7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 1/7

DOI: 10.1126/science.1194732, 341 (2011);331Science 

et al.Xiaohong WanGame ExpertsThe Neural Basis of Intuitive Best Next-Move Generation in Board

 This copy is for your personal, non-commercial use only.

 clicking here.colleagues, clients, or customers by, you can order high-quality copies for yourIf you wish to distribute this article to others

 here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): December 19, 2012 www.sciencemag.org (this information is current as of 

The following resources related to this article are available online at 

 http://www.sciencemag.org/content/331/6015/341.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

http://www.sciencemag.org/content/suppl/2011/01/18/331.6015.341.DC1.htmlcan be found at:Supporting Online Material

http://www.sciencemag.org/content/331/6015/341.full.html#relatedfound at:

can berelated to this articleA list of selected additional articles on the Science Web sites

http://www.sciencemag.org/content/331/6015/341.full.html#ref-list-1

, 7 of which can be accessed free:cites 39 articlesThis article

http://www.sciencemag.org/content/331/6015/341.full.html#related-urls1 articles hosted by HighWire Press; see:cited byThis article has been

http://www.sciencemag.org/cgi/collection/neuroscienceNeuroscience

subject collections:This article appears in the following

registered trademark of AAAS.is aScience 2011 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrighAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by thScience 

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 2/7

in vitro than those from control mice (Fig. 4F).

Therefore, the increased the proportion of  Clos-

tridium in the gut microbiota affected mucosal

inflammation and systemic antibody responses.

Although additional mechanisms and cell types

could be involved, our findings suggest that 

Clostridium-mediated induction of Tregs in the

colon may be responsible for these effects.

Our findings show that Tregs are abundant in

intestinal LP, and their accumulation in the SI and

colon is differentially regulated. The induction of colonic Tregs is dependent on commensal micro-

organisms with specialized properties. Among

the indigenous commensal bacteria, Clostridium

spp. belonging to clusters IV and XIVa are out-

standing inducers of Tregs in the colon. Although

alternative mechanisms may alsobe involved, our 

findings are consistent with a model in which the

 presence of  Clostridium induces the release of 

active TGF-b and other Treg-inducing factors

from IECs, which presumably cooperate with

dendritic cells to induce a general accumulation

of Tregs in the colon and at the same time affect 

the proportions of individual Treg subsets through

the preferential induction of IL-10+

CTLA4high

iTregs. Several recent studies have focused on

the microbial composition in the intestine during

health and disease.Notably, ClostridiumclustersIV

and XIVa constitute a smaller proportion of the

fecal community in patients with IBD than in

healthycontrols (15). Furthermore, some patients

with IBD have a specific reduction in Faecali-

bacterium prausnitzii, a bacterium belonging to

Clostridium cluster IV (16 ). These reports are

consistent with our findings and raise the possi-

 bility that indigenous Clostridium-dependent in-

duction of Tregs may be required for maintaining

immune homeostasis in mice and humans. The

factors derived from Clostridium that are required

for the induction of mucosal Tregs are currently

unknown. Because gnotobiotic mice colonized

with three strains of  Clostridium showed an in-

termediate pattern of Treg induction between GF

mice and mice inoculated with all 46 strains

(fig. S19), we speculate that a diverse set of 

metabolites that are most efficiently produced

 by the 46 strains of Clostridium as a whole may

 be required for the optimal induction of Tregs.

Identifying these metabolites and the molecular 

mechanisms underlying the Clostridium-host crosstalk will provide invaluable information to-

ward understanding how the gut microbiota reg-

ulates immune homeostasis and may suggest 

 potential therapeutic options for treating human

IBD and allergies.

References and Notes1. A. J. Macpherson, N. L. Harris, Nat. Rev. Immunol. 4, 478

(2004).

2. J. L. Round, S. K. Mazmanian, Nat. Rev. Immunol. 9, 313

(2009).

3. I. I. Ivanov et al., Cell 139, 485 (2009).4. V. Gaboriau-Routhiau et al., Immunity  31, 677

(2009).

5. H. J. Wu et al., Immunity  32, 815 (2010).

6. J. A. Hall et al., Immunity  29, 637 (2008).7. C. Di Giacinto, M. Marinaro, M. Sanchez, W. Strober,

M. Boirivant, J. Immunol. 174, 3237 (2005).8. K. Karimi, M. D. Inman, J. Bienenstock, P. Forsythe,

 Am. J. Respir. Crit. Care Med. 179, 186 (2009).

9. A. Lyons et al., Clin. Exp. Allergy  40, 811 (2010).10. J. L. Round, S. K. Mazmanian, Proc. Natl. Acad. Sci.

U.S.A. 107, 12204 (2010).

11. W. S. Garrett, J. I. Gordon, L. H. Glimcher, Cell 140, 859

(2010).12. I. I. Ivanov et al., Cell Host Microbe 4, 337 (2008).13. B. Min et al., Eur. J. Immunol. 37, 1916 (2007).

