The Psychology of Chess

Fernand Gobet

• Publisher: Routledge
• Year: 2018
• Print ISBN: 9781138216655
• eBook ISBN: 9781315441863
• Edition: 1
• Topic: Business
• Subtopic: Cognitive Psychology & Cognition
• Index: Business

Do you need to be a genius to be good at chess? What does it take to become a Grandmaster? Can computer programmes beat human intuition in gameplay?

The Psychology of Chess is an insightful overview of the roles of intelligence, expertise, and human intuition in playing this complex and ancient game. The book explores the idea of 'practice makes perfect', alongside accounts of why men perform better than women in international rankings, and why chess has become synonymous with extreme intelligence as well as madness.

When artificial intelligence researchers are increasingly studying chess to develop machine learning, The Psychology of Chess shows us how much it has already taught us about the human mind.

1
The Eye of the Master

Anybody who has seen chess masters playing bullet chess (1 minute per side for the entire game) or simultaneous exhibitions, where they play against 30 or 40 opponents at the same time, would have been struck by their amazing ability to play good moves very quickly. Indeed, the quality of moves played under these taxing conditions is surprisingly high, although not as high as with games played under normal conditions (on average, 3 minutes per move). It is as if masters see the board differently than weaker players. Where novices see wooden or plastic pieces, masters see trajectories, ideas, concepts and sequences of moves. In fact, the same applies in other fields: one of the hallmarks of experts in science, medicine and sport is the ability to rapidly perceive the key features of a problem.

The first person to have addressed this question empirically was Adriaan de Groot in his doctoral dissertation, originally published in Dutch in 1946 and translated in English in 1965. Because of the number of issues it addressed and its strong scientific impact, this work has become a classic in psychology.

A Better Understanding After 5 Seconds than After 15 Minutes!

De Groot’s main interest concerned the processes that allow chess players to choose a move. Specifically, he wanted to test the hypothesis that, compared to amateurs, chess masters considered more positions when they were looking ahead and that they anticipated longer sequences of moves – that is, they were searching deeper. In a first experiment, he gave chess players a board position unknown to them and asked them to select what they thought was the best move. He also asked them to say aloud what they were thinking about. The players consisted of amateurs, candidate masters and world-class grandmasters, including world champions. The transcripts of players’ utterances – called verbal protocols – were then analysed in great detail, both qualitatively and quantitatively.

The results did not support his expectations: although grandmasters played better moves, they did not differ substantially from other players with respect to structural variables such as the depth of search, the number of moves considered or the strategies used when carrying out search (see Chapter 3 for details). There was an important difference, however. The best players were able to pinpoint promising solutions very rapidly, which allowed them to narrow down their search drastically. As de Groot put it, the world champion understood the problem position better after 5 seconds than a candidate master after 15 minutes! This was fully unexpected. What was critical was not the detail of the way players analysed the position by trying out different moves, sometimes for more than 30 minutes. Rather, the difference resided in the very first few seconds of seeing a position: perception is central in chess expertise.

In a second experiment, de Groot directly tested this hypothesis. He presented a position briefly, from 2 to 15 seconds, took it away from participants’ view, and asked them to reconstruct it as precisely as they could. As expected, grandmasters did much better than candidate masters, who in turn did better than amateurs. Whilst a grandmaster could reconstruct nearly the entire position correctly, a strong amateur struggled to remember half of the pieces.

De Groot explored several variants of this experiment. In some versions, he asked players to think aloud, either during the presentation of the position, immediately after or 30 seconds after. From the protocols, it is clear that experts did not see individual pieces, but rather saw large complexes, in which perceptual aspects are intertwined with dynamic possibilities. In fact, they rarely perceived static groups of pieces, but almost always threats, probable moves and even sequences of moves. Note that this is the case even when players are told explicitly beforehand that the task is to recall a position, and not to find the best move.

I will have more to say about this experiment in Chapter 2, but for the time being it is important to realise that this task was in de Groot’s mind a perceptual task, the aim of which was to understand what grandmasters saw during the first seconds they looked at an unknown position. Later on, this task became highly popular and in more recent research has been predominantly used for studying memory.

Recording Eye Movements

A natural way to study skill differences in perception is to record eye movements. This technology was not available before the war when de Groot collected the data of his PhD research, and it is only in the 1960s that he was able to carry out such an experiment, with his PhD student Riekent Jongman. The task was again to reproduce a chess position presented briefly, this time uniformly for 5 seconds. The data were fully analysed even later, in a book de Groot wrote with Jongman and myself (see Further Reading). There were clear differences between weak players and masters, with the latter having shorter and less variable fixations. Masters’ fixations also covered more squares and landed more often on the squares that were important from a chess point of view. Another interesting result was that masters fixated more often on the intersection of squares than the weak players. This buttresses the hypothesis that masters perceive groups of pieces rather than individual pieces. Finally, it is likely that fairly simple visual cues – such as a White pawn that has penetrated Black’s defence – direct masters’ eye movements to significant squares. In chess, perceptually salient features correlate very often with the strategic and tactical meanings of a position.

