DOI: 10.1126/science.1102566 , 447 (2004); 306Science

et al.Paul W. Glimcher,DecisionNeuroeconomics: The Consilience of Brain and

www.sciencemag.org (this information is current as of December 4, 2006 ):The following resources related to this article are available online at

http://www.sciencemag.org/cgi/content/full/306/5695/447version of this article at:

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

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

http://www.sciencemag.org/cgi/content/full/306/5695/447#related-content

http://www.sciencemag.org/cgi/content/full/306/5695/447#otherarticles, 11 of which can be accessed for free: cites 25 articlesThis article

35 article(s) on the ISI Web of Science. cited byThis article has been

http://www.sciencemag.org/cgi/content/full/306/5695/447#otherarticles 5 articles hosted by HighWire Press; see: cited byThis article has been

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

: subject collectionsThis article appears in the following

http://www.sciencemag.org/help/about/permissions.dtl in whole or in part can be found at: this article

permission to reproduce of this article or about obtaining reprintsInformation about obtaining

registered trademark of AAAS. c 2005 by the American Association for the Advancement of Science; all rights reserved. The title SCIENCE is a

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

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

4. S. Ito, V. Stuphorn, J. W. Brown, J. D. Schall, Science302, 120 (2003).

5. K. Shima, J. Tanji, Science 282, 1335 (1998).6. J. O’Doherty, M. L. Kringelbach, E. T. Rolls, J. Hornak,

C. Andrews, Nature Neurosci. 4, 95 (2001).7. M. Ullsperger, D. Y. von Cramon, J. Neurosci. 23,

4308 (2003).8. C. B. Holroyd, M. G. H. Coles, Psychol. Rev. 109, 679

(2002).9. C. B. Holroyd, J. T. Larsen, J. D. Cohen, Psychophys-

iology 41, 245 (2004).10. H. Gemba, K. Sasaki, V. B. Brooks, Neurosci. Lett. 70,

223 (1986).11. M. Ullsperger, D. Y. Von Cramon, Cortex, in press.12. N. Picard, P. L. Strick, Cereb. Cortex 6, 342 (1996).13. C. B. Holroyd, S. Nieuwenhuis, R. B. Mars, M. G. H.

Coles, in Cognitive Neuroscience of Attention, M. I.Posner, Ed. (Guilford, New York, in press).

14. W. J. Gehring, B. Goss, M. G. H. Coles, D. E. Meyer,E. Donchin, Psychol. Sci. 4, 385 (1993).

15. M. Falkenstein, J. Hoormann, S. Christ, J. Hohnsbein,Biol. Psychol. 51, 87 (2000).

16. W. Schultz, Neuron 36, 241 (2002).17. C. B. Holroyd et al., Nature Neurosci. 7, 497 (2004).18. M. M. Botvinick, T. S. Braver, D. M. Barch, C. S. Carter,

J. D. Cohen, Psychol. Rev. 108, 624 (2001).19. N. Yeung, M. M. Botvinick, J. D. Cohen, Psychol. Rev.,

in press.20. Materials and methods are available as supporting

material on Science Online.21. R. Hester, C. Fassbender, H. Garavan, Cereb. Cortex

14, 986 (2004).22. The majority of activations fall into the border zone

between areas 8, 6, and 32, with some extension intoarea 24. Recent research in nonhuman primatesseems to suggest a functional-anatomical dissocia-

tion of regions subserving pre-response conflictmonitoring from structures sensitive to errors andomission of reward (1, 4). Although in humans thisview is still under debate (11, 13, 21), the presentmeta-analysis does not provide unequivocal evidencefor or against such a dissociation. Activations relatedto pre-response conflict and uncertainty occur moreoften in area 8 and less often in area 24 than dosignal increases associated with errors and negativefeedback (area 8, 32.5% versus 9.7%; area 24, 7.5%versus 25.8%), supporting the dissociation view.However, both groups of activations cluster primarilyin area 32 (pre-response, 42.5%; error, 41.9%),suggesting that pre- as well as post-responsemonitoring processes share at least one underlyingstructure. It seems that the currently available spatialresolution in fMRI, in conjunction with anatomicalvariability and differences in scanning and prepro-cessing methods between studies, limit the ability toresolve this debate about a possible dissociation inthe range of 10 mm or less.

23. T. Paus, Nature Rev. Neurosci. 2, 417 (2001).24. H. D. Critchley et al., Brain 126, 2139 (2003).25. E. K. Miller, J. D. Cohen, Annu. Rev. Neurosci. 24, 167

(2001).26. A. R. Aron, T. W. Robbins, R. A. Poldrack, Trends

Cogn. Sci. 8, 170 (2004).27. D. Badre, A. D. Wagner, Neuron 41, 473 (2004).28. S. A. Bunge, K. N. Ochsner, J. E. Desmond, G. H.

Glover, J. D. E. Gabrieli, Brain 124, 2074 (2001).29. M. Brass, D. Y. von Cramon, J. Cogn. Neurosci. 16,

609 (2004).30. K. R. Ridderinkhof et al., Science 298, 2209 (2002).31. K. R. Ridderinkhof, S. Nieuwenhuis, T. R. Bashore,

Neurosci. Lett. 348, 1 (2003).32. J. G. Kerns et al., Science 303, 1023 (2004).