14. Y. Momose, A. Maruyama, T. Iwasaki, Y. Miyamoto,

K. Itoh, J. Appl. Microbiol. 107, 2088 (2009).

15. D. N. Frank et al., Proc. Natl. Acad. Sci. U.S.A. 104,

13780 (2007).

16. H. Sokol et al., Inflamm. Bowel Dis. 15, 1183

(2009).

17. K. Itoh, T. Mitsuoka, Lab. Anim. 19, 111 (1985).

18. A. M. Thornton et al., J. Immunol. 184, 3433

(2010).

19. Y. Umesaki, H. Setoyama, S. Matsumoto, A. Imaoka,

K. Itoh, Infect. Immun. 67, 3504 (1999).

20. M. D’Angelo, P. C. Billings, M. Pacifici, P. S. Leboy,

T. Kirsch, J. Biol. Chem. 276, 11347 (2001).

21. G. Matteoli et al., Gut  59, 595 (2010).

22. M. J. Barnes, F. Powrie, Immunity  31, 401 (2009).

23. Y. P. Rubtsov et al., Immunity  28, 546 (2008).

24. M. Kamanaka et al., Immunity  25, 941 (2006).

25. C. L. Maynard et al., Nat. Immunol. 8, 931 (2007).

26. M. Boirivant, I. J. Fuss, A. Chu, W. Strober, J. Exp. M

188, 1929 (1998).

27. We thank Y. Ueda, J. Nishimura, and H. Saiga for

technical assistance, R. Eisenman for critically readi

the manuscript, and A. Miyawaki for Venus. We tha

the staff at the Sankyo Laboratories for gnotobiotic

handling of the mice. The work was supported by

Grants-in-Aid for Scientific Research from the Minis

of Education, Culture, Sports, Science and Technolo

Core Research for Evolutional Science and Technolo

(CREST), Japan Science and Technology, the Mochid

Memorial Foundation for Medical and Pharmaceutic

Research, the Kato Memorial Bioscience Foundation,

Mishima Kaiun Memorial Foundation, Kanae Founda

for the Promotion of Medical Science, and Inoue

Foundation for Science. The authors have applied fo

a patent for use of bacteria belonging to the genus

Clostridium or a physiologically active substance

derived from these bacteria to induce proliferation accumulation of regulatory T cells and suppress imm

functions (patent application no. JP 2010-129134).

Material transfer agreements are required for the u

of the IL-10 Venus mice, SFB, Clostridium spp. 46

strains, Bacteroides spp. 16 strains, and Lactobacill

spp. 3 strains.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/science.1198469/DC1

Materials and Methods

Figs. S1 to S19

References

29 September 2010; accepted 1 December 2010

Published online 23 December 2010;

10.1126/science.1198469

The Neural Basis of IntuitiveBest Next-Move Generation inBoard Game ExpertsXiaohong Wan,1 Hironori Nakatani,1 Kenichi Ueno,2 Takeshi Asamizuya,2

Kang Cheng,1,2 Keiji Tanaka1*

The superior capability of cognitive experts largely depends on quick automatic processes. Toreveal their neural bases, we used functional magnetic resonance imaging to study brain activityof professional and amateur players in a board game named shogi. We found two activationsspecific to professionals: one in the precuneus of the parietal lobe during perception of boardpatterns, and the other in the caudate nucleus of the basal ganglia during quick generation of thebest next move. Activities at these two sites covaried in relevant tasks. These results suggestthat the precuneus-caudate circuit implements the automatic, yet complicated, processes ofboard-pattern perception and next-move generation in board game experts.

Board games provide a good opportunity

to study the mechanisms underlying cog-

nitive expertise because these games are

 played in accordance with a set of well-defined

rules. There is a history of psychological studies

of chess over the last 100 years (1 – 3). In an early

study (4), both world-class and local-club play

were asked to think aloud while playing, and

difference was found in the depth or width

search between the two groups. Instead, a d

ference was found in the selection of game tr

 branches that the player put into the search p

cess: The best next move was always included

the first part of search in world-class play

whereas local-club players often missed it in th

large search. de Groot inferred that world-cl

 players generate one or a few best next mo

mainly by cued recall (4). Subsequently, on

 basis of the experts’ superior performance in

 board-pattern recall task it was proposed t

chess experts quickly perceive chess patter

using various stereotyped arrangements of seve

1Cognitive Brain Mapping Laboratory, RIKEN Brain ScieInstitute, 2-1 Hirosawa, Wako, Saitama 351-0198, Jap2Support Unit for Functional Magnetic Resonance ImagiRIKEN Brain Science Institute, 2-1 Hirosawa, Wako, Saita351-0198, Japan.

