 160)
There are several difficulties with this analysis. They appear below, roughly in order of increasing severity. First, Beckner’s problems with the system S are in some ways worse than Canfield’s; for Beckner explicitly wants to include arti facts, and in addition he says much more definite things about the relationship among P, F, and S. So in this case, when we say the function of a watch’s sweep hand is making seconds easier to read, we must not only find a system of which the sweep hand is a part, and in or of which “making seconds easier to read” is an activity, but this activity must be or contribute to one to which a number of the system’s critical parts definitionally contribute. In the case of the natural functions of the organs and other parts of organisms, the system S is typically a natural unit, easy to subdivide from the environment: the organism itself. But for the conscious functions of artifacts, such systems, if they can be found at all, must be hacked out of the environment rather arbitrarily. With no more of a guide than Beckner has given us, there is nothing like a guarantee that we can always find such a system. Accordingly, when our minds boggle—as I take it they do in trying to conceive of “making seconds easier to read” being an activity at all, much less one meeting all of
Functions
41
the other conditions of this analysis—we have to say that the anal ysis is at best too obscure to be applicable to such cases, and is per haps just mistaken. A second difficulty stems directly from the first. It is not at all clear that functions—even natural functions—have to be activities at all, let alone activities of the sort required by Beckner. Making seconds easier to read is an example, but there are many others: preventing skids in wet weather, keeping your pants up, or prop ping open my office door. All of these things are legitimate functions (of tire treads, belts, and doorstops, respectively); none are activi ties in any recognizable sense. Thirdly, we noticed in our discussion of Canfield that some thing could do a useful thing by accident, in the appropriate sense of “accident.” Similarly, a part of a system meeting all of Beckner’s criteria might easily make a contribution to an activity of that sys tem also quite by accident. For example, an internal-combustion engine is a system satisfying Beckner’s criteria for S. If a small nut were to work itself loose and fall under the valve-adjustment screw in such a way as to adjust properly a poorly adjusted valve, it would make an accidental contribution to the smooth running of that engine. We would never call the maintenance of proper valve adjustment the function, of the nut. If it got the adjustment right it was just an accident. But on Beckner’s formulation, we would have to call that its function. The nut does keep the valve adjusted; the engine is a complex system of parts that by definition aid in certain activities of the whole body, such as generation of torque and self maintenance (lubrication, heat dissipation); the concepts “nut," “valve,” and “valve adjustment” are components of the scheme we employ in describing this complex system; and proper valve adjust ment does contribute to the smooth running of the (whole) engine, which is an activity to which many of the other parts of the engine definitionally contribute (flywheel, connecting rod, exhaust ports). The final difficulty is also related to one we raised for Can field’s analysis. There we noticed that if an artifact was explicitly designed to do something, that usually determines its function, irre spective of how well or badly it does the thing it was supposed to do. An analogous point can be made here. If X was designed to do F, then Y is Ans function regardless of what contributions X does in fact make or fail to make. For example, the function of the federal automotive safety regulations is to make driving and riding in a car safer. And this is so even if they actually have just the opposite effect, through some psychodynamic or automotive quirk.
42
Larry Wright
So in spite of their enormous differences, this analysis and Canfield's fail for very similar reasons: problems with the notion of system S, failure to rule out some accidental cases, and general inability to account for the obvious role of design. There have been several other interesting attempts in the recent literature to provide an analysis of function. Most notable are those by Carl Hempel,8 Hugh Lehman,9 Richard Sorabji,10 Fran cisco Ayala," and Michael Ruse.12 The last two of these do a some what better job on the function/accident distinction than the ones we have examined. But other than that, a discussion of these anal yses would be largely redundant on the discussions of Beckner and Canfield. So I think we have gone far enough in clarifying the issues to begin constructing an alternative analysis.
An Alternative View
The treatments we have so far considered have overlooked, ignored, or at any rate failed to make, one important observation: hat functional ascriptions are—intrinsically, if you will—explanaory. Merely saying of something, X, that it has a certain function, is to offer an important kind of explanation of X. The failure to con sider this, or at least take it seriously, is, I think, responsible for the systematic failure of these analyses to provide an accurate account of functions. There are two related considerations which urge this observa tion upon us. First, the “in order to” in functional ascriptions is a teleological “in order to.” Its role in functional ascriptions (the heart beats in order to circulate blood) is quite parallel to the role of “in order to’ in goal ascriptions (the rabbit is running in order to escape from the dog). Accordingly, we should expect functional ascriptions to be explanatory in something like the same way as goal ascriptions." When we say that the rabbit is running in order to escape from the dog, we are explaining why the rabbit is run ning. If we say that John got up early in order to study, we are offering an explanation of his getting up early. Similarly in the functional cases. When we say that the distributor has that cover in order to keep the rain out, we are explaining why the distributor has that cover. And when we say the heart beats in order to pump blood, we are ordinarily taken to be offering an explanation of why the heart beats. This last sort of case represents the most trouble some problem in the logic of function, but it must be faced squarely,
Functions
43
and, once faced, I think its solution is fairly straightforward. The second consideration which recommends holding out for the explanatory status of functional ascriptions is the contextual equivalence of several sorts of requests. Consider:
1. What is the function ofX? 2. Why do C’s have JCs? 3. Why do Xs do Y! In the appropriate context, each of these is asking for the function of X. “What is the function of the heart?,” “Why do humans have a heart?,” “Why does the heart beat?" All are answered by saying, “To pump blood,” in the context we are considering. Questions of the second and third sort, being “Why?” questions, are undisguised requests for explanations. So in this context functional attributions are presumed to be explanatory. And why-form function requests are by no means bizarre or esoteric ways of asking for a function. Consider:
Why do porcupines have sharp quills? Why do (some) watches have a sweep-second hand?
Why do ducks have webbed feet?
Why do headlight bulbs have two filaments? These are rather ordinary ways of asking for a function. And if that is so, then it is ordinarily supposed that a function explains why each of these things is the case. The function of the quills is why porcupines have them, and so forth. Moreover, the kind of explanatory role suggested by both of these considerations is not the anemic "WTiat’s it good for?” sort of thing often imputed to functional explanations. It is rather some thing more substantial than that. If to specify the function of quills is to explain why porcupines have them, then the function must be the reason they have them. That is, the ascription of a function must be explanatory in a rather strong sense. To choose the weaker interpretation, as Canfield does,14 is once again to run afoul of the function-accident distinction. For, to use his example, if “Why do animals have livers?” is a request for a function, it cannot be ren dered “What is the liver good for?” Livers are good for many things
44
Larry Wright
which are not their functions, just like anything else. Noses are good for supporting eyeglasses, fountain pens are good for cleaning your fingernails, and livers are good for dinner with onions. No, the function of the liver is that particular thing it is good for which explains why animals have them. Putting the matter in this way suggests that functional ascription-explanations are in some sense etiological, concern the causal background of the phenomenon under consideration. And this is indeed what I wish to argue: functional explanations, although plainly not causal in the usual, restricted sense, do con cern how the thing with the function got there. Hence they are eti ological, which is to say “causal” in an extended sense. But this is still a very contentious view. Functional and teleological explana tions are usually contrasted with causal ones, and we should not abandon that contrast lightly: we should be driven to it. What drives us to this position is the specific difficulty the best-looking alternative accounts have in making the function/accident distinction. We have seen that no matter how useful it is for X to do Z, or what contribution A”s doing Z makes within a complex system,15 these sorts of consideration are never sufficient for saying that the function of A” is Z. It could still turn out that X did Z only by accident. But all of the accident counterexamples can be avoided if we include as part of the analysis something about how X came to be there (wherever): namely, that it is there because it does Z— with an etiological “because.” The buckle, the heart, the nose, the engine nut, and so forth were not there because they stop bullets, throb, support glasses, adjust the valve, and all the other things which were falsely attributed as functions, respectively. Those pseudo functions could not be called upon to explain how those things got there. This seems to be what was missing in each of those cases. In other words, saying that the function of X is Z is saying at least that
(1) X is there because it does Z. or Doing Z is the reason X is there. or ThatX does Z is why X is there. where because, reason,” and “why” have an etiological force. And it turns out that “X is there because it does Z,”16 with the proper
Functions
45
understanding of “because,” “does,” and “is there” provides us with not only a necessary condition for the standard cases of functions, but also the kernel of an adequate analysis. Let us look briefly at those key terms. “Because” is of course to be understood in its explanatory rather than evidential sense. It is not the “because” in “It is hot because it is red.” More importantly, “because” is to be taken (as it ordinarily is anyway) to be indifferent to the philosophical reasons/causes distinction. The “because” in “He did not go to class because he wanted to study” and in “It exploded because it got too hot” are both etiological in the appropriate way." And finally, it is worth pointing out here that in this sense “A because B” does not require that B be either necessary or sufficient for A. Racing cars have airfoils because they generate a downforce (negative lift) which augments traction. But their generation of negative lift is neither necessary nor sufficient for racing cars to have wings: they could be there merely for aesthetic reasons, or they could be forbid den by the rules. Nevertheless, if you want to know why they are there, it is because they produce negative lift. All of this comes to saying that “because” here is to be taken in its ordinary, conversa tional, causal-explanatory sense. Complications arise with respect to “does” primarily because on the above condition “Z is the function of X” is reasonably taken to entail “X does Z.” Although in most cases there is no question a all about what it is for X to do Z, the matter is highly context-depei dent and so perhaps I should mention an extreme case, if only £ notice that we should include it. In some contexts we will allow tha X does Z even though Z never occurs. For example, the button on the dashboard activates the windshield washer system (that is what it does, I can tell by the circuit diagram) even though it never has and never will. An unused organic or organismic emergency reaction might have the same status. All that seems to be required is that X be able to do Z under the appropriate conditions; for exam ple, when the button is pushed or in the presence of a threat to safety. The vagueness of “is there" is probably what Beckner and Canfield were trying to avoid by introducing the system S into their formulations. It is much more difficult, however, to avoid the diffi culties with the system S than to clarify adequately this more gen eral place-marker. “Is there” is straightforward and unproblematic in most contexts, but some illustrations of importantly different ways in which it can be rendered might be helpful. It can mean
46
Larry Wright
something like “is where it is,” as in "keeping food out of the wind pipe is the reason the epiglottis is where it is.” It can mean “C’s have them," as in “animals have hearts because they pump blood.” Or it can mean merely “exists (at all),” as in “keeping snow from drifting across roads (and so forth) is why there are snow fences.” Now, saying that (1), understood in this way, should be con strued as a necessary condition for taking Z to be the function of .V, is merely to put in precise terms the moral of our examination of the function/accident distinction. We saw above that, the accident counterexamples could not meet this requirement. On the other hand, this condition is met in all of the center-of-the-page cases. This is quite easy to show in the conscious cases. When we say the function of A’ is Z in these cases, we are saying that at least some effort was made to get A" (sweep hand, button on dashboard) where it is precise!}’ because it does Z (whatever). Doing Z is the reason X is there. That is why the effort was made. The reason the sweepsecond hand is there is that it makes seconds easier to read. It is there because it does that. Similarly, rifles have safeties because they prevent accidental discharge. It is only slightly less obvious how natural functions can sat isfy (i): We can say that the natural function of something—say, an organ in an organism—is the reason the organ is there by invoking natural selection. If an organ has been naturally differentially selected-for by virtue of something it does, we can say that the rea son the organ is there is that it does that something. Hence we can say animals have kidneys because they eliminate metabolic wastes from the bloodstream; porcupines have quills because they protect them from predatory enemies; plants have chlorophyll because chlorophyll enables plants to accomplish photosynthesis; the heart beats because its beating pumps blood. And each of these can be rather mechanically put in the “reason that” form. The reason por cupines have quills is that they protect them from predatory ene mies, and so forth. It is easy to show that this formula does not represent a suffi cient condition for being a function, which is to say there is some thing more to be said about precisely what it is to be a function. The most easily generable set of cases to be excluded is of this kind: oxy gen combines readily with hemoglobin, and that is the (etiological) reason it is found in human bloodstreams. But there is something colossalfy fatuous in maintaining that the function of that oxygen is to combine with hemoglobin, even though it is there because it does that. The function of the oxygen in human bloodstreams is
Functions
47
providing energy in oxidation reactions, not combining with hemoglobin. Combining with hemoglobin is only a means to that end. This is a useful example. It points to a contrast in the notion of because employed here which is easy to overlook and crucial to an elucidation of functions. As I pointed out above, if producing energy is the function of the oxygen, then oxygen must be there (in the blood) because it pro duces energy. But the “because” in “It is there because it produces energy is importantly different from the “because” in “It is there because it combines with hemoglobin.” They suggest different sorts of etiologies. If carbon monoxide, which we know to combine read ily with hemoglobin, were suddenly to become able to produce energy by appropriate (non-lethal) reactions in our cells and, fur ther, the atmosphere were suddenly!!) to become filled with CO, we could properly say that the reason CO was in our bloodstreams was that it combines readily with hemoglobin. We could not properly say, however, that CO was there because it produces energy. And that is precisely what we could say about oxygen, on purely evolu tionary-etiological grounds. All of this indicates that it is the nature of the etiology itself which determines the propriety of a functional explanation; there must be specifically functional etiologies. When we say the function of X is Z (to do Z) we are saying that X is there because it does Z but with a further qualification. We are explaining how X came t be there, but only certain kinds of explanations of how X came to b there will do. The causal/functional distinction is a distinction among etiologies; it is not a contrast between etiologies and some thing else. This distinction can be displayed using the notion of a causal consequence.18 When we give a functional explanation of X by appeal to Z (“X does Z”), Z is always a consequence or result of Xs being there (in the sense of “is there” sketched above).19 So when we say that Z is the function ofA’, we are not only saying that A is there because it does Z, we are also saying that Z is (or happens as) a result or consequence of A”s being there. Not only is chlorophyll in plants because it allows them to perform photosynthesis, photosyn thesis is a consequence of the chlorophyll’s being there. Not only is the valve-adjusting screw there because it allows the clearance to be easily adjusted, the possibility of easy adjustment is a consequence of the screw’s being there. Quite obviously, “consequence of" here does not mean "guaranteed by.” “Z is a consequence of A,” very much like “A does Z” earlier, must be consistent with Z’s not occur-
48
Larry Wright
ring. When we say that photosynthesis is a consequence of chloro phyll. we allow that some green plants may never be exposed to light, and that all green plants may at some time or other not be exposed to light. Furthermore, this consequence relationship does not mean that whenever Z does occur, happen, obtain, exist, and so forth, it is as a consequence ofX. There is room for a multiplicity of sufficient conditions, overdetermined or otherwise. Other things besides the adjusting screw may provide easy adjustment of the clearance. This (the inferential) aspect of consequence, as that notion is used here, can be roughly captured by saying that there are circumstances (of recognizable propriety) in which A' is nonredundant forZ. The aspect of “consequence” of central importance here, however, is its asymmetry. “4 is a consequence of B" is in vir tually every context incompatible with "B is a consequence of A.” The source of this asymmetry is difficult to specify, and I shall not try.20 It is enough that it be clearly present in the specific cases. Accordingly, if we understand the key terms as they have been explicated here, we can conveniently summarize this analysis as follows: The function of A' is Z means
(2) (a) X is there because it does Z, (6) Z is a consequence (or result) of X"s being there.
The first part, (a), displays the etiological form of functional ascrip tion-explanations, and the second part, (b), describes the convolu tion which distinguishes functional etiologies from the rest. It is the second part of course which distinguishes the combining with hemoglobin from the producing of energy' in the oxygen-respiration example. Its combining with hemoglobin is emphatically not a con sequence of oxygen’s being in our blood; just the reverse is true. On the other hand, its producing energy is a result of its being there. The very' best evidence that this analysis is on the right track is that it seems to include the entire array of standard cases we have been considering, while at the same time avoiding several very persistent classes of counterexamples. In addition to this, how ever, there are some more general considerations which urge this position upon us.21 First, and perhaps most impressive, this analy sis shows what it is about functions that is teleological. It provides an etiological rationale for the functional “in order to,” just as recent discussions have for other teleological concepts. The role of
Functions
49
the consequences of A' in its own etiology provide functional ascrip tion-explanations with a convoluted forward orientation which pre cisely parallels that found by recent analyses in ascription-expla nations employing the concepts goal and intention.22 In a functional explanation, the consequences ofAns being there (where it is, and so forth) must be invoked to explain why X is there (exists, and so forth). Functional characterizations, by their very nature, license these explanatory appeals. Furthermore, as I hinted earlier, (6) is often simply implicit in the “because” of (a). When this is so, the “because” is the specifically teleological one sometimes identified as peculiarly appropriate in functional contexts. The peculiarly func tional “because” is the normal etiological one, except that it is lim ited to consequences in this way. The request for an explanation as well will very often contain this implicit restriction, hence limiting the appropriate replies to something in terms of this “because”— that is, to functional explanations. “Why is it there?” in some con texts, and “What does it do?” in most, unpack into “What conse quences does it have that account for its being there?" The second general consideration which recommends this analysis is that it both accounts for the propriety of, and at the same time elucidates the notion of, natural selection. To make this clear, it is important first to say something about the unqualified notion of selection, from which natural selection is derived. Accord ing to the standard view, which I will accept for expository pur poses, the paradigm cases of selection involve conscious choice, per haps even deliberation. We can then understand other uses of “select” and “selection” as extensions of this use: drawing attention to specific individual features of the paradigm which occur in sub conscious or nonconscious cases. Of course, the range of extensions arrays itself into a spectrum from more or less literal to openly metaphorical. Now, there is an important distinction within the paradigmatic, conscious cases. I can say I selected something, X, even though I cannot give a reason for having chosen it: I am asked to select a ball from among those on the table in front of me. I choose the blue one and am asked why I did. I may say something like “I don’t know; it just struck me, I guess.” Alternately, I could without adding much give something which has the form of a rea son: “Because it is blue. Yes, I’m sure it was the color.” In both of these cases I want to refer to the selection as “mere discrimination,” for reasons which will become apparent below. On the other hand, there are a number of contexts in which another, more elaborate reply is possible and natural. I could say something of the form "I
■I
50
Larry Wright
selected X because it does Z,” where Z would be some possibility opened by, some advantage that would accrue from, or some other result of having (using, and so forth) X. “I chose American Airlines because its five-across seating allows me to stretch out. Oi T u) selected DuPont Nomex because of the superior protection it affords in a fire.”25 Let me refer to selection by virtue of resultant advantage of this sort as “consequence-selection. Plainly, it is this kind of selection, as opposed to mere discrimination, that lies behind conscious functions: the consequence is the function. Equally plainly, it is specifically tins kind of selection of which nat ural selection represents an extension. But the parallel between natural selection and conscious con sequence-selection is much more striking than is sometimes thought. True, the presence or absence of volition is an important difference, at least in some contexts. We might want to say that natural selection is really self-selection, nothing is doing the select ing; given the nature of A’, Z, and the environment, X will auto matically be selected. Quite so. But here the above distinction between kinds of conscious selection becomes crucial. For conse quence-selection, bj’ contrast with mere discrimination, de-emphasizes volition in just such a way as to blur its distinction from nat ural selection on precisely this point. Given our criteria, we might well say that X does select itself in conscious consequence-selection. By the very nature of X, Z, and our criteria (the implementation of which may be considered the environment), X will automatically be selected/’ The cases are very close indeed. Let us now see how this analysis squares with the desiderata we have developed. First, it is quite clearly a unifying analysis: the formula applies to natural and conscious functions indifferently. Both natural and conscious functions are functions by virtue of their being the reason the thing with the function “is there,” sub ject to the above restrictions. The differentiating feature is merely the sort of reason appropriate in either case: specifically, whether a conscious agent was involved or no. But in the functional-explana tory context which we are examining, the difference is minimal. When we explain the presence or existence ofX by appeal to a con sequence Z, the overriding consideration is that Z must be or cre ate conditions conducive to the survival or maintenance of X. The exact nature of the conditions is inessential to the possibility of this form of explanation: it can be looked upon as a matter of mere eti ological detail, nothing in the essential form of the explanation. In any given case something could conceivably get a function through
Functions
51
either sort of consideration. Accordingly, this analysis begs no the ological questions. The organs of organisms could logically possibly get their functions through God’s conscious design; but we can also make perfectly good sense of their functions in the absence of divine intervention. And in either case they would be functions in pre cisely the same sense. This of course was accomplished only by dis allowing explicit mention of intent or purpose in accounting for con scious functions. Nevertheless, the above formula can account for the very close relationship between design and function which the previous analyses could not. For, excepting bizarre circumstances, in virtually all of the usual contexts, X was designed to do Z simply entails thatX is there because it results in Z. Second, this analysis makes a clear and cogent distinction between function and accident. The things X can be said to do by accident are the things it results in which cannot explain how it came to be there. And we have seen that this circumvents the acci dent counterexamples brought to bear on the other analyses. It is merely accidental that the chlorophyll in plants freshens breath. But what it does for plants when the sun shines is no accident— that is why it is there. Furthermore, in this sense, “X did Z acci dentally” is obviously consistent with Xs doing Z having welldefined causal antecedents, just like the normal cases of other sorts of accident (automobile accidents, accidental meetings, and so forth). Given enough data it could even have been predictable that the belt buckle would deflect the bullet. But such deflection was still in the appropriate sense accidental: that is not why the buckle was there. Furthermore, it is worth noting that something can get a func tion—either conscious or natural—as the result of an accident of this sort. Organismic mutations are paradigmatically accidental in this sense. But that only disqualifies an organ from functionhood for the first—or the first few—generations. If it survives by dint of its doing something, then that something becomes its function on this analysis. Similarly for artifacts. For example, if an earthquake shifted the rollers of a transistor production-line conveyor belt, causing the belt to ripple in just such a way that defective transis tors would not pass over the ripple, while good transistors would, we could say that the ripple was functioning as a quality control sorter. But it would be incorrect to say that the ripple had the func tion of quality control sorting. It does not have a function at all. It is there only by accident. Sorting can, however, become its function if its sorting ability ever becomes a reason for preserving the ripple:
52
Larry Wright
if. for example, the company decides against repairing the conveyor belt for that reason. This accords nicely with Richard Sorabji’s com ment that in conscious cases, saying the function of A' is Z requires at least “that some efforts are or would if necessary be made” to obtain Z from A’.“ Third, the notion of something having more than one function is derivative. It is obtained by substituting something like "partly because”26 for "because” in the formula. Brushing dust off the num bers is one of the functions of the watch's sweep-second hand if that feature is one of the (restricted, etiological) reasons the sweep hand is there. Similarly in the case of natural functions. If two or three things that livers do all contribute to the sunaval of organisms which have livers, we must appeal to all three in an evolutionary account of why those organisms have livers. Hence the liver would have more than one function in such organisms: we would have to say that each one was a function of the liver. Happily, the analysis I am here proposing also accounts for the undoubted attractiveness of the other analyses we have exam ined. Beckner's first analysis is virtually included in this one under the rubric “X does Z.” The rest of the formula can be thought of as a qualification to avoid some rather straightforward counterexamoles which Beckner himself is concerned to circumvent in his more •ecent attempt. Canfield’s “usefulness" is even easier to accommo date: the usefulness of something, Z, which X does is very usually an informative way of characterizing why A has survived in an evo lutionary process, or the reason X was consciously constructed. The important point to notice is that this is only usually the case, not necessarily: not all useful Z’s can explain survival and some things are constructed to do wholly useless things. As for Beckner’s most recent analysis, the complex, mutually contributory relationship among parts central to it is precisely the sort of thing often respon sible for the survival and reproduction of organisms on one hand, and for the construction of complex mechanisms on the other. Again the valuable features of that analysis are incorporated in this one. There is still one sort of case in which we clearly want to be able to speak of a function, but which offends the letter of this anal ysis as it stands. In several contexts, some of which we have already examined, we want to be able to say that X has the func tion Z, even though X cannot be said to do Z. X is not even able to do Z under the requisite conditions. In the cases of this sort I have already mentioned (the defective washer switch and ineffective
Functions
53
governmental safety regulations), it has seemed necessary to itali cize (emphasize, underline) the word “function” in order to make its use plausible and appropriate. This is a logical flag: it signals that a special or peculiar contrast is being made, that the case departs from the paradigms in a systematic but intelligible way. Accord ingly, an analysis has to make sense of such a case as a variant. On the present analysis, the italic type signals the dropping of the (usually presumed) second condition. X does not result in Z, although, paradoxically, doing Z is the reason X is there. Of course, in the abstract, this sounds fatuous. But we have already seen cases in which it is natural and appropriate. That is the reason X(switch, safety regulations) is there. And a slightly more defensive formula tion of (2) will include them directly: a functional ascription-expla nation accounts for X*s being there by appeal to Xfs resulting in Z. These cases do appeal lox's resulting in Z to explain the occurrence of X, even though X does not result in Z. So the form of the expla nation is functional even in these peculiar cases. Interestingly, this account even handles the exotic fact that these italicized functions of X can cease being even italicized func tions without dispensing with or directly altering X. (Something that X did not do can stop being its function!) For example, if the ineffective safety regulations were superseded by another set, and merely left on the books through legislative sloth or expediency, we would no longer even say they had the (italicized) function of mak ing driving less dangerous. But, of course, that would no longer be the reason they were there. The explanation would then have to appeal to legislative sloth or expediency. This is usually done with verb tenses: that was its function, but is not any longer; that was why it was there at one time, but is not why it is still there. A sim ilar treatment can be given vestigial organs, such as the vermiform appendix in humans.