33. H. Garavan, T. J. Ross, K. Murphy, R. A. Roche, E. A.Stein, Neuroimage 17, 1820 (2002).

34. G. Bush et al., Proc. Natl. Acad. Sci. U.S.A. 99, 523(2002).

35. J. O’Doherty, H. Critchley, R. Deichmann, R. J. Dolan,J. Neurosci. 23, 7931 (2003).

36. R. Cools, L. Clark, A. M. Owen, T. W. Robbins, J.Neurosci. 22, 4563 (2002).

37. J. F. Bates, P. S. Goldman-Rakic, J. Comp. Neurol.336, 211 (1993).

38. M. Petrides, D. N. Pandya, Eur. J. Neurosci. 11, 1011(1999).

39. L. Koski, T. Paus, Exp. Brain Res. 133, 55 (2000).40. K. Fiehler, M. Ullsperger, D. Y. von Cramon, Eur. J.

Neurosci. 19, 3081 (2004).41. W. J. Gehring, R. T. Knight, Nature Neurosci. 3, 516

(2000).42. M. I. Posner, G. J. DiGirolamo, in The Attentive Brain,

R. Parasuraman, Ed. (MIT Press, Cambridge, MA,1998), pp. 401–423.

43. K. Matsumoto, K. Tanaka, Science 303, 969 (2004).44. K. Matsumoto, K. Tanaka, Curr. Opin. Neurobiol. 14,

178 (2004).45. This research was supported by a TALENT grant

(E.A.C.) and a VENI grant (S.N.) of the NetherlandsOrganization for Scientific Research and by thePriority Program Executive Functions of the GermanResearch Foundation (M.U.). Helpful comments byS. Bunge are gratefully acknowledged.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/306/5695/443/DC1Materials and MethodsTable S1References

R E V I E W

Neuroeconomics: The Consilience ofBrain and DecisionPaul W. Glimcher1* and Aldo Rustichini2

Economics, psychology, and neuroscience are converging today into a single, unifieddiscipline with the ultimate aim of providing a single, general theory of humanbehavior. This is the emerging field of neuroeconomics in which consilience, theaccordance of two or more inductions drawn from different groups of phenomena,seems to be operating. Economists and psychologists are providing rich conceptualtools for understanding and modeling behavior, while neurobiologists provide toolsfor the study of mechanism. The goal of this discipline is thus to understand theprocesses that connect sensation and action by revealing the neurobiologicalmechanisms by which decisions are made. This review describes recent develop-ments in neuroeconomics from both behavioral and biological perspectives.

The full understanding of utility willcome from biology and psychology byreduction to the elements of humanbehavior followed by a bottom-upsynthesis, not from the social sciencesby top-down inference and guessworkbased on intuitive knowledge. It is inbiology and psychology that econo-mists and social scientists will find the

premises needed to fashion morepredictive models, just as it was inphysics and chemistry that research-ers found the premises that upgradedbiology. (p. 206) (1)

Consider the famous St. Petersburg para-dox (2). Which of the following would youprefer, /40 or a lottery ticket that paysaccording to the outcomes of one or morefair coin tosses: heads you get /2 and thegame ends, tails you get another toss and thegame repeats, but now if the second tosslands heads up you get /4, and so on. If thenth toss is the first to land heads up, you get

2n dollars. The game continues, howeverlong it takes, until the coin lands heads up.We can assess the average objective, orexpected, value of this lottery by multiplyingthe probability of a win on each flip by theamount of that win:

Expected value 0 !0:5 " 2# $ !0:25 " 4# $

!0:125 " 8#I

0 1 $ 1 $ 1 $ I

This simple calculation reveals that theexpected value of the lottery is infinite eventhough the average person is willing to payless than /40 to play it. How could this be?

For an economist, any useful explanationmust begin with a set of assumptions thatrenders behavior formally tractable to coher-ent theoretical and mathematical analysis.Economists therefore explain this behaviorby assuming that the desirability of moneydoes not increase linearly, but rather growsmore and more slowly as the total amount atstake increases. For example, the desirabilityof a given amount might be a power function

1Center for Neural Science, New York University,New York, NY 10003, USA. 2Department of Econom-ics, University of Minnesota, Minneapolis, MN 55455,USA.

*To whom correspondence should be addressed.E-mail: glimcher@cns.nyu.edu

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

www.sciencemag.org SCIENCE VOL 306 15 OCTOBER 2004 447

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

of that amount, as shown by the black line inFig. 1. A decision-maker for whom thesubjective value, or utility, of money grewin this fashion would then determine thedesirability, or expected utility, of the St.Petersburg lottery by multiplying the proba-bility of a win on each flip by the utility ofthe amount won on that flip, and thus hemight well be willing to pay less than /40 toplay this game.

From the point of view of a psychologistattempting to understand and explain thissame phenomenon, it is the nature of riskaversion and the psychological mechanismsthat this set of preferences reveals thatbecome the subject of explanatory study.The psychological mechanism that accountsfor risk aversion in human subjects, forexample, has been shown to be moresensitive to monetary losses than to mone-tary gains. Further, psychologists have sug-gested that subjective utilities are computedwith regard to somewhat arbitrary andidiosyncratic monetary reference points, orframes, set by yet other psychological pro-cesses (3). Psychologists use observationslike these to argue that human choosers areendowed with a particularly strong fear oflosses and that they weigh the merits of allpossible gains and losses relative to apsychological benchmark: The psychologicalapproach seeks empirically to describe min-imally complex behavioral tendencies, mod-ules, or heuristics that can account for theactions of human choosers.

