ContentslistsavailableatScienceDirect
Neuroscience
Research
j ou rn a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / n e u r e s
Review
article
Development
of
artificial
empathy
Minoru
Asada
∗DepartmentofAdaptiveMachineSystems,GraduateSchoolofEngineering,OsakaUniversity,Yamadaoka2-1,Suita,Osaka565-0871,Japan
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received11August2014
Receivedinrevisedform11October2014 Accepted14October2014
Availableonline10December2014 Keywords: Artificialempathy Emotionalcontagion Motormimicry Cognitive/affectiveempathy
a
b
s
t
r
a
c
t
Wehavebeenadvocatingcognitivedevelopmentalroboticstoobtainnewinsightintothedevelopmentof
humancognitivefunctionsbyutilizingsyntheticandconstructiveapproaches.Amongthedifferent
emo-tionalfunctions,empathyisdifficulttomodel,butessentialforrobotstobesocialagentsinoursociety.
Inmypreviousreviewonartificialempathy(Asada,2014b),Iproposedaconceptualmodelforempathy
developmentbeginningwithemotionalcontagiontoenvy/schadenfreudealongwithself/other
differ-entiation.Inthisarticle,thefocusisontwoaspectsofthisdevelopmentalprocess,emotionalcontagion
inrelationtomotormimicry,andcognitive/affectiveaspectsoftheempathy.Itbeginswithasummary
ofthepreviousreview(Asada,2014b)andanintroductiontoaffectivedevelopmentalroboticsasapart
ofcognitivedevelopmentalroboticsfocusingontheaffectiveaspects.Thisisfollowedbyareviewand
discussiononseveralapproachesfortwofocusedaspectsofaffectivedevelopmentalrobotics.Finally,
futureissuesinvolvedinthedevelopmentofamoreauthenticformofartificialempathyarediscussed.
©2014TheAuthors.PublishedbyElsevierIrelandLtd.ThisisanopenaccessarticleundertheCC
BY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Contents
1. Introduction... 41
2. Developmentandevolutionofempathy... 42
2.1. Fromemotionalcontagiontoenvy/schadenfreude... 42
2.2. Schematicdiagramofempathy-relatedterms... 42
3. Affectivedevelopmentalrobotics... 43
3.1. ConceptsandapproachesofADR... 43
3.2. Developmentofself-othercognition... 43
3.3. Relationshipbetweendevelopmentofself-othercognitionandempathy... 43
4. Twokeyaspectstowardsartificialempathy... 44
4.1. Theconnectionbetweenemotionalcontagionandmotormimicry... 44
4.2. Therelationshipbetweenemotionalandcognitiveempathy ... 44
5. SeveralapproachesinADR/CDR... 45
5.1. Emotionalcontagionandmotormimicry... 46
5.2. Emotionalandcognitiveempathy... 46
5.3. Expressions ... 47
5.4. Socialbehavioranalysis... 47
6. Discussion... 48
Acknowledgements... 49
References... 49
1. Introduction
Empathicbehaviors towardshumansare expectedfor social robotstorealizetruecommunication,andseveralattemptshave
∗ Tel.:+81668797347.
E-mailaddress:[email protected]
been made to address specific contexts (e.g., see Leite et al., 2013 for survey of context specific behavior design). These attempts haverevealed thatthescopeforempathic interaction seems limited and is difficult to extend (generalize) to differ-ent contexts. The importance of “affectivity” in human–robot interactions (HRIs) was recently addressed in a brief sur-vey from the viewpoint of affective computing (Riek and Robinson,2009).However,adeeperunderstandingofitsmeaning http://dx.doi.org/10.1016/j.neures.2014.12.002
0168-0102/©2014TheAuthors.PublishedbyElsevierIrelandLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Empathic behaviors are thoughtto be learntthrough social interactionswithhumansintheframeworkof(cognitive) develop-mentalrobotics(Lungarellaetal.,2003;Asadaetal.,2009).Asada etal.(2009)hypothesizedthataffectivedevelopmentwasapart of cognitivedevelopment, and withregard toaffective aspects theyargued“artificialsympathy”(Asadaetal.,2012)andproposed classifyingaffectivedevelopmentalrobotics aspartofcognitive developmental robotics to accurately design artificial empathy (Asada,2014a,b).Amongtheseveralissuesarguedinthe previ-ousreview (Asada,2014b), Ifocus ontwoaspects indesigning thedevelopmentofartificialempathyalongwithseveralpossible approaches.
