Computer chess 0710097859

Citation preview

C O M P U T E R

Ludek Pachman & Vas I. Kuhnmund Translated from the German by John Littlewood and Keith Littlewood

Routledge & Kogan Paul London and New York

First published in English in 1986 by Routledge & Kegan Paul pic II New Fetter Lane, London EC4P 4EE Published in the USA by Routledge & Kegan Paul Inc. in association with Methuen Inc. 29 West 35"' Street, New York, N Y I ОООI Originally published in German as Computer Schach by Wilhelm Heyne Verlag, Munich, 1980. Set in Times by Hope Services, Abingdon, Oxon and printed in Great Britain by the Guernsey Press Co. Ltd Guernsey, Channel Islands У Copyright © 1980 by Wilhelm Heyne Verlag. Miinchen This translation © Routledge & Kegan Paul, /9X6 No part o f this book may be reproduced in any form without permission from the publisher, except for the quotation o f brief passages in criticism Library o f Congress Cataloging in Publication Data Pachman, l.udek. Computer chess. Translation of: Computer-Schach. I . Chess— Data processing. I. Kuhnmund, Vas. I. II. Title. GV1449.3.P3313 1986 794. Г 7 85-25669 British Library ( IP data also available ISBN 0 -7 100-9785-9 (pbk.)

Contents

Preface

vii

Introduction: the chess co m pu ter - master or apprentice?

1

Part 1 C o m p u te r Chess

5

Vas I. K uhnm und 1 H ow intelligent are chess com puters?

7

2 Tiresom e but essential mathematics

31

3 Chess c o m p u te r programs versus humans, 1951-1969

52

4 M od ern times, 1970-1979: ‘Chess’ - the new star

64

5 A dvantages and disadvantages of hum an thought processes

86

6 Choosing and buying the right chess com puter

96

7 T h e future of the game of chess

117

Part 2 G a m e s

123

Ludek Pachman

Preface

T h e chess c o m p u te r m arket has seen a great deal of change since the first G e r m a n language edition of this book in 1980. The rather tem p estu o u s sales boom in this field which sent sales figures rocketing in the late 1970s, and which surpassed the wildest expectations in this direction, was gradually dying down. Bitter competition flared up for shares in a seemingly saturated world m arket. Q uite a few c o m p u ter manufacturers sustained heavy losses after the sales boom (Applied Concepts, 1980; Scisys, 1981; Fidelity Electronics, 1983). Some manufacturers miscalculated in a m ood of pro nounced optimism and had to concede the field to their com petitors (Chafitz Ltd, 1980; Mattel Electronics, 1981; C o n su m e n ta Ltd, 1982; Conchess, 1983). But even for the remaining m arket leaders, there is no longer any reason for exaggerated optimism. W h a te v e r the reason for it, the sales prognoses for the chess c o m pu ter field in the financial year 1984-5 do not predict any ‘giant leaps in growth'; on e thing is certain, though: the customer has becom e choosier, more demanding, more spoilt; in short, he has come of age! ‘Long gone are the days,’ maintains Ossi Weiner, managing director of the first G e rm an chess com puter store, HCC in Munich, ‘when the custom er took the first, or the cheapest, piece of equipm en t he saw, and bought it straight aw ay.’ This is not totally unconnected to the num erous publications on computer chess which have since appeared. In G e rm an y alone, there are now at least three specialist magazines dealing in great detail with the most up-to-date trends and developments. In view of these facts, the authors of this book have faced a difficult task. We wanted to present a chronological and historical account of lasting validity and at the same time also convey the most up-to-date information. We hope that we have at least partly

viii / Preface managed to overcom e these difficulties, and that for this first English language edition we have succeeded in striking a happy medium between these two tasks. Some sections and passages in chapters 2 to 4 have been thoroughly revised; the table section has been rearranged and expanded. C h ap ter 6 has been completely updated and rewritten, and should now provide a satisfactory and useful guide to buying chess com puters - both for beginners and for experienced old hands. T h e authors hope that their English readers will find lasting pleasure in studying this book. Any suggestions for improvements would be gratefully welcomed. Vas I. K u hnm und For the authors