Just like some of the old experiments, the eye-movement experiments used verbal protocols. After reconstructing a position – more or less successfully – players were requested to retrospect on what they had seen during its presentation. In general, masters’ retrospective descriptions broadly agreed with the actual sequence of eye fixations. An interesting exception was that, in the cases where they had fixated the same square several times, players tended to remember only the first fixation. The same finding has been documented in experiments measuring memory for sequences of words, where repeated items tend to be recalled poorly – a phenomenon known as the Ranschburg effect. The protocols were also useful for providing information about where players directed their attention and about the way they dealt with atypical positions (e.g. positions that cannot be put in standard categories).

Later experiments measuring eye movements have also produced striking skill differences. For example, Charness and colleagues used a problem-solving task in which participants (intermediate players and candidate masters) had to find the winning move in a position. In addition to being faster and choosing better moves, the candidate masters had fewer fixations but larger eye movements than the intermediate players. Their first fixations tended to land on empty squares more often, and, when considering fixations on pieces only, they fixated on important pieces more frequently. In general, the data supported the idea that strong players combined perceptual knowledge with the information provided by peripheral vision to direct their eye movements.

Perception: Incremental and Anticipatory

A classic debate in psychology concerns the nature of perception: is it holistic, with objects perceived in their totality, as maintained by Gestalt psychologists, or is it constructed by incremental mechanisms, as argued by reductionists? Although not a Gestalt psychologist himself, de Groot proposed that strong players start with a “landscape view” of the board, which provides a global impression of the position, with the details omitted. In the 1960s, there was an interesting debate about this issue between Soviet psychologist Oleg Tikhomirov and Herbert Simon, with Tikhomirov defending the view that perception is holistic, while Simon argued that local mechanisms (e.g. perception of relations of defence and attack between pieces) were sufficient for explaining the data. More recently, Gobet and Chassy run computer simulations based on the idea of chunks and templates (see the next chapter), showing that experts’ perception, even though it might look holistic, can be accounted for by the incremental construction of an internal representation using patterns that are initially fairly small.

Another idea proposed by de Groot seems better supported by the empirical evidence. He suggested that chess masters used anticipatory schemas. These dynamic schemas contain information allowing players to anticipate potential actions. As experts have more and better developed schemas, they can anticipate actions better. Vincent Ferrari and colleagues at the University of Provence (France) tested this hypothesis. In a first experiment, players saw two positions in quick succession and had to say whether the second position was the same as the first one. The results showed that strong amateurs performed better when the two positions appeared as a normal sequence of moves, unlike beginners who could not use information about the normality of moves.

In a second experiment, Ferrari and colleagues studied whether players tend to recall positions as they were shown or, rather, the positions that would occur after the standard move is played, as predicted by the presence of anticipatory schemas. A recognition task was used. Players saw 10 chess positions displayed in succession; half of the positions were standard opening situations, while the other half were a different set of opening situations, this time with one additional move played. During the recognition phase, players were presented with 10 old positions (the positions they had seen in the first phase of the experiment) and 10 new positions (half were the positions they had seen plus one standard move, and the other half were the positions they had seen minus the standard move). The results showed that the strongest players (class A players) made many false recognitions, where they recalled not the position they had seen, but the position after the normal move had been made. The beauty of this experiment is that better players committed more false recognitions than weaker players, showing that in some circumstances expert perception can lead to errors. In sum, these two experiments back up the hypothesis that experts use anticipatory schemas in their perception: rather than recalling a scene the way they saw it, experts tend to recall it the way it normally unfolds in the near future.

Perception is Cognition

The importance and speed of perception is not limited to chess, but has been documented in many other domains of expertise, such as music, medicine, sports and driving. In all these domains, experts literally see a different problem situation and categorise it in a better way. Rather than being innate, experts’ perception is the product of many years of practice and study. One of de Groot’s major contributions is to have shown that there is no clear boundary between perception and cognition: in chess and in other domains, perception, memory and problem-solving are tightly interconnected.

2
Chunks!

De Groot uncovered some fascinating phenomena, but his work was essentially descriptive and his theoretical explanations were not compelling. The first theory to convincingly account for de Groot’s results was proposed by Herbert Simon and William Chase in 1973 in three classic papers, in what is known as the chunking theory. The theory was primarily aimed at explaining two phenomena: chess experts’ remarkable memory and their ability to find good moves rapidly – as de Groot put it, strong chess players automatically see the good move. Chunking theory’s strength is to have proposed fairly detailed mechanisms to explain these phenomena. The fact that Simon was one of the founders of artificial intelligence and modern cognitive psychology was not an incidental factor to the strength of chunking theory. Indeed, Simon had previously built several computational models that captured some of the ideas he developed with Chase. Chunking theory, as well as the empirical work that supported it, motivated a considerable amount of research on expertise in the following twenty years or so.

