Você está na página 1de 11

Search Heuristics of Chess Players of Different Calibers Author(s): Robert I.

Reynolds Reviewed work(s): Source: The American Journal of Psychology, Vol. 95, No. 3 (Autumn, 1982), pp. 383-392 Published by: University of Illinois Press Stable URL: http://www.jstor.org/stable/1422131 . Accessed: 10/02/2013 15:21
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

University of Illinois Press is collaborating with JSTOR to digitize, preserve and extend access to The American Journal of Psychology.

http://www.jstor.org

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

AMERICAN JOURNAL OF PSYCHOLOGY Fall 1982, Vol. 95, No. 3, pp. 383-392

Searchheuristicsof chess players of differentcalibers


ROBERT I. REYNOLDS Nasson College Information-processingmodels of chess-playing ability have distinguished between players of different calibers solely in terms of perceptual encoding and recognition of chess configurations. A reanalysis of deGroot's verbal protocols of 1965 indicates that players of different calibers direct their attention toward different areas of the board. Grandmastersand masters consider squares that are affected by many pieces, while lesser players direct their attention toward squareson which the pieces are located. In a recallexperiment, chess positions were presented that had been randomly generated so as to differ only in the degree to which the pieces converge on the same squares. Masters showed superior recall for briefly presented positions only when the material affects a highly centralized area of the board. The game of chess has been repeatedly selected as an ideal task for the investigation of problem-solving behavior. Chess has the complexity and variety of everyday problem solving, while the restricted domain of the chess board provides valuable experimental and descriptive controls, i.e. only one move can be made at a time, to only 1 of a possible 64 squares. In no other nonlaboratory activity is there such a precise measure of performance as that of the chess rating system. The progression from beginner to tournament chess player (0-1999 points), through expert (2,000-2,199), master (2,200-2,399), senior master (2,400 and above), and grandmaster (approximately 2,5002,700) levels, requires a minimum of 10 years of intensive study and practice. Chess is therefore a highly desirable task for investigating problem-solving strategies of problem solvers with different degrees of expertise. During the past 15 years there have been a number of studies investigating chess players' information-processing ability. These studies have been based on the pioneering research of the Dutch psyO 1982 Board of Trustees of the University of Illinois 0002-9556/82/0300-0383 $1.00/0

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

384

REYNOLDS

chologist Adrian deGroot (1965), who posed the question, What distinguishes the accomplished grandmaster from the average chess player? He presented novel middle-game positions to chess players of different calibers and elicited verbal protocols during their search for the best move. DeGroot concluded that he could find no quantitative difference between the thought process of the world's best players and the average chess player. The only difference deGroot discovered between strong and weak chess players was that the former have a quicker grasp of the position. "Within the very first five to ten seconds, the master subject is apt to have more relevant information about the position available to him than the lesser player can accumulate in, say, a quarter of an hour of analysis" (deGroot 1965, p. 324). In Chase and Simon's (1973a) theory, "the most important processes underlying chess mastery are those immediate visual-perceptual processes rather than the subsequent logical-deductive thinking processes" (p. 215). The speed with which a novel configuration is grasped is reflected in the stronger player's superior recall ability when chess positions are presented for brief durations (deGroot 1965, 1966; Jongman 1968). Chase and Simon (1973b) found that master chess players have superior recall for real chess positions, but not for positions with randomly generated configurations of pieces. They suggest that masters have superior recall and chess-playing ability because they can recognize a greater number of patterns of related pieces. They characterize these patterns as "local clusters of pieces of the same color that usually defend each other. These structures are built around visual features, such as color and spatial proximity" (p. 236). Simon and Gilmartin (1973) estimate that a master chess player would require storage of some 50,000 chess configurations. According to Chase and Simon (1973a), the chess master's superior playing ability arises as this vast number of patterns becomes associated with plausible moves, and stored in long-term memory. Despite Chase and Simon's success in inspiring research, their theory has not been strongly supported by other memory and reconstruction studies, nor does it account for data that directly analyzes chess players' search for the best move. If, as Simon and his collaborators suggest, masters are superior because of their experience with and storage of a large number of chess configurations, then we should expect them to be better able to reconstruct the configurations of highly typical positions. DeGroot (1966) has generated a position based on the highest frequency of piece distribution found in published games. He calls it "the stereotypic position par excellence."