Introduction: the chess computer - master or apprentice? Are robots about to checkmate us? T he artificial arm of a robot reaches for the black bishop, places it on the e3 square and stops the clock. The ro b o t’s baffled human o p p o n e n t can read the word ‘m a te ’ on the visual display unit. Is this just a science fiction gimmick after the style of the Stanley Kubrick films Star Wars or 2001 Space Odyssey ? Absolutely not! For International Master David Levy, at least, this vision turned into harsh reality during 1979 in Hamburg. T h e Scot’s fearful o p p o n en t was the ‘Chess 4.8’ program written by David Slate and Larry Atkin fed into a ‘Cyber 176’ m ade by the C D C firm, the fastest com puter on the market as well as being the World Champion among chess-programmed machines. In January 1979 this co m puter was on the verge of victory against Levy, and it was only with great difficulty that the Scottish F ID E title holder m anaged to obtain a draw from a losing position. In view of this arduously achieved draw, the experts are beginning to w onder how long human beings can compete with constantly improving co m puter technology and programming techniques. Even today at least 99 per cent of all chess players are hopelessly w eaker than the ‘Chess 4.8’ program, which relatively few highly qualified masters are capable of beating. The com puter must now be considered a clear threat to the future of our ‘game of kings’ which, as a creative art or exploratory science, has for centuries drawn man into intellectual combat. W e take it for granted that machines can swim or fly better than ourselves, yet it is for all o f us an alarming thought that they are beginning to provide serious competition in the exclusively human field of creative thinking. The very idea o f a Moliere comedy or a Bach suite produced by a co m puter is inconceivable to a man of culture! A b o u t 30 years ago a 30-ton monster com puter by the name of ‘Maniac 1', brought along by IBM experts, made its first entry into

2 / Introduction: the chess co m p u ter - m aster or apprentice? the chess arena to participate in a game which had hitherto been played solely by hu m ans. The experts were overjoyed when their ‘pupil’, on a smaller than usual 6 x 6 chessboard, m anaged to beat a confused lady who was not especially familiar with the finer points o f the game! It was not until the incredible increase in the speed of computers in the last decade (in the case of ‘Cyber 176’ approximately 18 million additions per second), along with the refinement of program m ing techniques, that the advance into this once exclusive dom ain of hum an thought was m ade possible. With the justly celebrated advent of the integrated circuit (IC), which enabled the old co m puter dinosaurs to be brought down to pocket size and operated within a time-scale of nanoseconds (i.e. one thousand millionths of a second), the game of chess, till then voracious in its d em ands on c o m p u ter time, finally became an acceptable proposition. This new development fulfilled the requirements for opening up chess co m puter technology to the general public by making it commercially viable. The ‘pocket grand masters’ imported from the USA cost about £250 in 1979 but by 1980 ‘Chess C ham pion Mark 2’, ‘Chess Challenger 7' or ‘Chessm ate C o m m o d o r e ’ could be purchased in a departm ent store for under half this price. As o pponents they had definite advantages, especially since they never felt insulted, nor did they lose their patience, and above all they proved to be good losers! However, in five or ten years' time, will a £50 pocket com puter with a vast calculating potential come along to spoil the pleasure people find in chess? Should we not ask too whether m an, in his pride, having recognised and exploited the physical superiority of machines, will learn to accept a m achine’s intellectual and even creative superiority? In fact, is the computerisation of chess to be equ ated with the gradual extinction of our noble game? It is beyond the scope of this volume to offer exhaustive answers to questions like these which b order on the realm o f philosophy. The a u th o rs ’ priorities lie ra th e r in helping the non-technical reader to find his feet in the dense forest of chess-computing, since even o u r beloved game cannot remain immune from the influences of cybernetics, information theory, psychology and other related disciplines. During the last few years the idea of c o m p u ter chess has spread

Introduction: the chess co m p u ter - master or apprentice? / 3 like wildfire, with the increasing interest in the subject only serving to fuel this developm ent. It is too early as yet to foresee the full effect this trend will have on the game of chess itself, but there is little d oubt that it will help to popularise it even more. A lthough it is impossible to provide a complete survey of the whole field of c o m p u te r chess, we shall nevertheless attempt to outline most of its elem ents and problems, to describe the present position and to offer suggestions for further possible developments. In addition, we shall en deav o u r to give the reader an insight into the worthwhile points o f knowledge to be found within the story of this two-hundred-year-old chess discipline.

Part 1

Computer Chess V as

I.

K u h n m u n d

Chapter 1

How intelligent are chess computers? Even today, experts cannot agree on the answer to our chapter heading. T h e question remains open as to what is actually meant by the term intelligence, a characteristic which is usually attributed to a wise and clever person with an astute mind. Books which attem pt a definition of intelligence do more to confuse than enlighten the reader. The dictionary informs us that intelligence is not only cleverness but also ‘the ability to conceptualise, c o m p re h e n d and evaluate information, as well as the intellectual capacity to adapt to new situations’. It also tells us that there is a so-called Intelligence Q u otient (IQ) to measure the above qualities. T o date, however, there is still no universally accepted definition, even am ong psychologists, who agree only upon understanding intelligence as a requirem ent or set of requirements for performing various tasks, without attempting a more specific definition. C la p a r e d e ’s and S te rn ’s view that intelligence is also the capacity to overcom e difficulties in new situations has found wide recognition. For the English c o m p u ter chess pioneer Alan Mathieson Turing, the solution to the whole problem was astonishingly simple: let so m eone unknowingly play a game of chess against a c o m p u ter, and if he cannot decide afterwards whether his o p p o n en t was a machine p rogram m ed for chess or a human being, then it would be possible to attribute to the com puter a certain degree of intelligence. Perhaps the English expert was correct in his assumption, since chess, by its very nature, seems to combine many of the above-m entioned aspects of this elusive concept of intelligence. Indeed, beyond its en te rta in m e n t, cultural and historical value, the erstwhile game of kings is most suitable for simulating actual thought processes. W hat m anufacturer can in fact resist the