*To whom correspondence should be addressed. E-mkeiji@riken.jp

www.sciencemag.org SCIENCE VOL 331 21 JANUARY 2011

REP

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 3/7

 pieces, called “chunks,” as units of perception

(5). Because chunks are associated with the

 best next moves in the long-term memory, the

 perception of chunks automatically generates an

idea of the best next move in the brain of chess

experts. Although previous imaging studies

found partial volume changes and specific ac-

tivities in the brains of experts (6  – 17 ), neural sub-

strates of the automatic processes have not been

understood. We aimed at revealing neural sub-

strates that underlie quick perception of board pattern and subsequent quick generation of the

 best next move by measuring brain activities of 

subjects playing shogi. Although more recent 

 psychological studies have shown that experts

are also efficient in conscious search and asso-

ciated evaluation (18 – 20), those are out of the

main scope of the present study.

The game shogi is popular in Japan (21).

Shogi is similar to chess in that pieces are moved

one at a time alternating between the two players,

the movability depends on the piece type, and

the objective is to capture the opponent ’s king

(fig. S1). Unique to shogi is the “drop rule”; the

captured opponent ’s pieces are held in reserve,

and if the player chooses during a turn, instead of 

moving a piece existing on the board the player 

may drop any reserve piece to any empty position

of the board as an ally piece. Mainly because of 

this reentrance ability, plays of shogi are much

more complex than those of chess. Professional

shogi players repeatedly note that the best next move comes to their mind “intuitively.” They use

the remaining time in the game for confirmatory

search and thinking about higher-level strat-

egies. Being “intuitive” indicates that the idea

for a move is generated quickly and automat-

ically without conscious search, and the process

is mostly implicit. This intuitive process oc-

curs routinely in experts, and thus it is different 

from inspiration, which occurs less frequently

and unpredictably.

Psychological studies show that the intuit

generation of next moves in board-game expe

is based on the superior, quick perception of

 patterns (5). Therefore, we first conducted a fu

tional magnetic resonance imaging (fMRI) exp

iment to explore neural circuits that are specifica

activated during the perception of shogi patte

(fig. S2) (22). In professional players (22),

contrast between five types of board game patte

(opening shogi, endgame shogi, random sho

chess, and Chinese chess) with other stimulus cegories (scenes, faces, other objects, and scramb

 patterns) showed significant activations at seve

sites in the posterior cortices ( P < 0.001 correct

(Fig. 1). Among these activations, only the ac

vation in the posterior part of precuneus show

 prominent selectivity for realistic shogi patte

(Fig. 1, B to D). The comparison between t

endgame and random shogi patterns is the m

relevant here because the patterns that we la

used in the next-move generation experime

Fig. 1. Specific activa-

tion of the posterior pre-cuneus associated withperception of shogi pat-terns (fig. S2). (A) Corticalregions (in a representa-tive professional player)that responded morestrongly to board gamepatterns than to othernon-game stimuli shownon a flattened corticalmap (yellow, P < 1.0 ×10−8 corrected; orange,P< 0.001 corrected). Forcomparison, shown are

the fusiform face area(FFA, blue), which re-sponded more stronglyto faces than to scram-bled images; the para-hippocampal place area(PPA, green), which re-sponded more stronglyto scenes than to scram-bled images; and themiddle temporal area(MT, turquoise), which re-sponded more stronglyto moving than station-ary random dot stimuli

(P < 0.005 corrected).paHG, parahippocam-pal gyrus; MTG, middletemporal gyrus; PreCun,precuneus; CingG, cingulategyrus;POS,parieto-occipitalsulcus; L, left; R, right. (B)Activations in the poste-rior part of theprecuneusin three professional players. (C) Time courses of fMRI responses to stimuli ofdifferent categories in the posterior precuneus of three subjects representingthe three subject groups, respectively. Gray bars indicate the stimuluspresentation period, and error bars indicate SEM across blocks. (D) Mean

fMRI responses in the four regions averaged among 11 professional play(Pro), eight high-ranking amateur players (Ama1), and nine low-rankamateur players (Ama2), respectively. Error bars indicate SEM acrsubjects.

21 JANUARY 2011 VOL 331 SCIENCE www.sciencemag.org42

REPORTS

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 4/7

were endgame patterns, and the pieces and low-

level features of random shogi patterns areidentical

to those of endgame shogi patterns. The random

shogi patterns here do not violate the rules of shogi,

 butthey never appear in real games.The precuneus

activation by the endgame shogi patterns, as well

as that by the opening shogi patterns, were sig-

nificantly larger than any of the activations evoked

 by chess, Chinese chess, or random shogi patterns

(Tukey’s t  tests, P  < 0.01) (Fig. 1, C and D, left 

column). None of the activations other than thosein the precuneus showed a significant difference

 between the endgame and random shogi patterns

(one-tailed t test, P > 0.05).