A traditional neurobiological perspectiveuses yet another approach: A hungry bird isshown a tray that contains five millet seedsand repeatedly permitted to fly to the trayand eat the seeds. At a neurobiological level,the study of this behavior begins with theassumption that the visual stimulus of thefive seeds must somehow propagate throughthe sensory system of the animal to triggeractivation in orienting circuits that move thebird to the seeds. Next, the same bird ispermitted to fly to a second tray covered by apiece of paper. When the bird displaces thecover, half of the time it reveals 12 seeds andhalf of the time it reveals nothing. Mecha-nistically, the visual stimulus must againtrigger an orienting response, and presum-ably in this case the strength with whichvisual signals connect synaptically to theorienting circuits reflects both the number ofseeds that the bird earns and the likelihoodthat seeds will be found under the paper.Lastly, both trays are presented, and the birdis observed to fly toward the tray that maycontain 12 millet seeds. A standard neurobi-ological explanation (4, 5) presumes thatunder these circumstances the two differentbehavioral circuits compete. In this case thesynapses that elicit an orienting response tothe covered tray are stronger and thus control

behavior. The neurobiological explanationspecifies the minimal neural circuitry re-quired to account for the observed behaviorof the bird.

What is striking about explanations ofchoice behavior by economists, psycholo-gists, and neurobiologists is the differentlevels at which they operate. The economicapproach attempts to describe globally allchoice behavior with a single logically con-sistent formalism. The psychological ap-proach examines the ways in whichsubjective and objective estimates of valuediffer and posits psychological modules thatcan account for these observed behavioralpreferences. The neurobiological explanationstarts with the simplest possible neural

circuitry that can account for the simplestmeasurable elements of behavior. It seemsobvious that these different levels of expla-nation should be linked, but how can such alinkage be accomplished? We argue that aunified explanation of decision-making isnot only possible but has recently begun andthat, when the linkage between these threelevels of explanation has matured, a new,more powerful decision science rooted in aneuroeconomic approach will have beendeveloped.

A second claim we will make is that oncethis reconstruction of decision science iscompleted, many of the most puzzlingaspects of human behavior, aspects thateconomic theory, psychological analysis, orneurobiological deconstruction have failed to

explain, will become formally and mecha-nistically explicable. The claim is, in essence,that a decision science that simultaneouslyengaged all three approaches would be moreheavily constrained and at the same timewould have much greater explanatory powerthan do any of these three approachesoperating alone. We will see examples ofhow this synthetic approach would operate inprinciple and early attempts at syntheticsolutions below.

This reconstruction of the study of de-cision is also going to be the appropriatebasis for a more ambitious theory that ex-plains not just how we make decisions butwhy. That such an explanation is necessaryand possible is indicated by the fact thatfundamental features of decision making arecommon to many species. For example, riskaversion as shown by the St. Petersburgparadox has been described in many species.Studies of birds making choices in riskyenvironments produce a behavior best de-scribed by a utility function (Fig. 1) (6, 7).We know that humans and birds deviatedfrom a common reptilian ancestor at least200 million years ago, but this basic functionfor choice has remained essentially un-changed. Such commonalities make a clearsuggestion: A utility function of this typeprobably is an efficient and evolved featureof vertebrate choice. For example, Robson(8) provides a justification of why a utilityfunction might be an evolutionary optimalresponse to changing environments. Just asinformation theory was used by Barlow (9)to explain why animals as diverse as horse-shoe crabs and cats use similar encodingschemes in their visual systems, an econom-ic theory that relates utility to Darwinian fit-ness must serve as an overarching tool forunderstanding vertebrate choice behavior.

Linking the Decision SciencesSubjective desirability. The central conceptin modern economic theory is the notion ofsubjective utility: Preferences must be de-scribed as subjective properties of thechooser. Surprisingly, the notion that prefer-ences are represented in the nervous system,that these preferences are subjective, andthat they guide the production of action hasonly recently entered the neurobiologicalmainstream. We believe that this has been acritical flaw in neurobiological studies, be-cause it is essential that economics, psy-chology, and neuroscience acknowledge acommon phenomenological base to achievea reductive unification of the decision sci-ences. The concepts that guide the behav-ioral study of decision-making must alsoguide the mechanistic study of that process.

In part, this preference-free approachmay have arisen from neurobiology’s rootsin the stimulus-response physiology of the

Value of a Gaindollar, calories, milliliters

Convex Utility Fn:Risk Seeking Subjective = Objective

Risk Neutral

Standard Utility Fn:Risk Averse

Objective Measure

Subj

ectiv

e De

sirea

bility

Fig. 1. Bernoulli’s notion of subjective value orutility. The black line plots the typical relation-ship between objective and subjective valua-tions of an action. As the objective value of again increases, the subjective desirability, orutility, grows more slowly. Bernoulli demon-strated that this relationship could account forthe observation that humans are typically risk-averse. The solid gray line plots a condition inwhich subjective value grows more quickly thanobjective value, a preference structure thatwould yield risk-seeking behavior.

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

15 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org448

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

early twentieth century (10). Working fromDescartes’ philosophy, Sherrington (11) pro-posed that physiologists should work tolink stimulus and response directly throughwhat Pavlov (12) would later call a ‘‘definitenervous path.’’ Scientists like Sherringtonand Pavlov proposed that it was the roleof neuroscience to chart these stimulus-response connections through the nervoussystem.