8 / C o m p u te r chess tem ptation to d em o nstrate the capacities of a co m puter by program m ing it to play chess? Many people now believe that chess can therefore be regarded as an ideal test-bed for the creation and further developm ent of artificial (i.e. com puter) intelligence. For almost two centuries now, scientists have been fascinated by the Faustian idea of constructing a chess-playing robot with a human-like ‘b rain’. We may smile knowingly at the early attem pts in this field of research, but it is worth asking ourselves if the present-day developm ent of a highly specialised pocket-sized chess co m puter would have been at all possible without the stimulus of imaginative creations such as the justly famous ‘T u r k ’ of Baron von Kempelen which first displayed its chess prowess at the court of Maria T heresa in 1780. The true secret of von K em p elen’s chess machine remained hidden from the public for over 50 years! Only a few shrewd observers suspected at the time that in reality a chess master was uncomfortably hidden behind the clockwork mechanism in the cabinet, whereas the majority, including even the master of macabre mystery, Edgar Allan Poe, were all too easily fooled by the ingenious contraption. Admittedly, von K e m p e le n ’s court adviser and, later, Johann Malzel devised a cunning m ethod of conveying to the spectator from the very beginning of the performance the illusion of a robot in perfect working order. The Asiatic could even roll his eyes, turn his head and utter the French for ‘check', whilst his left hand carried out the move by gripping a piece and releasing it on to the desired square. T he whole dem onstration was perform ed in a sort of ritual (see Figure 1.1). The crafty inventor would open door A, revealing a baffling array of wheels, levers, rods, hinges and weights. Above this lay a cylindrical drum similar to the ones used in musical boxes of the period. In o rd e r to ‘p ro v e ’ to the audience that no human being was in fact hidden in the cabinet, the d em onstrator would shine a candle through the back door B, illuminating that part of the cabinet so that the flame was visible from the front. During this operation, the player, huddled up inside, had to bend forwards. T o give the hidden player enough time to return to his normal position, the equipm ent needed by the challenger, such as the board and pieces, was taken out of a draw er G. Only after the completion of this intermediate step were the middle and side doors presumably allowed to be o p e n e d , when they in turn were

H ow intelligent are chess com puters? / 9

FIGURE 1.1

10 / C o m p u te r chess illuminated by the candle through the d o o r D. In this half-opened state, the cabinet was then wheeled through the rows of spectators, thus allowing everyone to glance inside the ‘T u r k ’, without being able to discover the real hum an agent. T h o usands of spectators, am ong whom were many prominent figures such as N apoleon B onap arte, flocked to see the skills of the T u r k ’ between 1780 and 1838 and went away duly impressed. In fact, only a handful of games were lost during the m achine’s E u ro p e a n and A m erican tours. We now know beyond doubt that the fingers of the ‘T u r k ’ were op erated by a lever mechanism manipulated by an experienced chess player. In particular, Johann Allgaier, William Lewis and Francois M ouret, renowned masters of the time, are supposed to have controlled the ‘r o b o t’ for an appropriate fee. O ne amusing example of the quirks of fate (or greed?) suffered by one Jo h ann Malzel, who at the time owned one-third of the contraption, is revealed in the following anecdote: one day when His Excellency the king of Holland had made a private reservation of the ‘T u r k ’s’ services for a sum of 3,000 francs, Malzel was horrified to find his current manipulator, M ouret, confined to bed with some sort of fever! It was not until a payment of 1,500 francs had been made that Monsieur Mouret finally consented to grace the royal occasion with his hidden presence. Two o th e r chess machines, ‘A j e e b ’ and ‘M ephisto’, continued to fool the public in both E u ro p e and America at the end of the nineteenth century, and two respected chess masters Pillsbury and G unsberg were at that time the manipulators of these machines which were constructed on the same lines as the ‘T u r k ’. Even during this past century the idea of a chess-playing robot has lost none of its fascination. Innumerable instances can be found in science fiction literature to support this view. O n e such example was provided by a French short story entitled ‘Check to the r o b o t ’, in which the leading role was assigned to an autom aton which, to ev ery o n e ’s am azem ent, played chess in superb fashion. However, the very quality of this chess style gave rise to suspicion, and it was eventually proved that the machine was in fact controlled by a brain which had been surgically removed from the skull of a leading chess player who had been brutally m urdered in his prime!