Notes 1. Richard Sorabji, "Function," Philosophical Quarterly, 14 (1964),
290.
2. Morton Beckner, The Biological Way of Thought (New York, 1959), ch. 6. 3. Beckner gives an alternative formulation in which we can speak of activities as having functions, instead of things. I have abbreviated it here for convenience and clarity. The logical points are the same.
Larry Wright
54
4. John Canfield. “Teleological Explanations in Biology'," The British Journal for the Philosophy of Science, vol. 14 (1964).
5. Morton Beckner, "Function and Teleology', Journal of the History of Biology, vol. 2 (1969). 6. As before. Beckner gives an alternative formulation so that we can speak either of a thing or of an activity having a function. My treatment will be limited to things, but again the logical points are the same.
7. Beckner seems to suggest (p. 160, top) that F must be an activity of the whole system S. which, of course, would conflict with part of 3. But his illustration, reproduced below, suggests the phrasing I have used here. 8. Carl Hempel, "The Logic of Functional Analyses." in L. Gross (ed.) Symposium on Sociological Theory (New York. 1959'.
9. Hugh Lehman, “Functional Explanations in Biology," Philosophy of Science, vol. 32 (1965). 10. Sorabji, op. tit. 11. Francisco J. Ayala. “Teleological Explanation in Evolutionary Biology,” Philosophy of Science, vol. 37 (1970).
12. Michael E. Ruse, “Function Statements in Biology’,” Philosophy of Science, vol. 38 (1970). 13. This is not to abandon, or even modify, the previous distinction between functions and goals: the point can be made in this form only given the distinction. Nevertheless, support is provided for the analysis I am pre senting here by the fact that the “in order to" of goal-directedness can be afforded a parallel treatment. For that parallel treatment see my paper “Explanation and Teleology," in the June 1972 issue ofPhilosophy ofScience.
14. Canfield, op. tit., p. 295. 15. It is sometimes urged that this sort of thing is all a teleological explanation is asserting; this is all “why?" asks in these contexts. 16. I take the other forms to be essentially equivalent and subject, mutatis mutandis, to the same explication.
17. Of course, it follows that the notion of a reason offered in one of the alternative formulations is the standard conversational one as well: the reason it exploded was that it got too hot. IS. The qualification “causal” here serves merely to indicate that this is not the purely inferential sense of “consequence.” I am not talking about the result or consequence of an argument—e.g., necessary conditions for the truth of a set of premises. The precise construction of “consequence” appropriate here will become clear below.
Functions
55
19. It is worth recalling here that “is there" can only sometimes, but not usually, he rendered "exists (at all).” So, contrary to many accounts, what is being explained, and what Z is the result of, can very often not be characterized as "that X exists” simpliciter.
20. It is often claimed that the asymmetry is temporal, but there are many difficulties with this view, Douglas Gashing, in "Causation and Recipes, Mind (Oct, 1955), attempts to account for it in terms of manipulabilily with some success. But manipulability is even less generally appli cable than time order, so, as far as I know, the problem remains. 21. The following considerations are intended primarily as support for the entire analysis considered whole. Since (a) has already been exam ined extensively, however, I have biased the argument slightly to empha size (6). 22. The primary discussions of this sort I have in mind are those in Charles Taylor’s Explanation of Behavior and the literature to which it has given rise. 23. Of course the advantage is not always stated explicitly; "I chose American because of its five-across seating.” But for it to be selection of the sort described here, as opposed to mere discrimination, something like an advantage must be at least implicit.
24. This is a version of the old problem about the tension between rationality and freedom. 25. Sorabji, op. cit., p. 290. 26. Again, it is worth pointing out that "partlv” here does not indicate that “because,” when not so qualified, represents a sufficient condition rela tionship. It merely serves to indicate that more than one thing plays an explanatorily relevant role in this particular case. More than one thing must be mentioned to answer adequately the functional "why?” question in this context. But that answer, as usual, need not provide a sufficient con dition for the occurrence of X.
1
CHAPTER 2
Functional Analysis* Robert Cummins
A survey of the recent philosophical literature on the nature of functional analysis and explanation, beginning with the classic essays of Hempel in 1959 and Nagel in 1961, reveals that philo sophical research on this topic has almost without exception pro ceeded under the following assumptions:'
(A) The point of functional characterization in science is to explain the presence of the item (organ, mechanism, process or whatever) that is functionally characterized. (B) For something to perform its function is for it to have certain effects on a containing system, which effects contribute to the performance of some activity of, or the maintenance of some condition in, that containing sys tem.
Putting these two assumptions together we have: a function-ascrib ing statement explains the presence of the functional!}’ character ized item i in a system s by pointing out that i is present in s because it has certain effects on s. Give or take a nicety, this fusion of (A) and (B) constitutes the core of almost every recent attempt to give an account of functional analysis and explanation. Yet these assumptions are just that: assumptions. They have never been sys* Reprinted from Robert Cummins, “Functional Analysis,” The Journal of Philosophy 72 (1975), pp. 7-11-765, with kind permission from The Journal of Philosophy and the author. © 1975 The Journal of Philosophy, Inc.
57
58
Robert Cummins
tematically defended; generally they are not defended at all. I think there are reasons to suspect that adherence to (A) and (B) has crip pled the most serious attempts to analyze functional statements and explanation, as I will argue in sections I and II below. In sec tion III, I will briefly develop an alternative approach to the prob lem. This alternative is recommended largely by the fact that it emerges as the obvious approach once we take care to understand why accounts involving (A) and (B) go wrong. I
I begin this section with a critique of Hempel and Nagel. The objections are familiar for the most part, but it will be well to have them fresh in our minds, for they form the backdrop against which I stage my attack on (A) and (B). Hempel’s treatment of functional analysis and explanation is a classic example of the fusion of (A) and (B). He begins by consid ering the following singular function-ascribing statement: (1) The heartbeat in vertebrates has the function of circu lating the blood through the organism.
He rejects the suggestion that ‘function’ can simply be replaced by ‘effect’ on the grounds that, although the heartbeat has the effect of producing heartsounds, this is not its function. Presuming (B) from the start, Hempel takes the problem to be how the effect the hav ing of which is the function of the heartbeat (circulation) is to be distinguished from other effects of the heartbeat (e.g., heart sounds). His answer is that circulation, but not heartsounds, ensures a necessary condition for the “proper working of the organ ism.” Thus, Hempel proposes (2) as an analysis of (1).
(2) The heartbeat in vertebrates has the effect of circulat ing the blood, and this ensures the satisfaction of cer tain conditions (supply of nutriment and removal of waste) which are necessary for the proper working of the organism. As Hempel sees the matter, the main problem with this anal ysis is that functional statements so construed appear to have no explanatory force. Since he assumes (A), the problem for Hempel is
functional Analysis
59
to see whether (2) can be construed as a deductive nomological explanans for the presence of the heartbeat in vertebrates, and, in general, to see whether statements having the form of (2) can be construed as deductive nomological explananda for the presence in a system of some trait or item that is functionally characterized. Suppose, then, that we are interested in explaining the occur rence of a trait i in a system s (at a certain time t), and that the fol lowing functional analysis is offered:
(a) At t, s functions adequately in a setting of kind c (char acterized by specific internal and external conditions). (b) s functions adequately in a setting of kind c only if a cer tain necessary condition, n, is satisfied. (c) If trait i were present in s then, as an effect, condition n would be satisfied.
(d) Hence, at t, trait i is present in s (Hempel, p. 310). (d), of course, does not follow from (a)-(c), since some trait i' differ ent from i might well suffice for the satisfaction of condition n. The argument can be patched up by changing (c) to (c'): “Condition n would be satisfied in s only if trait i were present in s,” but Hempel rightly rejects this avenue on the grounds that instances of the resulting schema would typically be false. It is false, for example, that the heart is a necessary condition for circulation in verte brates, since artificial pumps can be, and are, used to maintain the flow of blood. We are thus left with a dilemma. If the original schema is correct, then functional explanation is invalid. If the schema is revised so as to ensure the validity of the explanation, the explanation will typically be unsound, having a false third premise. Ernest Nagel offers a defense of what is substantially Hempel’s schema with (c) replaced by (c'). ... a teleological statement of the form, “The function of A in a system S with organization C is to enable S in the envi ronment E to engage in process P,” can be formulated more explicitly by: every system S with organization C and in environment E engages in process P; if S with organization C and in environment E does not have A, then S does not engage in P; hence, S with organization C must have A (Nagel, p. 403).
60
Robert Cummins
Thus he suggests that (3) is to be rendered as (4): (3) The function of chlorophyll in plants is to enable them to perform photosynthesis.
(41A necessary condition for the occurrence of photosynthe sis in plants is the presence of chlorophyll. So Nagel must face the second horn of Hempel’s dilemma: (3) is pre sumably true, but (4) may well be false. Nagel is, of course, aware of this objection. His rather curious response is that, as far as we know, chlorophyll is necessary for photosynthesis in the green plants (p. 404). This may be so, but the response will not survive a change of example. Hearts are not necessary for circulation, artifi cial pumps having actually been incorporated into the circulatory systems of vertebrates in such a wav as to preserve circulation and
life. A more promising defense of Nagel might run as follows. Although it is true that the presence of a working heart is not a nec essary condition of circulation in vertebrates under all circum stances, still, under normal circumstances—most circumstances in fact—a working heart is necessary for circulation. Thus it is per haps true that, at the present stage of evolution, a vertebrate that has not been tampered with surgically would exhibit circulation only if it were to contain a heart. If these circumstances are specif ically included in the explanans, perhaps we can avoid Hempel’s dilemma. Thus, instead of (4) we should have: (4 ) At the present stage of evolution, a necessary condition for circulation in vertebrates that have not been surgi cally tampered with is the operation of a heart (properly incorporated into the circulator}' system). (4 ), in conjunction with statements asserting that a given verte brate exhibits circulation and has not been surgically tampered with and is at the present stage of evolution, will logically imply that that vertebrate has a heart. It seems, then, that the Hempelian objection could be overcome if it were possible, given a true function-ascribing statement like (1) or (3), to specify “normal circumstances” in such a way as to make it true that, in those cir cumstances, the presence of the item in question is a necessary con dition for the performance of the function ascribed to it.
Functional Analysis
61
This defense has some plausibility as long as we stick to the usual examples drawn from biology. But if we widen our view a bit, even within biology, I think it can be shown that this defense of Nagel’s position will not suffice. Consider the kidneys. The function of the kidneys is to eliminate wastes from the blood. In particular, the function of my left kidney is to eliminate waste from my blood. Yet the presence of my left kidney is not, in normal circumstances, a necessary condition for the removal of the relevant wastes. Only if something seriously abnormal should befall my right kidney would the operation of my left kidney become necessary, and this only on the assumption that I am not hooked up to a kidney machine.2 A less obvious counterexample derives from the well-attested fact of hemispherical redundancy in the brain. No doubt it is in principle possible to specify conditions under which a particular duplicated mechanism would be necessary for normal functioning of the organism, but (a) in most cases we are not in a position actu ally to do this, though we are in a position to make well-confirmed statements about the functions of some of these mechanisms, and (b) these circumstances are by no means the normal circumstances. Indeed, given the fact that each individual nervous system develops somewhat differently owing to differing environmental factors, the circumstances in question might well be different for each individ ual, or for the same individual at different times. Apparently Nagel was pursuing the wrong strategy' in attempting to analyze functional ascriptions in terms of necessary’ conditions. Indeed, we are still faced with the dilemma noticed by Hempel: an analysis in terms of necessary conditions yields a valid but unsound explanatory schema; analysis in terms of sufficient conditions along the lines proposed by Hempel yields a schema with true premises, but validity is sacrificed. Something has gone wrong, and it is not too difficult to locate the problem. An attempt to explain the presence of something by appeal to what it does—its function—is bound to leave unexplained why something else that does the same thing—a functional equiva lent—isn’t there instead. In itself, this is not a serious matter. But the accounts we have been considering assume that explanation is a species of deductive inference, and one cannot deduce hearts from circulation. This is what underlies the dilemma we have been con sidering. At best, one can deduce circulators from circulation. If we make this amendment, however, we are left with a functionally tainted analysis; “the function of the heart is to circulate the blood
62
Robert Cummins
is rendered “a blood circulator is a (necessary/sufficient) condition of circulation, and the heart is a blood circulator.” The expression in italics is surely as much in need of analysis as the analyzed expres sion. The problem, however, runs much deeper than the fact that the performance of a certain function does not determine how that function is performed. The problem is rather that to “explain" the presence of the heart in vertebrates by appeal to what the heart. does is to “explain” its presence by appeal to factors that, are causally irrelevant to its presence. Even if it were possible, as Nagel claimed, to deduce the presence of chlorophyll from the occurrence of photosynthesis, this would fail to explain the pres ence of chlorophyll in green plants in just the way deducing the presence and height of a building from the existence and length of its shadow would fail to explain why the building is there and has the height it does. This is not because all explanation is causal explanation: it is not. But to explain in the presence of a naturally occurring structure or physical process—to explain why it is there, why such a thing exists in the place (system, context) it does—this does require specifying factors that causally determine the appear ance of that structure or process.3 There is, of course, a sense in which the question, “Why is x there?" is answered by giving x’s function. Consider the following exchange. X asks T, “Why is that thing there (pointing to the gnomon of a sundial)?” Y answers, “Because it casts a shadow on the dial beneath, thereby indicating the time of day.” It is exchanges of this sort that most philosophers have had in mind when they speak of functional explanation. But it seems to me that, although such exchanges do represent genuine explanations, the use of functional language in this sort of explanation is quite dis tinct from its explanatory' use in science. In section III below, I will sketch what I think is the central explanatory use of functional lan guage in science. Meanwhile, if I am right, the evident propriety of exchanges like that imagined between X and Y has led to prema ture acceptance of (A), and hence to concentration on what is, from the point of view of scientific explanation, an irrelevant use of func tional language. For it seems to me that the question, “why is x there?” can be answered by specifying x’s function only if x is or is part of an artifact. F’s answer, I think, explains the presence of the gnomon because it rationalizes the action of the agent who put it there by supplying his reason for putting it there. In general, when we are dealing with the result of a deliberate action, we may explain the result by explaining the action, and we may explain a
Functional Analysis
63
deliberate action by supplying the agent’s reason for it. Thus, when we look at a sundial, we assume we know in a general way how the gnomon came to be there: someone deliberately put it there. But we may wish to know why it was put there. Specifying the gnomon’s function allows us to formulate what we suppose to be the unknown agent’s reason for putting it there: he believed it would cast a shadow such that . . . , and so on. When we do this, we are elabo rating on what we assume is the crucial causal factor in determin ing the gnomon's presence, namely a certain deliberate action. If this is on the right track, then the viability of the sort of explanation in question should depend on the assumption that the thing functionally characterized is there as the result of deliberate action. If that assumption is evidently false, specifying the thing’s function will not answer the question. Suppose it emerges that the sundial is not, as such, an artifact. When the ancient building was ruined, a large stone fragment fell on a kind of zodiac mosaic and embedded itself there. Since no sign of the room remains, Y has mistakenly supposed the thing was designed as a sundial. As it happens, the local people have been using the thing to tell time for centuries; so Y is right about the function of the thing X pointed to.* But it is simply false that the thing is there because it casts a shadow, for there is no agent who put it there “because it casts a shadow.” Again, the function of a bowl-like depression in a huge stone may be to hold holy water, but we cannot explain why it is there by appeal to its function if we know it was left there by pre historic glacial activity. If this is right, then (A) will lead us to focus on a type of expla nation which will not apply to natural systems: chlorophyll and hearts are not “there” as the result of any deliberate action, and hence the essential presupposition of the explanatory move in ques tion is missing. Once this becomes clear, to continue to insist that there must be some sense in which specifying the function of chloro phyll explains its presence is an act of desperation born of thinking there is no other explanatory' use of functional characterization in science. Why have philosophers identified functional explanation exclusively with the appeal to something’s function in explaining why it is there? One reason, I suspect, is a failure to distinguish teleological explanation from functional explanation, perhaps because functional concepts do loom large in “explanations" hav ing a teleological form. Someone who fails to make this distinc tion, but who senses that there is an important and legitimate
64
Robert Cummins
use of functional characterization in scientific explanation, will see the problem as one of finding a legitimate explanatory role for functional characterization within the teleological form. Once we leave artifacts and go to natural systems, however, this approach is doomed to failure, as critics of teleology have seen for some time. This mistake probably would have sorted itself out in time were it not that we do reason from the performance of a function to the presence of certain specific processes and structures, e.g., from photosynthesis to chlorophyll, or from coordinated activity to nerve tissue. This is perfectly legitimate reasoning: it is a species of infer ence to the best explanation. Our best (only) explanation of photo synthesis requires chlorophyll, and our best explanation of coordi nated activity requires nerve tissue. But once we see what makes this reasoning legitimate, we see immediately that inference to an explanation has been mistaken for an explanation itself. Once this becomes clear, it becomes equally clear that (A) has matters reversed: given that photosynthesis is occurring in a particular plant, we may legitimately infer that chlorophyll is present in that plant precisely because chlorophyll enters into our best (only) explanation of photosynthesis, and given coordinated activity on the part of some animal, we may legitimately infer that nerve tis sue is present precisely because nerve tissue enters into our best explanation of coordinated activity in animals. To attempt to explain the heart’s presence in vertebrates by appealing to its function in vertebrates is to attempt to explain the occurrence of hearts in vertebrates by appealing to factors that are causally irrelevant to its presence in vertebrates. This fact has given “functional explanation” a bad name. But it is (A) that deserves the blame. Once we see (A) as an undefended philosophi cal hypothesis about how to construe functional explanations rather than as a statement of the philosophical problem, the correct alternative is obvious: what we can and do explain by appeal to what something does is the behavior of a containing system.5 A much more promising suggestion in the light of these con siderations is that (1) is appealed to in explaining circulation. If we reject (A) and adopt this suggestion, a simple deductive-nomological explanation with circulation as the explicandum turns out to be a sound argument. (5) a. Vertebrates incorporating a beating heart in the usual way (in the way s does) exhibit circulation.
