Академический Документы
Профессиональный Документы
Культура Документы
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The University of Chicago Law Review is collaborating with JSTOR to digitize, preserve and extend access to
The University of Chicago Law Review.
http://www.jstor.org
Wanting,Liking,and Learning:
Neuroscience and Paternalism
Colin F Camerert
A large number of young childrendie every year from ingesting
poisons by accident.Revealed-preferencetheory,the foundationof microeconomics,has only two conceptsto explainthese accidents:beliefs
and preferences (that is, utilities revealed by choices). In the beliefpreferencelanguage,all we can say is that a child who died of accidental poisoningeither believed the poison was safe, and made a mistake,
or that the child preferreddeath to life, and committedsuicide.While
economic theory may be comfortablecallingthese accidentsonly mistakes or expressionsof preference,I am not. My goal in this Essay is to
acquaintthe legal audience with ideas emerging in neurosciencethat
could potentiallybe a richerlanguagefor talking about cases like accidental child poisoning and, more broadly,about welfare and paternalism in some limited cases. Furtherdevelopment of the framework
could lead to a broaderview with wider applicability.
The idea is that three separable neural systems are relevant for
choice and welfare:a hedonic "liking"system (welfare), a "wanting"
system that guides choice, and a learning system that, ideally, links
informationstored in the other systemsso that people choose to learn
what they truly like. Other multiple-processneuroeconomicaccounts
use differentcomponentsbut have similarimplications.'
t Rea A. and Lela G. Axline Professor of Business Economics,CaliforniaInstitute of
Technology.Thanksto conferenceparticipants,to MeghanaBhatt, MingHsu, and Ian Krajbich
for researchassistance,andto Alice Lin for a presentationon the Berridgework.
1 See Drew Fudenbergand David K. Levine,A Dual Self Modelof ImpulseControl(Aug
17, 2005), online at http://post.economics.harvard.edu/faculty/fudenberg/papers/dualself.pdf
(visin selfited Jan17,2006) (arguingthat a simple"dual-self'modelexplainsempiricalirregularities
controlproblemsand can give a valuefor commitmentin decisionproblems);IsabelleBrocasand
Juan D. Carillo,The Brain as a HierarchicalOrganization(Aug 2005), online at http://www(visitedJan 17, 2006) (constructinga model of conrcf.usc.edu/-juandc/PDFpapers/wp-brain.pdf
sumptionand time allocationthat is solved usingtools frommechanismdesignand economicsof
Decision
information);B. Douglas Bernheimand Antonio Rangel,Addictionand Cue-Triggered
Processes,94 Am Econ Rev 1558,1561(2004)(proposinga modelof addictionthatis premisedon a
pathologicaldivergencebetweenchoiceandpreferenceconsistingof mistakes,environmentalcues,
and attemptsby users to minimizetheir mistakes);George Loewensteinand Ted O'Donoghue,
and DeliberativeProcessesin EconomicBehavior(July27,2004),onlineat
AnimalSpirits:Affective
(visited Jan 17, 2006)
http://gsbwww.uchicago.edu/research/workshops/behavioral/lowenstein.pdf
(discussinga model of humanbehaviorbased on the interactionof a deliberativesystem and an
affectivesystem);George Loewenstein,Out of Control:VisceralInfluenceson Behavior,65 Org
87
88
[73:87
Neuroscience
andPaternalism
2006]
89
formation.7
Remember that Pareto did not win this argumentabout how to
proceed by mathematicalproof or data. The equation of utility with
choice was not a scientific discovery on par with a powerful theorem
or solid empiricalregularity.Pareto simplyassertedthat,as a matterof
convenience,it was okay to give up on understandingthe "essence of
things."Pareto thought economics could start with "secondaryprinciples" like utility-maximization(which, ironically,are now taught to
graduate students as "firstprinciples").Pareto's turn-the definition
of utility as a quantity revealed by expressed preference equation-5
90
[73:87
8 See Daniel Kahneman,Peter P.Wakker,and RakeshSarin,Back to Bentham?Explorationsof ExperiencedUtility,112 Q J Econ 375,376-77 (1997);Daniel Kahneman,New Challenges
2006]
Neuroscienceand Paternalism
91
92
TheUniversity
of ChicagoLawReview
[73:87
fact conformsperfectlyto the objectivefact.This can be done because we will consider only repeatedactions to be a basis for
claimingthatthereis a logical connectionunitingsuch actions.1
Pareto clearly implicates learning as a mechanism that brings
wantingand likingtogether:
A man who buys a certain food for the first time may buy more
of it than is necessary to satisfy his tastes, price taken into account. But in a second purchasehe will correct his error,in part
at least, and thus,little by little, will end up by procuringexactly
what he needs. Wewill examinethis actionat the timewhenhe has
reachedthis state. Similarly,if at first he makes a mistake in his
reasoningabout what he desires,he will rectifyit in repeatingthe
reasoningand will end up by makingit completelylogical."