How intelligent are chess com puters? / 1 1 The first real chess machine It was the Spanish genius L eo n ard o Torres у Q uevedo who built the first real chess playing machine in 1890. Without any human intervention, this electromechanical autom aton could mate with rook and king against rook within 63 moves! Even though the F I D E rules do not allow a win in this num ber of moves with the given material, it is worth examining a little more closely the imaginatively conceived mating geometry (Figure 1.2). Play had to take place within certain grooves so that the moves could be registered by the machine, and for this purpose the board was divided into two rook zones and a central zone. Moreover, for the solution to work, the starting position had to be the one indicated. Beginning with the zone in which the black king is situated, the machine proceeds along the following lines: •9

(1) If the black king is in the same zone as the rook, the latter is moved away. (2) If the vertical distance between the rook and the black king is more than one rank, the latter’s space is further restricted. (3) If the distance between both kings is not small enough for direct opposition to be achieved immediately, the white king is brought nearer. Consider the following mating sequence in the light of the above com m ents (black king on a3, white king on a l and rook on b2) and note that the bracketed numbers correspond to one of the six p ro g ram m ed steps. 1 2 3 4 5 6 7 8 9 10 11 12

Rd2 Re2 Kbl Rh2 Rg2 K cl Kdl Ra2 Rb2 Kel Kfl Kgl

(5) (5) (6) (1) (5) (6) (6) (1) (5) (6) (6) (6)

Kb3 Kc3 Kd3 Kc3 Kd3 Ke3 Kf3 Ke3 Kf3 Kg3 Kh3 Kg3

13 Rb3+ (4) Kg4 14 Kg2 (3) Kf4 15 Ra3 (5) Ke4 16 Kf2 (6) Kd4 17 Ke2 (6) Kc4 18 Rh3 (1) Kd4 19 Rg3 (5) Kc4 20 Kd2 (6) Kb4 21 Kc2 (6) Ka4 22 Kb2 (6) Kb4 (4) Kb5 23 Rg4+ with mate on move 63

12 / Com puter chess rook zone

rook zone

8

7

6 5 4

3

The wo starting positior for mat with roi

2

1 a

b

c

d

e

f

g

h

BK = black king WR = white rook WK = white king

FIGURE 1.2

The model sequence of moves 14 to 23 is repeated four times more (= 5 x 10), and if we add the first 13 moves we arrive at a grand total of 63 moves. (To any good chess player who appreciates the value of the ‘waiting move’ it may seem surprising that Quevedo did not restrict all the action to the left-hand zone exclusively, with a pattern such as 1 Rc2 Kb3 2 R c 1 Ka3 3 Rc3+ Ka4 4 Ka2 Kb4 5 Rcl Ka4 6 Rc4+ with mate in 18 moves = 6 x 3 , although of course the quickest mate is by 1 R b 1 Ka4 2 Ka2 Ka5 3 КаЗ Ka6 4 Ka4 Ka7 5 Ка5 Ка8 6 Кb6! Кb8 7 Rcl Ка8 8 Rc8 mate, where the danger of stalemate appears! Translators.] As can be seen, the machine uses only five criteria: defence, approach, tempo, opposition and check(mate), but this simple question and answer end-game strategy is not so easily

How intelligent are chess com puters? / 13 applied to the complicated king, knight and bishop against king ending. Nevertheless, at this distance in time, we must admire the ingenuity of the attem pt. M oreover, the actual machine can still be found in working condition at the Polytechnic University in Madrid if one enquires of Senor D on Miguel Toros у Gallino. On the principle of the early g ra m o p h o n e , the m achine’s loud speaker announces ‘ch eck’ and ‘m a t e ’!