Functional Analysis
65
b. Vertebrate incorporates a beating heart in the usual way. Hence, .s exhibits circulation.
Though by no means flawless, (5) has several virtues, not the least of which is that it does not have biologists passing by an obvious application of evolution or genetics, in favor of an invalid or unsound “functional” explanation of the presence of hearts. Also, the redundancy examples are easily handled; e.g., the removal of wastes is deduced in the kidney case. The implausibility of (A) is obscured in examples taken from biology by the fact that there are two distinct uses of function state ments in biology. Consider the following statements. (a) The function of the contractile vacuole in protozoans is elimination of excess water from the organism. (b) The function of the neurofibrils in the ciliates is coordi nation of the activity of the cilia.
These statements can be understood in either of two ways, (i) They are generally used in explaining how the organism in question comes to exhibit certain characteristics or behavior. Thus (a) explains how excess water, accumulated in the organism by osmo sis, is eliminated from the organism; (b) explains how it happens that the activity of the cilia in paramecium, for instance, is coordi nated. (ii) They may be used in explaining the continued survival of certain organisms incorporating structures of the sort in question by indicating the survival value that would accrue to such organ isms in virtue of having structures of that sort. Thus (a) allows us to infer that incorporation of a contractile vacuole makes it possible for the organism to be surrounded by a semi-permeable membrane, allowing the passage of oxygen into, and the passage of wastes out of, the organism. Relatively free osmosis of this sort is obviously advantageous, and this is made possible by a structure which solves the excess water problem. Similarly, ciliates incorporating neu rofibrils will be capable of fairly efficient locomotion, the survival value of which is obvious.6 The second sort of use occurs as part of an account which, if we are not careful, can easily be mistaken for an explanation of the presence of the sort of item functionally characterized, and this has perhaps encouraged philosophers to accept (A). For it might seem
66
Robert Cummins
that natural selection provides the missing causal link between what something does in a certain type of organism and its presence in that type of organism. By performing their respective functions, the contractile vacuole and the neurofibrils help species incorpo rating them to survive, and thereby contribute to their own contin ued presence in organisms of those species, and this might seem to explain the presence of those structures in the organisms incorpo rating them. Plausible as this sounds, it involves a subtle yet fundamental misunderstanding of evolutionary theory. A clue to the mistake is found in the fact that the contractile vacuole occurs in marine pro tozoans that have no excess-water problem but the reverse prob lem. Thus the function and effect on survival of this structure is not the same in all protozoans. Yet the explanation of its presence in marine and fresh-water species is almost certainly the same. This fact reminds us that the processes actually responsible for the occurrence of contractile vacuoles in protozoans are totally insensi tive to what that structure does. Failure to appreciate this point not only lends spurious plausibility to (A) as applied to biological exam ples; it seriously distorts our understanding of evolutionary theory. .Vhether an organism o incorporates s depends on whether s is specified” by the genetic “plan' which o inherits and which, at a certain level of abstraction, is characteristic of o’s species. Alter ations in the plan are due to mutation. If a plan is altered so that it specifies s' rather than s, then the organisms inheriting this plan will incorporate s', regardless of the function or survival value of s, in those organisms. If the alteration is advantageous, the number of organisms inheriting that plan may increase, and, if it is disad vantageous, their number may decrease. But this has no effect whatever on the plan, and therefore no effect whatever on the occurrence of s, in the organisms in question. One sometimes hears it said that natural selection is an instance of negative feedback. If this is meant to imply that the rel ative success or failure of organisms of a certain type can affect their inherited characteristics, it is simply a mistake: the charac teristics of organisms which determine their relative success are determined by their genetic plan, and the characteristics of these plans are utterly independent of the relative success of organisms having them. Of course, if s is very disadvantageous to organisms having a plan specifying s, then organisms having such plans may disappear altogether, and s will no longer occur. We could, there fore, think of natural selection as reacting on the set of plans gen-
Functional Analysis
67
erated by mutation by weedingout the bad plans: natural selection cannot alter a plan, but it can trim the set. Thus, we may be able to explain why a given plan is not a failure by appeal to the functions of the structures it specifies. Perhaps this is what some writers have had in mind. But this is not to explain why, e.g., contractile vacuoles occur in certain protozoans, it is to explain why the sort of protozoan incorporating contractile vacuoles occurs. Since we can not appeal to the relative success or failure of these organisms to explain why their genetic plan specifies contractile vacuoles, we cannot appeal to the relative success or failure of these organisms to explain why they incorporate contractile vacuoles. Once we are clear about the explanatory role of functions in evolutionary theory, it emerges that the function of an organ or pro cess (or whatever) is appealed to to explain the biological capacities of the organism containing it, and from these capacities conclusions are drawn concerning the chances of survival for organisms of that type. For instance, appeal to the function of the contractile vacuole in certain protozoans explains how these organisms are able to keep from exploding in fresh water. Thus evolutionary biology does not provide support for (A) but for the idea instanced in (5): identi fying the function of something helps to explain the capacities of a containing system.’ (A) misconstrues functional explanation by misidentifying what is explained. Let us abandon (A), then, in favor of the view that functions are appealed to in explaining the capacities of con taining systems, and turn our attention to (B). Whereas (A) is a thesis about functional explanation, (B) is a thesis about the analysis of function-ascribing statements. Perhaps when divorced from (A), as it is in (5), it will fare better than it does in the accounts of Hempel and Nagel.
II
In spite of the evident virtues of (5), (5a) has serious shortcom ings as an analysis of (1). In fact it is subject to the same objection Hempel brings to the analysis that simply replaces ‘function’ by ‘effect’: vertebrates incorporating a working heart in the usual way exhibit the production of heartsounds, yet the production of heart sounds is not a function of hearts in vertebrates. The problem is that, whereas the production of certain effects is essential to the heart’s performing its function, there are some effects production of which is
68
Robert Cummins
irrelevant to the functioning of the heart. This problem is bound to infect any “selected-effects" theory, i.e., any theory' built on (B). What is needed to establish a selected-effects theory is a gen eral formula that identifies the appropriate effects.’ Both Hempel and Nagel attempt to solve this problem by identifying the funct ion of something with just those effects which contribute to the main tenance of some special condition of, or the performance of some special activity of, some containing system. If this sort of solution is to be viable, there must be some principled way of selecting the rel evant activities or conditions of containing systems. For no matter which effects of something you happen to name, there will be some activity of the containing system to which just those effects con tribute, or some condition of the containing system which is main tained with the help of just those effects. Heart activity, for exam ple. keeps the circulatory system from being entirely quiet, and the appendix keeps people vulnerable to appendicitis.’ Hempel suggests that, in general, the crucial feature of a con taining system, contribution to which is to count as the functioning of a contained part, is that the system be maintained in “adequate, or effective, or proper working order" (p. 306). Hempel explicitly declines to discuss what constitutes “proper working order,” pre sumably because he rightly thinks that there are more serious aroblems with the analysis he is discussing than those introduced oy this phrase. But it seems clear that for something to be in work ing order is just for it to be capable of performing its functions, and for it to be in adequate or effective or proper working order is just for it to be capable of performing its functions adequately or effec tively or properly. Hempel seems to realize this himself, for in set ting forth a deductive schema for functional explanation, he glosses the phrase it quest) n as “functi ns adequately” (p. 310). More gen erally, if we identify the function of something x with those effects of x which contribute to the performance of some activity a or to the maintenance of some condition c jf a containing system s, then we must be prepared to say as well that a function of s is to perform a or to maintain c. This suggests the following formulation of “selected-effects" theories. (6) The function of an F in a G is f just in case (the capac ity for) f is an effect of an F incorporated in a G in the usual way (or: in the way this F is incorporated in this G), and that effect contributes to the performance of a function of the containing G.
Functional Analysis
69
It seems that any theory based on (B)—what I have been calling “selected-effects” theories—must ultimately amount to something like (6).'" Yet (6) cannot be the whole story about functional ascrip tions. Suppose we follow (6) in rendering, “The function of the con tractile vacuole in protozoans is elimination of excess water from the organism.” The result is (7):
(7) Elimination of excess water from the organism is an effect of a contractile vacuole incorporated in the usual way in a protozoan, and that effect contributes to the performance of a function of a protozoan.
In order to test (7) we should have to know a statement of the form, 7" is a function of a protozoan.” Perhaps protozoans have no func tions. If not, (7) is just a mistake. If they do, then presumably we shall have to appeal to (6) for an analysis of the statement attribut ing such a function, and this will leave us with another unanalyzed functional ascription. Either we are launched on a regress, or the analysis breaks down at some level for lack of functions, or perhaps for lack of a plausible candidate for containing systems. If we do not wish simply to acquiesce in the autonomy of functional ascriptions, it must be possible to analyze at least some functional ascriptions without appealing to functions of containing systems. If (6) can be shown to be the only plausible formulation of theories based on (B), then no such theory can be the whole story. Our question, then, is whether a thing’s function can plausibly be identified with those of its effects contributing to production of some activity of, or maintenance of some condition of, a containing system, where performance of the activity in question is not a func tion of the containing system. Let us begin by considering Hempel’s suggestion that functions are to be identified with production of effects contributing to proper working order of a containing system. I claimed earlier on that to say something is in proper working order is just to say that it properly performs its functions. This is fairly obvious in cases of artifacts or tools. To make a decision about which sort of behavior counts as working amounts to deciding about the thing’s function. To say something is working, though not behaving or disposed to behave in a way having anything to do with its func tion, is to be open, at the very least, to the charge of arbitrariness. When we are dealing with a living organism, or a society of liv ing organisms, the situation is less clear. If we say, “The function of
70
Robert Cummins
the contractile vacuole in protozoans is elimination of excess water from the organism," we do make reference to a containing organ ism. but not, apparently, to its function (if any). However, since contractile vacuoles do a number of things having nothing to do with their function, there must be some implicit principle of selec tion at work. Hempel’s suggestion is that, in this context, to be in “proper working order” is simply to be alive and healthy. This works reasonably well for certain standard examples, e.g„ (1) and (3): circulation does contribute to health and survival in verte brates, and photosynthesis does contribute to health and survival in green plants.11 But, once again, the principle will not stand a change of example, even within the life sciences. First, there are cases in which proper functioning is actually inimical to health and life: functioning of the sex organs results in the death of individu als of many species (e.g., certain salmon). Second, a certain process in an organism may have effects which contribute to health and survival but which are not to be confused with the function of that process: secretion of adrenalin speeds metabolism and thereby con tributes to elimination of harmful fat deposits in overweight humans, but this is not a function of adrenalin secretion in over weight humans. A more plausible suggestion along these lines in the special context of evolutionary biology is this:
(8) The functions of a part or process in an organism are to be identified with those of its effects contributing to activities or conditions of the organism which sustain or increase the organism’s capacity to contribute to sur vival of the species. Give or take a nicety, (8) doubtless does capture a great many uses of functional language in biology. For instance, it correctly picks out elimination of excess water as the function of the contractile vac uole in fresh-water protozoans only, and correctly identifies the function of sex organs in species in which the exercise of these organs results in the death of the individual.12 In spite of these virtues, however, (8) is seriously misleading and extremely limited in applicability even within biology. Evi dently, what contributes to an organism’s capacity to maintain its species in one sort of environment may undermine that capacity in another. When this happens, we might say that the organ (or what ever) has lost its function. This is probably what we would say
Functional Analysis
71
about the contractile vacuole if fresh-water protozoans were suc cessfully introduced into salt water, for in this case the capacity explained would no longer be exercised. But if the capacity explained by appeal to the function of a certain structure continued to be exercised in the new environment, though now to the individ ual’s detriment, we would not say that that structure had lost its function. If, for some reason, flying ceased to contribute to the capacity of pigeons to maintain their species, or even undermined that capacity to some extent," we would still say that a function of the wings in pigeons is to enable them to fly. Only if the wings ceased to function as wings, as in the penguins or ostriches, would we cease to functionally analyze skeletal structure and the like with an eye to explaining flight. Flight is a capacity that cries out for explanation in terms of anatomical functions regardless of its contribution to the capacity to maintain the species. What this example shows is that functional analysis can prop erly be carried on in biology quite independently of evolutionary considerations: a complex capacity of an organism tor one of its parts or systems) may be explained by appeal to a functional anal ysis regardless of how it relates to the organism’s capacity to main tain the species. At best, then, (8) picks out those effects which will be called functions when what is in the offing is an application of evolutionary theory. As we shall see in the next section, (8) is mis leading as well in that it is not which effects are explained but the style of explanation that makes it appropriate to speak of functions. (8) simply identifies effects which, as it happens, are typically explained in that style. We have not quite exhausted the lessons to be learned from (8). The plausibility of (8) rests on the plausibility of the claim that, for certain purposes, we may assume that a function of an organism is to contribute to the survival of its species. What (8) does, in effect, is identify a function of an important class of (uncontained) con taining systems without providing an analysis of the claim that a function of an organism is to contribute to the survival of its species. Of course, an advocate of (8) might insist that it is no part of his theory to claim that maintenance of the species is a function of an organism. But then the defense of (8) would have to be simply that it describes actual usage, i.e., that it is in fact effects con tributing to an organism’s capacity to maintain its species which evolutionary biologists single out as functions. Construed in this way, (8) would, at most, tell us which effects are picked out as func-
72
Robert Cummins
tions; it would provide no hint as to why these effects are picked out as functions. We know why evolutionary biologists are interested in effects contributing to an organism's capacity to maintain its species, but why call them functions? This is precisely the sort ot question that a philosophical account of function-ascribing state ments should answer. Either (8) is defended as an instance of (6) maintenance of the species is declared a function of organisms oi it is defended as descriptive of usage. In neither case is any philo sophical analysis provided. For in the first case (8) relies on an unanalyzed (and undefended) function-ascribing statement, and in the second it fails to give any hint as to the point of identifying cer tain effects as functions. The failings of (8) are, I think, bound to cripple any theory that identifies a thing's functions with effects contributing to some antecedently specified type of condition or behavior of a containing system. If the theory is an instance of (6', it launches a regress or terminates in an unanalyzed functional ascription; if it is not an instance of (6), then it is bound to leave open the very question at issue, viz., why are the selected effects seen as functions? III
In this section, I will sketch briefly an account of functional explanation which takes seriously the intuition that it is a gen uinely distinctive style of explanation. The assumptions (A) and (B) form the core of approaches that seek to minimize the differences between functional explanations and explanations not formulated in functional terms. Such approaches have not given much atten tion to the characterization of the special explanatory strategy sci ence employs in using functional language, for the problem as it was conceived in such approaches was to show that functional explanation is not really different in essentials from other kinds of scientific explanation. Once the problem is conceived in this way, one is almost certain to miss the distinctive features of functional explanation, and hence to miss the point of functional description. The account of this section reverses this tendency by placing pri mary emphasis on the kind of problem that is solved by appeal to functions. 1. Functions and Dispositions. Something may be capable of pumping even though it does not function as a pump (ever) and
Functional Analysis
73
even though pumping is not its function. On the other hand, if something functions as a pump in a system s or if the function of something in a system s is to pump, then it must be capable of pumping in s." Thus, function-ascribing statements imply disposi tion statements; to attribute a function to something is, in part, to attribute a disposition to it. If the function of x in s is to , then x has a disposition to in s. For instance, if the function of the con tractile vacuole in fresh-water protozoans is to eliminate excess water from the organism, then there must be circumstances under which the contractile vacuole would actually manifest a disposition to eliminate excess water from the protozoan that incorporates it. To attribute a disposition d to an object a is to assert that the behavior of a is subject to (exhibits or v/ould exhibit) a certain law like regularity: to say a has d is to say that a v/ould manifest d (shatter, dissolve) were any of a certain range of events to occur (a is put in water, a is struck sharply). The regularity associated with a disposition-—call it the dispositional regularity—is a regularity that is special to the behavior of a certain kind of object and obtains in virtue of some special fact(s) about that kind of object. Not every thing is water-soluble: such things behave in a special way in virtue of certain (structural) features special to water-soluble things. Thus it is that dispositions require explanation: if x has d, then x is sub ject to a regularity in behavior special to things having d, and such a fact needs to be explained. To explain a dispositional regularity is to explain how mani festations of the disposition are brought about given the requisite precipitating conditions. In what follows, I will describe two distinct strategies for accomplishing this. It is my contention that the appropriateness of function-ascribing statements corresponds to the appropriateness of the second of these two strategies. This, I think, explains the intuition that functional explanation is a special kind of explanation. 2. Two Explanatory Strategies (i) The Subsumption Strategy. Suppose a has a disposition d. The associated dispositional regularity consists in the fact that cer tain kinds of events would cause a to manifest d. One way to explain this fact would be to discover some feature of a which allowed us to represent the connection between precipitating events and manifestations as instances of one or more general laws, i.e., laws governing the behavior of things generally, not just things having d. Brian O’Shaughnessy has provided an example which
74
Robert Cummins
allows a particularly simple illustration of this strategy'.'5 Consider the disposition he calls elevancy: the tendency of an object to rise in water of its own accord. To explain elevancy, we must explain why freeing a submerged elevant object causes it to rise.16 This we may do as follows. In every case, the ratio of an elevant object s mass to its nonpermeable volume is less than the density (mass pei unit volume) of water. Archimedes' principle tells us that water exerts an upward force on a submerged object equal to the weight ol the water displaced. In the case of an elevant object, this force evi dently exceeds the weight of the object by some amount, f. Freeing the object changes the net force on it from zero to a net force of mag nitude/'in the direction of the surface, and the object rises accord ingly. Here, we subsume the connection between freeings and ris ings under a general law connecting changes in net force with changes in motion by citing a feature of elevant objects which allows us (via Archimedes’ principle) to represent freeing them under water as an instance of introducing a net force in the direc tion of the surface. (ii) The Analytical Strategy. Rather than subsume a disposi tional regularity under a law not special to the disposed objects, the analytical strategy proceeds by analyzing a disposition d of a into a number of other dispositions . d- had by a or components of a such that programmed manifestation of the di results in or amounts to a manifestation of d.'-: The two strategies will fit together into a unified account if the analyzing dispositions (the di) can be made to yield to the subsumption strategy. When the analytical strategy is in the offing, one is apt to speak of capacities (or abilities) rather than of dispositions. This shift in terminology’ will put a more familiar face on the analytical strategy',18 for we often explain capacities by analyzing them. Assembly-line production provides a transparent example of what I mean. Production is broken down into a number of distinct tasks. Each point on the line is responsible for a certain task, and it is the function of the workers/machines at that point to complete that task. If the line has the capacity to produce the product, it has it in virtue of the fact that the workers/machines have the capacities to perform their designated tasks, and in virtue of the fact that when these tasks are performed in a certain organized way—according to a certain program—the finished product results. Here we can explain the line’s capacity to produce the product—i.e., explain how it is able to produce the product—by appeal to certain capacities of
functional Analysis
75
the workers/machines and their organization into an assembly line. Against this background, we may pick out a certain capacity of an individual exercise of which is bis function on the line. Of the many things he does and can do, his function on the line is doing what ever it is that we appeal to in explaining the capacity of the line as a whole. If the line produces several products, i.e., if it has several capacities, then, although a certain capacity c of a worker is irrele vant to one capacity of the line, exercise ofc by that worker may be his function with respect to another capacity of the line as a whole. Schematic diagrams in electronics provide another obvious illustration. Since each symbol represents any physical object what ever having a certain capacity, a schematic diagram of a complex device constitutes an analysis of the electronic capacities of the device as a whole into the capacities of its components. Such an analysis allows us to explain how the device as a whole exercises the analyzed capacity, for it allows us to see exercises of the ana lyzed capacity as programmed exercise of the analyzing capacities. In this case, the “program” is given by the lines indicating how the components are hooked up. (Of course, the lines are themselves function symbols.) Functional analysis in biology is essentially similar. The bio logically significant capacities of an entire organism are explained by analyzing the organism into a number of “systems”—the circu latory system, the digestive system, the nervous system, etc.—each of which has its characteristic capacities.^ These capacities are in turn analyzed into capacities of component organs and structures. Ideally, this strategy is pressed until pure physiology takes over, i.e., until the analyzing capacities are amenable to the subsumption strategy. We can easily imagine biologists expressing their analy ses in a form analogous to the schematic diagrams of electrical engi neering, with special symbols for pumps, pipes, filters, and so on. Indeed, analyses of even simple cognitive capacities are typically expressed in flow charts or programs, forms designed specifically to represent analyses of information processing capacities generally. Perhaps the most extensive use of the analjtical strategy in science occurs in psychology, for a large part of the psychologist’s job is to explain how the complex behavioral capacities of organisms are acquired and how they are exercised. Both goals are greatly facilitated by analysis of the capacities in question, for then acqui sition of the analyzed capacity resolves itself into acquisition of the analyzing capacities and the requisite organization, and the prob lem of performance resolves itself into the problem of how the ana-
76
Robert Cummins
lyzing capacities are exercised. This sort of strategy has dominated psychology ever since Watson attempted to explain such complex run a maze by capacities as the t.„ ability ------ .. to —----------„ analyzing „ „ the .performance into a series of conditioned responses, the stimulus for each response being the previous response or something encountered as the result of the previous response?” Acquisition of the complex capacity is resolved into a number of distinct cases of simple condi tioning. i.e., the ability to learn the maze is resolved into the capac ity for stimulus substitution, and the capacity to run the maze is resolved into abilities to respond in certain simple ways to certain simple stimuli. Watson's analysis proved to be of limited value, but the analytic strategy' remains the dominant mode of explanation in behavioral psychology.-'1 3. Functions and Functional Analysis. In the context of an application of the analytical strategy, exercise of an analyzing capacity emerges as a function: it will be appropriate to say that x functions as a 0 in s, or that the function of x in s is d>-ing, when we are speaking against the background of an analytical explanation of some capacity of s which appeals to the fact that x has a capacity to in s. It is appropriate to say that the heart functions as a pump against the background of an analysis of the circulatory system’s capacity to transport food, oxygen, wastes, and so on, which appeals to the fact that the heart is capable of pumping. Since this is the usual background, it goes without saying, and this accounts for the fact that “The heart functions as a pump’’ sounds right, and “The heart functions as a noise-maker” sounds wrong, in some contextfree sense. This effect is strengthened by the absence of any actual application of the analytical strategy which makes use of the fact that the heart makes noise.We can capture this implicit dependence on an analytical con text by entering an explicit relativization in our regimented recon struction of function-ascribing statements:
(9) x functions as a in s (or: the function of x in s is to t. In the case of biological functions and in other cases of func tions where human intentions are not obviously causally active, this kind of description of function has been difficult to assimilate into our scientific view of the world. There are several reasons, but we shall here concentrate on one which arises directly from the fact that, in describing a present structure in terms of its function, we mention a future outcome of some sort. The future outcome may be, in many cases, nonexistent. A structure may never be called upon to perform that function. The function of a bee’s sting, for instance, is relatively clear; yet most bees never use their stings. Likewise teeth may never pulp food, just as nutcrackers may never crack nuts. Thus, when we describe the function of something in the pre sent, we make reference to a future event or effect which, in some cases, will never occur. Hence, prima facie, we cannot really be describing any genuine, current property of the character. * Reprinted from John Bigelow and Robert Pargetter, "Functions,” The •Journal of Philosophy 81 (1987). pp. 181-196, with kind permission from The •Journal of Philosophy and the authors. © 1987 The Journal of Philos ophy, Inc.