A critical step beyond this initial frame-work was a recent effort to explain morecomplicated behaviors and to focus onactions for which deterministic sensory-to-movement mapping approaches were insuf-ficient. Newsome and his colleagues (13, 14)made that step in the late 1980s when theyexamined perceptual decision-making bymonkeys viewing ambiguous sensory stim-uli. In those experiments, monkeys stared ata display of chaotically moving spots oflight. On training trials, a subset of the spotsmoved coherently in a single direction,whereas the remaining spots moved ran-domly (15). The direction of this coherentmotion indicated which of twopossible saccadic eye move-ments would yield a fruit juicereward, and at the end of eachtrial animals were free to make asaccade. If they made the correctmovement, they then receivedthe reward. On a critical sub-set of trials, however, monkeysviewed displays in which noneof the dots moved in a singlecoherent direction, and thus thedisplay provided no informationfrom which the location of therewarded eye movement couldbe deduced. Under these conditions, Newsomeand his colleagues found that the firingrates of single neurons in the middle tem-poral visual area (area MT) were still cor-related with the behavior of the animals,even when that behavior could not be pre-dicted from the properties of the stimulus.Newsome and his colleague Shadlen’s sub-sequent studies revealed the basic neuro-biological substrate for perceptual decision-making and showed convincingly that thiscircuit could not be modeled simply as asingle ‘‘definite nervous path’’ from stimu-lus to response (16).

This work, in turn, accelerated studies ofthe posterior parietal cortex, an area inter-posed between many of the sensory circuitsand motor circuits of the primate brain,which appeared to play a critical role in theperceptual decision-making Newsome hasstudied (17, 18). Platt and Glimcher (19)made an important advance when theyextended Newsome’s approach by proposingthat posterior parietal cortex might play arole in decision-making in an economic

sense and that it might encode the desirabil-ities of making particular movements.

In Platt and Glimcher’s experiments,trained rhesus monkeys were allowed toparticipate in repeated rounds of a simplelottery while the activity of nerve cells in theposterior parietal cortex was monitored. Atthe beginning of each round, two yellowspots were illuminated on a screen, one tothe left and one to the right of where themonkey was looking. This began the lotteryphase of the round, a period during whichthe monkey did not know whether the left orright light would be offered as a prize at theend of that round. At the end of this phase, athird light changed color to red or green,indicating which of the two initial lights hadbeen randomly selected to yield a fruit juicereward on that particular round. The monkeyreceived the fruit juice if he oriented to theselected light at the end of the round. Whilemonkeys played hundreds of rounds of thisgame, Platt and Glimcher systematicallyvaried either the relative probabilities thatthe left or right lights would be selected at

the end of each round or the size of thereward associated with each. These twovariables were selected because economictheories assess desirability by combining thevalue and likelihood of gain in somesubjective manner. Platt and Glimcher foundthat some parietal neurons did indeed encodethe value and likelihood of reinforcementduring the lottery phase of each round.Under these conditions, the brains of themonkeys explicitly encoded something verymuch like the economically defined expectedvalue or expected utility of each light in thissimple lottery task.

Subsequent studies of human decision-making using functional magnetic resonanceimaging (fMRI) have yielded similar con-clusions. Knutson and colleagues (20) haveshown, for example, that activity in the hu-man striatum is correlated with the magni-tude of the monetary reward subjects earnduring lotteries, and Paulus and colleagues(21) have shown a similar result in thehuman posterior parietal cortex. In a partic-ularly interesting study, Breiter and col-

leagues (22) (Fig. 2) presented humansubjects on sequential rounds with one ofthree possible lotteries. In lottery one (thegood lottery), they faced equal chances ofwinning /10, /2.50, or /0. In lottery two (theintermediate lottery), they faced equal chan-ces of winning /2.50, winning /0, or losing/1.50. In lottery three (the bad lottery), theyfaced equal chances of winning /0, losing/1.50, or losing /6. At the beginning of eachround, the subjects were told which lotterythey would be playing, and the averageactivity in many brain areas was simulta-neously measured. After that measurementwas complete, the lottery was actuallyplayed and the humans were then told howmuch real money they had earned on thatround. Importantly, all three of these lotteriespresent a one-third possibility of winning/0, but they do so under different conditions.In the good lottery, winning /0 is the worstpossible outcome, whereas in the bad lotteryit is the best. The psychologists Kahnemanand Tversky (23) have shown that, when ahuman participates in this good lottery, they

find winning /0 to be an intenselynegative outcome whereas whena human participates in the badlottery, they find winning /0 to bea positive outcome; subjectiveutilities are computed with regardto a reference frame. Breiter andcolleagues found that the activityof the sublenticular extendedamygdala encoded the desirabil-ity of each lottery, taking into ac-count this behaviorally describedframing effect.

Other recent neurobiologicalstudies have revealed yet other

neurally encoded variables that include thelog likelihood that a given eye movementwill result in a reward (24, 25), the veryclosely related integral of perceptual signalsindicating which saccade will be rewarded inthe Newsome task (26, 27), the average rateat which a saccade has been rewarded in therecent past (28), the instantaneous likeli-hood, or hazard, that a reinforced saccadewill be instructed (29), and combinations ofthese variables (30).

Strategic thinking. All of these resultssuggest that classical utility theory can beused as a central concept for the study ofchoice in economics, psychology, and neu-roscience. In the middle of the twentiethcentury, however, economists pushed utilitytheory beyond this boundary, enhancing itto include the study of the strategicinteractions which arise when decisionmakers confront intelligent opponents.Extending the concept of subjective utility,VonNeumann and Morgenstern (31) andNash (32, 33) developed a formal utility-based economic approach in the theory of

Good

$10

$0

$2.50

Int.