Who thinks and learns faster, the human being or the computer? For almost 170 years after the ‘T u r k ’s’ world premiere, the task of coping with the e n o rm o u s am ount of variations arising from this sophisticated board game seemed to be the exclusive preserve of hum an intelligence. Before analysing this concept more closely, we shall q u o te a few statistics. Assuming that we have the choice of 20 available moves from the starting position of a gam e, then after the first 10 moves played by both sides we arrive at the inconceivable total of 169, 518, 829, 100, 544 all to the power of 15, of possible positions. The English journalist Paul Mason came up with this result which, measured in seconds, is longer than the life span of our planet, formed in the cosmic gases about 4,600 million years ago. The hum an brain, or rather the cerebral cortex, with its enormously complicated network of about 14,000 million cell combinations, was therefore viewed as unique in its ability to work out correct move sequences from this astronomical num ber of possibilities. The handling and storage capacities of modern largescale com puters have multiplied more than 200 times compared with the early valve or relay monsters of the 1950s. Why is it then that even the fastest am ong them are still in no position to compete successfully with the brain of a grand master? Some researchers have tried, with the help of a cybernetic model, to m ake comparisons between the memory capacity of the hum an brain and that of a com puter. If we solely consider speed and the capacity to acquire information, it has to be conceded that the machine is far superior to the human being. After all, what is the pitiful speed of 100 metres per second, at which a signal is conveyed along h u m a n nerve fibres, com pared with the impressive 3,000 m etres covered by an electronic signal in the same time?

14 / C o m p u ter chess Scientists have calculated that the human nervous system can acquire only about 30 ‘bits’ per second (‘bit’ = binary digit - the smallest unit of communication), whereas the best com puters are capable of processing as many as 50 million ‘bits’ per second!

dialogue calculator central memory

FIGURE 1.3

H ow intelligent are chess com puters? / 15 not carried out; only tested

2nd step

1st step

movable panels

artificial mouse

FIGURE 1.4 The electronic mouse finding its piece of bacon. The whole learning process can be broken down into six steps: 3 forward, 1 to left, 1 to right, 1 backwards. If it meets an obstacle, it takes 1 step back and repeats the sequence until the correct way is found (from Rolf Lohberg and Theo Lutz, Was denkt sich ein

Elektronengehirn? [How Does an Electronic Brain Think?], Heyne, Munich, 1963, page 132)

How ever, examples such as these fail to take into account vital qualitative differences. The blueprint of the human brain, with its proliferation of neurons and nerve fibres, is so intricate that no fully efficient model has yet been able to compete with it on a purely qualitative level. In comparison, the blueprint of a typical c o m p u te r is very much determ ined by the separate functional areas of its electronic brain, involving control, calculation and storage (see Figure 1.3). The electronic memory stores data systematically, which m eans that a desired piece o f information can as a rule only be found under a specific heading, clearly a limiting factor.

16 / C o m p u te r chess A lthough the neurons of the central nervous system react in similar fashion to the logical elements of a com puter, hum an nerve fibres are interconnected in a far more intricate way. It almost seems as if the physiological composition of these allows them to adopt at will the above-m entioned functions of control, calculation and storage. Despite all this, however, if we assumed that a fully functioning brain were the sole measure of intelligence within the context of thinking and learning, then chess program s would have long ago achieved superiority over human beings. In particular, impressively convincing examples provided by cyberneticians have demonstrated the capacity of electronic brains to learn much more quickly and accurately than we can ever hope to do. For instance, observe the ‘skill’ with which an electronic mouse finds its piece of ‘bacon’ on its second trip through the corridors of a maze, and you can hardly doubt the potential o f a c o m p u te r ’s learning capacity (see Figure 1.4). Its biological counterpart, a fairly experienced animal guided by its nose (see Figure 1.5), has to keep repeating the experiment until it has finally learnt to be right first time, a method known to psychologists as conditioning. Snack

FIGURE 1.5 The electronic m ouse’s living counterpart uses its nose (from Lohberg and Lutz, op. cit., page 130)

H ow intelligent are chess com puters? / 17 T h e c o m p u ter simulates a hotel m anager receiving a booking for a room. Dialogue

Explanations

(H = hum an; M = machine) M H M H M

H M H M H

M H M H M

H M H M

H M

Hotel Alsterblick. Good evening. Good evening, Briegel speaking from Bonn.

Recognition of name

W hat can I do for you, Mr Briegel? I would like a double room with bath for 3 nights. W ord order Certainly. I'll just check. Double room w ith bath Conversion until 15 November, that's right? Assumption: from today Yes, please. We can do that. Anything else? Taking the initiative Is there a radio or TV in the room? Double question There's no TV available but the room has a radio Double answer Syntactically Is there a phone? incomplete question. Yes. Doubt expressed Is a desk available? Confirmation Yes, there is one. How many single beds are there in the room? Syntactically Two. incomplete answer Assumption I hope one is fairly hard. Vague answer Yes, one is reasonably hard. Metalinguistic W hat do you mean by that? dialogue It has a new mattress. Shall I book the room for Logical answer based on you, Mr Briegel? knowledge Yes, please. Right then: a double room for Mr Briegel from Summary of facts 12 to 15 November. Many thanks. Goodbye.