97
98
John Bigelow and Robert Pargettcr
I. The Problem
Even when a character does perform its supposed function, the future events that result from it cannot play any significant “scientific’ role in explaining the nature and existence of the character. The char acter has come into existence, and has the properties that it does hav e, as a result of prior causes. It would still have existed, with just the i ui rent properties it does have, even if it had not been followed by the events that constitute the exercise of its alleged function. Hence its existence and properties do not depend on the exercising of its func tion. So it is hard to see what explanatory' role its functions could have. Crudely put: backwards causation can be ruled out—structures always have prior causes—hence reference to future events is explana torily redundant. Hence functions are explanatorily redundant. Of course, there is nothing inappropriate about describing a character and mentioning its future effects. But describing a charac ter as having a function is not just mentioning that it has certain effects. Not every effect counts as part of its function. And some func tions are present even when there are no relevant effects to be men tioned—as with some bees and their stings. Future events are not unmentionable; but they are explanatorily redundant in characteriz ing the existence and current properties of a character. Hence, what role can functions have in a purely scientific description of the world: how can they be “placed" within the framework of current science? There are three main theories that attempt to construe func tions in a way that allows them to fit smoothly into the scientific, causal order. We believe that each is nearly right or partly right. Yet they are all unsatisfactory in one crucial respect: they do not restore to functions any significant explanatory power. In particu lar, they' deny to functions any causal efficacy. So, for instance, they will not permit us to explain the evolution of a character by saying that it evolved because it serves a specific function. We will offer an account of biological function and of functions generally which, although it shares much with the most promising extant theories, is crucially different from them in that it bestows greater explanatory’ power upon functions.1 But first we will briefly consider the three theories.
II. Eliminativism There are three responses that arise naturally in the face of the tension between functions and the scientific standpoint. The
Functions
99
first is eliminativist. II is assumed that functions, if there were any, would have to be important, currently existing, causally active, and explanatory properties of a character or structure. It is also assumed that functions, if there were any, would essentially involve reference to future, possibly nonexistent, events. Yet something involving essential reference to future, possibly nonexistent events could not possibly characterize currently existing, explanatory properties. It is thus concluded that there really are no functions in nature. To add functions to the scientific biological picture, on this view, is parallel to adding final causes to physics. Final causes have no place in the scientific account of the physical universe, and, if the psychological pressures are resisted, we find we can do without them and final causes just fade away. To the eliminativist, the same will be true of functions; as the biological sciences develop, any need for function talk will vanish, and the psychological natu ralness of such talk will fade away with time and practice. A variant on this eliminativist view adds an account of why the attributions of function seem to serve a useful purpose in everyday and scientific discourse. The eliminativist does not believe in functions as genuine, currently existing properties of a character. But the eliminativist does believe in future effects of a character. And nothing stops us from mentioning whichever future effects we take an interest in. Consequently, an eliminativist can interpret talk of “functions” as being merely the specification of effects one happens to be interested in. Which effects are deemed to relate to “functions” of a character, will depend not on the nature of the char acter itself, but on our interests. The function of kidneys is differ ent for the anatomist from what it is for the chef. Insofar as func tion talk makes sense, it does not describe the current nature of a character (there are no functions in nature); rather, it relates a cur rent character to a future outcome, in an interest-dependent, extrinsic manner.2 The best answer to an eliminativist theory is to come up with an adequate analysis of functions still within the scientific view. This is what we attempt to do later in this paper. But a motive for seeking such a noneliminativist account can be cited; and this motive will also provide a less than conclusive, but nevertheless weighty, argument against eliminativism. In the biological sciences, functions are attributed to characters or struc tures, and these attributions are intended to play an explanatory role which cannot be squared with the eliminativist’s account of
100
John Bigelow and Robert Pargetter
function talk. For instance it is assumed that biological structures would have had the functions they do have even if we had not been here to take an interest in them at all. And some of the effects of structures that we take an interest in have nothing to do with their function. And some functions are of no interest to us at all. Fur thermore. biology standardly treats function as a central, explana tory concept. None of this rests easily with an eliminativist theory. A powerful motive for resisting eliminativism, then, is that an ade quate analysis of functions, if we can find one, will enable us to take much biological science at face value: we will be relieved of the necessity of undertaking a radical reformation of the biological sci ences. The eliminativists’ vision, of functions “fading away,” is as yet just a pipe dream; and their explanation of the apparent use fulness of function talk fails to explain away more than a fragment of the uses functions serve in biological science. It is not the biological sciences alone which could be cited here. Psychology could be canvassed too. And even physics has facets that raise problems for an eliminativist view. Suppose some one were to suggest that the function of water is to refract light and the function of mists to create rainbows. Presumably this is plainly false. Yet it describes something in terms of future effects in which we take an interest . . . which is exactly what the eliminativist takes to be the business of function talk. So eliminativists have no good explanation of why the physicist takes such attributions of function to be plainly false. They cannot explain the manifest dif ference between “The function of mists is to make rainbows,” and "The function of teeth is to pulp food.”2
III. Representational Theories
There is a response to the tension between the scientific view and functions which rejects any role for future events in the char acterizing of functions. The future effects of a character do not themselves play an explanatory role in characterizing that charac ter. Yet sometimes there exists, prior to the character, a plan, a representation of that character and of its future effects. Such a representation of future effects may exist, whether or not those effects ever come to pass. And this representation exists prior to the character and so contributes to the causal processes that bring that character into being—by the usual, forward-looking, causal pro cesses that rest comfortably within our over-all scientific image of
Functions
101
the world. On this view, we can account for a function not by direct reference to any future event, but rather by reference to a past rep resentation of a future event. Theories of this sort have frequently been called “goal theories”; but they are best construed as a sub category within the class of goal theories. They are goal theories in which the identification of a goal depends on the content of prior representations.1 This kind of account of functions fits best with attribution of functions to artifacts. The idea of breaking open nuts seems to have played a causal role in the production of the nutcracker. It does not fit so neatly into the biological sciences. Of course, it used to provide a persuasive argument (the teleological argument) for the existence of a Creator: there are functions in nature; functions require prior representations; yet the creatures themselves (even when creatures are involved) have no such foresight or were not around at the right time; hence the prior representations must have been lodged in some awfully impressive being . . . etc. Nowadays, however, in the clear and noncontroversial cases of functions in nature, it is taken that they can be accounted for from the standpoint of a theory of evolution by way of natural selection—and in such a theory there can be no room for any analysis of biological functions which rests on prior representations. Even if God foresaw the functions of bio logical structures, that is a matter outside biology; functions, how ever, are a biological and not a theological matter. It is worth noting that, even though the representational the ory seems to rest comfortably with attributions of functions to arti facts, nevertheless, some artifacts prove more problematic than might first appear. Many artifacts evolve by a process very like nat ural selection. Variations often occur by chance and result in improved performance. The artisan may not understand fully the reasons why one tool performs better than others. Yet, because it performs well, it may be copied, as exactly as possible. The repro duction of such tools may occur for generations. The features of the tool which make it successful and which lead it to be selected for reproduction are features that have specific functions. But they were not created with those functions in mind. They may have been produced with an over-all function in mind (say, hitting nails); but the toolmaker may not have in mind any functions for the compo nents and features of the tool, which contribute to the over-all func tion. For instance, the toolmakers may copy a shape that has the function of giving balance to the tool—but they need not foresee, or plan, or represent any such function. They know only that tools like
102
John Bigelow and Robert Pargetter
this work well at banging nails, or sawing wood, or whatever the over-all function might be. Consequently, even with artifacts, structures can serve specific functions even though there exists no prior representation of that function." There is a further reason for uneasiness about the represen tational theory. The theory analyzes the apparent forward direct edness of functions by an indirect, two-step, route. The forward directedness of functions is analyzed as comprising a backward step to a representation, which in turn has a forward directedness toward a possibly nonexistent future state. Thus the seeming forward directedness of functions is reduced to another sort of forward directedness: that of representations— plans, beliefs, intentions, and so on. And this is worrying. The worry is not just that these are “mentalistic." and just as problem atic as functions—and just as hard to assimilate into the scientific picture of the world. Rather, an even greater worry for many will be that of vicious circularity. Many find it plausible that the notion of representation will turn out to be analyzable in terms that at least include functional terms. And functional terms presuppose functions. Hence the future directedness of representations may turn out to presuppose the future directedness of functions. This threatens to do more than just restrict the scope of representational theories; it undermines such theories even in their home territory, as applied to artifacts. IV. Etiological Theories
The third response to the tension between the scientific view of the world and the concept of function again involves rejecting any role for future events in the characterization of functions. Yet etiological theories also eschew reference to prior representations of future effects, as well as reference to the future effects them selves.6 Representational theories and etiological theories have an important feature in common. Both shun any genuine, direct refer ence to future effects, and refer instead only to past causes. Both construe the attribution of function as supplying information about the genesis of the character, that is, about how the character came into existence. The difference between the two theories recalls the distinction Charles Darwin drew between artificial selection and natural selec-
functions
103
lion. When animal breeders select, they represent to themselves the characters they wish to develop. Natural selection has closely analogous results, hut it operates in the absence of (or at least with out any need for) any conscious or unconscious representations of future effects. 'flu: etiological theory of functions explains biological func tions by reference to the process of natural selection. Roughly: a character has a certain function when it has evolved, by natural selection, because, il. has had the effects that constitute the exercise of that function. Clearly, there is room here for an overarching, disjunctive the ory, which unites the representational with the etiological. Such an overlapping theory would say: a character has a certain function when it has been selected because that character has had the rele vant effects. In the case of artifacts, the selection involves conscious representations (mostly); in the case of (Darwinian) sexual selec tion, representations may enter the picture, but reproduction, heredity, and evolution also play a part; and in the case of natural selection, representations drop out altogether. But, on the etiological theory, a character has a biological function only if that character has been selected for by the process of natural selection because it has had the effects that constitute the exercise of that function. This is the only kind of selection com patible with the dictates of modem biological science. We take this etiological theory of biological function as the main alternative to the account of biological function we shall prof fer. The big plus for the etiological theory is that it makes biologi cal functions genuinely’ explanatory, and explanatory in a way most comfortable with the modern biological sciences. But we shall argue that this explanatory power is still not quite right: it offers expla nations that are too backward-looking. The theory we offer will be more forward-looking in its explanatory nature. But, before we turn to this matter, we should note another worry with the etiological theory, a worry that extends even to the overarching disjunctive theory of which it is a part. This worry is that there is too great a dependence of the intrinsic nature of func tions on contingent matters—matters which, had they been (or even if they are) otherwise, would rob the theory’ of any viability as an account of functions. The etiological theory of biological functions has such functions characterized in terms of evolution by natural selection. Most theo rists take evolutionary theory to be true, but contingently so. What
104
John Bigclou' and Robert Pargetter
if the theory of evolution by natural selection were to be (or had been' false? Clearly then, on the etiological theory of biological func tions, as we have specified it, there would be no biological funct ions. Whether or not there are biological functions at all, on the overar ching, disjunctive theory, will depend on what replaces the theory of evolution by natural selection. Suppose it is creationism. Then the representational theory would apply; for we have the representa tions in the mind of the creator that would have the appropriate causal role in the development of biological structures, so the repre sentational theory would become a general theory of functions. We noted earlier that the representational theory had prob lems with the functions of some artifacts: artifacts that seemed to evolve over time by a process similar to natural selection. If this analogous process is also not available, along with natural selection proper, then our representational theory’ will not bestow functions upon such artifacts. But perhaps this is small change. For creationism, there would of course be an enormous epis temological problem of discovering what the functions of biological structures were; for this would depend on discovering what the cre ator had in mind. So we would be stuck with great difficulty in dis covering whether the function of the heart is to produce the sound of a heart beat, in line with the creator’s idea of a beating rhythm in nature, with the circulation of the blood as a nonfunctional effect; or whether the reverse is true. It would be much like an anthropologist discovering the nature of ancient artifacts without any presuppositions about the intentions of the earlier cultures. But suppose creationism is not the alternative. Consider the possible world identical to this one in all matters of laws and par ticular matters of fact, except that it came into existence by chance (or without cause) five minutes ago. Now, on even the over-arching theory, there are no functions; for there are no biological functions on the etiological theory, and no causally active representations as required by the representational theory, and hence no functions at all. We have the intuition that the concept of biological function, and views about what functions biological characters have, are not thus contingent upon the acceptance of the theory of evolution by natural selection and on discovering what led to the evolutionary development of particular characters. In parallel, we are also inclined to think that the representational account of the function of artifacts gives too much importance to the representations or ideas of the original planner, even in cases when there is one. As we
Functions
105
indicated earlier, we believe a satisfactory account of functions in general, and of biological functions in particular, must be more for ward-looking. We now turn to this.
V. Fitness, Function, unci Looking Forward It emerges from our discussions that the tension between functions and modern, causal science has generated, fundamen tally, two stances on the nature of functions. The first is the eliminativist stance. This has the merit of giv ing full weight to the forward-looking character of functions, by specifying them in terms of future and perhaps nonexistent effects; and also to the explanatory importance of functions. It is mistaken only in its despair of reconciling these two strands. The second stance is backward-looking. This embraces theo ries which look back to prior representations and those which look back to a prior history of natural selection and those which look back to a history of either one sort or the other. We will argue for a forward-looking theory. Functions can be characterized by reference to possibly nonexistent future events. Fur thermore, they should be characterized that way, because only then will they play the explanatory’ role they need to play, for instance, in biology. The way to construe functions in a forward-looking manner, we suggest, is (roughly) to construe them in the manner of disposi tions. The shift we recommend, in our conception of functions, has a precedent: the analysis of the evolutionary concept of fitness. One wrongheaded, but at times common, objection to the Dar winian theory of evolution is that its central principle—roughly, “the survival of the fittest"—is an empty tautology which cannot possibly bear the explanatory weight Darwin demands of it.7 This objection assumes that fitness can be judged only retrospectively: that it is only after we have seen which creatures survived that we can judge which were the fittest; moreover, it assumes that the fact that certain creatures have survived, whereas others did not, is what constitutes their being the fittest. The etiological theory of biological functions rests on the same sort of misconception as that which underlies the vacuity objection to Darwin. On this theory, we can judge only’ retrospectively that a character has a certain function, when its having had the relevant effect has contributed to survival. Indeed, on the etiological theory, that an effect is part of the function of a character is constituted by
106
John Bigelow and Robert Pargetter
the fact that having this effect has contributed to the survival of t he character and of the organisms that bear it. Consequently, the notion of function is emptied of much explanatory potential. It is no longer possible to explain why a char acter has persisted by saying that the character has persisted because it senes a given function. To attempt to use function in that explana tory role, would be really to fall into the sort of circularity often alleged (falsely) against the explanatory use of fitness in Darwinism. This comparison with fitness serves another purpose. It has dis played why functions would lack explanatory power on the etiologi cal theory, but it also shows how to analyze functions so as not to lose this explanatory power. Fitness is not defined retrospectively, in terms of actual survival. It is, roughly, a dispositional property of an individual (or species) in an environment, which bestows on that individual (or species) a certain survival potential or reproductive advantage. This is a subjunctive property: it specifies what will hap pen or what is likely to happen in the right circumstances, just as fragility is specified in terms of breaking or being likely to break in the right circumstances. And such a subjunctive property supervenes on the morphological characters of the individual (or species).6 Hence there is no circularity involved in casting fitness in an explanatory role in the Darwinian theory of evolution. In the right circumstances fitness explains actual survival or reproductive advantage, just as in the right circumstances fragility explains actual breaking. In each case the explanation works by indicating that the individual has cer tain causally active properties that in such circumstances will bring about the phenomena to be explained. What holds here of fitness holds, too, of biological functions. The etiological theory is mistaken in defining functions purely ret rospectively, in terms of actual survival. Hence there need be no circularity in appealing to the functions a character serves, in explaining the survival of the character. Fitness is forward-looking. Functions should be forward-looking in the same way and, hence, are explanatory in the same way.