Therestofthearticleisorganizedasfollows.Thenextsection providesasummaryofthereview(Asada,2014b)beginningwith emotionalcontagiontoenvy/schadenfreude,followedbyan intro-ductiontoaffectivedevelopmentalroboticsasanapproachtothe developmentaldesignofartificialempathy.Next,areviewand dis-cussiononseveralapproachesfortwofocusedaspectsofaffective developmentalrobotics,emotionalcontagioninrelationtomotor mimicryandcognitive/affectiveaspectsofempathyfollow.Finally, futureissuesregardingdevelopmentofamoreauthenticformof artificialempathyarediscussed.
2. Developmentandevolutionofempathy
Aprevious studyattempted todefine“empathy” and “sym-pathy”toclarifytheapproach fordesigning anartificialsystem becausethesetwo emotions areoften mistakenfor each other (Asada et al., 2012). However, their attempt did not accu-rately discriminate between these terms from a neuroscience andbiobehavioralperspective.Asadapublisheda moreaccurate review of artificial empathy (Asada, 2014b), based on reviews fromneuroscienceperspectivesthatincludeontogeny,phylogeny, brainmechanisms,context,andpsychopathologyasoutlinedby Gonzalez-Liencresaetal.(2013).
First, theyclaimed that themanifold facets of empathy are explored in neuroscience from simple emotional contagion to higher cognitive perspective-taking, and a distinct neural net-workof empathy comprises both phylogenetically olderlimbic structuresandneocorticalbrainareas. Thesesuggestthat emo-tionalcontagionismainlybasedonphylogeneticallyolderlimbic structures,whilehighercognitiveperspective-takingisbasedon neocorticalbrainareas.Next,theypointedoutthatneuropeptides suchasoxytocinandvasopressin,aswellasopioidergicsubstances, playaroleinmodulatingempathy.Thesekindsofneuromodulation mayregulatelevelsofempathybothpositivelyandnegatively.
Awidedefinitionofempathymayencompassfromemotional contagion to envy/schadenfreude as mentioned in Section2.1. However,arestricteddefinitionofempathyissimplytheability toformanembodiedrepresentationofanother’semotionalstate whilesimultaneouslybeingawareofthecausalmechanismthat inducedthatemotionalstate(Gonzalez-Liencresaetal.,2013).This suggeststhattheempathizerhasinteroceptiveawarenessofhisor herownbodilystatesandisabletodistinguishbetweentheself andothers,whichistheunderlyingprincipleofempathy-related termsfromanevolutionaryviewpoint.Inthisregards,thenarrow definitionmayholdtrueonlyforcognitiveempathy.
2.1. Fromemotionalcontagiontoenvy/schadenfreude
Emotionalcontagionisanevolutionaryprecursorthatenables animals to share their emotional states. However, animals do not understand what aroused an emotional state in another.
suchasemotional/cognitiveempathy,sympathy,andsoon. Bothemotionalandcognitiveempathy(EEandCE)mayonly occurinanimalswithself-awarenesssuchasprimates,and per-haps,elephants,anddolphinsalthoughpresenceofself-awareness inthesespeciesremainscontroversial.1Neuralrepresentationsfor
suchcomplexemotionsand self-awarenessare localizedinthe anteriorcingulatecortexandtheanteriorinsula(Craig,2003).EEis anolderphylogenetictraitthanCE,andallowsindividualstoforma representationofothers’feelingsbysharingthesefeelingsthrough embodiedsimulation,aprocessthatistriggeredbyemotional con-tagion.While,EEconsiderablyoverlapsindefinitionaltermswith “theoryof mind”(Premack andWoodruff, 1978)which present inapesandhumans(Edgaretal.,2012),andrequiresperspective takingandmentalizing(deWaal,2008).
Unlike emotional contagionthat does notrequire reasoning about the cause of aroused emotions in others, both EE and CE require distinction between one’s own and others’ mental states and forms of a representation of one’s own embodied emotions.
Sympathyandcompassionseemsimilartoempathyinterms ofemotionalstates,butdifferintermsoftheresponsesproduced in reference to others’ emotional states. Both requirethe abil-itytoformrepresentationsofothers’emotions,eventhoughthe emotionisnotnecessarilyshared;however,inempathy,the emo-tionalstatesaresynchronized(Goetzetal.,2010).Thisimpliesthat sympathyandcompassionmayrequirethecontrolofone’sown emotionsin additionto this self-otherdiscrimination (emotion regulation).
Thereseemtobetwokindsofemotionregulationextensions. In in-group/out-group cognition sympathy and/or compassion is shown toward in-group members, whereas opposite feel-ingsareexperiencedinresponsetoout-groupmembers.Thisis calledenvy/schadenfreudeandevolvedinresponsetothe selec-tionpressure ofsocial coherenceamongearlyhunter-gatherers (Gonzalez-Liencresaetal.,2013).