In both high- and low-rank amateur players

(22), the same set of areas as those in professional

 players were activated by board games as com-

 pared with other non-game stimuli (fig. S6). How-

ever, none of the areas showed significantly

stronger activations to the endgame shogi patterns

as compared withrandom shogi, chess, or Chinese

chess patterns in either amateur group (one-tailed

t test, P > 0.05) (Fig. 1,C and D,middle and right 

columns). The high-rank amateur players showed

significantly stronger activations to the openingshogi patterns in the precuneus (Tukey’s t tests,

 P  < 0.01), but the activations evoked by the

endgame shogi patterns were comparable with

those evoked by the random shogi, chess, and

Chinese chess patterns (Fig. 1, C and D). Opening

shogi patterns are usually stereotyped, whereas

endgame patterns vary extensively.

We then searched neural circuits responsible

for intuitive generation of the best next move

itself that may occur after the perception of 

 board patterns. We used spot games of shogi,

called checkmate and brinkmate problems,

which usually require the player to find a series

of moves leading to a checkmate (capturing

the opponent ’s king) even when the opponent 

makes optimal counter moves. To emphasize

the intuitive generation of the best next move

and reduce the amount of search processing,

we used a short interval (1 s) for the presen-

tation of board patterns and asked the subjectsto only report the first move (fig. S3A) (22).

The presentation of the pattern was followed

 by presentations of four response choices, from

which the subject selected his answer within 2 s

(fig. S5A). After answering to questions about 

confidence and memory, the subject was en-

gaged, during the remaining time in each trial,

in a simple detection task: detecting a “Gold”-

 piece (fig. S1A) among serially presented shogi

 pieces (fig. S5B). When the activity locked to

the shogi-pattern presentation and subsequent re-

sponse time was contrasted with that locked to

the Gold-piece detection, professional players (22)

showed significant activations ( P  < 0.001 cor-rected with a minimum cluster size threshold of 

15) in the head of the caudate nucleus as well

as in several association cortical areas, includ-

ing the posterior dorsolateral prefrontal cortex,

 pre-supplementary motor area, premotor cortex,

and precuneus (Fig. 2A, fig. S7, and table S1).

The caudate activation was bilateral, although

it was stronger in the right side. During the

sensory-motor control task (fig. S3A), in which

the subject reported the king’s position in ope

ing shogi patterns only composed of the o

 ponent ’s pieces, an identical set of associati

cortical foci was activated, whereas there w

no activation in the caudate nucleus (Fig. 3A

Thus, the activation in the caudate head w

specific to the generation of the best next mo

(Fig. 2B). There were no significant differen

in eye position traces between the two con

tions [one-way analysis of variance (ANOVA

 P  = 0.65 and P  = 0.82 for the mean and Srespectively].

To compare the activations during the qu

generation of the best next move with those d

ing deliberative search, we conducted an ad

tional experiment on each subject (fig. S4) (2

in which board patterns of spot game proble

were shown for a longer time (up to 8 s). T

subject pressed a button when he found the c

rect next move, and then four response choi

appeared, from which he selected his answer. T

subject performed the Gold-piece detection ta

during the remaining part of each trial. The a

erage period of the problem presentation was

and 7.3 s for professional and amateur playerespectively (fig. S5D). In post-experimental

terviews, both professional and amateur play

reported that they examined branches step

step, starting from a promising next move

check whether checkmate could be reached

gardless of the opponent ’s moves; if not, th

moved the search to branches starting from

other promising next move. When brain activ

locked to the problem presentation period w

A Quick generation vs. Gold -piece detection control

X=-44!  Y=15!  Z=48!p<0.0001 p<0.001 

R L!

B Quick generation vs. sensory-motor control

C Deliberative search vs. Gold -piece detection control

Professional Professional

 X=9!  Y=15!  Z=12!p<0.0001 p<0.001 

R L!

p<0.0001 p<0.001 X=-44!  Y=15! Z=48!

R L!

p<0.0001p<0.001 X=9!  Z=12!Y=15!

R L!D Professional vs. amateur

Fig. 2. Activations associated with quick generation of the best nextmove (fig. S3) and deliberative search (fig. S4) in professional players.The results shown in this figure are all group data averaged across sub-jects. (A) Activations associated with quick generation of the best nextmove as compared with the activity during Gold-piece detection in thequick-generation task. (B) Only the activations in the head of the caudate(right hemisphere, peak pattern at 9, 15, and 12 mm in the Talairachcoordinates with 38 significant voxels; left hemisphere, –12, 11, and 12 mm

with 26 voxels) remained when the activations during the quick geeration of the best next move was contrasted to the activity during tsensory-motor control task. (C) Activations associated with the delibative search as compared with the activity during Gold-piece detectionthe deliberative search task. (D) Direct comparison of the activity durthe quick generation of the best next move in the professional playwith that in the amateur players. The activations are shown accordingthe P values after cluster-size correction.

www.sciencemag.org SCIENCE VOL 331 21 JANUARY 2011

REP

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 5/7

contrasted with that locked to the Gold-piece de-

tection period, there were significant activations

at the association cortical foci largely overlapping

with those activated during the quick generation

of the best next move. However, there was no

activation in the caudate nucleus (Figs. 2C and

3A and table S1). Even with a focus on the last 

1 s before the button press, there was no caudate

activation.