VI. The Propensity Theory Here is one way to derive a “forward-looking” theory of func tions. Let us begin with the etiological theory. Consider a case in which some character has a specific effect and has been developed
Functions
107
and sustained by natural selection because it had that effect. In such a case the etiological theory deems that it is (now) a function of the character to produce that effect. Look more closely, then, at tin.' past process that has “con ferred” a function, according to the etiological theory. The character in question must have bad the relevant effect, on a sufficient num ber of occasions—and in most cases, this will have been not on ran domly chosen occasions, but on appropriate occasions, in a sense needing further clarification. (For instance, sweating will have had the effect of cooling the animal—and it will have had this effect on occasions when the animal was hot, not when it was cold.) The history that confers a function, according to the etiologi cal theory, will thus display a certain pattern. The effect that will eventually be deemed a function must have been occurring in appropriate contexts; that is to say, it must have been occurring in contexts in which it contributes to survival, at least in a statisti cally significant proportion of cases. Further, this contribution to survival will not, in realistic cases, have been due to sheer accident. One can imagine individual incidents in which a character contributes to survival by sheer chance. Laws of probability dictate that a long run of such sheer accidents is conceivable; but it will be very unlikely—except for characters with relatively short histories. The only cases in which such long runs are likely to occur are cases in which the character confers a standing propensity upon the creature, a propensity that increases its chances of survival. If we imagine (or find in the vast biological record) a case in which a character is sustained by a chance sequence of accidents, rather than by a standing propensity, then it would not be appro priate to describe that character as having a function. This can hap pen when a character is linked with another character that does bestow a propensity and where variations in the character just have not occurred to allow selection against the inoperative charac ter. It can also happen by sheer chance, a long-run sequence of sheer flukes. Such a sequence is very improbable; but biologj' offers a stunningly large sample. It is very probable that many improba ble events will have occurred in a sample that large. Consequently, what confers the status of a function is not the sheer fact of survival-due-to-a-character, but rather, survival due to the propensities the character bestows upon the creature. The etiological theory describes a character now as serving a function, when it did confer propensities that improved the chances
108
John Bigeloir and Robert Pargetter
of survival. We suggest that it is appropriate, in such a ease, to say that the character has been serving that function all along. Even before it had contributed (in an appropriate way) to survival, it had conferred a survival-enhancing propensity on the creature. And to confer such a propensity, we suggest, is what constitutes a function. Something has a (biological) function just when it confers a survival-enhancing propensity on a creature that possesses it. Four features of this propensity theory of biological functions should be made explicit. First, like the corresponding account of fitness, this account of functions must be relativized to an environment. A creature may have a high degree of fitness in a specific climate—but a low degree of fitness in another climate. Likewise, a character may confer propensities which are survival-enhancing in the creature’s usual habitat, but which would be lethal elsewhere. When we speak of the function of a character, therefore, we mean that the character generates propensities that are survival-enhancing in the crea ture’s natural habitat. There may be room for disagreement about what counts as a creature's “natural habitat”; but this sort of vari able parameter is a common feature of many useful scientific con cepts. Ambiguities will arise especiallj' when there is a sudden change in the environment. At first, we will refer the creature’s “natural habitat” back to the previous environment. But eventually we will transfer the term to the current environment. The thresh old at which we make such a transference will be vague. The notion of natural habitat will also be ambivalent as applied to domestic animals. In its most obvious use, the term ‘habitat’ applies to the phys ical surroundings of a whole organism. But we can also extend its usage, and apply the term “habitat’ to the surroundings of an organ within an organism. Or to the surroundings of a cell within an organ. In each case, the natural habitat of the item in question will be a functioning, healthy, interconnected system of organs or parts of the type usual for the species in question. When some of the organs malfunction, then other organs, which go on performing their natural functions, may no longer be contributing to survival. We still say they are performing their natural function, even though this does not enhance the chance of survival. Why? Because it would enhance survival if the other organs were performing as they do in healthy individuals. Consequently, functions can be ascribed to components of an
Functions
109
organism, in a descending hierarchy of complexity. We can select a •subsystem of the organism, and we can ascribe a function to it when it enhances the chances of survival in the creature’s natural habi tat. Within this subsystem, there may be a subsubsystem. And this may be said to serve a function if it contributes to the functioning of the system that contains it— provided all other systems are func tioning “normally” (that is, provided it is lodged in its own “natural habitat”). And so on. Similar hierarchies may also occur in the opposite direction: a microscopic organism has a function in a pond, which has a func tion in a forest, which has a function in the biosphere, and so on. Secondly, on the propensity theory, functions are truly disposi tional in nature. They are specified subjunctively: they would give a survival-enhancing propensity to a creature in an appropriate man ner, in the creature’s natural habitat. This is true even if the crea ture does not survive or is never in its natural habitat. Likewise, fragility gives a propensity to break to an object in an appropriate manner, in the right circumstances—and of course some fragileobjects never break. And fitness gives a propensity to an individual or species to survive, in an appropriate manner, in a specified envi ronment and in a struggle for existence, even if there is no struggle for existence or if the individual or species fails to survive. Of course, when functions do lead to survival—just as when dispositions are manifested—the cause will be the morphological structural form of the creature and the relationship between this form and the environment. Functions supervene on this in the same way that dispositions supervene on their categorical bases. But the functions will be explanatory of survival, just as dispositions are explanatory of their manifestations; for they will explain survival by pointing to the existence of a character or structure in virtue of which the creature has a propensity to survive. Thirdly, in the long run. it will be necessary to spell out the notion of a “survival-enhancing propensity" in formal terms, employing the rigors of the probability calculus. Clearly, there will be a spectrum of theories of this general form. These theories will vary in the way they explicate the notion of “enhancement : whether they construe this as involving increasing the probability of survival above a certain threshold, or simply increasing it signif icantly above what it would have been, and so on. We are not attempting to find and defend the correct propensity theory, but only arguing that a propensity theory offers the most promising theory of functions.
110
John Bigelow aiul Robert Pargetter
Fourthly, there is the question as to whether the scope of the propensity theory is limited to biological functions or whether it can be extended, in some sense, to artifacts. Obviously, like the etiological theory, the propensity theory could be part of an overarching, disjunctive theory which analyses biological functions in terms of bestowing propensities for survival and the function of artifacts in terms of prior represent at ions. Wo noted earlier some problems for a backward-looking theory, even for artifacts. Yet surely representations should have some causal role in the case of consciously produced artifacts. We are attracted here to a general, overarching theory, but one that concentrates on the propensity for selection. So a character or structure has a certain function when it has a propensity for selec tion in virtue of that character or structure"s having the relevant effects. In the case of biological functions, we have a propensity for survival in the natural habitat, and so in such a habitat natural selection is likely to be operative. In the case of artifacts, we have a selection process clearly involving representations. But the repre sentations are those at the time of selection, at the time of bestow ing the function; now, so to speak. They need not be blueprints that antedate the first appearance of the prototype. Thus we feel there is a sense in which all functions have a commonness of kind, whether they be of biological characters or of artifacts.
VII. Comparisons We hope to have led our readers to appreciate the attractive ness of the propensity theory of biological functions (and of func tions generally). We conclude by making some direct comparisons between the etiological theory and the propensity theory. On most biological examples, the etiological theory and our propensity theory' will yield identical verdicts. There are just two crucial sorts of case on which they part company. One sort of case which distinguishes the theories is that of the first appearance of a character that bestows propensities conducive to survival. On our theory', the character already has a function, and by bad luck it might not survive, but with luck it may survive, and it may' survive because it has a function. On the etiological theory, in contrast, the character does not yet have a function. If it survives, it does not do so because it has a
Functions
Ill
function; but, after time, if it Jias contributed to survival, the char acter will have a function. We think our theory gives a more intuitively comfortable description of such cases, at least in most instances. But there are variants on this theme, on which our theory gives less comfortable results. Suppose- a structure exists already and serves no purpose at all. Suppose then that the environment changes, and, as a result, the structure confers a propensity that is conducive to survival. Our theory tells us that we should say that the structure now has a function. Over all, this seems right, but there are cases where it seems counterintuitive. Consider, for instance, the case of heart beats—that is, the sound emitted when the heart beats. In this cen tury, the heartbeat has been used widely to diagnose various ail ments; so it has come to be conducive to survival. The propensity theory deems the heartbeat to have the function of alerting doctors. That sounds wrong. The etiological theory says the heartbeat has no such function because it did not evolve for that reason. That sounds plausible. And yet, we suggest, the reason we are reluctant to grant a function to the heartbeat is not that it lacks an evolutionary past of the required kind. Other characters may lack an evolutionary- past, yet may happily invite attribution of a function. Rather, our reluc tance to credit the heartbeat with a function stems from the fact that the sound of the heartbeat is an automatic, unavoidable by product of the pumping action of the heart. And that pumping action serves other purposes. Although the heartbeat does (in some countries, recently) contribute to the survival of the individual, it does not contribute to survival of the character itself. The charac ter—heartbeat—will be present in everyone, whether or not doctors take any notice of it. Although it "contributes'' to survival, it is a redundant sort of contribution if it could not fail to be present whether it was making any contribution or not. Perhaps the propensity theory should be carefully formulated in such a way as to rule out such "automatic" contributions to survival. Nevertheless, we will note only that, although the example of heart beats seems initially to count against a propensity theory, there are other examples, and wider theoretical considerations, which count in its favor. Further, the case introduces many complications. For these reasons, it cannot be regarded as, in any sense, conclusive. So much for cases of new survival-enhancing characters. There is a dual for these cases: that of characters that were, but are no longer, survival enhancing. These cases, like the former cases,
112
John Bigelow and Robert Pargetter
serve to distinguish between the etiological and propensity theo ries. If a character is no longer survival-enhancing (in the natural habitat), the propensity theory deems it to have no function. The etiological theory, in contrast, deems its function to be whatever it was that it used to be, and was evolved for. In general, we think the propensity theory gives the better verdict in such cases. Under some formulations, our judgment may be swayed in favor of the etiological theory. We may be inclined to say that the function of a character is to do such and such, but unfortunately this is harmful to the creature these days. Yet surely the crucial fact is, really, that the function mas to do such and such. It serves no pressing purpose to insist that its function still is to do that. Especially not, once we have passed the threshold over which we redefine the creature's natural habitat. If a character is no longer survival-enhancing, because of a sudden and recent change in environment, we may continue to refer its natural habitat to the past. Consequently, our propensity theory will continue to tie func tions to what would be survival-enhancing in the past habitat. In such cases, there will be no conflict between the judgments of our theory and those of the etiological theory. The test of examples and counterexamples is important. Yet in this case, in the analysis of functions, there is a risk that it will decat’ into the dull thud of conflicting intuitions. Similarly with intuitions as to how unified should be the analyses of functions for biological characters and for artifacts. For this reason, we stress the importance of theoretical grounds for preferring the propensity theory. A propensity can play an explanatory causal role, whereas the fact that something has a certain historical origin does not, by itself, play much of an explana tory, causal role. Consequently, the propensity theory has a theo retical advantage, and this gives us a motive for seeking to explain away (or even overrule) apparent counterintuitions. In a similar way, Darwinian evolutionary theory provides strong theoretical motives for analyzing fitness in a certain way. Our intuitions—our unreflective impulses to make judgments— have a role to play, but not an overriding one.
Notes
1. Our theory will be a cousin of theories which are sometimes called “goal theories” and which have been advocated, for instance, by Christo-
Functions
113
pher Boorse, “Wright on Functions," P/uZosop/iIca/ Review, LXXXV, 1 (Jan uary 1976); 70-86, and "Health as a Theoretical Concept," Philosophy of Science, XLIV, . Let me explain how I believe an etiological theory is best understood. In my view, the central ele ment of the etiological approach should be seen as the simple idea that a function of a trait is the effect for which that trait was selected. This is the general, everyday notion of a proper function. It is, according to this idea, the function of the swatch on the wall to turn on the light, because the switch was put there for that pur pose. I will come back to artifacts in a moment, but first I will explain how this everyday notion of a function is precisified for the purposes of modem biology. It is a common mistake to suppose that a good theory of proper functions will be univocal with respect to both the functions of artifacts and biological, evolved functions. There are important differences of detail that vary with the kind of selection process involved. Where natural selection is the relevant selection process, there are two constraints on function ascriptions that do not apply in the case of intentional selection by an agent. (1) Selection is always of types, not tokens. So function attributions belong pri marily to types and only derivatively to tokens; your opposable thumb, for example, has a proper function in virtue of the fact that this type of trait has a proper function. Natural selection operates over whole populations and not on individual items—there is no way that an individual instance of a trait, such as Margaret Thatcher’s nose, can be selected. Just what the unit of selection is, is controversial, and there may be a number of equally suitable can didates. However, it is clear that the unit of selection is not a trait of an individual organism. Genotypes might properly be seen as the
130
Karen Neandcr
unit of selection, since genotypes can proportionally increase or decrease in the gene-pool. Phenotypes arc also indirectly selected, through their role as intermediary between genes and environ ment. But however the debate about units of selection is settled, function attributions in biology belong primarily to types—geno types and their phenotypic expressions, perhaps—because it is
types, not tokens, that are selected for their effects. Furthermore, (2) since natural selection lacks foresight and is a blind causal process operating over random mutations, it can only operate on actual past causal contributions to inclusive fitness. It is always past performances of the functional effect (the effect that becomes the function) by the parts and processes of ancestral organisms that play the causal role in the etiology of present items. Let’s see how this works in a particular case—your opposable thumb. In so far as your opposable thumb is the result of selection, it is. indirectly, the result of natural selection for thumbs of that type.9 According to the theory I favor, the function of your opposable thumb is to assist in grasping objects, because it is this effect which opposable thumbs contributed to the inclusive fitness of your ances tors, and which caused the underlying genotype, of which opposable thumbs are the phenotypic expression, to be selected. In brief, grasping objects was what the trait, the opposable thumb, was selected for, and that is why it is the function of your thumb to help you to grasp objects. This is the logical form of most biological functions, since most biological functions are evolved functions. Even those functions that have resulted from human cultivation or selective breeding practices are evolved functions. After all, being cute, useful, tasty, or good company for humans is just another way to be more fit. Only those physiological traits that result from genetic engineering will be exceptions to the general rule that biological functions are evolved functions. Where a trait results from intentional selection by an agent it will have an intended function, along the lines of the standard artifact functions that I will shortly describe. But first, notice the neat fit between this understanding of evolved biological functions and the general pattern of teleological explanations outlined earlier. Since evolved biological functions belong principally to types, not tokens, the forward-reference to a trait’s function, to what the trait is supposed to do, serves as an implicit reference to past selection of that type of trait for that typo of effect. We have here, in common with other teleological explana tions, an explanans that explicitly refers to something that post-
The Teleological Notion of ’Function’
131
dates the explanandum. In this case the explanans is a biological function attribution, so it explicitly refers to what the trait is sup posed to do. In common with other teleological explanations, we also have an explanans that implicitly refers to a selection process that pro-dates the explanandum, for the function of the trait is whatever it did that caused it to be selected by natural selection. So understood, biological function attributions generate teleological explanations that are unobjectionable, and so understood they con form to the general pattern of teleological explanations. So far my position may seem to coincide with that of Paul Grif fiths, who takes himself to be agreeing with Ruth Millikan,’' when he says that “. . . the central element of the etiological approach [is] the idea that the past effects of a type provide an explanation for the existence of current tokens of that type” 11993, p. 413). How ever, I disagree with this claim. It is true for evolved functions, because that is the v/ay that natural selection works, but it is not true for functions which are the result of intentional selection. I suggest that the function of an artifact is the purpose or end for which it was designed, made, or ^minimally) put in place or retained by an agent. Once again, its function is the effect for which it was selected, but the selection is usually intentional selection by an agent.11 Since there will often be several agents involved, and each might have a different purpose, we might want to distinguish between “design functions,” “user functions.” "occasion functions," and so on. The everyday notion of an artifact’s function is context sensitive, and in some contexts one intentional agent can take precedence over another. However, although the particular context might highlight the intentions of the user, rather than the designer, say, the function of an artifact is always whatever it was selected for. A frying pan intentionally seized to hit someone over the head literally becomes a weapon on that occasion, because the user selected it for that use on that occasion, but its standing function is frying food because that is the effect for which it was designed and bought. Because intentional design has a very different modus operandi from natural selection the constraints differ. Whereas evolved functions must be generalizable over types, artifact func tions may be idiosyncratic. Unique inventions, like the additions to James Bond’s briefcase, can have proper functions peculiar to them because they can be individually selected for particular effects. Also, because intentional agents do have foresight, there need be no past performances of the functional effect, nor any “ancestral arti-
132
Karen Neu nder
facts to do any performing. It is enough, in the case of intentional selection, if the designer believes or hopes that the artifact will have the desired effect and selects it for that purpose. Griffiths has suggested that, where intentional selection is involved, the agent minimally imagines an artifact and some hypo thetical alternatives performing the functional effect. He claims that this imagined relative success in performance plays a causal role in producing the item. But intentional selection need not always involve an explicitly imagined performance of the func tional effect by the artifact and hypothetical alternatives. It needn’t always involve comparative trial and error testing in the imagination. For one thing, inventing might sometimes be a purely cognitive and inferential affair—so that reason alone, rather than imagination, sometimes leads to the design outcome. For another, in some cases there may be only one apparent way to achieve a desired outcome, so there may be no comparison with alternatives. Griffiths’ suggestion has the merit of attempting a univocal analysis of the notion of “selection’’ as found in “natural selection" and the “selection of artifacts"12 but unfortunately it does not succeed. Natural selection is one type of selection process, which counts literally as such, through the death of a rewarding metaphor. It has both important differences and similarities to intentional selection by an agent. Some of the important differences have been noted in this section. One feature they share, is that the products of both lend themselves to teleological explanations. That is, they both give rise to what Wright called “consequence-etiologies,” which are causal explanations of a special kind, those in which an effect of the trait being explained “plays a role.” The role played, however, depends upon the kind of selection process. In the case of natural selection, effects of past instances of that type of trait causally con tribute to increased replication of the trait. While in the case of intentional selection, it may be that a mental representation of the effect is what plays the causal role. To summarize the main point of the last two sections: teleo logical explanations of the functional variety are like purposive explanations, in that they explicitly refer to a future effect of a trait for which that trait was selected. In doing so they explain the trait by implicitly referring to the causally efficacious selection process from which it resulted. We do not understand teleological explana tions correctly, as a species of ordinary causal explanation, unless we understand that they are not only explicitly “forward-looking”
The Teleologic.il Notion of'Function'
133
but also implicitly “backward-looking,” and it is only in virtue of this implicit looking back to prior causes that teleological explana tions are explanatory.