F IG U R E 1.6 (from Jorg Siegmann, ‘Kunstliche Intelligenz 25 Jahre

verspatet’ [‘Artificial intelligence 25 years late’], Computer Woche, 19/1981, page 58)

Chess masters think differently C o m p u t e r s can then com pete successfully with humans in most straightforward learning exercises. When applied to the easier board games such as backgam m on or draughts, computer

18 / C o m p u ter chess programs with a learning element showed themselves to be at least on a par with the best players. Chess, however, makes higher demands on a player’s intellectual capacity. O f course, a good mem ory is also an essential require­ ment for this game. It was, for instance, astounding to witness the confidence with which form er World C ham p io n, Robert Fischer, could faithfully reproduce games he had played long before. Yet machines possess excellent m emories as well. Consider, for example, the opening repertoire of the top chess programs which can store up to 100,000 positions. However, leading masters also stand out because of their ability to think logically and associatively. In scientific language this is term ed the ability to classify, generalise, recognise patterns and make inductions or deductions. Computers also can be programmed with these high-level thought processes. From the following dialogue between the hum an being (BEN -user) and model (H A M RPM ), the electronic interlocutor made in H am burg, we can see what enorm ous progress has been made in the processing of natural language (see Figure 1.6). The co m puter simulates a hotel manager receiving a booking for a room. According to Dr Jorg Siegmann’s observation, the main emphasis of such research has shifted from language comprehension alone to the structural formation of speech.

Imagination is more important than knowledge In order to explain why chess masters defend our hum an skills so tenaciously against the onslaught o f c o m p u ter giants, we must rem em ber some fundamental features of human intelligence. ‘Imagination is more important than knowledge' was an opinion often reiterated by Albert Einstein. At that time no machine was capable of working in a totally creative way. It is clear that the faculties of intuition and ju dgm ent (i.e. the ability to distinguish the particular from the general) play a huge part in the game of chess. By intuition, following Lasker's definition, we refer to the chess master's ability to create new concepts and m ethods from old ones by using more than just logical deduction. It is purely these latter qualities which distinguish hum an intelligence from that of the machine, since such imaginative qualities can hardly form part

H ow intelligent are chess com puters? / 19 of a m achine’s logic patterns. Figure 1.7 illustrates the memory, learning and thought processes of the human being and the com puter. Martin Kulp had similar trains of thought: ‘A problem becomes extremely difficult, when the existing rules are not sufficient to solve it without revision . . . . H ere, there is the great danger that the new definition conflicts with some already established fact, so that even the freedom to contradict has its restrictions’ (Human and Mechanical Thought , 1968, pp. 226-7). Consider the highly PROPERTIES

Person

A. m em ory (quantitative) intake capacity 0 30 bit/sec (conscious) speed reaction tim e w ord count retrieval m em ory (qualitative)

kind

B. thinking, learning kind

learning performance

qualitative

100 m/sec seconds 10 billion incomplete, patchy short-term m em ory long-term m em ory associative according to place, time, etc. forgetful

trial and error successful subject to error w ro ng ly remembered varying according to situation (fatigue, 'w arm ing up', etc.) intuitive, associative, subjective

(Chess) computer

up to 50 million bit/sec and more 300,000 m/sec nanoseconds and more complete, flawless C.P.U. storage (nano­ seconds) external storage strictly hierarchical according to listing flawless

programmable free of error never forgotten stable

objective, simulates human thought pro­ cesses, programmable.

FIGURE 1.7 Table of comparisons between human and artificial thinking processes. At first sight, man appears to be inferior to the machine on most counts. However, this does not mean that quantitative properties such as speed and intake capacity are necessarily beneficial to chess playing. Even apparent human defects such as forgetfulness and insufficiency may prove advantageous by preventing overloading of the brain, thus allowing economy in chess thinking.

20 / C o m p u ter chess imaginative style of ex-World C ham pion Tal who succeeded in shattering many accepted positional concepts with the dynamic power of his tactical skill. Martin Kulp continues, ‘This includes the intuitively daring leap into the unknow n, without any support from logic. The likelihood of this distinctly human quality being reproduced by an efficient, yet in this connection blind, autom aton is so small that in practice it would never h a p p e n ’ (p. 227). T h e international grand master and chess theoretician, Ludek Pachman, in his discourse in the FID E Revue (1968), gave similar reasons for the failure to produce a perfect chess playing machine: In the most advanced stages of the pursuit of chess truths, we step beyond the field of logic and e n ter a sphere in which not even the most highly perfected machines can operate and which therefore will remain for a long time the exclusive preserve of the hum an brain. Indeed, the main value of chess is that it allows the expression of this special capacity of the hum an brain to discover for itself inconsistencies in the nature of p h en o m e n a, to explore these in the most complex and dynamic situations, to combine cool mathematical prediction with brilliant artistic intuition, and constantly to seek out and find new paths which deviate from previous experiences. (FI DE Revue) Although Pachman himself now admits that he has had to revise some of his early reactions to this subject, his above com m ents retain their validity for the time being. W hen he first wrote the article, his thesis was that even a machine equipped with the most complete program could at best attain the standard o f the average player. Since then, however, chess com puters have come into existence which b order on master level. In contrast, the father of chess programming, Konrad Zuse, in 1972 cautiously ventured the opinion that we should not deny the machine specific possibilities in the processing of information: These matters are still in a state of flux . . . . I t must not be forgotten that the qualitative superiority attributed to the human brain largely comes down to a quantitative one. Admittedly, we are not yet in a position to group billions of data processing elements into a small area, but such quantitative aspects are simply a question of improved technology.