The magnitude of the fMRI signals during the

quick generation of the best next move in theanatomically determined extent of the caudate

head in an individual subject positively correlated

with the percentage of correct responses of the

subject in the quick generation [correlation co-

efficient (r ) = 0.77, P = 0.00015] (Fig. 3B, left)

within the professional group.

In amateur players (22), a very similar set of 

association cortical areas, as in the professional

 players, was activated by quick generation of 

the best next move as well as by the deliberative

search (Fig. 3A, middle and right, fig. S8, and

table S2). However, there was no significant acti-

vation in the caudate nucleus during quick gen-

eration of the best next move when the datawere analyzed asa whole ( P > 0.01)(Fig. 3A, left,

fig. S8, and table S2). Direct comparison between

the professional and amateur groups demon-

strates the specificity of the caudate activation in

the professional players (Fig. 2D).

Activation occurred in the caudate head of the

amateur players only in a particular group of trials.

The percentage of correct responses was nega-

tively correlated with the response time (Fig. 3C,

left, and table S3), whereas it did not depend on

the number of steps to reach checkmate or the

number of possible next moves. When the trials

were divided into four quarters in individual sub-

 jects according to response time, the fMRI sig-nals in the anatomically determined extent of the

caudate head showed significant dependency on

the response-time group (one-way ANOVA, P =

0.046), and the activation in the shortest response-

time group was significant (one-tailed t  test, P =

0.022) (Fig. 3C, right). This group of trials large-

ly overlapped with the trials in which the subject 

was “confident ” about the move selection (con-

fident trials constituted 62% of the shortest trials),

and a significant activation of the anatomically

defined caudate head was also detected in the

confident trials (one-tailed t test, P = 0.043) (Fig.

3D). There was no such dependency of caudate

activation on the properties of trials in the pro-

fessional players: The activation was always

strong [one-way ANOVA, P  > 0.36 (Fig. 3C,

right) and P > 0.23 (Fig. 3D)].

The problems to which the subject responded

with his shortest response times varied between

subjects. Post-experimental interviews of the am-

ateur players revealed common features among

the problems. We showed 10 examples of the

 problems to which the player responded with

response times in the shortest quarter range, as

well as six examples of the problems to which

they responded with response times in the longer 

half range. They reported that in many of the

former problems, the board pattern had features

matching the situations in which particular strat-

egies or schemas are useful or those contained

in games that they had experienced before. Most 

of the problems that required longer response

times did not have such connections to their 

knowledge. Thus in the amateur players, the cau-

date head was activated only when the problem

had features that matched those of the situation in

which some popular schemas were useful or 

those of the patterns personally experienced

the player.

Lastly, to examine the relation between t

 precuneus activation in perception of shogi p

terns and the caudate activation in quick ge

eration of the best next move we analyzed t

correlation of trial-by-trial variance betwe

the precuneus and caudate activations (Fig

top left). The correlation was significantly high

in the quick-generation task than in the senso

motor control and deliberative search tasks in p

Fig. 3. Variety in the strength of activity in the caudate head across trial types and subject groups. Activity in the caudate head (Caudate), posterior dorsolateral prefrontal cortex (DLPFC), and dorpremotor cortex (PMd) in professional and amateur players. Region of interest (ROI) for the cauda

head was determined by anatomical structure, whereas those for the dorsolateral prefrontal cortex adorsal premotor cortex were determined by means of conjunction analysis of functional activity acrthe professional and amateur players. The error bars indicate SEM across subjects. The activity wcalculated in comparison with that during the Gold-piece detection in individual tasks. (B) Correlatbetween the strength of caudate head activation and the percentage of trials with correct responacross subjects for professional and amateur players. (C) (Left) Percentage of correct responses plotagainst averaged response time differences and (right) the magnitude of activity in the caudate headthe quarter groups of trials, to which trials were divided in individual subjects according to the respotime (RT1 to -4, RT1 with the shortest response times). The response time difference was obtained subtracting the average response time in the sensory-motor control task from the response timeindividual trials. (D) Activity in the caudate head in trials with confidence (Conf) and without confiden(Unconf). In (C) and (D), the error bars indicate SEM across subjects, and the activity was calculatedcomparison with that during the sensory-motor control.

21 JANUARY 2011 VOL 331 SCIENCE www.sciencemag.org44

REPORTS

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 6/7

fessional players (one-tailed t  test, P  < 0.01).