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

SEARCH

HEURISTICS

385

/Z

:E
7+7 4f
7/7,

4/4 mill ANN


7o

Figure 1. DeGroot's stereotypic position This position (see Figure 1) was used in the memory experiment of Holding and Reynolds (1982) and yielded approximately the same recall scores for players ranging from class C to expert. Goldin (1978) found that positions taken from real games and judged to be "highly typical" were better recalled and recognized than "atypical" positions. However, chess players ranked "high"(between class A and master) were not significantly different (p < .06) in recall performance from those ranked "low"(classes B, C, and D). Finally, deGroot (1966) had subjects make blind guesses as to the distribution of material in highly typical positions selected from published games. He found no difference between masters and "weak" players in guessing the placement of chess positions in such highly typical positions. Tikhomirov and Poznyanskaya (1966) monitored eye movements in order to determine the area of the chess board actually considered during search. "One of the basic characteristics in the selection of a move is that the subject considers only a part of formally possible continuations. Out of all the elements a zone of orienting emerges, expressed by the total number of elements of the situation on which the eye concentrates. These elements, in turn, do not have equal value in the frequency and duration of fixation" (Tikhomirov & Poznyanskaya, 1966, p. 7). They observed that the "zone of orienting" did not necessarily involve squares connected with important pieces or objectively powerful moves. The protocol of eye fixations from an expert subject indicates that of the squares attended to most frequently, six consisted of his own pieces, four of his opponent's, and nine empty squares. During the 106 sec used to make a move, the square e5 was fixated 23 times. Though this square is unoccupied and unaffected by any piece, it was fixated more often than 40 other squares fixated during the session. Simon and Barenfeld (1969) developed a computer

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

386

REYNOLDS

simulation of eye movements during chess perception. The basic consistent with the subsequent theory of assumption of PERCEIVER, Chase and Simon (1973a), was that "the information being gathered during the perceptual phase is information about pieces" (p. 475). "fixations" of the Tikhomirov and Consequently, all of PERCEIVER's fell on squares occupied by pieces, when, in Poznyanskaya position concerned with spares and primarily a was human the fact, subject affected even not by a piece. immediately square I believe that deGroot's analysis and Simon's extrapolations have both failed to notice that chess players of different calibers concentrate their attention on different areas or "zones of orienting." Such differences may, in fact, be found in deGroot's protocols and will be analyzed in the following section. What we will find is that the stronger players do not attend so much to piece configurations as to critical distributions of affected squares.

REANALYSIS OF DEGROOT'S PROTOCOLS


The following analysis has been performed on one of deGroot's (1965) positions (Figure 2), which was taken from an actual game. This position provides the largest sample of protocols: 5 grandmasters, 4 masters, 5 experts, 2 chess players. We may plot two objective distributions on any chess board: one consists of the squares on which the pieces are situated (P distribution), the other consists of the squares affected by the pieces (A distribution). The geometric midpoint of the P distribution is the point midway between all of the pieces. The midpoint of the A distribution is derived by assigning to each square the value equivalent to the number of times that the square is affected and then calculating the (weighted) center of affected squares. A similar analysis can be performed on the protocols in order to determine the subjects' center of attention or "zone of orienting." This is done by calculating the midpoint of squares mentioned in response to deGroot's position (Figure 2). Table 1 presents the midpoints in terms of x-y coordinates for grandmasters (G), masters (M), experts (E), and class players (C), as well as the center of pieces (P), and affected squares (A) in Figure 2. Table 2 presents the distance between the midpoints of the objective distributions and the players' area of attention. On the average, both the grandmasters' and masters' center of attention is closer to the center of affected squares than it is to the piece center. The experts' center