Thesecondtyperequiresmetacognition,whichrealizesakind ofvicariousness,thatis,animaginationoftheselfasothers.A typ-icalexampleisasituationwhereweenjoysadmusic(Kawakami etal.,2013a,b).Theobjective-(virtualized)selfperceivessadmusic assad(perceivedemotion),whilethesubjective-selffeels pleas-antemotionsbylisteningtosuchmusic(feltemotion).Thisseems to be a form of emotion control by metacognition of the self asothers.
2.2. Schematicdiagramofempathy-relatedterms
Fig.1 shows a schematicdepiction of theterminology used in the context of empathy thus far. The horizontal axis indi-cates the “conscious level” spanning from “unconscious” (left) to “conscious with self-other distinction” (right). The vertical axisindicates“physical/motor”(bottom)and“emotional/mental” (top) contrasts. Generally,theseaxes showdiscrete levels such as “conscious/unconscious”or “physical/mental.”However, ter-minology in the context of empathy could be distributed in the zones where discrimination between these dichotomies could be difficult. In addition, there are three points of mention:
• Inthisrepresentation,thelocationindicatestherelativeweight betweenbothdichotomies,andthearrowtotheleft(thetop)
1e.g., visit http://www.world-of-lucid-dreaming.com/10-animals-with-self-awareness.html.
Fig.1. Schematicdepictionofempathyterminology. AdaptedfromFig.2inAsada(2014b).
impliesthat theconscious(mental)levelincludes the uncon-scious(physical)one.Inotherwords,theconscious(mental)level existsontheunconscious(physical)levelbutnotviceversa. • The direction from left (bottom) to right (top) indicatesthe
evolutionaryprocess,as wellasthedevelopmental processif “ontogenyrecapitulatesphylogeny.”Therefore,awholestoryof empathyfollowsagentleslopefromthebottom-lefttothe top-right.
• Theabove arrow alsoindicatesthedevelopmentalprocess of self/otherdifferentiation(deWaal,2008), thedetailsofwhich arediscussedinAsada(2014b).
3. Affectivedevelopmentalrobotics
Asadaetal.advocatedcognitivedevelopmentalrobotics(CDR) (Asadaetal.,2001,2009),assumingthatempathydevelopment couldbeapartofCDR.Actually,onesurvey(Asadaetal.,2009) introduceda study of empathicdevelopment (Watanabe et al., 2007)asanexampleofCDR.Forourpurposes,wewillrephrasea partofCDRasaffectivedevelopmentalrobotics(ADR).2Therefore,
ADRfollowstheapproachofCDR,particularlyfocusingonaffective development.First,wegiveabriefoverviewofADRfollowingCDR andthendiscusshowtoapproachissuesofempathicdevelopment. 3.1. ConceptsandapproachesofADR
Based on theassumptions of CDR, ADR aims to understand humanaffectivedevelopmentalprocesseswithsyntheticor con-structive approaches. Its core idea is “physical embodiment,” andmoreimportantly,“socialinteraction”thatenables informa-tionstructuringthroughinteractionswiththeenvironment.This includesother agents, and affective developmentis thoughtto seamlessly connect both aspects. The most related disciplines areneuroscience(microstructureforinternalmechanisms) and developmental psychology(macro structure based on behavior observations),aspectsbothhavebeenoverlappingwitheachother insocialneuroscience.
Two main approaches are (A) computational model con-struction for affective development including verification with virtual/realagentsand(B)offeringnewmeansordatatobetter understand the human developmental process, including pro-vidinga robotasa reliablereproductiontool in(psychological) experiments.
2ADRstartsfromapartofCDRbutisexpectedtoextendbeyondthecurrentscope ofCDR.
Fig.2.Thedevelopmentalprocessofestablishingtheconceptoftheselfandother(s) andtheirexpectedunderlyingmechanisms.
3.2. Developmentofself-othercognition
Thetop panelof Fig.2shows thedevelopmental processof establishing the concepts of self and other(s), partially follow-ing Neisser’s definition of the “self” (Neisser, 1993). The term “synchronization”is usedtoexplainhow thisconceptdevelops through interactions with the external world, including other agents. AsFig.1 shows, thetargetfor synchronization changes startingfromphysicalobjects,then,other’smovements,andfinally other’smentalstates.Accordingly,thebehaviorchangesfrom phys-icalprimitivemovementssuchasearlyrhythmicmotions,actions consistingofprimitivemovements,andactionscausedbymental statessuchasempathic/sympatheticfacialexpressionsandeven altruisticbehaviour.Wehypothesizethatthethreestagesof self-developmentthatareactuallyseamlesslyconnected.Moredetails ofthesestagesprocessarediscussedin(Asada,2011).
3.3. Relationshipbetweendevelopmentofself-othercognition andempathy
DuringthethreestagesinFig.2,synchronizationsindifferent levelsoccur.