IV. 77/e Propensity Theory and Functional Explanations These claims run contrary to claims made by Bigelow and Par getter in their paper on functions (1987). They claim, in effect, that if biological function is a genuinely teleological notion then the the ory required to explicate the notion is some kind of propensity the ory. According to their propensity theory, a function is a disposition apt for selection. They suggest that a biological function confers upon the creature that possesses it a survival-enhancing propen sity, and that a biological function is a disposition to systematically contribute to the creature’s survival in its “natural habitat. Bigelow and Pargetter concede that their theory is “revisionary but argue that it is motivated by two major considerations. (1) They believe that the etiological theory is the best alternative to their theory, but that there are serious objections to the etiological the ory, and (2) they believe that their theory restores explanatory power to functions. The objections to the etiological theory have been dealt with elsewhere (as mentioned in the introduction to this paper). Here I will concentrate on (2)—the alleged success of the propensity theory in restoring explanatory power to functions. I will give their claim a little more flesh, and then assess it. Bigelow and Pargetter believe that the propensity theory cap tures the “forward-looking" nature of functional explanations because, according to the propensity theory, functions are disposi tions apt for selection in the future. Its main advantage, as they see it, is related to this. They believe the propensity theory restores sig nificant scientifically respectable explanatory power to functions by allowing us to explain the evolution of a trait by saying that a trait evolves because it serves a function. They do not explicitly talk of teleology, but the puzzle they address at the start of their paper is the teleological puzzle with which we are familiar. They describe it as follows:
Even when a character does perform its supposed function, the future events that result from it cannot play any signif icant “scientific” role in explaining the nature and existence of the character. The character has come into existence, and
134
Karen Ncandcr has the properties that it does have, as a result of prior causes. ... So it is hard to see what explanatory role func tions could have. Crudely put—backwards causation can be ruled out—structures always have prior causes—hence ref erence to future events is explanatorily redundant. Hence functions arc explanatorily redundant, (pp. 181-182)
They complain that although some alternative theories are “nearly right or partly right,” ... they do not restore to functions any significant explana tory power. In particular, they deny to functions any causal efficacy. So. for instance, they will not permit us to explain the evolution of a trait by saying that it evolved because it serves a specific function, tp. 182) We are now in a position to see that the claim that the propen sity theory restores significant explanatory power to functions is false. To begin with, the propensity theory does not capture the gen eral pattern of teleological explanations. It is true that the propen sity theory is forward-looking, but the trouble is that it only looks forward. Pes, we have seen, teleological explanations can explain present items because talk of their purposes or functions involves implicit reference to prior causes, specifically past selection pro cesses. However, understood according to the propensity theory, talk of functions involves no implicit reference to prior causes. According to it, talk of functions only has implicit reference to future Selection, but future selection can only explain future instances cf traits, not presen' ones. So the propensity theory clearly fails to permit functions explain all they seem to explain. When we say that koalas have pr uches to protect their young, we take ourselves to be explaining w hy presently existing koalas have pouches, not why future generations of koalas will have pouches. Still, it is independently of interest whether the propensity theory bestows on functions some other significant form of explana tory power. Bigelow and Pargetter claim that the propensity theory allows us to explain the evolution of a trait by saying that it evolves because it serves a function. But this claim fares no better when considered independently, apart from its relevance to teleologyThere is a distinction between a trait serving a function and having a function. The parson’s pocket bible that fortuitously stops the bul-
The Teleological Notion of‘Function’
135
let from killing the parson sert/es as a bullet-stopper, but does not have bullet-stopping as its function. A trait serves a function Zjust as long as the trait does Z, and if a trait cannot do Z, it does not serve the function of Z-ing, even if Z-ing is its function, tit is the function of all kidneys, diseased or otherwise, to filter the blood, but if they are very diseased they will be incapable of performing, or serving, this function.) So having a function and serving a function are different matters. It follows that any account of what it is to have a function will allow us to say that traits evolve because they serve a function, simply because a trait need not have a function in order to serve it. The claim made on behalf of the propensity theory, if taken literally, is thus a claim to which all extant theories of func tion are entitled. There is, perhaps, a more interesting and less trivial interpre tation of the claim made on behalf of the propensity theory. What may have been meant faithough it was not said) was that the propensity theory permits us to say that characters evolve because they perform their proper function. This is more substantial. Now the claim is that traits evolve because they have a function which they also serve. I want to make two points relating to this. First, it approximates the truth only if we accept the propen sity theory’s revision of our function talk. According to current usage, a new mutation has no function. At best it has accidentally beneficial effects. This accords with the etiological account, which says that new mutations have no function until they have been selected for their fitness-enhancing effects. Of course we cannot rule out the possibility that some revision of our function talk may be necessary, but I question the alleged necessity for revision in this case. We have no theoretically driven need to say that a newly mutated trait evolves because it has a function, it is sufficient to say that it evolves because it serves it. At best, and given a certain amount of revision, the propensity theory permits us to paraphrase evolutionary findings that can be expressed in other terms. How much is paraphrased is unclear. Sometimes Bigelow and Pargetter say that functions are survival enhancing propensities, and sometimes they say they are disposi tions apt for selection in the creature's "natural habitat. But a dis position to enhance survival and a disposition to be selected are not quite the same thing: reproductive functions (for example, giving birth or chest-beating by male gorillas) often endanger life and so are not survival-enhancing, but they are fitness-enhancing, and so arc apt to be selected. Aptness for selection must also include heri-
136
Karen Neandcr
tability, and competitiveness with current alternatives. More or less can be built into the propensity account of “function” depend ing on how we interpret it. Reading the propensity theory thinly, it will often be true that a trait evolves (in part.) because it has a “function" (i.e. because it systematically enhances survival, or less thinly, because it systematically enhances fitness in the creature’s “natural habitat"). The problem is, however, that we can easily express this in other terms, and we can be more exact in the pro cess—that is, simply by saying that a trait evolves (in part) because it systematically enhances fitness, or in other words, because it is adaptive in the actual environment. Reading the propensity theory richly, if a trait has a “function" (i.e. is apt for selection in the crea ture’s “natural habitat") then it will generally be true that the trait will evolve—it will evolve, accidents (etc.) aside, if the creature is in its “natural habitat." But once again. I can see no advantage of an explanatoiy nature in our being able to express the matter this way. The propensity theory permits us to say that a trait evolves because it has a “function (barring accidents and given that the creature is in its “natural habitat") which is just to say that a trait evolves because it is apt to evolve. I sometimes think that the appeal of the propensity theory is due to a piece of faulty reasoning that can be captured as follows: (P,) Biologists say the bee-sting evolved because it pro tected the hive. . t, applicable for causal explanation. 6. The differences between flies in these two situati interesting ways when flies from warm and cool areas arc gle temperature in the laboratory (Levins, 1963, PP bo
a gjn_
CHAPTER 10
A Modern History Theory of Functions* Peter Godfrey-Smith
I. Introduction ulU or effects a trait has Biological functions are dispositions of the trait under natural which explain the recent maintenance c. selection. This is the “modern history approach to '’"’’hPV
approach is historical because to ascribe a function iclaim about the past, but the relevant past is the recent pa_t, m ern history rather than ancient. . „,h,>h The modern history view is not new. It i= a pom up much of the functions literature has been com erging " , part of two decades, and there are implicit or parti ? , the the view to be found in many writers. This paper aim? of position entirely explicit, to show how it emerge? om hi_iotTj. other authors, and to claim that it is the right approac cj cal functions. . , „]] the Adopting a modern history position does no philosophical problems about functions. It dea ? wt. ot)jer jjf. questions concerning time and explanation, but t erear Acuities which are quite distinct. The most importan comcern the extent to which functional characterization re^ 1979). mitment to some form of adaptationism (Gould & "° writers These issues will not be addressed here. Further, as! m variety note, “function" is a highly ambiguous term. It is u.e' 1 erv(]ay of scientific ami philosophical theories, several omaim * Reprinted from A'mls 28 (1994). pp. 344-362. withi kind Pc'™1”‘l'n Blackwell Publishers and the author. © 1994 Basil Blackwell,
199
200
Peter Godfrey-Smith
discourse, and there is probably even a plurality of senses current within biology. This paper is concerned with one core biological sense of the term, which is associated with a particular kind of explanation. In this sense a function has some link to an explana tion of why the functionally characterized thing exists, in the form it does. Cummins (1975) argued that functions are properly associ ated with a different explanatory project, that of explaining how a component in a larger system contributes to the system exhibiting some more complex capacity. Following Millikan (1989a) I suggest that both kinds of functions should be recognized, each associated with a different explanatory project. If it is claimed, for instance, that the function of the myelin sheaths round some brain cells is to make possible efficient long distance conduction of signals, it may not be obvious which explanatory project is involved—that of explaining why the sheath is there, or that of explaining how the brain manages to perform certain tasks. Often the same functions will be assigned by both approaches, but that does not mean the questions are the same. The aim of this paper is to analyze an existing concept of func tion, which plays a certain theoretical role in biological science. So the aim is a certain sort of conceptual analj’-sis, a conceptual analysis guided more by the demands imposed by the role the concept of function plays in science, the real weight it bears, than by infor mal intuitions about the term's application. Also, though I will defend the modern history view within the context of a particular theory of functions which draws on the work of Larry Wright and Ruth Millikan, the overall value of the modem history approach stands independently of many of the details of my theory.' II. The Wright Line Our point of departure is a simple formula proposed by Larry Wright in 1973 and 1976: ‘‘The function ofX is that particular con sequence of its being where it is which explains why it is there” (1976, p. 78). That is:
The function of X is Z iff:
(i) Z is a consequence (result) of %’s being there, (ii) X is there because it does (results in) Z. (1976, p. 81)
A Modern History Theory of Functions
201
Wright argued that his theory dealt with a broad range of cases, handling both the functions of artifacts and biological enti ties without significant modification. The function of spider webs is catching prey, because that’s the thing they do that explains why they are there; the function of tyre tread is improving traction because that’s also the thing it does that explains why its there; and the function of the newspaper under the door is to prevent a draft, for the same reason. However, Wright’s analysis covers more cases than these. Boorse (1976) notes that when a scientist sees a leak in a gas hose, but is rendered unconscious before it can be fixed, on ng schema the break has the function of releasing gas. The break is there because it releases gas, keeping the scientist immobilized, and the leaking gas is a consequence of the break in the hose. Sim ilar cases take us even further from the plausible realm o purp One might see a small, smooth rock supporting a larger roc in a fast-flowing creek, and note that if it did not hold up t a rock, it would be washed away, and no longer be there. u i >not the function of the small rock to support the largvr one. ' problem here is with the broad range of "X and Z, Mt en• to restrict the kinds of things to which the schema can e app • ArestrictionofthiskindisakeycomponentofRuth.il i ana
ory (1984, 1989c). Before moving on however, it is important to rec: „ Wright’s aims. Wright’s strategy is to avoid convoluted analy=, J trusting many details to pragmatic factors which wi app J explanation, and case. For Wright, function hinges directly on mltitude of ways. explanation is pragmatically sensitive in a m> "analysis" of There is a sense in which Wright’s theory is not an writers were function in the sense that earlier accounts are. Earlier L.....-• largely concerned with how it can ever be be that that something ?omein ’s= existence can be inferred from its function, given that ® ® could often have done the same job (Hempel, 196ai. 1 . inference, it was thought there could be no functional explanmio^ -•» realistic Wright simply insists that with a less demanding, mor •explain the picture of explanation, it becomes clear that peop e o ex presence of t hings in terms of what they do, and a unc effect that operates in such an explanation. Wright also hopes. I suspect, that some naturei djng notion of function will be mirrored and explaine . h nd where slack in the notion of explanation, that the fonnulation the concept analyzed naturally bends, w right s
202
Peter Godfrey-Smith
of the relevant explanandum—“why its there"—is intended to wrap unsystematically around a variety of explanatory projects, in biol ogy, engineering and everyday life. Nonetheless, counterexamples such as Boorse's do suggest that Wright has backed off too early, and a sensitivity to pragmatics should not prevent us from pushing an analysis as far as we profitably can. Millikan's analysis, like Wright's, is historical. It locates functions in actual selective histories. The most important sophis tication of the historical approach in Millikan (1984) is her detailed treatment of functional co/egor/es. The first concept she defines is that of a “reproductively established family." A reproductively established family is a group of things generated by a sort of copying. Family members can be copied one off the other, or be common copies off some template, or be generated in the per formance of functions by members of another family. These differ ent kinds of copying are all distinguished by Millikan, but the finer divisions are not important here. Call any entities which can be grouped as tokens of a type by these lines of descent by copying, members of a “family." Understand "copy" as a causal matter involving common properties and counterfactuals. The copy is like the copied in certain respects, though it is physically distinct, and jf the copied had been different in certain ways, then, as a conse quence of causal links from copied to copy, the copy would have been different in those ways too (1984, p. 20).2 So two human hearts are members of the same family', as are two frill-necked lizard aggressive displays, two AIDS viruses, and two instances of the acronym “AIDS," assuming that acronym was hit upon only once. But two planets, and two time-slices of a rock or hose are not, as one was not copied off the other, nor are they produced off a common template, and so forth. Functions are only had by family members, and the performance of a function must involve the action of one of the properties copied, one of those properties defin ing the family. This restriction deals with many of Boorse’s counterexamples, such as the gas hose case. It also removes from the realm of func tion some cases Wright was concerned to capture, such as the news paper under the door. However, our project here is to capture the biological usage. Preserving a continuity between biological cases and other domains can be sacrificed.3 The next step is to add to this an explanation-schema in the style of Wright. The explanandum is the existence of current mem bers of the family. The explanans is a fact about prior members.
A Modern Hi dory The ory of Function s
20.3
(Fl) The function ofm is to/' iff; (i)
m is a member of family T, and
(ii)
among the properties copied between members of /’ is property or property cluster C, and
(Hi) one reason members of T such as m exist now is the fact that pa st. members of T performed F, through having C. member’s function whatever prior Most simply, a family members Junction is vnaie.ci y members did that explains why current members exist . • • It°?s one^f^hVstrengths of the historical approach
with an appeal to “families” that it can say without strain Particular thing which is in principle unable to do F new, less has the function to do F. It has this function >" membership in a family which has that function. member can do F is irrelevant to its family meniinonna) it was produced by lines of copying tha. - B p enough. A genetic defect may produce a heart unab" ;arnf way blood, but if this token was produced in morfr-or --as others, it has the function characteristic o 1 " . history. At this point we must confront an issueMil]ikan’s It is striking that while analyses such as »> bjo]ogjsts do not more expensive to get rid of than to retain. u f ething with describe junk DNA like this; it is the genetically on no function. Similarly, characters which hi c v .tabi|jty dike useful traits or persist through ^eveloPm5n..'jojng’ things which male nipples) might, in extended senses, e . . gome restriclead to their survival. So we might const er
204
Peter Godfrey-Smith
tion on the selective processes relevant to functional status. This will not be easy. A simple requirement, that the trait do something positive, that the null power is not a power, will not suf fice. Beside the cases where biological entities persist through doing nothing, there are positive and selectively salient powers which seem unlikely candidates for functions. As well as junk DNA, which does nothing, there is “selfish DNA" (Orgel & Crick, 1980). Selfish DNA can move around within the genome, replicating itself as it goes, and proliferate in a population despite having deleteri ous effects on individuals carrying it. Similarly, segregation distorter genes disrupt the special form of cell division (meiosis) which produces eggs and sperm (gametes). Meiosis usually results in a cell with two sets of chromosomes giv ing rise to four gametes with one set each, and on average a par ticular type of chromosome will be carried by half the gametes pro duced. Segregation distorters lever their way into more than their fair half share of gametes, by inducing sperm carrying the rival chromosome to self-destruct as they are formed (Crow, 1979). Fruit flies, house mice, grasshoppers, mosquitoes and a variety of plants are known to have segregation distorters in their gene pools. Now, disrupting meiosis is something that segregation distorter genes io, that explains their survival (Lewontin, 1962). Further, this ■xplanation appeals to natural selection, at the gametic level; the problem can not be solved by disqualifying traits that survive for non-selective reasons. Disrupting meiosis is not generally claimed to be the genes’ function though. Should we restrict the powers which can become functions, to exclude these subversive cases?4 There are two attitudes we might have to this issue. First, as a question of conceptual analysis, there is not much doubt that biologists typically restrict the powers that can qualify as func tions. Many might say we should then change the selective theory of functions to include this factor. An obvious move is to bring in some reference to the goals of some larger system. Disrupting meio sis makes no contribution to the goals of individuals bearing segre gation distorter genes, so this is not a function. An appeal to goals is certainly a step backwards however. So we might consider a more aggressive attitude to the problem. It may be that many biologists reserve “function” for powers with some intuitively benign nature, and withhold it from more subver sive activities, with there being no theoretically principled reason for this distinction. Some hold that biology since the 1960’s has produced, for better or worse, an increasingly cynical view of the
A Modern History Theory of functions
205
coalitions that, make up organisms (Dawkins, 1982; Buss, 1987), families (Trivcrs, 1974), and larger groups (Williams, 1966; Hamil ton, 1971). 'J'he feeling that functions must involve harmonious interactions may, from this point of view, be a holdover from an ear lier, more truly teleological view of nature. It might be claimed that •he theoretically important category of properties, the category our concept of function should be tailored to, is simply the category of selectively salient powers and dispositions.' If so, we should remain w>th the simpler analysis that allows any survival-enhancing power, however subversive, to qualify as a function. Although some may favor this more heartless approach I will adopt a third, intermediate position. Consider first another coun terintuitive consequence of an unembellished selective account: whole organisms, like people, have functions. Past tokens of people did things—survived and reproduced—that explain why current tokens are here. Hence, we have the function to survive and repro duce. This usage seems odd—note that these are not functions peo ple might have with respect to some social group, they are functions people just have, individually. One way to exclude both people as bearers of functions and also exclude disruption of meiosis as a function of segregation distorters is to stipulate that (i) the func tionally characterized structure must reside within a larger biolog ically real system, and (ii) the explanation of the selection of the functionally characterized structure must go via a positive contriution to the fitness of the larger system. My account here resem bles that of Brandon, who requires that a functional trait increase the relative adaptedness of [its] possessors 1990, p. 188). Bran don requires not just selective salience, but selective salience which goes via the fitness of a larger system “possessing the trait. The catalog of “real systems" is taken from biology, and clari fying the catalog is part of the units of selection problem. Individ uals, kin groups and perhaps populations and species might be examples of these systems. Thus hearts reside within people, and survive by aiding people’s fitness. But people, considered individu ally, reside within no such systems. There may, however, be groups within which people do things which contribute to the selection of the group, and then people would have functions. Similarly, segregation distorter genes do not have the function °f disrupting meiosis, because their proliferation under selection does not occur through a positive contribution to the fitness of indi viduals bearing these genes. Indeed, many segregation distorters, when present in two copies, greatly impede the fitness of their car-
206
I’cler Godfrey Smith
tiers. On the other hand, as some readers may have felt earlier, there could well be functional characterization of parts of segrega tion distorter genes or gene combinations. Some part of t he gene or combination might have its current presence explained by the fact that it has been selected for carrying out some part of t he segrega tion distortion project. Crow (1979) distinguishes two genes which cooperate to produce segregation distortion in fruit Hies. The "S" gene produces sabotage in sperm, and the “/?" gene stops the chro mosome that the S and 7? are on from sabotaging itself. So a chro mosome with S but no 7? sabotages itself, and a chromosome with 7? but no S does not distort, but is immune to distortion by its rival. Here the segregation distorting chromosome is the larger system, and the selective explanation of S goes via the explanation of the success of the whole chromosome. S has the function of sperm sab otage, and it has this function with reference to the segregation dis tortion gene complex. The selection of R is only partly an explana tion in terms of the selection of the distorter chromosome, as R is useful without S, once the population contains some chromosomes with 8. So R has the function of preventing sabotage, and it has this function with reference to two larger units, the segregation dis torter complex and the individual. It is important that not all failures on the part of evolution to produce intuitively well-engineered animals disqualify selective episodes from bestowing functions. A question sometimes arises concerning the status of traits which are explained in terms of some forms of sexual selection. If it is true that sexual selection can oper ate through females favoring characteristics in males which have no other benefit or use to the male (Fisher, 1930; Lande, 1981), then the explanation of a bird's long tail is not an explanation in terms of anjdhing intuitively useful the tail does. The explanation is simply that females prefer long tails (Andersson, 1982). Once a female preference gets established, for any reason, it can be sus tained and made stronger through the association of the gene for the preference in females (unexpressed in males) and the gene for the preferred trait (unexpressed in females). The preference leads to the selection of long tails, and the selection of long tails leads to the strengthening of the associated preference. The Jong tail could be a hindrance elsewhere in life. Consequently, some biologists hes itate to describe the tail as an adaptation, and functional in the ordinary sense: “Runaway sexual selection is a fascinating example of how selection may proceed without adaptation” (Futuyma, 1986, p. 278). On the present account how'ever the tail has the function to
A Modern History Theory of Functions
207
attract females. 11, has been selected because of that power, and this explanation goes via the augmentation of the individual's fitness.7 Here is an amended definition: (b'2) Th': function of m is to F iff: fi)
ni is a member of family T,
Hi)
member:-; of family T are components of biologi cally real systems of type .S’,
(Hi) among the properties copied between members of 7' is property or property cluster C,
(iv) one reason members of T such as m exist now is the fact that past members of T were successful under selection, through positively contributing to the fitness of systems of type S, and
(v)
members of T were selected because they did F, through having C.
HL Looking Forward Although philosophers have discussed a variety of intuitive problems with the view that functions derive from a selective his tory (Boorse, 1976), the most damaging charge against this view derives from the biological literature, from the wide acceptance o the distinctions made in “Tinbergen’s Four Questions. It is common in ethology and behavioral ecology to distingui= four questions “why?” we can ask about a behavior. Someone vv o asks why frill-necked lizards extend the skin around their necks so spectacularly might want an answer:
1. In terms of the physiological mechanisms and the phys ical stimuli that lead to the behavior. 2. In terms of the current functions of the behavior. 3. In terms of the evolutionary history of the behavior.