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

SEARCH

HEURISTICS

387

4HA~

~1v/
~~Dj

1e~i~~

Figure 2. Position used in deGroot's protocols of attention is mid-way between the piece and affected centers, while the class players' center of attention is closer to the piece center. The weaker players appear to be captivated by the existing placement of pieces, while the stronger players concentrate their attention about the functional distribution of affected squares. Table 1. Weighted midpoint of squares mentioned along X-Y coordinates Sx Y X Sy Protocol Grandmasters Masters Experts Class players Center Affected space (A) Piece (P) 3.70 3.87 4.41 5.08 4.29 4.61 .34 .50 1.12 .20 4.06 3.78 4.40 4.265 3.99 4.68 .58 .30 .54 .02

Table 2. Distance between midpoints on deGroot's position (Figure 2) for different protocols Protocol Grandmaster Master Expert Class player
Note. 1.00 = 1 square

Affected squares .59 .47 .42 .84

Piece 1.10 1.17 .34 .63

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

388

REYNOLDS

A RECALL EXPERIMENT
It is notable that in deGroot's (1965) study, the grandmasters, masters, and experts most often mention first a square near to the center of their particular zone of orienting. We might expect that when a position is presented for only a few seconds, players will concentrate on those squares most often attended to in free-viewing situations. It follows that stronger players will have superior recall for the location of briefly presented chess positions when those pieces are associated with and concentrated upon a small area of affected squares. The random assignment of pieces to the chess board in the recall experient of Chase and Simon (1973b) diminished not only the number of recognizable configurations, but also the central clustering of affected squares. Similarly, deGroot's (1966) guessing experiment precluded subjects from grasping the centralization of the hidden positions. In that experiment, masters still did not outperform lesser players, even though configurations familiar to them were used.

METHOD Stimuli
In the following experiment, players of different calibers were presented chess positions generated by randomly assigning pieces according to the degree of affect grouping. The conditional randomization described below was designed to control the distribution of affected squares, while equating and reducing familiarity of piece configurations. A first-order grouping is generated by first listing all of those squares on which the pieces may affect (i.e., attack or defend) the three inner perimeters. The pieces are then randomly assigned to one of those 34 squares, with the condition that neither king is in check. A second-ordergrouping is generated by first listing all of the squares on which a piece may affect one of the 16 squares in the two inner perimeters, followed by random assignment to one of those squares. A third-ordergrouping is generated by randomly assigning pieces so that each piece affects one of the four central squares. Three positions of each grouping were so generated. An expert chess player (United States chess finalist = 2100) selected one position from each grouping so as to equate, as much as possible, the familiarityof configurationsof pieces. Figure 3 represents these three positions.

Subjects and procedure


The subjects in this study were three masters, three class A players, and three class C players(mean ratings: 2,235, 1,861, 1,533, respectively). Each subjectfirst viewed and reconstructeda practiceposition- deGroot'sstereo-

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

SEARCH

HEURISTICS

389

Position

MAN
~Fmmm/ ,

Position

Ig

-RA

x@

1R .II*

Pos i t ion C

/1 N

/_Ni

Figure 3 (A, B, C). Positions used in recall experiment

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

390

REYNOLDS

typic chess position (Figure 1). They then viewed each of the three test positions. The order of presentation of positions was randomized for each subject. Subjects were presented a position for 8 sec (comparable to the experiments of deGroot and of Simon), after which time the pieces were abruptly removed from the board. Their task was to reconstructthe position, trying "new positions until you feel satisfied with the reconstruction, or until you can no longer remember the position."Approximately 2 min. separatedthe presentation of successive positions.

RESULTS
Table 3 presents the number of pieces placed correctly in relation to the degree of affect grouping and caliber of player. The masters show a graded increase of 82% in recall performance with the increase from the first-order to the third-order affect grouping. The class C and class A players, in contrast, actually show poorer recall for the higher levels of affect grouping (though this is not statistically significant). All classes of players, including the masters, have the same recall performance when the affect grouping is low. The higher the affect grouping, the better does the position serve as a discriminator for different calibers of chess players. The third-order grouping yields a statistically significant difference between groups (KruskalWallace: H = 6.44, p < .05). In relation to the earlier analysis of search heuristics, we may say that master chess players show superior recall for positions with distributions concentrated about the area they most often attend to in free-viewing situations, namely the center of affected space.