Thus,the standardassumption,that choices reveal "true"utilities,
is clearly endorsed by Pareto only for choices that are repeated often
enough to ensure that learning has occurred.The wanting-learningliking frameworkcan therefore be seen as a proper generalizationof
revealed-preferencetheory and, in fact, it is perhaps the kind of general theory Pareto had in mind in the first place. Bruni and Sugden
note that the "discoveredpreferencehypothesis"is a revivalof Pareto's
idea.14They assert that rationalchoice theory appliesafterpreferences
have been "discovered"by trial-and-errorlearningand reflection.But
this leaves a large and importanthole: what happensbeforelearning?
Besides generalizing revealed-preference theory, the wantinglearning-likingframeworkprovides a potential way to talk scientifically about how restricting or imposing choices (paternalistically)
could improve welfare.The core idea is simple:if there are separate
systems for recordingliking, expressingwanting,and for learning to
want what the brain likes, then paternalismcould be justified if the
wanting system produces choices that are not later liked, and if a paternalisticcorrectionproduces choices that are unwantedby an agent
but will be liked by her, or that are wanted but not liked, and if the
correction does not cause other harms (or much harm to rational
agents).
12 Pareto,Manualof PoliticalEconomyat 103 (citedin note 6) (emphasisadded).
13 Id (emphasisadded).
14 See Bruniand Sugden,TheRoadNot Takenat 7 (cited in note 2) (labelingthe claimthat
rational-choicetheorycan predictbehaviorwhen certaincriteriaare satisfiedas the "discovered
preferencehypothesis").See also CharlesR. Plott,RationalIndividualBehaviourin Marketsand
Social ChoiceProcesses:The DiscoveredPreferenceHypothesis,in KennethJ.Arrow,et al, eds,
The RationalFoundationsof EconomicBehavior225, 226 (St. Martin's1996) (suggestinga theory thatrationalchoicesevolve throughthree stagesreflectingexperienceandpractice).
2006]
Neuroscienceand Paternalism
93
But
MutantMice Have Higher"Wanting"
15 See SusanaPecifia,et al, Hyperdopaminergic
Not "Liking"forSweetRewards,23J Neuroscience9395 (2003)(findingthatchronicallyelevated
butdoes not increase"liking").
extracellulardopaminein mutantmicefacilitates"wanting"
94
[73:87
equally fast after a lot of training,this difference duringlearningsuggests that extra dopaminefacilitatesfaster learning.16
These experiments show that the DAT-knockdownmice eat
more, and learn there's food waiting in the goal box faster than the
control mice do, but do they actuallylike food and liquid more? Here
we have no choice but to skate out onto thin empiricalice:how do you
measureliking?
If you are open-minded about the possible difference between
choice, which is observable,and liking (true utility), which is not as
readily observable,then you should accept that any measure of true
utility won't be as sharp and agreeable as observingchoice is. If you
aren'twillingto accept this sad fact,stop readingnow.
Pecifia and colleagues measureliking and dislikingusing physical
movementsof the mice--tongue protrusionsand paw lickingto represent liking,and gaping (mouth-opening,like a yawn),screwingup the
face, and arm flailing to representdisliking.Your first reaction might
be that these are not serious scientific measures.But they are taken
very seriouslyby people who study animals,because they are the best
availablemeasuresof liking"(puttingobservedchoice aside,of course;
if you use choice to infer likingwe're back to squareone with Pareto).
These expressions and movements are also very similar across mice,
primates,and humaninfants;in fact, they are so similarthat researchers use an equation (d = .26(w)32)to predict the duration(d) of a facial expression(in milliseconds)in termsof a species' body weight (w)
(in kilograms).'8
In any case, these measures of liking and dislikingshow that the
DAT-knockdownmice do not like the sucrose solution more than the
control mice. In fact, 1.0 M sucrose,the largest concentration,is liked
significantlyless by the knockdownmice."
16 See SamuelM. McClure,NathanielD. Daw, and P. Read Montague,A Computational
Substratefor IncentiveSalience,26 TRENDSin Neurosciences423,424 (2000).