Although there were no such differences in ama-

teur players when trials of the quick-generation

task were analyzed as a whole, the precuneus-

caudate correlation in the quick-generation task 

was significantly higher when the amateur player 

responded with response times in the shortest 

quarter range than those in the sensory-motor 

control and deliberative search tasks ( P  < 0.05).

The correlation of activities in the caudate head

with those in the posterior dorsolateral pre-frontal cortex, dorsal premotor area, and pre-

supplementary motor area was also higher in

the quick-generation task than in the sensory-

motor control and deliberative search tasks in

 professional players ( P < 0.01), but not in ama-

teur players ( P > 0.05) (Fig. 4, left, bottom three).

The present results show two activations that 

are prominent in professional players. One is the

activation in the posterior part of the precuneus

during the perception of shogi patterns, and the

other is the activation in the head of the caudate

nucleus during the quick generation of the best 

next move. In the quick-generation task, the play-

ers quickly generated the idea of the best next move without conscious search. This situation

matches the “intuitive generation of the best next 

move” that has been discussed in the literature of 

 board-game psychology. There was no activation

in the caudate head during the sensory-motor 

control task or during the conscious search. Thus,

we conclude that the caudate head activation was

associated with the intuitive generation of the best 

next moves.

Recent imaging studies show activation

of the posterior part of the precuneus in visuo-

spatial imagery and in episodic memory retrieval

(23 – 26 ). Precuneus activation has also been

shown in relation to the working memory of 

chessboard patterns (12). In our perception ex-

 periment, the posterior precuneus was gener-

ally more activated by board games than other 

visual stimuli in both professional and ama-

teur players. In addition to this general selec-

tivity, realistic shogi patterns activated the area

more strongly than the other board game pat-

terns in professional players. Higher features

specific to realistic shogi patterns might evoke

neuronal activities corresponding to their repre-sentations or computations about them in the

 posterior precuneus. The posterior precuneus

 projects to the dorsolateral prefrontal cortex, which

in turn projects to the caudate head (27  – 29). In

addition, there is direct projection from the

 precuneus to the dorsomedial part of the cau-

date head (30 – 32). The information sent from

the posterior precuneus to the caudate head

about key features of the shogi pattern might 

help generate the best next move. The strong

correlation between the activation of the pre-

cuneus and that of the caudate head in pro-

fessional players during the quick-generation

task supports this view.How does the caudate head activity contribute

to the intuitive generation of the best next move?

We may get insights from other situations in

which the caudate head is activated in animals

and humans. The caudate nucleus is a part of the

 basal ganglia, which is thought to be responsible

for the formation and execution of habit, or 

stimulus-response association (33 – 38). The pu-

tamen may be involved when the response is

motor, whereas the caudate nucleus is involved

when the response is cognitive (34), as in the

 present case. The generation of the best next 

move to a given shogi pattern is similar to a habit 

in that it is quick and implicit. However, the p

cessing cannot be a simple stimulus-respon

association, in which an identical stimulus

repeatedly presented and a single response

evoked. Each of the 180 problems used in t

quick-generation task was given only once in

experiment for each subject. Moreover, 82

them were probably seen for the first time

the subjects because they were newly creat

for this study. Even for the remaining proble

taken from published books, because of huge variety of such problems the subject cou

not have experienced each of the problems f

quently. The variety of inputs and outputs in o

case was several orders of magnitude larger th

that in typical habits. Therefore, the generati

of the best next move had to be based on t

 perception of key features extracted from

 pattern but not that of the pattern itself. In ot

words, the mapping from inputs to outputs had

 be categorical. In chess experts, chunks of pie

are associated with the best next move or b

series of moves in the long-term memory, and t

the perception of chunks automatically evo

the idea of the best next move or best seriesmoves (5). This idea is similar to that of stimulu

action association. It was later pointed out t

chunks defined by combinationsof several pie

are too small to specify the best next move.

solve this difficulty, Gobet and his colleagu

(39, 40) have proposed that chunks evolve

larger assemblies (“templates”) in experts. T

templates have simple variables that can be

stantiated. The relation between habits and chun

ortemplate-based next-move generation sho

 be further studied.