1.Emotionalcontagion:asdescribedinlatersections,emotional contagioniscloselyrelatedtomotormimicry,anautomatic reac-tionofimitatingtheother’sprimitivemovements.Therefore, thelevelofthistypeofsynchronizationisintermsofprimitive movementssuchascyclicones.
2.Emotionaland cognitiveempathy: through emotional conta-gion(motormimicry)anddevelopmentofself-awarenessand perspective taking,a varietyofmental states areinducedby observing (estimating) theother’s mental state. Theinduced mentalstateisthesameasthatoftheothers.Inthissense,this typeofthesynchronizationisintermsofthementalstate. 3.Sympathy and compassion: unlike emotional and cognitive
empathy,theinducedemotionalstatesofsympathyand com-passionaredifferentfromtheother’sone.Thismeansthatonce theagentunderstandstheother’sstate(synchronization),and thenheorshefeelsdifferentones,suchaspityorsorrow (de-synchronization).
The formation of these different (de-)synchronizations accom-panies the developmental process of self-other cognition as mentionedbelow.
Fig. 2 (bottom) indicates the mechanisms corresponding to these three stages. The common structure is a mechanism of “entrainment.”Thetargetwithwhichtheagentsynchronizesmay changefromobjectstoothers,andalongwiththesechanges,more
In the first stage, a simple synchronization with objects is realized,whileinthesecondstage,acaregiverinitiatesthe syn-chronization.Arepresentationoftheagent(agency)isgradually established,and a substructureof inhibition is added for turn-taking. Finally, more synchronization control skill is added to switch from one person (e.g., father) to another (e.g., mother) asmentionedabove.Imaginaryactionstowardobjectscouldbe realizedbasedonthesophisticatedskillofswitching.These sub-structuresarenotaddedbutareexpectedtoemergefromprevious stages.
Table1showsthesummaryoftherelationshipsamong self-development, self/other discrimination (and its requirements), empathyterminology,andimitationterminology(Asada,2014b). The imitation terminology comes from de Waal (2008) which claimstheparallelevolutionofimitationandempathyalongwith self/otherdiscrimination,andweapplythisprocesstothe devel-opmentalprocessofempathy.
4. Twokeyaspectstowardsartificialempathy
Fromtheperspectiveofaffectivedevelopmentalrobotics,the earlytwo stages areimportantfor designingartificialempathy. Theseareemotionalcontagionandemotional/cognitiveempathy. 4.1. Theconnectionbetweenemotionalcontagionandmotor mimicry
AsmentionedinSection2.1,emotionalcontagionisautomatic (unconscious). An example is an experiment with mice where onemouse-Aobservesanothermousereceivinganelectricshock accompaniedbyatone.Eventually,mouse-Afreezesinresponse tothetoneeventhoughithadneverexperiencedtheshock(Chen etal.,2009).Here,thefreezingbehavioristriggeredbyitsemotional reactionandmightbeinterpretedasasignofemotionalcontagion. deWaal(2008)proposedtheevolutionaryprocessofempathy inparallelwiththatofimitationstartingfromemotionalcontagion andmotormimicry.Bothemotionalcontagionandmotormimicry arebasedonatypeofmatchingreferredtoasperception–action matching(PAM).Beyondtheprecisedefinitions of otherterms, motormimicryrequiresasortofresonancemechanismfromthe physicalbodythatsuppliesafundamentalstructureforemotional contagion(Uitholetal.,2011),andtherelationshipbetweenmotor mimicryandemotionalcontagionisthefirstkeystagetoconnect physicalandmentalsynchronizations.
Hatfieldet al. (2000)defined primitive emotional contagion asthe tendencyto automaticallymimic and synchronize facial expressions,vocalizations,postures,andmovementswiththoseof anotherperson’sand,consequently,toconvergeemotionally.Later, Hatfieldetal.(2009)proposedamechanismofemotionalcontagion consistingofthreephases:mimicry,feedback,andcontagion.Based onthismechanism,peopletendtoautomaticallymimicthefacial expressions,vocalexpressions,postures,andinstrumental behav-iorsofthosearoundthem,andtherebytofeelaweakreflectionof others’emotions,andfinallyexperienceothersemotions.
Sonnby-Borgstrom(2002)proposedthatmimicryenablesone toautomatically(unconsciously)shareandunderstandanother’s emotionsregardlessofconsciousinterpretationoftheemotional situation. Upon classification of subjects into high- and low-empathic groups and measuring facial electromyogram (EMG) responsesuponexposureto“happy”and“angry”facepictures,the high-empathicgroupsexhibitedapositivecorrelationbetweenthe exposedpictureandtheirinducedemotion.Similarsituationscan beseenincaseofthechameleoneffect,whichreferstounconscious
passivelyandunintentionallychangestomatchthatofothersin one’scurrentsocialenvironment.Thiseffectissupposedtoactasa socialgluebecauseunconsciousbehavioralmimicryincreases affil-iation,whichservestofosterrelationshipswithothers(Chartrand andBargh,1999;Lakinetal.,2003).Ontheotherhand,subjects ofthelow-empathicgroupexhibitedinverted“smilingreactions” inresponseto“angry”facepicturesindicatingthatthesereactions throughfacialfeedbackmayactasadefenseagainstnegative feel-ings.