$2.50 $0

$1.50loss

Bad

$0 $1.50loss

$6loss

Fig. 2. The three lotteries used in the Breiter and colleagues experiment.

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

www.sciencemag.org SCIENCE VOL 306 15 OCTOBER 2004 449

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

games. Recently, Lee and his colleagues (34)and Dorris and Glimcher (35) have begun tolink the neurobiological corpus to thisliterature by examining the activity of singleneurons in awake-behaving monkeys en-gaged in decision-making during strategicconflicts. In Dorris and Glimcher’s study,two opponents face each other, an employerand an employee. On each round of the gamethe employee must decide whether to go towork, in which case he earns a fixed wage,or whether to shirk, in hopes of earning hiswage plus a bonus. The goal of the employeeis simply to maximize his gains in terms ofsalary and bonus. The employer, on the otherhand, must decide between trusting his em-ployee to arrive for work or spending moneyto hire an inspector who can actually checkand see whether the employee arrived forwork that day. The goal of the employer is tospend as little as possible on inspections whilemaximizing the employee’s incentive to work.

The inspection game isof particular interest to gametheorists and economists be-cause rational strategies forutility maximization duringthis strategic conflict leadto predictable outcomes ac-cording to an equilibriumtheory originally developedby John Nash in the 1950s.Nash (36) equilibrium theo-ry describes how, when thecost of inspection to theemployer is set high, theefficient strategy for bothplayers converges on a solu-tion in which the employ-ee manages to shirk fairlyoften. Conversely, a low in-spection cost to the employer defines atheoretical equilibrium solution in whichshirk rates are low. One of the fundamentalinsights this formal analysis reveals is that ata mixed strategy equilibrium, a situation inwhich a rational player should distribute hisactions amongst two or more alternatives inan unpredictable fashion, the desirability ofthe two or more actions in equilibrium mustbe equivalent. The Nash approach argues,essentially, that a behavioral equilibrium oc-curs when the desirability of working andshirking are rendered equal by the behaviorof one’s opponent irrespective of how oftenthat equilibrium requires that one work.When Dorris and Glimcher examined theactivity of neurons in the posterior parietalcortex of monkeys playing the inspectiongame, they found neurons that carried a sig-nal that behaved like relative expected util-ity. When the monkey’s behavior was wellpredicted by the Nash equations, neural ac-tivity was fixed at a single level irrespectiveof the frequency with which the monkey

chose to make a particular response, eventhough these same neurons were stronglymodulated by changes in the value of ac-tions during lottery tasks.

Research on human-human strategicinteractions that are well described byclassical game theory are also now underway in a number of laboratories (37). Likethe earlier fMRI studies of simple decision-making tasks, these experiments are alsobeginning to shape the common groundbetween economics, psychology, and neu-roscience. Taken together, these findingssuggest that at least under some circum-stances decisions may actually be made inthe primate neuro-architecture in a mannerlong suspected by economists and nowbeing actively analyzed by psychologistsand neuroscientists: Neural circuits maycompute and represent the desirability ofmaking a response. Economics, psycholo-gy, and neuroscience do seem to be

converging around a common conceptualframework. All three disciplines are begin-ning to acknowledge that decision-makinginvolves the representation of subjectivedesirabilities. The challenge that this con-vergence around a single concept poses,however, is to leverage the intersection ofthese three disciplines to explain choicebehavior that cannot be described with thecommon framework of utility theory; theseare classes of behaviors which have sty-mied traditional economics and which havelain far beyond the reach of traditionalneuroscience. If it is to be of value, the goalof a unified decision science will have to beto use all three sets of approaches simul-taneously to gain traction in this newterritory.

Beyond Classical ConceptsChoice under risk. As we have seen, theintroduction of the concept of expected utilitysolved the St. Petersburg’s puzzle and formedthe core of neoclassical economics. Sub-

sequent puzzles and paradoxes, however, haveplagued this solution. In Ellsberg’s (1961)paradox (38) (Fig. 3) you are presented withan urn, and you are told that it contains 90balls. Of these, 30 are blue, and 60 are eitherred or yellow; any proportion is possible.You are then offered a choice between alottery that pays /100 if a blue ball is drawn(a 1/3 probability) and one that pays /100 if ared ball is drawn. The probability of a reddraw is unspecified or ambiguous: It is achoice between an event with a knownprobability and an event with an unknownprobability. Under these circumstances, peo-ple typically choose the first lottery, whichwins if a blue ball is drawn. According toexpected utility theory they could only do soif they believe that there are fewer than 30red balls in the urn or, equivalently, thatthere are more than 30 yellow balls. Then(before any balls are actually drawn, but withthe same urn standing in front of you) you

are asked to choose again,this time between a lotterythat pays /100 on eitherblue or yellow and one thatpays /100 on either red oryellow. Now the likelihoodof winning is clear in thesecond case (a 2/3 proba-bility of winning /100) butunclear in the first case (aprobability between 1/3and 1). People this timetypically choose the secondlottery. The first lotteryseems less attractive, be-cause there might be toofew yellow balls. Is thereanything wrong with thisbehavior? If expected util-

ity theory is correct, then there certainly is:You cannot think that there are too few andtoo many yellow balls in the urn at the sametime.