17 Methodslike this are used by pediatriciansto judge how much
pain childrenare in.An
"oucher"scale is used by pediatriciansto map facialexpressionsonto a numericalscale.See, for
example,How to Use the Oucher,online at http://www.oucher.org/differences.html
(visitedJan
17,2006). See also,for example,Rita Rubin,Kids' DistressIs No SmallThing,USA TodayD1-2
(May 9, 2005). This can be thoughtof as a utilityfunctionthat mapsfacial expressionsinto numerical"truedisliking."If the goal is to prescribean optimaldose of painkiller,you can'tjust let
a child "choose"a dose. Therefore,inferringchildren'sdisutility- pain--from choice is a bad
idea.Inferringtheirpain fromfacialexpressionsis a good idea.
18 See BerridgeandRobinson,26 TRENDS in Neurosciencesat 509 (citedin note 11).
19
A concern in the study is that the eating task used regular food (Purina Rat Chow) and
water;the learningtask used Froot Loops;and the likingmeasuresused sucrosesolution.I presume Pecifiaand colleaguesare comfortablewith the workinghypothesisthat knockdownmice
do not have differenttastes for any of these substancesthan controlmice do. See also Kent C.
Berridge and Terry E. Robinson, What Is the Role of Dopamine in Reward: Hedonic Impact,
2006]
Neuroscienceand Paternalism
95
RewardLearning,or IncentiveSalience?,28 BrainRsrch Revs 309, 350 (1998) (findingthat dopamine-depletedratsstill"like"rewardsandsimplyfail to "want"rewards).
20 See Berridgeand Robinson,26 TRENDS in Neurosciencesat 508 (cited in note 11)
(discussinghow drug addictswill work for low doses of stimulantsor morphinedespite no subby Humans,157
Self-administration
jective effect). See also CarlL. Hart,et al, Methamphetamine
75, 80 (2001) ("[S]everalinvestigatorshave reporteddata demonstrating
Psychopharmacology
that drug-related'positive'subjectiveeffects do not entirely accountfor drug-takingbehavior
andsome have suggestedthatotherfactorsshouldalso be considered.").
and
21 See LisaH. Brauer,et al, HaloperidolReducesSmokingof Both Nicotine-containing
DenicotinizedCigarettes,159 Psychopharmacology31, 34-35 (2001) (discussingexperimental
resultswhere haloperidol,a dopamine-receptorantagonist,was administeredand significantly
reducedsmokingbehaviorbut did not affecthow muchthe subjectslikedsmoking).
22 See BerridgeandRobinson,26 TRENDS in Neurosciencesat 508 (citedin note 11).
23 See id at 510 (containinga schematicexplanationof regionsof the brainthat are linked
to wanting,liking,and learning).Theiranalogoustermsare motivation(wanting)and emotionor
affect(liking).Id at 508.
96
[73:87
understandingof wanting,learning,and liking could emerge for humans,probablyat firstfor a specializeddomainlike food.24
Having separate systems for wanting,learning,and liking makes
sense from an evolutionarypoint of view.Take food as a simple case.
First note that unless tastes for specific foods are truly genetically
hardwired,some kind of learning system is necessary to connect the
sensory properties of food-how it smells, feels, and looks-to how
much an animallikes specific foods.This basic learningis what infants
do for months-putting everythingin sight in theirmouths to see how
things taste.An infant who tastes dirt doesn't anticipateliking dirtshe just exhibits wanting,which is really just wanton exploration (to
create inputs for the learning system).Then she graduallylearns the
difference between brown, crumbly stuff that tastes bad-dirt-and
brown,crumblystuff that tastes good - gooey brownies.
The fact that the learning system exists at all is prima facie evidence that wanting and liking are not automaticallylinked at birth;
learningis there for a reason,and the reason is to educate the wanting
system about what is liked. But why would wantingand likingoccur in
separatebrainregions?Berridgeand Robinsonpropose that likingfor
food, sex, and warmthrely, at least to some extent, on quite different
sensory systemsfor processingsmell, heat, and taste.A separatewanting system is needed to combine these signals into a "common currency,"25in order to make tradeoffs between goods that are liked in
(sensorily)differentways.
Wantingand liking may also deviate, usefully,in the special case
of pregnancy.When my wife was pregnant,she had a fierce cravingfor
lamb-one of her favorite foods-then almost vomited after taking
one bite. Her wantingsystem had tuned to her own tastes but the circuitryprotectingthe delicate fetus registeredrapiddislike.