Initialevidenceforinvoluntarypupillarycontagionwasfound inafunctionalmagneticresonanceimaging(fMRI)study(Harrison etal.,2006).Participantswerepresentedwithphotosofsadfaces withvariouspupil sizes.Their ownpupil sizewassignificantly smallerwhentheyviewedsadfaceswithsmall,ratherthanlarge pupils,andtheEdinger–Westphalnucleusinthebrainstem,which controlspupilsize, wasspecificallyengaged bythis contagious effect.Activationin thissubcorticalstructure providesevidence that pupillary contagion occurs outside of awareness and may representa precursor ofempathy. Theirdata identifythe neu-ralsubstratesthroughwhichsubconsciousmirroringofanother’s pupilsizemayenhanceempathicappraisalandunderstandingof theirfeelingsofsadness.Thisstudyalsoshowsthestrongoverlap betweenmimicryandemotionalcontagion.
Theseobservationsseem toassume thatmimicry represents somesortofautomaticorhardwiredmotorresonancewithanother person’saffectivedisplay.However,aninfluenceoftop-down pro-cessesonmimicryissuggested.Therefore,mimicryseemstoserve asocialfunctioninincreasingrapportandfondnessbetweenself andothers(SingerandLamm,2009).
Therearecasesinwhichmimicryoccurswithoutanemotional component and other cases in which emotions are automati-callyelicitedby observingothers’ emotional stateswithoutthe involvementofmotormimicry.Therefore,mimicryandemotional contagionareregardedasimportant,yetdistinctandneither neces-sarynorsufficientprocessesfortheexperienceofempathy(Singer andLamm,2009).
Sperryarguedthattheperception–actioncycleisthe fundamen-tallogicofthenervoussystem(Sperry,1952).Perceptionandaction processesarefunctionallyintertwined;perceptionisameansto actionandviceversa.
Thediscoveryofmirrorneuronsintheventralpremotorand parietalcorticesofthemacaquemonkey(Galleseetal.,1996) pro-videdneurophysiologicalevidence fordirect matching between actionperceptionandactionproduction(RizzolattiandSinigaglia, 2008).TheMNSseemscloselyrelatedtomotormimicrybecauseit recognizesanactionperformedbyanotherandproducesthesame action,whichisreferredtoasmotorresonancethatcouldinduce emotionalcontagion.Furthermore,this relatestoself-other dis-crimination,actionunderstanding,jointattention,imitation,and theoryofmind(adetaileddiscussionisgiveninAsada,2011).
In humans, motor resonance in thepremotor and posterior parietalcorticesoccurswhenparticipantsobserveorproduce goal-directedactions(Grezesetal.,2003).Thistypeofamotorresonance systemseemshardwiredoratleastfunctionalveryearlyinlife (Sommervilleetal.,2005).
4.2. Therelationshipbetweenemotionalandcognitiveempathy Accordingtode Waal’smodel (de Waal, 2008), EEdevelops (evolves)earlierthanCE;therefore,thelattermightinvolvethe formerinatypeofhierarchy(seeFig.1).However,severalstudies havereportedthattheirrelationshipisnotinclusiveandthatthey aredifferentsystems,havingdifferentroles,and locatedwithin differentbrainregions.
Table1
Summaryoftherelationshipamongself-development,self/otherdiscrimination,empathyterminology,andimitationterminology(Asada,2014b).