The Ellsberg paradox is just one of manydemonstrations presented in the last half ofthe twentieth century that were consideredformal falsifications of expected utilitytheory. An even earlier example is Allais’paradox (39), based on the idea that a certainoutcome may be perceived as more desir-able, in a qualitatively different way, thanany random outcome, even if very likely(40). These examples proved that expectedutility theory as originally proposed couldnot be globally correct; at best it could onlypredict choices under some circumstances.This has led economists and social psychol-ogists both to attempt modifications toexpected utility theory and to replace itoutright. Both the modifications and replace-ments have provided important and econom-ically powerful insights into choice behaviorbut have not yet provided a global theory of

Ellsberg’s Paradox

If blue is preferred then blue/yellow should also be preferredbut it is not.30 blue

60 red+yellow

Choice 1

blue$100

-or-

Number of Balls in Urn:

red$100

30 x

Choice 2

blue/yellow$100

-or-

Number of Balls in Urn:

red/yellow$100

30+(60-x) 60

Fig. 3. Ellsberg’s paradox. If blue is preferred in choice one, then blue/yellow shouldlogically be preferred in choice two. Surprisingly, this is rarely the case.

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

15 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org450

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

choice that can truly replace expected utilitytheory.

The emerging discipline of neuroeconom-ics offers a new strategy both for testingexisting models of all types and for devel-oping new models with empirical techniques.If we succeed in understanding mechanisti-cally how choices that violate expectedutility theory are made at a neural level,then a new global theory of choice will bedeveloped. To that end, a number of labo-ratories are now beginning to reexamine theconditions under which expected utilitytheory fails.

The reason that expected utility theoryfails under some conditions may be thatchoosers use more than one evaluativemechanism at a neurobiological level (41).For example, in Dickhaut et al. (42), theprocesses involved when a certain outcome isone of the options are different fromthose involved when only random out-comes are at stake, providing an ex-planation of the Allais’ paradox citedabove.

Under many conditions thesemechanisms may work together toyield choices similar to those pre-dicted by expected utility theory butmay produce odd results when usedin isolation, in novel combinations,or in situations for which they areill suited. Recent work by Damasioand colleagues [for example, (43)]on the class of behavioral paradoxesfrom which the Ellsberg example isdrawn seem to support this conclu-sion. These studies suggest that anambiguity-sensitive mechanism as-sociated with the expression of emo-tion may reside, at least in part, inthe ventromedial prefrontal cortex(VMPFC) (Fig. 4) and may be re-sponsible for choice under some but not allconditions. These researchers and othershave shown that patients with damage tothis area have an impaired ability to makesome classes of decisions and have diffi-culties planning their work and choosingfriends. Further, the actions these individu-als do elect to pursue often lead to financialas well as personal losses. Yet despite thesespecific failures, patients with damage tothe VMPFC show normal performance onmultiple-choice tests of intelligence.

These observations and others like themhave led Damasio to propose that theinability of patients with VMPFC lesionsto make advantageous decisions undersome circumstances is caused by damageto an emotional mechanism that stores andsignals the value of future consequences ofan action, the somatic marker hypothesis.The hypothesis proposes that, because theylack this emotional mechanism, the patients

must rely on other brain mechanisms thatachieve a different analysis of the numer-ous and often conflicting options involvingboth immediate and future consequences.This other mechanism, operating alone, ishypothesized to produce decisions that areless efficient and slower than those pro-duced by a normal, intact, system.

The importance of the emotion-relatedVMPFC for regular decision-making hasbeen confirmed by experiments wheresubjects were asked to make choices amonga group of alternatives that carry a mone-tary reward (typically by selecting one cardat a time from four different decks ofcards), but for which the probability ofreward is unspecified (44). This is precise-ly the ambiguous situation that producesEllsberg’s paradox. Under these conditions,patients with VMPFC lesions seem to lack

an aversion to ambiguity or losses thatnormal subjects have, an aversion that maybe quite advantageous under many condi-tions. Further support for this hypothesiscomes from brain imaging studies. Forexample, O’Doherty et al. (45) have shownthat the VMPFC is relatively more activewhen human subjects are actively learningabout the availability of rewards and punish-ments during one of these ambiguous choicetasks.

The process may be very different,however, when subjects simply choose be-tween options without any feedback orlearning taking place at the same time. Forexample, Rustichini et al. (46) asked normalsubjects to make choices among ambiguouslotteries, risky lotteries, and certain out-comes while their brain activity was moni-tored. Subjects were paid for the outcome oftheir choices, but the outcome was commu-nicated only after the experiment was over.

Under these conditions, the VMPFC did notshow any activation; it was actually lessactive when choices were being made thanwhen subjects waited between trials. Theseresults suggest that emotional circuits maybe important in learning and processinginformation, rather than in selecting amongalternatives.

Together, these data may begin to ex-plain, in a mechanistic way, how informationis analyzed when at least one class ofbehavior which is not predicted by theexpected utility theory is produced. Theprocess of learning and evaluating feedbackmay involve emotion-related areas. Ambigu-ity aversion, whether advantageous or disad-vantageous in a particular situation, maybecome explicable as we learn more aboutthe computations that brain areas like theVMPFC perform.

Strategic cooperation. As with theEllsberg paradox, challenges have alsobeen raised recently to classical gametheory. In a path-breaking study, Guthet al. (47) analyzed the behavior ofsubjects playing the ultimatum game.In this game, a first player, the pro-poser, has /10 to split with a secondplayer. He can offer any amount be-tween zero and /10. The second playeris informed of the offer and can ac-cept or refuse. If she accepts, the splitis made. If she refuses, both playersget nothing. The prediction of a re-strictive concept of game theory, thesubgame perfect equilibrium, is thatfor any positive amount offered by theproposer, the second player knows thatshe faces a choice between gainingnothing (if she refuses the offer) orsomething (if she accepts). The pro-poser should therefore always offer theminimum possible split to player two,

who should always accept. Contrary to thisprediction, the robust experimental finding isthat low offers (typically /2 or even /3) areconsistently refused. The second player ap-pears to prefer, under these conditions, togain nothing. Anticipating this, proposers typ-ically avoid low offers.