According to the basic structure of the ba

ganglia circuit, the thalamus and brainstem mo

centers are inhibited by the substantia nigra preticulata and the internal segment of glob

 pallidus,which are in turn inhibitedby the striatu

which is composed of the caudate nucleus and p

tamen. This inhibition-of-inhibition (disinhibiti

circuit, together with the parallel excitation-

inhibition circuit via the subthalamic nucleus

thought to be advantageous for efficient selecti

of one action plan among multiple possibili

(41). In checkmate problems, there are multi

next moves that check the opponent ’s king b

will not reach checkmate if the opponent mak

appropriate counter moves. The move that w

reach checkmate regardless of the opponen

moves is often accompanied with losses of on

own important pieces (“sacrifice”). In this sen

the generation of the best next move inclu

selection procedures to turn down the moves th

appear advantageous in the short term in favor

the one that is most promising in view of the fi

goal (checkmate). The way to resolve competit

among multiple possible moves has not b

much discussed in cognitive psychology of ch

experts, although a computer model nam

CHUMP has a function of such resolution (39

 Neuropsychological, imaging, and anim

studies suggest that the caudate nucleus is

Fig. 4. Coefficients of cor-relation, across trials, of thefMRI responses in the cau-date head and posteriorprecuneus with those inthe precuneus,DLPFC, PMd,andpre-supplementarymo-tor area (preSMA) duringthequickgeneration(Quick),sensory-motor control (Con-trol), anddeliberativesearch(Search) tasks. Trials of thequick-generation task were

divided into four quartersaccording to the responsetime of the subject (RT1to -4). Error bars indicateSEM across subjects in eachsubject group. *P < 0.05,**P < 0.01, significantdifference with the coeffi-cients in thesensory-motorcontrol task.

www.sciencemag.org SCIENCE VOL 331 21 JANUARY 2011

REP

7/29/2019 341.full.pdf

http://slidepdf.com/reader/full/341fullpdf 7/7

volved in goal-directed behavior. The caudate is

activated in set switching, selection of action in

response to the expectation of the action out-

come, and judgment accompanied by positive or 

negative outcome (42 – 45). In the checkmate and

 brinkmate problems used in the present study,

multiple moves (1 to 19, including the opponent ’s

counter moves) are required to reach checkmate.

In interviews after the experiments, professional

 players reported that, although they did not figure

out the whole sequence of moves to reach check-mate within the short period they often got an

idea of the arrangement of key pieces at the final

checkmate in addition to that of the next move.

The intermediate moves did not come to their 

mind, but the next move might be generated as

a step in the direction to realize the final check-

mate pattern. The perception of key features in

a given pattern might evoke the idea of a final

checkmate pattern, which in turn evoked the

idea of a next move that eventually produced

the final pattern. Thus, the quick generation of 

the best next move has an element of goal-

 based action selection. A theory (3, 46 ) about 

chess experts has indicated a goal-directed fac-tor in the expertise. It proposes that experts first 

obtain a temporal goal that is based on higher-

level perception of the board pattern and then

find a path connecting from the current board

 pattern to the goal pattern. Although generation

of an idea of the best next move may be self-

rewarding, a simple explanation of the caudate

activation by this self-rewarding is not compat-

ible with the generality of the caudate activation

in professional players common to trials with and

without confidence (Fig. 3D).

Thus, the quick generation of the best next 

move in professional players appears to recruit 

the precuneus-caudate circuit. Professional shogi players have conducted daily concentrated train-

ing for 3 or 4 hours per day over a number of 

years. This long-term training might result in the

general recruitment of the posterior precuneus

and the caudate head. The strong correlation be-

tween the caudate head and prefrontal and

 premotor cortical areas during the quick genera-

tion of the best next move suggests that the cau-

date head in professional players might efficiently

coordinate the prefrontal and premotor circuits

for this particular purpose. The present results

should initiate extensive interactions between ex-

 pert psychology and neuroscience concerning ba-

sal ganglia. The many data concerning anatomical

connections and functional aspects of the basal

ganglia may provide clues for the further eluci-

dation of the information processing modes in-

cluded in the process. For example, we here

suggest that the process of quick generation of 

the bestnext movemaybe not a simplerecall ofa

move associated with the perceived feature of 

 board pattern but also includes the competitionamong possible moves and the goal-directed

 processing.

References and Notes1. K. A. Ericsson, A. C. Lehmann, Annu. Rev. Psychol. 47,

273 (1996).

2. F. Gobet, A. de Voogt, J. Retschitzki, Moves in Mind: The

Psychology of Board Games (Psychology Press, Hove and

New York, 2004).3. P. Saariluoma, Chess Players’  Thinking: A Cognitive

Psychological Approach (Routledge, London, 1995).

4. A. D. de Groot, Het Denken van Den Schaker 

(Nooord-Hollandsche Uitgevers Maatschappij, Amsterdam,

1946).

5. W. G. Chase, H. A. Simon, Cognit. Psychol. 4, 55

(1973).

6. P. Nichelli et al., Nature 369, 191 (1994).

7. M. Onofrj et al., Neurosci. Lett. 198, 169 (1995).

8. O. Amidzic, H. J. Riehle, T. Fehr, C. Wienbruch, T. Elbert,

Nature 412, 603 (2001).

9. H. J. Volke, P. Dettmar, P. Richter, M. Rudolf, U. Buhss,

 J. Psychophysiol. 16, 23 (2002).

10. O. Atherton, J. Zhuang, W. M. Bart, X. P. Hu, S. He,

Cognit. Brain Res. 16, 26 (2003).