Learningis the mechanismthat trainswantingabout what is actuliked
However,in the modem economy,manyof
ally
by trial-and-error.
the goods people purchaseare muchmore complexto processand rep24 A similarbody of researchin humansfocuses attentionon "actor-critic"
modelsof temporal difference(TD) learning.In TD learning,an organismlearnsa value functionaccordingto
an adaptive updatingequation driven by the predictionerror (or temporaldifference).John
O'Doherty and colleagues have located areas of striatum(in the brain'stemporallobe) that
correspondto "critic"temporal differences,and other areas that correspondto an "actor's"
expectedrewardor forecast.JohnO'Doherty,et al, DissociableRoles of Ventraland DorsalStriatum in InstrumentalConditioning,304 Science 452 (2004). In terms of this Essay,the actor is
wanting,andthe criticsystemis learning,whichadjustswantingso it learnsthe truevalue(liking).
25 See Peter Shizgal,On the Neural Computationof Utility:Implicationsfrom Studiesof
Brain StimulationReward,in Daniel Kahneman,Ed Diener, and Norbert Schwarz,eds, WellBeing:The Foundationsof HedonicPsychology500, 509-10 (RussellSage 1999) (discussingthe
necessity and existence of a currencyfunction to break down differentstimuli into decision
utility(wanting)).
Neuroscienceand Paternalism
2006]
97
A. Disorders
The most compellingexamples of wanting-likinggaps are pathological disordersthat people clearlywould like to changeif they could.
Obsessive-compulsivedisorder (OCD) is a psychiatricdisorder
characterizedby obsessive thoughts and compulsive actions,such as
cleaning, checking, counting, or hoarding.26"Cleaners"report an intense feeling that their hands are not clean, and must be washed over
and over to reduce that feeling. "Checking"is similar.My father used
to compulsivelycheck the four oven stove dials in our house before
leaving the house-he would point to each and say "off, off, off, off,"
partly coping with embarrassmentby makingfun of his mild checking
compulsion.Sometimes he would stop the car, halfwayout the driveway,and get out to double-checkthat the front door was locked.
In the revealed-preferencesapproach,we wouldinfer that compulsive cleanershave a high utilityfor clean hands.Some sensiblepredictions may come from this inference--for example,cleanersspend a larger incomeshareon cleaningthannoncleanersdo.
But the revealed-preferencesview only gets you so far. If you
give compulsive cleaners a serotonin reuptake inhibitor (SSRI) like
fluoxetine (Prozac) or paroxetine (Paxil), a majorityof cleaners find
26 Note that the DAT knockdownmutant mice exhibit "sequentialsuper-stereotypy"
in
groomingthat is remarkablysimilarto OCD cleaningin humans.See Kent C. Berridge,et al,
Mutant
of an InstinctiveFixedAction Patternin Hyper-dopaminergic
SequentialSuper-stereotypy
Mice:A Model of ObsessiveCompulsiveDisorderand Tourette's,3(4) BMC Biology (2005),
online at http://www.biomedcentral.com/content/pdf/1741-7007-3-4.pdf
(visitedJan17,2006).
98
[73:87
27 See LorrinM. Koran,et al, CitalopramTreatmentof CompulsiveShopping:An OpenLabel Study,63 J Clin Psychiatry704, 704 (2002) (showingthat treatmentwith the antidepressantcitalopram(Celexa)reducesshoppingcompulsion).
28 StefanoDella Vigna and Ulrike Malmendier,PayingNot to Go to the Gym22 (unpublished manuscript2005), online at http://emlab.berkeley.edu/users/sdellavi/wp/gymemp05-0420.pdf (visited Jan 17, 2006) (explainingthe irrationalbehaviorof gym users' contractchoices
basedon theiroverestimationof futureefficiencyor self-control).
2006]
Neuroscienceand Paternalism
99
100
[73:87
2006]
Neuroscienceand Paternalism
101
102
[73:87
system likes and if the interventioncreates more liking than the person would achieve on his own or with market-suppliedhelp.40
Ideally,such paternalismwould be "libertarian"in the sense that
it would only intervenein choice when some kind of regulationis necessary (for example,setting a default),4'and would be "asymmetric"in
the sense that little harmwould be done to people whose wantingand
likingsystemsare well-synched.42
Paternalismcould take two mild forms,licensingand dramatizing.
"Licensing"is checkingwhetherwantingand likingare sufficientlysynchronizedto permit a person to make choices on his own (presumably
in a particulardomain)."Dramatizing"is using mechanismsto remind
the wantingsystem,at the time of choice,aboutlikingthat may be overlooked or missynchronizedwithwanting.