H ow intelligent are chess com puters? / 21 Chess computers imitate human intelligence Whilst cautious c o m p u ter experts and cyberneticians do not claim that com puters can actually think or have hum an intelligence at their disposal, we can for the m om ent agree with them that com puters simulate hum an intelligence. M oreover, it is a fact that this imitated form of intelligence is sufficient to play chess. Chess playing au to m a to n s are nothing more than computers which have been ‘ta u g h t’ by a hum an being which possible moves to make in o rd e r to achieve the aim of mating the o p p o n e n t’s king. The most d em an d in g technical part of the work is then taken over by the calculating ap p aratu s without any human assistance. An imaginatively played game will be the exception rather than the rule with such a machine, as it will usually proceed rationally and not creatively. A chess program can therefore only be viewed as partially intelligent, since it is totally unsuited to performing o th e r intellectual tasks at the same time, such as solving mathematical equations. According to the American scientist D. Michie, hum an reason is distinguished by its versatility and ability to o p erate simultaneously in many different fields of activity. This is not true of computers. H ow ever, we would not care to state definitively that a c o m p u ter program will never be able to play creative chess. The Soviet academic A. N. Kolmogorov half jokingly voiced the opinion that ‘to teach a c o m p u ter the art of poetry you would have to feed into it the whole history of m a n kin d ’. Judge for yourself w hether the following lines reveal the presence of a creative mind: All snow is cold. Every angel is white. Not all snow is still A n d no peace is cold. No angel is bright A nd all peace is still . . . . D oes this represent a new trend in poetry? By no means! The lines originate from a Zuse Z 22 computer! On closer examination we find that the ‘p o e m ' actually consists of the three nouns ‘snow', ‘angel’, ‘p e a c e ’, the descriptive adjectives ‘cold’, ‘white’, ‘still’, ‘bright’ and the delimiting adjectives ‘all’, ‘every’, ‘n o ’ simply linked at random by using the verb ‘to b e ’. We should not however disregard the fact that this was an attem pt made several years ago.

22 / C o m p u ter chess The first chess program in action On the subject of hum an and mechanical intelligence, we could pose a n u m b er of vitally important questions such as the one about creative play we have just m entioned, or w hether a co m puter will ever manage to beat a grand master. However, in view of the scope and complexity o f such questions about which very little research has been done, we shall not attem pt to answer them until later in the book but instead offer the reader two practical examples of com puter chess which illustrate the enorm ous advance made by chess programs over the relatively short space of time of 30 years. The first game was played in 1951, or possibly a year later, the author of the program being Allan Mathieson Turing, a brilliant mathematician and co m p uter expert who is today considered one of the founders of modern co m puter chess. Some time before the American Claudwood Shannon published his first principles of chess programming, Turing had already begun to develop his own chess program. Incidentally, there turned out to be many interesting similarities between the problems posed by chess playing and those involved in the decoding of the G e rm a n radar defence system during the Second World War. When Turing drew up his first chess program , there was unfortunately no suitable co m puter available to carry it out, so its creator decided to calculate and execute the moves himself, following the program m ed instructions. His oppo n en t Allick Glennie, an experienced co m puter expert but only an average chess player, won the game in 29 moves against the program christened ‘T u r o c h a m p ’. The game lasted almost three hours, since T uring had to calculate all the moves of the program by using a chart, there being no suitable calculator on the market at the time. As we can gather from the actual game (Figure 1.8), T u r o c h a m p ’ and his human o p p o n e n t remained on level par for a long time. The decisive erro r on move 29 resulting in the loss of the queen (Figure 1.9) was attributed to the ‘shortsightedness’ of the program, because ‘T u r o c h a m p ’ could see no more than one move ahead! The best program s of today, in comparison, can see on average three to four moves ahead (six to eight half-moves), whilst in forced variations the world's master program ‘Belle’ can