GENERAL DISCUSSION
The game of chess is a task so complex that an algorithmic search of all possible continuations would require a modern computer's continTable 3. Mean number of pieces correctlyrecalled for positions of increasing affect grouping (see Figure 3) Mean class rating 1533 (C) 1861 (B) 2235 (A) First-order grouping (A) 5.00 5.00 5.67 Second-order grouping (B) 3.67 4.00 6.33 Third-order grouping (C) 3.67 4.33 10.33

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

SEARCH

HEURISTICS

391

uous operation for well over 100 years. Only in positions of extremely reduced material will a human chess player attempt an exhaustive search. In fact, the grandmaster relies on heuristic search even where algorithmic search through the tree of possible moves is psychologically feasible. The reanalysis of deGroot's protocols indicates that master and grandmaster chess players direct their attention to a different area of the board from that of players of lesser expertise. While the beginning tournament player is captivated by configurations of black and white pieces of wood, the masters and grandmasters center their attention on those squares affected by the pieces. In the Tikhomirov and Poznyanskaya (1966) study, the expert primarily fixated upon a square unoccupied and unaffected by any piece. By calculating the piece and affective centers for that position, we find that that square (e5) is the center of all the material on the board (i.e., the P distribution). The midpoint of all his fixations is located between the piece and affective centers, as in the deGroot protocols from experts. Simon and Gilmartin (1973) developed a computer simulation of It incorporates their memory for chess positions which they call MAPP. simulation of eye movements during chess perception (PERCEIVER) and a simple theory of rote verbal learning (EPAM). MAPP's recall for real chess positions was compared with a class A and a master chess player. The program's recall performance closely matched that of the class player, but was 24% lower than that of the master. Class players, unlike masters, encode chess positions according to local clusters of pieces. In recent years, computer programs have become relatively successful in human competition. Their steady increase in performance has been a function of increasing reliance on the computer's large storage capacity. In fact, the most successful programs today use no heuristic search, but calculate all possible first moves to a depth much further than that of the human chess player. The move generators of existing programs are based on the storage of configurations of pieces, much as described by Simon and his collaborators. The above analysis indicates that move generation based on configurations of pieces may inadvertantly lead to the simulation of expert class chess players. The leading chess-playing programs are, in fact, performing at the level of expert. The analysis of search heuristics suggests the feasibility of programming a simulation of different calibers of chess players. The progression to master and grandmaster levels will, I believe, require an adoption of search heuristics directed toward the distribution of affected space.

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

392 Notes
June 23, 1981; revision received December 21, 1981.

REYNOLDS

Requests for offprints may be sent to Robert Reynolds, Department of Psychology, Nasson College, Springvale, ME 04083. Received for publication

References Chase, W. G., & Simon, H. A. The mind'seye in chess. In W. G. Chase New York: Academic Press, processing. (Ed.), Visualinformation
1973. (a) Chase, W. G., & Simon, H. A. Perception in chess. CognitivePsychology, 1973, 4, 55-81. (b) Goldin, S. E. Memory for the ordinary: Typicality effects in chess memory. Human Learningand Memory, 1978, 4, Journal of ExperimentalPsychology.605-616. deGroot, A. D. Thoughtand choicein chess. The Hague: Mouton, 1965. deGroot, A. D. Perception and memory versus thought: Some old ideas and method recent findings. In B. Kleinmuntz (Ed.), Problem solving.-Research and theory.New York: Wiley, 1966. Holding, D., & Reynolds, R. Memory does not determine chess skills. Memory & Cognition, 1982, 10, 237-242. Jongman, R. W. Het Oog van de Meester. (Doctoral dissertation, University of Amsterdam). Assen: Van Gorcum & Company, 1968. Simon, H. A., & Barenfeld, M. Information-processing theory of perceptual processes in problem solving. Psychological Review, 1969, 76, 473-477. Simon, H. A., & Gilmartin, K. A simulation of memory for chess positions. CognitivePsychology, 1973, 5, 29-46. Tikhomirov, O. K., & Poznyanskaya, E. D. An investigation of visual search as a means of analyzing heuristics. Soviet Psychology, 1966, 5, 3-15.

This content downloaded on Sun, 10 Feb 2013 15:21:19 PM All use subject to JSTOR Terms and Conditions

Você também pode gostar