11. P. Saariluoma, H. Karlsson, H. Lyytinen, M. Ters,

F. Geisler, Eur. J. Cogn. Psychol. 16, 753 (2004).

12. G. Campitelli, F. Gobet, K. Head, M. Buckley, A. Parker,

Int. J. Neurosci. 117, 1641 (2007).

13. E. A. Maguire et al., Proc. Natl. Acad. Sci. U.S.A. 97,

4398 (2000).

14. S. Tanaka, C. Michimata, T. Kaminaga, M. Honda,

N. Sadato, Neuroreport 13, 2187 (2002).

15. C. Gaser, G. Schlaug, J. Neurosci. 23, 9240 (2003).

16. T. Hanakawa, M. Honda, T. Okada, H. Fukuyama,H. Shibasaki, Neuroimage 19, 296 (2003).

17. Y. Ouchi et al., Brain Res. Cogn. Brain Res. 23, 164

(2005).

18. N. Charness, J. Exp. Psychol. Hum. Percept. Perform. 2,

467 (1981).

19. D. H. Holding, Simul. Games 10, 207 (1979).

20. D. H. Holding, H. D. Pfau, Am. J. Psychol. 102, 103

(1985).

21. About 160 active professional players belong to the

Japan Shogi Association (www.shogi.or.jp), an organization

that participated with us in the present study.

22. Materials and methods are available as supporting

material on Science Online.

23. R. J. Dolan et al., Nature 389, 596 (1997).

24. A. E. Cavanna, M. R. Trimble, Brain 129, 564

(2006).

25. A. D. Wagner, B. J. Shannon, I. Kahn, R. L. Buckner,

Trends Cogn. Sci. 9, 445 (2005).

26. R. Cabeza, E. Ciaramelli, I. R. Olson, M. Moscovitch,

Nat. Rev. Neurosci. 9, 613 (2008).

27. G. E. Alexander, M. R. DeLong, P. L. Strick, Annu. Re

Neurosci. 9, 357 (1986).

28. C. Cavada, P. S. Goldman-Rakic, Neuroscience 42, 6

(1991).

29. J. M. Kemp, T. P. S. Powell, Brain 93, 525 (1970).

30. L. D. Selemon, P. S. Goldman-Rakic, J. Neurosci. 5, 7

(1985).

31. E. H. Yeterian, D. N. Pandya, J. Comp. Neurol. 352,

(1995).

32. G. R. Leichnetz, Anat. Rec. 263, 215 (2001).

33. M. Mishkin, B. Malamut, J. Bachevalier, in Neurobio

of Human Learning and Memory , G. Lunch,

J. L. McGaugh, N. M. Weinberger, Eds. (Guilford Pres

New York, 1984), pp. 65–77.

34. R. A. Poldrack, V. Prabhakaran, C. Seger, J. D. E. Gabr

Neuropsychologia 13, 564 (1999).

35. A. M. Graybiel, Neurobiol. Learn. Mem. 70, 119

(1998).

36. A. M. Graybiel, Annu. Rev. Neurosci. 31, 359 (2008)

37. M. G. Packard, B. J. Knowlton, Annu. Rev. Neurosci.

563 (2002).

38. M. D. Lieberman, Psychol. Bull. 126, 109 (2000).

39. F. Gobet, P. J. Jansen, in Advances in Computer 

Chess 7, H. J. van den Herik, I. S. Herschberg,J. W. Uiterwijk, Eds. (Univ. of Limburg Press,

Maastricht, Netherlands, 1994), pp. 35–60.

40. F. Gobet, H. A. Simon, Cognit. Psychol. 31, 1 (1996

41. G. S. Berns, T. J. Sejnowski, J. Cogn. Neurosci. 10, 1

(1998).

42. J. A. Grahn, J. A. Parkinson, A. M. Owen, Prog. Neurob

86, 141 (2008).

43. H. H. Yin, B. J. Knowlton, Nat. Rev. Neurosci. 7, 464

(2006).

44. J. Crinion et al., Science 312, 1537 (2006).

45. S. Han, S. A. Huettel, A. Raposo, R. A. Adcock,

I. G. Dobbins, J. Neurosci. 30, 4767 (2010).

46. P. Saariluoma, Psychol. Res. 54, 17 (1992).

47. This work was supported by the Fujitsu Laboratorie

and a grant-in-aid for Young Scientists from the

Ministry of Education, Culture, Sports, Science and

Technology of Japan (19700312, X.W.). We thank

Japan Shogi Association for cooperation and T. Ito

K. Okuma, and T. Kawatsuma for discussion.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/331/6015/341/DC1

Materials and Methods

Figs. S1 to S8

Tables S1 to S3

References

7 July 2010; accepted 16 December 2010

10.1126/science.1194732

21 JANUARY 2011 VOL 331 SCIENCE i46

REPORTS