How intelligent are chess com puters? / 23

1 e4 2 Sc3 3 d4 4 Nf3 5 Bd2 6 d5 7 h4 8 a4 9 gxf3 10 Bb5+ 11 dxc6 12 Cxb7 13 Ba6 14 Qe2 15 Tg1

White

Black

TUROCHAMP

ALLICK GLENNIE

(2.6) (2.0) (3.5) (0.2) (1.1!) ( 1 .0 !) (2.4)

(-1.5!) (0.6) (1.2!)

e5 Sf6 Bb4 d6(?) Nc6 Nd4 Bg4 Nxf3+ Bh5 c6 0-0 Rb8 Qa5 Nd7 Nc5

16 17 18 19 20 21 22 23 24 25 26 27 28 29

Rg5 Bb5 0-0-0 Bc6 Bd5 ВхсЗ Kd2 Rg4 Qd3 Bb3 Bc4 Rg3 Bxb5 Qxd6

(0.4) (3.7!)

(0.3)

Bg6 Nxb7 Nc5 Rfc8 ВхсЗ Qa4 Ne6 Nd4 Nb5 Qa6 Bh5 Qa4 Qxb5 Rd8

and the human player won. The figure in brackets gives the computer's evaluation of each move.

F IG U R E 1.8 T U R O C H A M P v. Allick Glennie

F IG U R E 1.9

The position after W hite’s 29 О x d6

calculate as many as 15 half moves. In the opening phase, the amateurish moves 7 h4 and 8 a4 are immediately noticeable, and opening or middle-game strategy is conspicuous by its absence, the program merrily developing the pieces without following any plan.

24 / C o m p u ter chess However, it has u n d o u b ted com m and of the rules of the game and just about knows the relative values of the individual pieces (Figure 1.10). It would be interesting to know how far ‘T u r o c h a m p ’s’ playing strength actually surpassed that of its creator. Turing probably knew little of the most important principles of chess, with the result that his program would certainly have failed to meet his own intelligence criteria which were quoted earlier. King Queen Rook Bishop Knight Pawn

1000 10 5 3.5 3 1

points points points points points point

FIGURE 1.10 T U R O C H A M P ’S piece evaluation

Which is the game played by the computer? Some 30 years later there was already a whole range of chess programs being put forward as fulfilling the requirements laid down by Turing, and in the spring of 1980 the producers of the N R D television program m e ‘Culture and science’ decided to stage an interesting test. A. Folsing and F. Friedel, again in the spirit of Turing, put to chess co m puter experts the question, ‘Can one easily recognise a game played by a c o m p u te r? ’ In a television programme entitled ‘C o m p u ter on the way to becoming a grand m aster’, filmed in Linz during the Third World Chess C o m p u ter Championship, the following experiment was carried out: D r Helmut Pfleger, a grand master from Munich, was to give a simultaneous exhibition against 26 players, with three of them, unknown to the grand master, simply following the moves supplied by three computers. Each of these players was equipped with a radio receiver through which the suggested moves of the com puter were secretly transmitted. The grand m aster’s moves were observed through binoculars from a glass cabin and immedi­ ately telephoned to the respective computer. T he experts in charge of the com puter programs were entitled to offer draws or declare a game lost, as appropriate. O f course, in order not to attract too

How intelligent are chess com puters? / 25 much attention an d to keep to the rules of simultaneous play which de m an d a prom p t reply as soon as the grand master arrives at the b o ard , drastic cuts in calculating time were often necessary, resulting in a clear reduction of the c o m p u te r’s playing strength.

Game 1 1 d4 99 i! 99 !i 99 гю о Г101 ;ю г Ti03 99 | 99 j 99 | 99 90

8

7

99

■ ip A

80

-

з

Л

95

m

- 6

"i«?7 ii 99 ;107 1 99 l

TTl8 119 S i 99 99 ;108 109 i 99 99 l 99"

fcH

ш

-2

Ш л Ш " - Bxf6 3 h %g5 v/ith a queen mate on h7 or (if 3 , . . RtH) on f7 1____ R e 8 2 N x f 6 - B / f 6 3 Bxf7+ K f 8 4 Bb3 Ke7

(4 . , . Qe75 Nh7m ate)5 O f7- Kd66 Ne4 mate. H ow ever

after '

( ) ' V- ( .

K n < ’ л r..'e car or.. . v. n \v; excr.ar.i-e ' • /

w

У

\ v r~ Rxf~ etc Vi r. logical terms rr.e rno’.e 1 Qg" beautiful as it mas be. js no better ’ nan the computer г choice! We can dra* tv/o important conclusions from the above examples first!) >/ell programmed computers have a Jread) shov.n that tne'. are ni^Jil- efficient /.hen it comes to caJcuterinz precise uv.tical sequences А г >