4. In terms of the development of the behavior in the life of the individual lizard.
This four-way distinction is usually attributed to Tinbergen (1963). Tinbergen in turn credits Julian Huxley with distinguishing
208
Peter Godfrey-Sm itli
questions 1-3, and adds question 4. Tinbergen, it must, be admit ted. uses the term “survival value" rather than “function" in the official formulation of question 2. But generally he uses these two expressions interchangeably (1963, pp. 417 & 420). Tinbergen's distinctions are often endorsed in the opening pages of books about animal behavior (Krebs & Davies, 1987, p. 5; Halliday & Slater, 1983, p. vii; and see also Horan, 1989). This is clearly an embarrassment for any historical theory of function which seeks to capture biological usage: on the historical view there should be three questions, not four, as the functional question is a question about evolutionary history, as long as the rest oi (F2) above is satisfied. Related distinctions with this separation between function and history are found elsewhere in evolutionary writings as well. Mayr (1961) distinguishes ••functional” from “evo lutionary biology, and Futuyma’s widely used textbook echoes Mayr in dividing the study of biology into functional and historical "modes" (19S6, p. 286).s There are various ways to respond to this problem. Many ahistorical usages of “function" are probably best understood as refer ring to Cummins functions. However, it is common for writers to both regard functions as ahistorical and regard them as intrinsi cally tied to natural selection, sometimes via the expression sur' ival value. This supports the proposal of a number of writers that functions involve not actual selective histories, but probable futures of selective success, or atemporal dispositions to succeed. Tinbergen may have accepted such a view: “the student of survival value, so-to-speak, looks 'forward in time’" (1963, p. 418). Tinbergen ip. 428) also casts the question about a structure’s function as a question about how deviations from the actual structure would lower the fitness of the bearer. John Staddon concurs (1987, p. 195). One way to develop this approach is with an appeal to propensities. Bigelow and Pargetter (1987; develop a theory of functions modelled explicitly on the widely accepted propensity view of fit ness (Mills & Beatty, 1979). The propensity view of fitness claims that the fitness of an individual is not the actual fact of its repro ductive success, but its propensity to have a certain degree of repro ductive success. Similarly, Bigelow and Pargetter claim, functions should be understood as dispositions or propensities to succeed under natural selection. “Something has a (biological) function just when it confers a survival-enhancing propensity on a creature that possesses it” (1987, p. 192). The propensity' view’ is not satisfactory, though its failure per-
A Modern History Theory of Functions
209
forms the valuable service of narrowing the discussion down, along with Tinbergen’s Four Questions, to a point where the modern his tory view will become compelling. I will discuss first some internal difficulties with the propensity view and then argue that the whole forward-looking approach is on the wrong track. The central internal problem is that as one tries to fill in some more details, the theory tends to go in one or other of three unde sirable directions. It can become enmeshed in strong counter-factual commitments. Alternately, it draws on the historical facts it sought to avoid. Or thirdly it makes the wrong kinds of demands on the future. Putting it briefly: propensities to be selected and survive bestow functions, but, the questions swarm: survive where? be selected over what? Bigelovz and Pargetter address the first ques tion, admitting that their account “must be relativized to an envi ronment” (p. 192). The context assumed is the creature’s “natural habitat.” “Natural habitat,” it appears, is understood historically by Bigelow and Pargetter. The statistically most common context for a trait now might be odd and unnatural 'Meander, 1991a). More worrying is the question of the competitors that have a propensity to be ousted from the population by the trait we are inter ested in. Bigelow and Pargetter make no mention of the fact that claims about propensities to do well under natural selection are surely always comparative claims. A trait does not have a propen sity to be selected and survive simpliciter, but always a propensity to be selected over some range of alternatives. Evolution is driven by differences in relative fitness. Bigelow and Pargetter cannot claim that current useful traits would triumph over any possible alterna tives. Which are the relevant ones? Those alternatives genetically attainable (given mutation rates, population structure, other con straints . . .) now? Those that could enter the fray during the next thousand years? Those that could enter the fray if the ozone layer goes and mutation rates are elevated? If Bigelow and Pargetter think there is a range of alternatives, and circumstances of selec tion, appropriate to the trait in question independently of history, they are making strong modal commitments. These might be avoided with an appeal to what is most likely to happen in the actual future, but then problems are created by (what appear to be) irrele vant contingent features of this future. If a trait is adaptive, but doomed because of linkage to something bad, then it is not likely to survive. But this should not make a trait itself non-functional. So, though the propensity theory is tailored to avoid dragging up the past, the propensities involved must either make tacit refer-
210
J
Peter Godfrey-Sm ith
once to millennia gone by, inappropriate predictions about the future, or questionable modal commitments about relevant ranges of alternatives and circumstances of selection. These internal prob lems are important, because it is easy to think that propensity views are somehow more economical than analyses appealing to the past. Still, the propensity view has recommendations. It does seem to be a way to accommodate the intuition that functions derive from selection with the observation that many biologists keep functional and historical questions separated. In addition, I am often told that no matter how questionable philosophers may find the modal commitments outlined above, many biologists con stantly talk as if these facts are quite unproblematic and accessi ble. It is difficult to work out the right attitude to such a datum. Further, one principled way to deal with these internal problems is to fashion a mixed theory, using the basic propensity format with an appeal to history to answer the objections raised above. (This mixing was suggested to me by Elisabeth Lloyd). The mixed theory claims that functions derive from propensi ties to be selected, but all the factors that Bigelow and Pargetter left vague are understood historically. The relevant ecological con ditions are the actual ones that obtained during the development of the trait. The range of alternatives the trait has a propensity to be selected over are the ones it actually triumphed over, and continues to be selectedI over. The propensity that bestows functions is strictly atemporal; a trait 1 is held to have a certain advantage under certain conditions over .’CT certain rivals. But these conditions and rivals are determined by the actual world. So it does seem likely that the propensity approach can be developed in a coherent way, at the price of narrowing the gap between it and the historical view. This is the general form of the contemporary functions debate: each theorj is made more plausible by setting it on a course of convergence with its rivals. There is, however, a more important problem with propensity theories, and other forward-looking views. These theories inevitably distort our understanding of functional explanation. In the first section I claimed that the sense of function under discus sion is a sense linked in some way to explanations of why the func tionally characterized entity exists, or exists in the form it does. The most straightforward way to envisage this link, which I have been assuming, is to say that functions are used in explanations of v hj the functionally characterized thing exists now. If this is granted, and the explanation is understood causally, then there is
I A
A Modern Hi-lory Theory of Functions
211
a simple argument against propensity views. The only events that can explain why a trait is around now are events in the past. Forward-looking accounts claim that functions are not bestowed by facts about the past, hut rather by how things are in the present. But then appealing Io a function cannot itself explain the fact that the trait exists now. If the environment is uniform, then present propensities to do well under selection may be a good guide to actual prior episode ; of selection. But this epistemological point docs not alter the fact that it is not the present propensities, but the prior episodes, that arc causally re sponsible for how things are now (see also Millikan, 1989a; Neander, 1991b).: I do not claim that Bigelow and Pargetter have missed this straightforward point. On their view, there is a problem with the background assumptions I have made about the explanatory role of functions, and which the argument above assumes. Bigelow and Pargetter claim that if the fact that some effect is a function itself depends on the fact that this effect explains the survival of the trait in question, if the assignment of a function is always retrospective in this way, “then it is no longer possible to explain why a charac ter has persisted by saying that the character has persisted because it serves a given function” (1937, p. 190). This vacuity problem can be solved, according to Bigelow and Pargetter, if functions arc understood as propensities. These propensities can be used tc explain the existence of a trait in the present if we claim, in addi tion, that the propensities in question did exist in the past, and were causally active in the past. This postulation of the past action of the propensities is an extra claim: it is not guaranteed by the mere fact that the effects in question are functions. Bigelow and Pargetter’s claims about explanatory vacuity and the historical view have been criticized effectively by Sandra Mitchell (1993). She points out that if we say “Trait A’ persisted because it had a consequence responsible for its selection and con sequent evolution,” this is only vacuous if we read "persisted" as meaning “evolved by natural selection." That is, it is only vacuous if we assume that the only mechanism which could explain some trait being around today is natural selection, though in fact there are alternative evolutionary forces which could play this explana tory role (1993, pp. 253-54). This is correct, and it shows that Bigelow and Pargetter’s argument about the vacuity of the histori cal view assumes an implausible adaptationism. There is. also another objection to Bigelow and Pargetter's claim, which is com patible with even the strongest adaptationism. On the historical
212
Peter Godfrey-Smith
view and with the assumption of adaptationism, it will he truly vac uous to say that A' persisted because it serves some function, because we are assuming that this is the only possible type of explanation. But even against this background it will of course not be vacuous to say that A' persisted because it. provided effective camouflage, or because it attracted mates, or because it conserved heat. Neither is it vacuous to say that the trait persisted because some specific effect was its function. If the historical theorist, says “A'persisted because its function was to conserve heat," this is to be translated into something which is ungainly, and contains a redun dancy—"A' persisted because its actually-selected effect, was that of conserving heat." But this is not vacuous; it does contain a real explanation, though to express it this way mentions the explanatoriness of the effect twice. So this is not the most natural mode of expression for the historical view; on that view the sentence “The function of A" is to conserve heat" is itself explanatory, and if some one is asked “Why is A' there?" they can reply by simply citing the function. This is not possible at all on the propensity view. On the propensity view, a functional explanation must give a function and also make an additional claim that the function was causally active in the past. So despite what Bigelow and Pargetter claim, as long as “a given function” is understood to refer to some specific task or ben efit, it is not trivial to say that “the character has persisted because it serves a given function," even assuming adaptationism. This, along with Mitchell’s argument, shows that there is no vacu ity problem with the background assumptions about explanation that proponents of the historical view make. It is possible to retain the explanatory force of function ascriptions, along with the philosophically attractive view, argued by Wright, that actual explanatory salience is exactly what distinguishes functions from mere effects. A “forward-looking” approach to functions has also been endorsed bj’ Barbara Horan (1989>, but the claims she makes about explanation are more problematic than those of Bigelow and Par getter. Horan says “questions about the function of a given pattern of social behavior are a way of asking how that behavior enhances the fitness of an individual who engages in it” (1989, p. 135). Nev ertheless, she claims soon after that the presence of a trait like a social behavior can be explained by an attribution of a function to that behavior. The model of explanation she applies, citing G. A. Cohen, is called a “consequence explanation.” Consequence expla-
A Modern History Theory of Functions
213
nations use laws of the form: “If (if C then E), then C.” In the pre sent context: “if a behavior pattern would increase individual fit ness, individuals will come to display that behavior’ (1989, p. 136). This is trying to have it both ways. It is true that useful things a behavior does now can lead to its prevalence in the future. So for ward-looking functions may predict and explain the future preva lence of a trait. But if the explanandum is how things are now, nothing present or future can be the explanans. Only the past will do. Of course, traits that are useful now were often useful then, so we can often infer that a propensity existing now was also causally active then. But if so, it is explanatory with respect to the present because it was causally active then. To claim that present useful ness in itself explains the morphologies and behaviors organisms presently display, and to build this into an account of functions, is to distort the explanatory structure of evolutionary theory.
IV- The Modern. History Theory
It might appear that we are painting ourselves into an ana lytical corner. Historical analyses are unacceptable because they fail to respect an apparently important distinction in biology between functional and evolutionary explanation. Forward-looking analyses are unacceptable because they distort our understanding of functions’ explanatory role. In fact there are several options available at this point. Bechtel (1989) suggests that we retain a forward-looking account of functions while giving up our prior concep tion of functional explanation. We might, alternately, claim that functional explanation just is evolutionary explanation, and banis other notions of function (except for Cummins ) as creatures of tele ological darkness. A third option is to analyze functional explana tion as a particular kind of evolutionary explanation. One alterna tive here is to regard a functional explanation as a selective explanation which satisfies (F2) above, hence a subset of evolution ary explanation. The option I prefer, however, is to construe unc tional explanation more narrowly still. This brings us, at last, to the modern history view: functions are dispositions and powers which explain the recent maintenance of a trait in a selective context. Several people have already sai , in effect, that this is the answer, but these people either make the sug gestion in passing (Kitcher, 1990), or more often, they only say it some of the time. Horan says “to explain the maintenance of a rai
214
Peter Godfrcy-Sinitli
in a species, one gives a functional explanation" (1989, p. 135), but insists on an atemporal construal of this explanation. And consider this remark of Millikan’s, in response to Horan:
If natural selection accounts for a trait, that is something that happened in the past, but the past may have been, as it were, “only yesterday." Indeed, usually the relevant past is only yesterday: the main business of natural selection is steady maintenance of useful traits against new intruders in the gene pool. But only yesterday is not outside of t ime. (1989a, p. 173) We need not endorse the claim about the “main business” of natural selection; whether or not maintaining traits is the main business of selection, it is one important kind of selection. It might be important enough to make this a constitutive part of the concept of function. Millikan does not take this step: her historical account does not build into functions the historically recent nature of the relevant selective episodes. Indeed, in her 1984 treatment she explicitly allows powers which were important in ancient history, but not in modern history, to be functions (1984, p. 32). In the 1989a treatment her emphasis is different, and she claims the rel evant past is “usually' only yesterday. But perhaps, as far as func tions go, it must be only yesterday. The modern history view does not respect the letter of Tinber gen's Four Questions, but it is faithful to their spirit. Tinbergen makes the modem/ancient history distinction himself (1963, pp. 428-29), but he regards both these explanations as “evolutionary” rather than functional. This puts two distinct questions under one head, however, as well as leaving the explanatory significance of functions in the dark. From the present viewpoint, the “evolution ary" question is the question about the forces which originally built the structure or trait in question. This may' or may not be a selec tive explanation, and this explanation might be different from the explanation of why the trait has recently been maintained in the population. Some might wonder how recent the selective episodes relevant to functional status have to be. The answer is not in terms of a fixed time—a week, or a thousand years. Relevance fades. Episodes of selection become increasingly irrelevant to an assignment of func tions at some time, the further away we get. The modern history view does, we must recognize, involve substantial biological com-
A Modern History Theory of Functions
215
mitments. Perhaps traits are, as a matter of biological fact, retained largely through various kinds of inertia. Perhaps there is not constant phenotypic variation in many characters, or new vari ants are eliminated primarily for nonselective reasons. That is, per haps many traits around now are not around because of things they have been doing. Then many modern-historical function statements will be false. If functions are to be understood as explanatory, in Wright’s sense, there is no avoiding risks of this sort. One way to support the modern history view of function is to demonstrate that the category of explanation it distinguishes is a theoretically principled one. This can be done by focusing on traits for which the modern historical explanation and the ancient histor ical explanation diverge, so the -elective forces salient in the origin of the trait are different from those salient in the recent mainte nance of the trait. Here is where a distinctively functional style of characterization—in the modem history sense—can be seen to be useful. The importance of the distinction between modern and ancient evolutionary explanations is discussed, in support of an analysis of function quite opposed to mine, in Gould and Vrba (1982). The central concern of Gould and Vrba is a distinction between adaptations and “exaptations" their coinage). They under stand adaptations as characters shaped by natural selection for the role they perform now. Exaptations are characters built originally by selection for one job, or characters with no direct selective expla nation at all, which have since been coopted for a new use. This analysis has consequences for their concept of function; only adap tations have functions, and exaptations have "effects.’’ Gould and Vrba do not discuss the recent past, as distinct from the present, so I am uncertain how they would classify modem-historical func tions. Generalfy they seem to understand effects-of-exaptations as propensities (19S2, p. 6). Their effects-of-exaptations correspond to the functions of Bigelow and Pargetter, and Horan. It should be clear why I think their way of dividing the cases is inadequate: modern history and ancient history can both furnish genuine expla nations, which we should distinguish, for why something exists now, while present propensities cannot themselves furnish such explanations. Gould and Vrba’s central point is the importance of cases where a trail's original and current uses diverge. But these are also cases where the selective forces that built a character and th.- actjvities; but lions with respect to some analysis ot an indhi u.
260
Denis M. Walsh and Andre Anew
as function is usually understood, only the former of these is the heart's function, while the latter is a mere accident . A heart which (in propitious conditions) cannot pump blood still has the function of doing so; it merely malfunctions. The richer, normative notion of function is clearly in use within evolutionary biology. There has been considerable energy expended recently on cobbling together a definition of function, understood in this way, from the concepts belonging to evolutionary biology. To our knowledge, there are three general approaches; call them the “historical theory” (or the “etiological theory"), the “for ward looking theory,” and the "relational theory." They share one central idea: that a trait’s function—its evolutionary function—is determined by its contribution to fitness. Fitness, as we shall use that notion, is a property of an individual, its propensity to survive and/or reproduce/ It is the fitness differences between individuals that natural selection works upon. Thus, in all three theories, evo lutionary function is tied in some way to natural selection. Though the central idea is the same in all three, it is the nuances that make the difference. We offer a survey of the salient features of each which, we hope, is sufficiently detailed to justify our claim that one, the relational theory, is clearly superior. The received view of evolutionary functions originated with Millikan and has since been embellished by Neander and GodfreySmith.5 The biological function of a trait, x, is to do m just in case individuals possessing x have been favored by natural selection in the past because their xs have med.'-5 The crux of this account is that the function of a trait is the way it has contributed to fitness in the past. This is known as the etiological, or historical, theory of evolutionary functions. A number of advantages have been claimed for this theory. First, it captures the sense of utility inherent in a function attribu tion; the historical function of a trait explains what that trait is for, by invoking the benefit to its possessors in past environments. Sec ond, by appeal to selection in the past, historical function also explains a trait type’s current presence or prevalence in a popula tion." Third, function attributions on the etiological account appear to be normative. They tell us what a trait ought to be able to do. Hearts ought to pump blood because that is what hearts have been selected for in the past. Because historical function in biology is normative in this way, it supports the distinctions between func tion and accident and between function and malfunction. Traits generally do more than they have been selected to do, but not
A Taxonomy of Functions
261
everything a trait docs is a function. For instance, hearts have been selected in Un- past for their blood pumping capacities, so pumping blood is a heart’s function. Some of the things heart- do, like cause traces on cardiograms, are mere accidents. If a heart fails to pump blood (in propitious conditions), it is malfunctioning. The etiological theory is undoubtedly powerful, but we con tend that it captures only part of the concept of evolutionary func tion. Not only rlo v/c think it is incomplete, we also think that the prevalent, yet mistaken thought that, all biological functions are historical has prevented a clear picture of the relation between Cfunctions and E-functions from emerging. As we see it, there are three problems with the historical theory. One concerns the notion of utility inherent in a function ascription. The second concerns the explanatory role function attribution plays in biology. The third involves normativity. First, the etiological account doesn’t always properly capture the notion of utility inherent in the concept of function. Consider that the contribution a trait has made to fitness in the past may in some cases diverge from its current contribution to fitness. That is to say the utility of a trait in the past does not determine its cur rent utility. In such cases it is often customary in biology to associ ate function with current utility. For example, Liem demonstrates how novel uses of the pharyngeal jaw apparatus of cichlid fishes have contributed to their subsequent adaptive radiation.” He calls these innovations “novel functions' 1432f. . There is a common intv ition among physiologists and functional anatomists, for exampl that if a trait type generally contributes to fitness in a novel v.. within a population, then the novel way may well constitute a fum tion,13 For the same reason, we ought to be able to ascribe functions to novel characters which generally benefit their bearers.- The eti ological theory' explicitly withholds function ascriptions in such cases. A second problem concerns the presumed explanatory role of function attributions in evolutionary biology. Etiological functions explain whyr a particular biological character is currently preva ent a population.15 This is an important application of E-function ascriptions, but there are others. Explaining a trait's expected per sistence or maintenance in future populations by describing w at i is for is another. For example, we may want to explain why we would expect melanic moths (of the species Biston betulana) to replace light morph individuals in a newly polluted forest, the darK coloration happens to camouflage the moths against soot encrus e
262
Denis M. Walsh and Andre Ariew
trees. Dark moths are less visible to predators in the newly polluted forests and consequently have a greater expected number of off spring than their lighter conspecifics. Camouflage, it. appears, ought to be considered the (or a) function of melanic coloration in the newly polluted environment. In this case the function, camou flage, explains why we should expect the persistence or mainte nance of melanism in future populations, but doesn't explain its etiology.16 Alternatively, the current function of melanism may be invoked to explain perceived fitness differences between light and dark morph moths. The ascription ot merely etiological E-functions does not have these explanatory consequences.17 Third, the historical theory does not account for the norma tivity of function attributions as it claims to do. The historical the ory tells us, for instance, that vertebrate hearts ought to pump blood because that is what hearts have been selected for in the past.'8 We take it that this is intended to imply that vertebrate hearts ought to pump blood now for the same reason. But that seems wrong. An individual's heart ought to be able to pump blood note (in part’ because if it cannot she'll die. That is not a fact about hearts’ contribution to fitness in the past, it’s a fact about the way hearts generally contribute to fitness now. Hearts generally con tribute to fitness now by pumping blood. Those individuals whose hearts can't pump blood suffer a fitness decrement on average on account of it. More abstractly, a token of a trait ought to be able to contribute to fitness in the way that tokens of their type generally do now. Current contribution to fitness determines what a trait ought to do now, but current contribution to fitness is no part of the etiological theory of E-functions. A common thread runs through each of the above three criti cisms of the historical theory; that the general contribution that a trait type has made to fitness in the past does not determine its current contribution to fitness and that current contribution to fit ness ought to be at least part of the concept of evolutionary func tion. Because the etiological theory ties function exclusively to past contribution to fitness, it fails to encompass what might be called “current function.” This brings us to the second theory of evolutionary function. Bigelow and Pargetter have proposed the propensity theory, often called the “forward looking theory'.” The gist of the forward looking thesis is that the function of a trait is what traits of its type do to contribute to fitness currently. Function, according to this view, is not a historical notion; it has to do with the way a trait benefits its
A Taxonomy of Functions
263
possessors now, irrespective of what it has done in the past. To its credit, the forward looking theory permits the ascription of strictly current functions, as we have suggested an adequate theory of Ehmction should." But, like the historical theory, it is incomplete. It fails to recognize historical functions. As a result, it fails to account J°r an important da. : of explanations for which functions are invoked in biology. Forv/ard looking functions explain the future Persistence of a trait, but. not its etiology. We have canvassed two theories of evolutionary function: the historical and the forv/ard looking. Each captures what the other misses, but neither is complete. There is a need for both historical and current evolutionary functions in biology, and a need for the distinctive kinds of explanations that each issues in. The require ment for both current and historical functions brings us to our third theory, relational function. The relational theory, like its predecessors, ties function to fit ness contribution, but in a novel way. It has been developed, and explained in detail, by Walsh ("Fitness and Function"). The leading idea is that the way a trait contributes to fitness may' vary wildly according to the environment." Thus one cannot specify the contri bution that a trait makes to fitness simpliciter. Instead, one must specify the contribution to fitness with respect to a selective regime.-1 Consequently, one may’ not specify the function of a trait sim pliciter; instead one may only specif.' the function of a trait with respect to some selective regime or other. V>e give the definition of relational function first and then take a moment to discuss how it yields a satisfactory’ theory of E-function. Relational function is defined in the following way. RF: The/an evolutionary function of a token ot type A with respect to selective regime R is to m it and only if A s doing m positively (and significantly) contributes to the average fitness of individuals possessing A in R."