Although expected utility theorists haveproposed some explanations for this be-havior, it may well be that by analyzingthe neural circuits active during the ulti-matum game we may be able to both ex-plain the causes of this behavior and topredict it. Studying the ultimatum game insubjects undergoing brain scans, Sanfeyet al. (48) found that offers refused by thesecond players activated specific brain cir-cuits in those players, and interestinglythese brain circuits are also associatedwith emotional arousal: the anterior insula(AI, associated with disgust, both physicaland emotional), the dorsolateral prefrontal

Fig. 4. Medial view of the left half of a human brain, with thefront of the brain on the right side of the image. The humanventromedial prefrontal cortex is shown in red.

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

www.sciencemag.org SCIENCE VOL 306 15 OCTOBER 2004 451

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from

cortex (DLPFC, associated with goal main-tenance and executive control), and the an-terior cingulate cortex (ACC, associated withdetection of cognitive conflict). Also sig-nificant is the correlation of activation withchoices: An activation of the AI is posi-tively correlated with rejection, suggest-ing that an emotional arousal associatedwith a low offer is correlated with rejec-tion. The overall picture is that offers wemight consider unfair may activate emo-tional circuits of the brain involved in thedecision to reject an offer. If we can cometo more fully understand how these cir-cuits reach this conclusion, then a behav-ior that was difficult for classical gametheory to predict may become fully explica-ble with the synthetic approach that neuro-economics provides.

A similar line of investigation has exam-ined interplayer cooperation during singlerounds of the trust game. In this game, twoplayers move sequentially. The first playercan decide to transfer a sum of money out ofan initial endowment that she receives intoan investment pool that immediately triplesin value. The second player then gains con-trol of the investment and can divide it be-tween the two players in any way he chooses.In this game, the only Nash equilibriumchoice for the first player is to transfernothing into the investment. Were she tomake any transfer, the second player shouldtake all of the money for himself. However,in real experiments the first player typicallydoes transfer a significant amount into theinvestment, and the second player recipro-cates by returning part of the pool. McCabeet al. (49) had subjects play the trust gameboth against a human opponent and againsta computer program which, they were told,would play a human-like strategy. Underthese conditions McCabe and colleaguesfound that subjects were more likely tocooperate with real humans than with com-puters and that cooperators have a signifi-cantly different brain activation in the twoconditions. Cooperation is associated withactivation of the anterior paracingulate cor-tex, a brain region associated with (50, 51)interpreting and monitoring the mental stateof others.

Although these studies are in their earlystages, they suggest the existence of specificbrain components that make specializedcontributions to decision making. The chal-lenge that these studies face is to derivedetailed computational models of the neuralmechanisms, which will make neuroeco-nomic models broadly predictive as well asexplanatory.

SummaryEconomics, psychology, and neuroscienceare converging into a single, unified field

aimed at providing a theory of humanbehavior. In this enterprise, the method andthe standard set by neuroscience is the finalgoal: a reconstruction of the process andmechanism that goes from a stimulus pre-sented to the subject to his final action inresponse. Economics provides the conceptu-al structure and the object of the analysis. Inthis emerging view, people are seen asdeciding among options on the basis of therelative desirability of each option. This istrue when they are in isolation as well aswhen they are in strategic (interaction withfew persons) and market (interaction with alarge number) environments. The recentresearch we have been surveying describeshow desirability is realized as a concreteobject, a neural signal in the human andanimal brain, rather than as a purely theoret-ical construction. Desirability is computedand is represented in the brain, and we nowhave the means to test, measure, and rep-resent this activation.

But the complete reconstruction of thedecision process, and hence of humanbehavior, is not going to be easy, becausetwo of the cornerstones of economicanalysis, subjective utility theory and Nashequilibrium, provide, even from the de-scriptive point of view, an incompletepicture. For example, desirability as repre-sented by the simple economic formalismof expected utility may be appropriate onlyin simple conditions, where ambiguity isexcluded. A more general notion is neededand, as we have seen, is beginning to beinvestigated and developed by psycholo-gists and economists working together. Thegoal of the emerging neuroeconomic pro-gram will have to be a mechanistic,behavioral, and mathematical explanationof choice that transcends the explanationsavailable to neuroscientists, psychologists,and economists working alone. Although itis unclear today how complete this expla-nation will ultimately be, neuroeconomicapproaches have already begun to yieldsubstantial fruit and to fuse natural andsocial scientific approaches to the study ofhuman behavior.

References and Notes1. E. O. Wilson, Consilience (Knopf, New York, 1998).2. D. Bernoulli, Econometrica 22, 23 (1954).3. D. Kahneman, P. Slovic, A. Tversky, Eds, Judgement

Under Uncertainty: Heuristics and Biases (CambridgeUniv. Press, Cambridge, 1982).

4. C. S. Sherrington, The Integrative Action of theNervous System (Scribner, New York, 1906).

5. I. P. Pavlov, Conditioned Reflexes an Investigation ofthe Physiological Activity of the Cerebral Cortex(Oxford Univ. Press, Oxford, 1927).