Self-development(based onNeisser,1993)
Self/otherdiscrimination(Asada, 2014a)anditsrequirements(–)
Empathyterminology (Gonzalez-Liencresaetal.,2013)
Imitationterminology(deWaal, 2008)
Ecologicalself(Kuniyoshi andSangawa,2006;Mori andKuniyoshi,2007)
Nodiscrimination
–Primaryemotions(Russell,1980) Self/non-self-discrimination –MNS/motorresonancearchitecture
Emotionalcontagion(Chenetal., 2009;deWaal,2008;Chartrand andBargh,1999)
Motormimicry(Galleseetal., 1996;RizzolattiandSinigaglia, 2008)
Motorresonance(Sommerville etal.,2005;Agnewetal.,2007) Interpersonalself
(Meltzoff,2007;Nagai andRohlfing,2009;Inui, 2013;Kuhletal.,1997)
Self-awareness
–Differentiationofprimaryemotion Completeself/others
discrimination
–Perspectivetaking(Molland Tomasello,2006)andToM (PremackandWoodruff,1978)
Emotionalempathy(Craig,2003; Shamay-Tsooryetal.,2009) Cognitiveempathy(Premackand Woodruff,1978;Edgaretal.,2012; deWaal,2008;Smith,2006)
Coordination Sharedgoals
Socialself(Asada,2011) Emotionregulationofselfasothers –Emotionregulation
Metacognitionofselfasothers –Metacognition(Schraw,1998) In-group/out-groupemotion control
–Developmentofsocialandmore vicariousemotion(Amodioand Frith,2006)
Sympathy/compassion(Goetz etal.,2010)
Perceived/feltemotion(Kawakami etal.,2013a,b)
Envy/schadenfreude
(Gonzalez-Liencresaetal.,2013)
Emulation
Imitation(deWaal,2008)
AsmentionedinSection2.1,neuralrepresentationsforcomplex emotionsandself-awarenessarelocalizedintheanterior cingu-late cortexand anterior insula(Craig, 2003).Bushet al.(2000) reviewed neuroimagingstudiesof theanterior cingulatecortex (ACC)andhypothesizedthattheACCisapartofacircuitinvolved inaformofattentionthatservestoregulatebothcognitiveand emotional processing. Its two major subdivisions have distinct functions.Theseincludeadorsalcognitivedivision(ACcd)anda rostral–ventralaffectivedivision(ACad,rostralandventralareas). Theycarriedoutameta-analysisof reviewedstudies,and illus-tratedthatcontrolsforaffectiveandcognitivetasksareseparately locatedinthesetworegions.
Shamay-Tsooryetal.(2009)foundthatpatientswithlesionsin theventromedialprefrontalcortex(VMPFC)exhibitdeficitsinCE andtheoryofmind(ToM),whilepatientswithlesionsintheinferior frontalgyrus(IFG)showimpairedEEandemotionrecognition.For instance,Brodmannarea44(inthefrontalcortex,anteriortothe premotorcortex)wasfoundtobecrucialforEE,andthesamearea waspreviouslyidentifiedaspartoftheMNSinhumans(Rizzolatti, 2005).Shamay-Tsooryet al.(2009)summarized thedifferences betweenthesetwoseparatesystems(seeTable2).
Fanaet al.(2011)reportedsomedifferencesintherecruited brain regions when subjectswere unaware of being tested for empathy(whichtheycalledaffective-perceptual),thanwhenthey wereaskedtopayattentiontoempathy-relatedcues (cognitive-evaluative).Whilebothconditionsactivatedtheabove-mentioned
Table2
Twoseparatesystemsforemotionalandcognitivebasedempathy.
Emotionalempathy Cognitiveempathy
Simulationsystem Mentalizingsystem
Emotionalcontagion Perspective-taking
Personaldistress Imagination(ofemotionalfutureoutcomes)
Empathicconcern Theoryofmind
Emotionrecognition
Corestructure Corestructure
IFGBA44 VMBA10,11
Development Development
Infants Children/adolescents
Phylogenetics Phylogenetics
Rodents,birds Chimpanzees
AdoptedfromFig.6inShamay-Tsooryetal.(2009)).
areas(i.e.,thedorsalanteriorcingulatecortex[dACC],anterior mid-cingulate cortex [aMCC],supplementary motor area[SMA] and bilateralinsula), the cognitive-evaluativecondition additionally activatedthedorsalaMCC,whiletheaffective-perceptualcondition activatedtherightanteriorinsula.
These studies discussed above claim that emotional and cognitiveaspectsareinseparatebrainregionsandtherefore inde-pendentlyprocessed.However,theyseemcloselyrelatedtoeach other.Onecaseismetacognitionbywhichonecanobserve him-/herselffromanother’sperspective.Therefore,theindividualself isseparatedintotwostates:theobservingandobservedself.The formermaycorrespondtothesubjective(real)selfandthelatter theobjective(virtualized)self.Atypicalphenomenoninthiscase canbeseenwhenenjoyingsadmusic.Sadmusicisperceivedas sadbytheobjectiveself,whilelisteningtothismusicitselfisfeltas pleasantbythesubjectiveself(Kawakamietal.,2013a,b,2014).In thiscase,therelationshipbetweenemotionalandcognitiveaspects iscomplicated.Theperceivedemotionitselfisatargetofthefelt emotion,andthesituationitselfisorganizedbyacognitiveprocess (metacognition).
Pessoa published a book entitled “The cognitive-emotional brain” (Pessoa, 2013) that described how severalbrain regions committedtodifferentfunctionswithfunctionaldiversitymaps andfingerprintsofbrainregions.Heclaimed thattherewasno dichotomy of cognitive versus emotional; rather, he proposed a dynamicnetworkstructure. Thisseemssuggestive forADRto design artificialempathy becausea unified architectureofsuch a dynamic network is more attractive than the integration of independentfunctionmodules.Whilewehavenotyetdevisedand developedsuchanetwork,thefollowingsectionsdescribeseveral attemptsinthisdirection.