A. Licensing
Driving tests provide a model for how licensingmight occur,as a
way to certify sufficient education of the wanting system. Because
adolescents develop the hand-eye coordinationand patience to drive
safely at different ages, and learningto drive takes time and supervision, the currentpolicy makes sense: prohibit drivingby all children
up to a certain age;43 issue learner's permits that allow driving only
when a supervisingadult is present (so the learnerscan learn); then
force would-be driversto pass written and physicaldrivingtests,preferably with periodic retesting and other constraints(such as license
suspensionafter accidents).
One can imagine a "credit-cardpermit"test that is conceptually
similarto a drivingtest.Treata creditcardas a piece of equipmentthat
40 A shadowis alwayscast on this
type of analysisby the naggingquestion,"Howcan we
be so sure that theywon'tlike it?"Thereare two centralproblems:a slipperyslope,and capture.
The slipperyslope argumentis that interferingwith OCD patientsmightmake it legally or politicallyeasier for more interferencein similarbehaviorsthat are less scientificallygroundedas
disorders.The captureconcernis that paternalisticpolicies will be guided or shaped (in their
detail) by special interestswho either profit from interference(for example,prohibitingprice
advertisingfor eyeglasses)or who want to "saveothers"(for example,religiousfanatics).A full
analysisof any policyshould,of course,anticipateboth slipperyslope escalationandcapture.It is
too muchto ask at this earlystage,however,when detailsof even a singlepolicy are speculative,
to anticipatethese problemsandrespondto them,at least in this shortEssay.
41 See generallyRichardH. Thalerand Cass R. Sunstein,LibertarianPaternalism,93 Am
Econ Rev 175 (2003);CassR. Sunsteinand RichardH. Thaler,LibertarianPaternalismIs Not an
Oxymoron,70 U Chi L Rev 1159(2003).
42 See generallyColin
BehavioralEconomics
Camerer,et al, Regulationfor Conservatives:
and the Casefor "Asymmetric
151 U Pa L Rev 1211(2003).
Paternalism,"
43 Some states have a "farmlicense"that allows very young childrento drive in certain
situations.For example,in North Dakota,fourteen-year-olds
can drive farmvehicleswithin150
miles of their farm.LicenseClassifications(North Dakota Dept of Transportation
2005), online
at http://www.state.nd.us/dot/dlclass.html
(visitedJan17,2006).
Neuroscienceand Paternalism
2006]
103
104
[73:87
2006]
Neuroscienceand Paternalism
105
106
[73:87
2006]
Neuroscienceand Paternalism
107
108
[73:87
2006]
Neuroscienceand Paternalism
109
The idea in this highly speculativeEssay is simple:there is evidence from mice with gene "knockdowns"that wanting (choice) and
liking (hedonic reaction) are dissociated.There are many types of evidence with humanstoo--like addictswho report a compulsionto use
drugs,but say they get no pleasure (admittedly,all this evidence relies
on directmeasurementof liking).
This type of evidence invites a wanting-learning-liking
interpretais
and
tion of choice and utility.Wantingguides choice,
liking the same
as the hedonic (Benthamite)conceptof utility.Learningis a processby
whichthe wantingsystemcomes to know whatthe likingsystemlikes.
A historicaldiscursionon the developmentof revealed-preference
theory54is a reminderthat equating unobservedutility with observed
choice--assuming wanting forecasts liking correctly--was a conventional simplification,not the result of any proof or data. At the same
time, Pareto, who advocated such a simplification,readily admitted
that assumingthat choices match the "subjectivefact"of utility is justified by restrictingattention to repeated actions,where learning has
had a chance to work. (His view is echoed by recent advocates of a
"discoveredpreference hypothesis.")By emphasizingrepetition and
learning as the conditions under which wanting and liking coincide,
Pareto is actuallya surprisingprogenitorfor the ideas in this Essay.
Distinguishingwanting, liking, and learning does two things: It
generalizesrevealed-preferencetheory,because conventionalrevealedpreferenceis the specialcase where learninghas taughtwantingwhat is
liked. And it provides a languagefor talkingabout paternalism.Government paternalismis justified if there is a verifiable gap between
wantingand liking,marketsdo not close the gap, and no other harms
are created.
Here are a few very roughwayspaternalismmightwork to equate
wantingand liking:
53 See GruberandKoszegi,116 Q J Econ at 1285-93(citedin note 35).
54 See Bruniand Sugden,TheRoadNot Taken1 (citedin note 2).
110
[73:87