6. T. Caraco, S. Martindale, T. Whittam, Anim. Behav.28, 820 (1980).

7. A. Kacelnik, M. Bateson, Trends Cogn. Sci. 1, 304(1997).

8. A. Robson, J. Polit. Econ. 109, 900 (2001).9. H. B. Barlow, in Current Problems in Animal Behavior,

W. H. Thorpe, O. L. Zangwill, Eds. (Cambridge Univ.Press, Cambridge, 1961).

10. P. W. Glimcher, Decisions, Uncertainty and the Brain:The Science of Neuroeconomics (MIT Press, Cam-bridge, MA, 2003).

11. C. S. Sherrington, The Integrative Action of theNervous System (Scribner, New York, 1906).

12. W. T. Newsome, K. H. Britten, J. A. Movshon, Nature341, 52 (1989).

13. W. T. Newsome, K. H. Britten, C. D. Salzman, J. A.Movshon, Cold Spring Harbor Symp. Quant. Biol. 55,697 (1990).

14. See also (50) and (51) for a parallel set of studiesthat presaged and continues to parallel many ofthese developments.

15. M. N. Shadlen, K. H. Britten, W. T. Newsome, J. A.Movshon, J. Neurosci. 16, 1486 (1996).

16. R. A. Andersen, L. H. Snyder, D. C. Bradley, J. Xing,Annu. Rev. Neurosci. 20, 303 (1997).

17. C. L. Colby, M. E. Goldberg, Annu. Rev. Neurosci. 22,319 (1999).

18. M. L. Platt, P. W. Glimcher, Nature 400, 233(1999).

19. B. Knutson, C. M. Adams, G. W. Fong, D. Hommer,J. Neurosci. 21, RC159 (2001).

20. M. P. Paulus, N. Hozack, X. Zauscher, Neuroimage 13,91 (2001).

21. H. C. Breiter, I. Aharon, D. Kahneman, A. Dale, P. Shizgal,Neuron 30, 619 (2001).

22. D. Kahneman, A. Tversky, Econometrica 47, 263(1979).

23. J. I. Gold, M. N. Shadlen, Nature 404, 390 (2000).24. J. I. Gold, M. N. Shadlen, Trends Cogn. Sci. 5, 10

(2001).25. M. N. Shadlen, K. H. Britten, W. T. Newsome, J. A.

Movshon, J. Neurosci. 16, 1486 (1996).26. M. N. Shadlen, W. T. Newsome, J. Neurophysiol. 86,

1916 (2001).27. L. P. Sugrue, G. S. Corrado, W. T. Newsome, Science

304, 1782 (2004).28. P. Janssen, M. N. Shadlen, Soc. Neurosci. Abstr. 767,

2 (2003).29. B. Coe, K. Tomihara, M. Matsuzawa, O. Hikosaka,

J. Neurosci. 22, 5081 (2002).30. J. V. Von Neumann, O. Morgenstern, Theory of

Games and Economic Behavior (Princeton Univ. Press,Princeton, NJ, 1944).

31. J. F. Nash, Proc. Natl. Acad. Sci. U.S.A. 36, 48(1950).

32. J. F. Nash, Econometrica 18, 155 (1950).33. D. J. Barraclough, M. L. Conroy, D. Lee, Nat. Neurosci.

7, 404 (2004).34. M. C. Dorris, P. W. Glimcher, Neuron, (in press).35. J. F. Nash, Ann. Math. 54, 286 (1951).36. P. R. Montague et al., Neuroimage 16, 1159

(2002).37. D. Ellsberg, Q. J. Econ. 75, 643 (1961).38. M. Allais, Econometrica 21, 503 (1953).39. For a review of these issues, see D. Luce, Utility of

Gains and Losses, vol. 8 of Scientific Psychology Series(L. Erlbaum, Mahwah, NJ, 2000).

40. G. Gigerenzer, R. Selten, Bounded Rationality: TheAdaptive Toolbox (MIT Press, Cambridge, MA,2002).

41. J. Dickhaut et al., Proc. Natl. Acad. Sci. U.S.A. 100,3536 (2003).

42. See for example A. Bechara, H. Damasio, D. Tranel,A. Damasio, Science 275, 1293 (1997).

43. J. O’Doherty, M. L. Kringelbach, E. T. Rolls, J. Hornak,C. Andrews, Nat. Neurosci. 4, 95 (2001).

44. A. Rustichini, J. Dickhaut, P. Ghirardato, K. Smith,J. Pardo, in Games and Economic Behavior, inpress.

45. W. Guth, R. Schmittberger, B. Schwarze, J. Econ.Behav. Organ. 3, 347 (1982) .

46. A. Sanfey, J. K. Rilling, J. Aronson, L. E. Nystrom,J. Cohen, Science 300, 1755 (2003).

47. K. McCabe, D. Houser, L. Ryan, V. Smith, T. Trouard,Proc. Natl. Acad. Sci. U.S.A. 98, 11832 (2001) .

48. C. D. Frith, U. Frith, Science 286, 1692 (1999).49. H. L. Gallagher, C. D. Frith, Trends Cogn. Sci. 7, 77

(2003).50. C. R. Gallistel, Cognition 50, 151 (1994).51. K. Conover, P. Shizgal, in Games and Economic

Behavior, in press.

C O G N I T I O N A N D B E H A V I O RC O G N I T I O N A N D B E H A V I O R

15 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org452

SPECIA

LSECTIO

N

on

Dece

mbe

r 4, 2

006

www.

scie

ncem

ag.o

rgDo

wnlo

aded

from