5. SeveralapproachesinADR/CDR
FromanADR/CDRperspective,thefollowingissuesshouldbe discussed:
• Whatisthefundamentalmechanismconnectingemotional con-tagionandmotormimicry?
• EEcouldbeanextensionofemotionalcontagionwithmore capa-bilitiesintermsofself-awarenessandself-othercognition.
Inthissection,wereviewpreviousstudies,someofwhichwere notcategorizedasADRbutseemrelatedtothetopicsdiscussed here.
5.1. Emotionalcontagionandmotormimicry
Damasioand Carvalho (2013) stated that a lack of homeo-stasisinthebodytriggersadaptivebehaviorviabrainnetworks, suchas attention toa stranger’s next move. This implies that a homeostasis-like structure is needed to design embodied emotional representations. One of the pioneering WAMOEBA (Waseda-Ameba,WasedaArtificialMindonEmotionBase)studies (OgataandSugano,2000)proposedarobotemotionalmodelthat expressesemotionalstatesconnectedtoself-preservationbased onself-observingsystemsandhormoneparameters.Thissystem wasadaptivetowardexternalstimuli tomaintainbodilyfeeling stability.Therefore,thebestactionissleepingtominimizeenergy consumptionunlessexternalstimuliariseorbatterychargingis needed.
Thisstudyispioneeringinterms ofemotionbeinglinkedto self-preservationorinstinct,withrobotsbeingcapableof display-ingemotionalexpressionsbasedontheseemotionmodels.Here, theterm“instinct”meansthesurvivalparadigmembeddedbythe robotdesignerinadvancetocorrespondtobiological evolution. However,itisnotclearhowarobotcanshareanemotionalstate withhumans.Because almost allrobot behaviorsare explicitly specifiedbythedesigner,thereislittlescopeforrobotstolearnor developthecapacitytosharetheiremotionalstateswithhumans. Emotionalcontagionandmotor mimicryare related toeach other,andmotorresonanceseemstoplayakeyrolein connect-ingthetwo.KuniyoshiandSangawa(2006)proposedoneofthe mostfundamental structures for behavior generationbased on interactionsamongmanydifferentcomponentsincluding198 neu-raloscillators,amusculoskeltalsystemwith198muscles,andan endometrialenvironment in thecase of fetal simulations.Mori andKuniyoshi(2007)extendedtheenvironmenttothehorizontal planeundertheforceoftheEarth’sgravityinthecaseof neona-talsimulation.Fetalorneonataloscillatorymovementsoccurin theseexternalworlds,andself-organizationoforderedmovements is expectedthrough theseinteractions. Thisleads toadditional interactionswithotheragentsthroughmultiplemodalities,suchas visionoraudition(motorresonance).Theneuralarchitectures pro-posedinpreviousstudies(KuniyoshiandSangawa,2006;Moriand Kuniyoshi,2007)aretypicalexamplesfortheunderlying mecha-nismatthefirststageofthedevelopmentalprocessoftheself/other cognition (see Fig. 2, bottom left). This type of architecture is expectedtobeexpandedinlaterstagesbyaddingmore substruc-turesfocusedonsocialinteractions.
Mimicryisonesuchinteractionthatmayinduceemotional con-tagion,which is linkedto EE.A partof theMNS couldalsobe includedinthisprocess(Shamay-Tsooryetal.,2009).Mirror neu-ronsinmonkeysonlyrespondtogoal-orientedactions(actionsof transitiveverbs)withavisibletarget,whileinthecaseofhumans, theMNSalsoseemstorespondtotheactionsofintransitiveverbs withouttargets(RizzolattiandSinigaglia,2008).Thisisstilla con-troversialissuethatneedsmoreinvestigation(Agnewetal.,2007). Nagaietal. proposeda computational model forearly MNS developmentthatoriginatesfromimmaturevision(Nagaietal., 2011), bywhich twotypesofassociations werelearned:one is betweenmotorcommandsandself-observation,andtheotheris betweenmotor commandsandother-observation. Their experi-mentsdemonstratethat themodelachievesearlydevelopment of the self-other cognitive system, which enables a robot to
oftheself/othercognitionshowninFig.2,fromthelefttothe cen-teratthebottom.Byaddingvisionmodalitytotheneuralsystem (KuniyoshiandSangawa,2006;MoriandKuniyoshi,2007),wemay expecta seamlessconnectionbetweenthefirstandthesecond stagesofthedevelopmentofself/othercognition.