The function of a trait token i.r' with respect to R is the w ay ^ra‘" of its type (A') positively contribute to average fitness ot individuals >n R- That is equivalent to saving that the function of (trait token) * in R is just what (trait tvpe) A" is being selected for in Rr ms is •'I the utmost importance. If there is selection (in R) for trait type, in virtue of its (tokens’) capacity to do m, then the capacity o ‘‘xplains t hi- general utility to an individual of having a trait ot type It also explains the persistence ofA under natura se cc ion.
264
Denis M. Walsh and Andre Ariew
thermore, what a trait type is being selected for in R determines conditions of proper functioning for tokens of the trait’s type. Tokens of X are working properly when they are capable of doing what X is being selected for. Within the definition RF, we have the basis of all we require of a theory of E-function. Because of its rela tional nature, RF can recognize both current and historical func tions. Because the function of a trait is tied to selection for that, trait (in a regime), relational function preserves both the explana tory consequences and the normativity inherent in ascriptions of evolutionary function. We take these points in turn. Relational function, as defined by RF. is neither strictly his torical nor strictly current but instances of relational function may be either. When the selective regime. 7?, is a past regime, the func tion will be historical. When the relevant regime, R, is the current regime, the function is current/* RF also does justice to the variety of explanations for which evolutionary functions are invoked. A relational function explains the persistence of a trait type from one time t, to a later time t. within a regime R. When t. is the present, the attribution of a function to a trait tells us why that trait type is prevalent in the current population, that is, the function ascription provides us with an etiological explanation for the trait in question. When t, is the present, the attribution of a function to a trait tells us why we should expect traits of that type to be prevalent at t2. In addition, invoking function with respect to the current regime allows a biologist to explain observed fitness differences between individuals. Finally, relational function is fully normative. If the function of X (in R) is to do zn, then X is being selected for in R because of its (tokens’) capacity to do m. That just means that indi viduals whose x’s are capable of doing m are, on average, favored by selection. Individuals whose x’s cannot do m, on average, suffer a fitness decrement as a result. From the perspective of the individ ual, all in all, it’s a good thing if its x can do what x is being selected to do. If a token x is incapable of doing m, (in propitious conditions) it is malfunctioning. Traits (tokens and types) have many effects which do not dispose the trait to be selected for in R. These effects are accidents. Thus the function/malfunction distinction and the function/accident distinctions are preserved intact under the rela tional account of E-function. The difference between the etiological theory and the rela tional theory of E-functions can be thought of as follows. According to the etiological theory, the function of a trait is not a causal power or a disposition of the trait. It is the product of a past process. Con-
A Taxonomy of Functions
265
sider Millikan’s claim that the “basic form” of the definition of evo lutionary function, “looks to the history of an item to determine its function rather than to the item's present properties or disposi tions. In contrast, according to the relational theory, the E-func tion of n trait is a property or disposition which has some causal role within a selection process. The difference manifests itself in the following way. Suppose that we ascribe to x the function m with respect to R. By the relational theory, this entails that in any regime matching the description of R, tokens of x would causally contribute to the average fitness of their bearer’s by doing m. There is no such implication under the historical theory. It is because RF casts evolutionary function as a causal power or disposition of a trait to contribute to fitness that it allows us to elucidate the rela tion between evolutionary function and Cummins function. 3. Relation of C-Function to EFunction. A number of recent works have attempted to determine the relation between C-function and E-function within biology.-' There are two general approaches. One is to minimize, or deny, the apparent differences between the C-function and E-function concepts , of the body’s tissues, in circulate blood. Enf’s account is intentionally vertebrate hem description of the heart's C-function: reminiscent oi v t„ to sav that the heart functions as a pump H is appi oP • ind ofan analysis of the circulatory sysagmnstthe b“L \ran rt food, oxygen. wastes, and so on,
to the tact that the heart is capable of pump ing. (“Functional Analysis, 64)
, where the E-function of a trait and its CWe clearly have a case
Denis HI. Walsh and Andre Anew
270
The point here is a methodological one; evolutionary funct ions are discovered by conducting C-function analysis." It seems quite evident that m could not be designated the evolutionary function of a trait token, x. unless there were a Cummins-style functional anal ysis which established that it is (or was) the C-function of some token ofx’s type to do m.” This is a necessary condition for a C-func tion analysis to reveal an E-function. but it is not sufficient: some thing more is needed to make a C-tunction an E-function. Recall that the E-function of a trait token is determined by t he contribu tion to the average fitness of individuals possessing tokens of its type. So, in addition the C-function analysis would have to be such that:
(i)
the most inclusive system under analysis, s, is the individual (or the relevant unit of selection),
(ii) the capacity of s. V', to which x contributes is survival or reproduction or both: i.e. the contribution of x is to s’s fitness, and that,
(iii) doing T significantly contributes to the average fitness of those individuals possessing the trait type to which x belongs. Conditions (iHiii) are intended to stipulate that the causal role revealed by the C-function analysis is also the (or a) contribution of that trait type to average individual fitness. Not every C-function analysis of fitness contribution reveals this typical contribution to fitness. Consider a C-function analysis of poor Sally’s heart. Sally’s heart does not make the same causal contribution to her fitness that properly functioning vertebrate hearts make to the average fit nesses of their bearers. Hence whatever contribution Sally’s heart makes to her survival cannot be its E-function. A Cummins-style analysis of her heart would not reveal the E-function of vertebrate hearts. So what is the relation between the C-function concept and the E-function concept? As we mentioned, there is no straight reduction of one concept to the other. However, conditions (i)-(iii) above suggest that we can reduce E-function to C-function plus the concept of contribution to average individual fitness (with respect to a selective regime) employed in the definition of relational function, RF. The following relation holds:
A Taxonomy of Functions
271
7’/;^ E-function of ft trait token (with respect to a regime) is that C-function which constitutes the (positive) contribution to average individual fitness for tokens of the trait’s type (tinth reject to that regime).
Notice that this reduction of E-function to a kind of typical C-funct'on does not entail that if it is the E-function of (token) x to do tn, th 'ri x actually performs m in the economy of the individual. All it 'tails is that traits of x':> type significantly and positively con tribute to the average fitness of individuals by doing m. One might be tempted to object that if there is to be any reducf one kind of function to the other it should go the other way, b°n use to quote Dennett, . - lt]he biologist who helps himself CCaUto 'such an obviously safe functional category as eye, leg, or even a|rcady committed to assumptions about what is good.” '^ly one cannot ascribe a C-function to a lung without recogvi en ]ungi and iUng is an E-functional category. So, E-funcruzing i on is prior to C-function ascription. The objection fails ti°n ascro^e of the salient differences between C-functions and Eto nee above. C-functional analyses hold over trait tokens functions t0 tfae types of which they partake. Thus, unlike witnou ascrintions, C-function ascriptions are made to a trait E-func io .ts e membership. Consequently, one may perform a irrespec i token lung without first recognizing it as VMe i.k.y well he fa hnrbot trait types in biology, such eye. leg, or lung, are defined according to their E-function, that does not entail that the E-funct on of a trait token must be recognized prior to an analysis of any
°fltSw fUmay°now turn briefly to Amundson and Lauder's claim
functions w anatomists attribute C-functions to traits, on the Physiologists an t-unetional analyses. But they don't basis of C unctions. The important questions are: -which attribute jusW interested in?- and -wh, these rather C-functions • undson and Lauder's response is correct as far as than othms. ' tha(. -functional anatomists typically choose to it goes; ley ■" • iricter complexes which have significant bio analyze integr at d C0J5titutes a "significant biological X?
.»
that a trait has a
MM
Denis Id. Walsh and Andre Ariete
272
when it contributes to the survival and (or) reproduction of indi viduals, and that this role is typical for the trait type. That is to say that typically a C-function has a significant biological role insofar as it meets the conditions (iHiii) above. When that occurs, the Cfunction is also an E-function. We claim that the C-functions most, commonly ascribed by anatomists and physiologists are also Efunctions and are interesting because they are E-funetions.11’ The above discussion suggests that every E-function is a Cfunction. More precisely, if m is the E-function of trait .v, then m is a C-function of a significant proportion of (most of?) the tokens of x’s type. This tells us the relation between the respective exten sions of the C-function and E-function concepts. Consider the matrix [below] . . . ; it encompasses the full range of function ascriptions in biology. Selective Salience for s
Explanatory
Enhances
Does not enhance
average fitness of s
average fitness of s
etiological
1
3
non-etiological
2
4
Role
Each of the cells represents a legitimate kind of C-function in biolog}’. We have argued, contrary to the historical theory of evolu tionary' function, that the concept of E-function comprises cells 1 and 2. When the selective regime is set as a past regime, and con ditions (i>—(iii) hold, the result of a Cummins-style functional anal ysis will be an historical E-function (cell 1). It will entail an expla nation of the etiology’ of the trait. When the selective regime is current, and conditions (iMiii) hold, the result will be an ahistorical E-function (cell 2). It will entail an explanation of what a trait is for, but may or may not entail an explanation of its etiology."' Cfunctions may fail to be evolutionary functions and yet may entail an explanation of a trait’s prevalence (cell 3). The example of junk DNA belongs here.11 Some cases falling within cell 4 have already been discussed: the function of krill in a marine ecosystem and the function of B-cells in the production of the symptoms of rheumatoid
A Taxonomy of Functions
273
arthritis. These are rnon-etiological non-E-functions. All four kinds of C-function ascriptions’ are important in biology. Only two of these, 1 and 2, are E-functions.
Hl. Etiological, Teleological, and Causal Explanation
The relation between E-function and C-function is mirrored by the relation between their respective explanatory roles. As we dis cussed, the Cummins approach to functions and the evolutionary approach arc motivated by sharply divergent opinions on the explana tory role that functions ought to play. The theory of evolutionary func tions we present is motivated by the view that E-function ascriptions are teleological and sometimes etiological. For Cummins, functional explanations are strictly causal, and neither teleological nor etiological. It would seem that Cummins functions and evolutionary functions are appropriate to different kinds of questions. Cummins functions answer how-questions, e. g. “how do hearts contribute to the activities of the circulatory system?” Evolutionary functions answer why-questions, e. g. “why do vertebrates have hearts?” or “why will vertebrates continue to have hearts?” E-functions explain the prevalence or) persistence of trait types. C-functions explain the effects of tokens. Within evolutionary' biology, however, the distinction between why-questions and how-questions is not so trenchant: why-ques tions just are a certain kind of how-question and certain answers to how-questions constitute answers to why-questions. To explain what a trait is for (i. e. to give a teleological explanation) in evolu tionary biology is to explain why that trait type persists from one time to the next under natural selection. In turn, why a trait persists is explained by citing how the trait type contributes in general to the fitnesses of its bearers (with respect to some selective regime?. Tele ological explanation in biology is a special case of causal explanation which invokes contribution to fitness. E-function attribution has teleological import precisely because it identifies a certain typical causal role (i. e. a typical C-function) in the determination of indi vidual fitness. When that causal role has occurred in past regimes, the teleological explanation that function attribution issues in is also etiological. Thus, within evolutionary biology, etiological func tional explanation is a special kind of teleological explanation, which, in turn, is a special kind of causal explanation. The relation between causal, teleological and etiological explanations reflects the relation between Cununins-style, evolutionary and historical func-
Denis M. Walsh and Andre Ariciv
TIA
tions. The subsumption of teleological explanation under causal explanation, via E-function, requires that E-function is a causal power or disposition. It is one of the virtues of the relational theory that it identifies E-function as a causal power or disposition.
IV. A Taxonomy Of Functions
We propose a taxonomy of functions as depicted in Figure 1.
C-Function
1 E-Function I
Causal Role
Current
Historical
(Cummins)
(Bigelow & Pargener)
(Millikan)
Etiological
Teleological
Explanatory Role
Causal
FIGURE 1. A proposed taxonomy of functions. See text for details.
a
A Taxonomy of functions
275
Figure 1 is to be interpreted in the following way. We recog nize three “nested" categories of function, each is distinguished by its explanatory role. The end of each branch represents a cat egory of function (from left to right: causal, non-teleological Cfunctions; current E function:-: and historical E-functions). Each is marked by the authorities who have proposed it as a more or less complete account of biological function. Our conception of the relation of C-function and E-function is represented by the square brackets along the top of the diagram (labeled “C-func tion” and “E-function"). Historical E-functions (The “Millikan" functions) constitute a proper subclass of E-functions, as decur rent E-functions (the “Bigelow and Pargetter" functions). E-func tions as a whole constitute a proper subclass of C-functions (the “Cummins” functions). The characteristic explanatory roles which distinguish the various types of function are shown along the main branch of the diagram (labeled "Causal," “Teleological,” and “Etiological”). As discussed in the previous section, all func tion ascriptions explain (a certain kind of: causal role. A subset of these, E-functions, further explains the prevalence and (or) persistence of trait types by citing their causal contribution to average fitness of individuals. Such explanations are teleological. A more exclusive subset of evolutionary functions, historical Efunctions, are etiological. They explain the current presence of a trait by adverting to the causal contribution to average fitness within the history of the lineage.
V. Conclusion
We endorse the relational theory of evolutionary function. Doing so permits us to elucidate the relation between the two con cepts of function employed within biology. E-functions are a kind of C-function typical for traits of a type where the causal contribu tion is to fitness. Consequently, the E-function of a trait is discov ered by conducting a C-function analysis of its traits typical con tribution to the survival and reproduction of the individuals possessing it. Finally, the explanatory roles played by E-functions are special cases of the explanatory role of C-functions. The pres ence and maintenance of traits is explained by citing that trmts causal role in the suni■ival and reproduction of individuals within a selective regime. •
.
’
Denis M. Walsh and Andre Arietv
276 Notes
1. We wish to thank the following people for comments: Henry Byerly, Robert Cummins, Berent En?, D. D. Kohn, .Joel 1’ust, Larry Shapiro, Elliott Sober, Chris Stephens, and two anonymous referees. This work was funded in part by SSHRC Postdoctoral fellowship 756 91 0750 to DMW.
2. R. Cummins, "Functional Analysis.” Journal of Philosophy 72 (1975) 741-65. Reprinted in E. Sober. ed„ Conceptual Issues in Evolution ary Biology 2nd ed. (Cambridge, MA: The MIT Press, 199 I). All quotations from this paper will be taken from the Sober reprint. 3. P. Kitcher, "Function and Design.” in P. A. French, T. E. Uehling, and H. K. Wettstein. eds.. Midwest Studies in Philosophy XVIII (Min neapolis: University of Minnesota. 1993) 379—97. 4. R. Amundson and G. Lauder. "Function without Purpose: The Uses of Causal Role Function in Evolutionary Biology," Biology and Phi losophy 9 (19941 443-69; P. Godfrey-Smith. "Functions: Consensus With out Unity,” Pacific Philosophical Quarterly 74 (1993) 196-208. 5. In The Nature of Psychological Explanation (Cambridge, MA: The MIT Press. 1983). Cummins amends his account of function to allow that s and x may also range over processes themselves (e. g. multiplying 27 by 32) and their component processes le. g. multiplying 2 by 7, adding 5 and 1), irrespective of how these capacities are instantiated.
6. It will become apparent that there is some dispute over how to interpret the question of why a feature is present. Some (B. Enc, “Function Attributions and Functional Explanation,” Philosophy of Science 46 (1979) 343-65; and we) believe that function ascription yields genuinely teleolog ical explanations of the presence of a feature (i. e. it explains why a feature persists by citing its effects?; others 'K. Neander, "The Teleological Notion of ’Function,”’ Australasian Journal of Philosophy 69 (1991) 454-68; P. Kitcher, "Function and Design”: P. Godfrey-Smith, “A Modern History The ory of Functions,” Nous 28 (1994? 344—62; R. G. Millikan “In Defense of Proper Functions,” Philosophy of Science 56 (1994) 288-302) believe that function ascriptions merely explain the current presence of a trait by cit ing its causal history. 7. L. Wright, “Functions,” Philosophical Review 82 (1973) 139-68. 8. S. Mills and J. Beatty, “The Propensity Interpretation of Fitness,” Philosophy of Science 46 (1979) 263-86; see also Sober’s Conceptual Issues in Evolutionary Biology. 9. R. G. Millikan, Language, Thought, and other Biological Categories (Cambridge, ALA The MIT Press, 1984) and “In Defense of Proper Functions”;
A Taxonomy of Functions
277
K. Neandcr, "Functions as Selected Effects: The Conceptual Analyst's Defense,” Philosophy of.‘science 58 (1991) 168-84 and “The Teleological Notion of'Function"’; I’. < lodfrey-Smith, “A Modem History Theory of Functions "
10. We . .hall ir ori.., ' being published hither and yon, it's been very difficult to follow. Buller has netted very readable and important essays, ones that track the course of the debate quite nicely. I was delighted to have them all collected, edited, and organized for me." —Barbara L Horan, Georgia Southern University
David J. Buller is Assistant Professor of Philosophy at Northern Illinois University.
A volume in the SUNY series in Philosophy and Biology David Edward Shaner, editor
STATE UNIVERSITY OF NEW YORK PRESS Visit our web site at http://www.sunypress.edu
ISBH 0-7914-4211-X
90000>
s 9 780791 442111
• PHILOSOPHY
• Function, Selection and Design David J. Buller, editor This authoritative book, written by the leading experts in the field of the philosophy of biology, brings together the defining literature in the debate concerning proper analysis of teleological concepts in biology. The introduction provides a clear and coherent overview to the philosophical progress regarding the nature of function in biology, and the book's chronological structure offers historical insight and perspective. This anthology is well-planned, representative, and current in its orientation. All of the major positions and figures are represented, and the volume is framed by Buller's essays, an organization that serves to consolidate many themes introduced by the diverse slate of authors. The scientific revolution ushered in a picture of the universe as governed solely by mechanical causation working forward in time, which appeared to leave no room in nature for teleological (or goal-directed) processes. But within the last decade a near-consensus has emerged among philosophers that the theory of evolution by natural selection provides the framework for a wholly naturalist analysis of the concept of function in biology, and thus solves the traditional philosophical problems regarding teleology. Function, Selection, and Design illustrates this growing consensus and the recent debate concerning the details of a fully adequate analysis of the concept of function. "The debate about functions has become rather complex, and with articles being published hither and yon, it's been very difficult to follow. Buller has selected very readable and important essays, ones that track the course of the debate quite nicely. I was delighted to have them all collected, edited, and organized for me." —Barbara L Horan, Georgia Southern University
David J. Buller is Assistant Professor of Philosophy ot Northern Illinois University.
A volume in the SUNY series in Philosophy and Biology David Edward Shaner, editor
STATE UNIVERSITY OF NEW YORK PRESS Visit our web site at http://www.sunypress.edu
ISDN 0-7914-4211-X
90000)
9 '780791 '442111