Chunking theory assumes that chess players encode most of their long-term memory knowledge as chunks – perceptual units that can be treated as wholes. The first chunks are small, but then larger chunks are incrementally built using these smaller chunks. In chess, chunks consist, at the beginning, of individual pieces on a given square, and then grow into groups of pieces. An analogy with reading will make the process clear. At the beginning, a reader learns to recognise individual letters, such as “t” and “h”. With practice, these letters form chunks, such as “th”, and later “the”. The power of chunking is that very large units indeed can be created by this mechanism. So, for example, assuming much practice with reading, the following chunk may be learnt: “To be or not to be, that is the question”. Such a chunk is then a unit of both perception and meaning, and can be processed as a whole. Figure 2.1 provides an example of the kinds of chunks learnt by a weak chess player and a master.

Figure 2.1 Chunks learnt by a weak player (left diagram) and a master (right diagram).

In addition to mechanisms explaining how a network of chunks is constructed, the theory made several assumptions about learning and memory. It takes a fairly long time to learn a new chunk (8 seconds) and to add information to a chunk already in long-term memory (2 seconds). But once learnt, chunks can be retrieved rapidly, in a few hundred milliseconds. Short-term memory capacity, which is limited to seven items, is the same for experts and non-experts. Thus, the main difference between weak and strong chess players is the number and size of the chunks they have acquired. A final assumption is that chunks can be linked to information. In chess, this information can be a move or a sequence of moves, strategic ideas or tactical motifs. For example, given a chunk encoding a pawn structure with a weak square, the suggested action could be: “Place a knight on this square!” In psychology and artificial intelligence such condition-action pairs are known are productions.

Chunking theory explains the skill effect in recalling chess positions by assuming that strong players are more likely to recognise chunks on a board position, since they have stored many more chunks in long-term memory. Once a chunk is recognised, a pointer to it is placed in short-term memory. Although only seven pointers can be placed in short-memory, due to its limited capacity, pointers can be asso -ciated to small, medium or large chunks. Since strong players have acquired larger chunks than weaker players, they can encode a position with a smaller number of units than weaker players, and can memorise an entire position despite the limited capacity of short-term memory. In sum, while novices perceive a position as a collection of individual pieces, stronger chess players perceive it as collection of familiar configurations.

In work with Kevin Gilmartin, Simon used computer simulations and mathematical extrapolations to estimate that a chess master must have learnt about 50,000 chunks. Simon and Chase note that this number is roughly the same as the number of words that American college students have in their vocabulary. Given the time needed to learn these chunks, they estimated that it takes a minimum of 10 years, or 10,000 hours, to become a chess master.

Simon and Chase also addressed the question of how players use mental imagery to visualise board positions and anticipate moves in their mind’s eye. The central idea, already mentioned above, is that potential moves are proposed by pattern recognition: patterns on the board elicit chunks in long-term memory, which in turn suggest possible actions. Chunks also provide information allowing players to reconstruct groups of pieces in their mind’s eye. An important aspect of the theory is that pattern recognition occurs not only when looking at a physical board, but also when looking at a board position imagined in the mind’s eye. In both cases, recognised chunks might elicit information about what to do in a given situation, including what kind of moves are likely to be good. Thus, this look-ahead search consists of recognising chunks, using the information they provide to update the position in the mind’s eye and repeating this process several times. Because chunks provide useful information about moves, plans, tactical motifs, etc., they enable a highly selective search. In sum, chunking theory explains the speed at which expert players find good moves by assuming that chunks allow them (a) to identify patterns on the board; (b) to use these patterns to access useful information, including potentially good moves; and (c) to repeatedly update, after a move has been carried out, the board constructed in their mind’s eye and explore the consequences of moves and sequences of moves.

Simon and Chase’s theory was very ambitious, since it accounted for data not only about memory, but also about problem-solving. Moreover, it is not limited to chess expertise, but can also be applied to other domains of expertise, and indeed to the study of cognition in general. This explains why it had considerable impact, not only on expertise research, but also on cognitive psychology more generally.

Recall Experiment

Simon and Chase’s genius was not only to have proposed a powerful theory, but also to have supported it empirically with elegant experiments. They focused on de Groot’s perception task – which they considered as a recall task – providing both a replication and an extension. A chessboard was shown for five seco...

Table of contents

• Cover
• Title
• Copyright
• Dedication
• CONTENTS
• Preface
• Acknowledgements
• Opening
• 1 The eye of the master
• 2 Chunks!
• 3 The best move
• 4 Practice makes (almost) perfect
• 5 Men vs. women
• 6 Style and intuition
• 7 Errare humanum est
• 8 Psychological warfare and training techniques
• 9 The magic bullet?
• 10 Costs of playing chess
• 11 Endgame
• Further reading