Unlikefromnon-humanprimates,thehuman’sMNScanwork fornon-purposeful actionssuchasplay. Kuriyamaet al.(2010) revealedamethodforinteractionrulelearningbasedon contin-gencyandintrinsicmotivationforplay.Thisstudypartiallydepends onthefundamentalMNScapabilityandthereforeseemstobea suc-cessoroftheNagai’smodel(Nagaietal.,2011).Thismaycorrespond toasubstructuretobeaddedatthesecondstageofthe develop-mentalprocessofself/othercognition(seeFig.2,bottomcenter) althougha directconnectionwithNagai’smodelorKuniyoshi’s model(KuniyoshiandSangawa,2006;MoriandKuniyoshi,2007) doesnotseemstraightforward.
5.2. Emotionalandcognitiveempathy
Theabovestudieshavenotbeendirectlyrelatedtoemotional statessuchaspleasure(unpleasant)orarousal(sleep),whichare regardedasthemostfundamentalemotionalaxes(Russell,1980). Assumingthathumaninfantsarebornwiththisfundamentalform ofemotion,howcantheyhavevariationsinemotionalstatessuch ashappinessandanger?Indevelopmentalpsychology,intuitive parentingisregardedasamaternalscaffoldingonwhichchildren developempathy(GergelyandWatson,1999).Atypicalexampleis whencaregiversmimicorexaggerateachild’semotional expres-sions.Thisisconsideredagoodopportunityforteachingchildren howtofeelinrealtime(Rochat,2001),andmostadultspossessthis skill.Childrenarethusabletounderstandthemeaningoffacial expressionsanddevelopsympathytowardothersastheprocessis reinforcedthroughemphasisontheircaregivers’facialexpressions. Thisisbecausechildrenempiricallylearntheconnectionbetween theirinternalstateandthefacialexpressionsofothers.Watanabe etal.(2007)modeledhumanintuitiveparentingusingarobotthat associatesacaregiver’smimickedorexaggeratedfacialexpressions withtherobot’sinternalstatetolearnanempathicresponse.The internalstatespaceandfacialexpressionsaredefinedusing psy-chologicalstudiesandchangedynamicallyinresponsetoexternal stimuli.Afterlearning,therobotrespondstothecaregiver’s inter-nalstatebyobservinghumanfacial expressions.Therobotthen faciallyexpressesitsowninternalstateifsynchronizationevokes aresponsetothecaregiver’sinternalstate.
Consideringtheneuralsubstratesrelatedtoempathyreported inpaststudies(e.g.,Fanaetal.,2011;Shamay-Tsooryetal.,2009; Liddelletal.,2005),adraftoftheneuroanatomicalstructureforthe abovecomputationalmodelisdepictedinFig.3.Thefindingsofpast studiesmaynotbeconsistentastheexperimentswereconducted withdifferenttaskparadigmsandmeasures.Rather,thisstructure isintendedtogiveanapproximatenetworkstructure.During learn-ing,thecaregivers’facialexpressions,whichthelearnerhappens toencounterduringaninteraction,aresupposedtobeprocessed intheinferiorfrontalgyrus(IFG)and/orinsulaandthenmapped ontothedorsalanteriorcingulatecortex(dACC).ThedACC main-tainsthelearner’s emotionalspacethatdrivesfacialmusclesto expressone’sownemotionalstates.Afterlearning,the correspond-ingfacialexpressionisimmediatelydrivenbythecaregiver’sfacial expression.
Theaboveprocessmaycorrespondtothesecondstageofthe developmentalprocessofself/othercognition(Fig.2).More pre-cisely,itmaycorrespondtotheearlyphaseofthesecondstages sincethereactionseemstobelikethatofanMNS-like system. Moreover,mappingfromthecaregiversfacialexpressiontoone’s
Fig.3.AneuroanatomicalstructureforthecomputationalmodelinWatanabeetal. (2007).
ownemotionalstaterequiresatleastself-awarenessasshownin Table1.However,atthisstage,perspectivetakingandor mental-izingdonotseemcomplete,yet.Theseissuesarearguedinthe followingsection.
In additiontotheMNS, CErequires “perspective-takingand mentalizing”(deWaal,2008),bothofwhichsharefunctionswith “theoryofmind”(PremackandWoodruff,1978).Thisisanother difficultissuefornotonlyempathydevelopmentbutalsogeneral humandevelopment.
Early perspective-taking development can be observed in 24-month old children as visual perspective taking (Moll and Tomasello,2006)whileitcannotbeobservedin18-monthold chil-dren.Thisimpliesthattherecouldbeadevelopmentalprocessthat takesplacebetweentheseages(MollandTomasello,2006).
The3-Dcoordinatetransformationsbasedongeometric recon-struction of self, others and object locations is a conventional engineeringsolution,butitdoesnotseemrealistictoestimatethe preciseparametersneededtoreconstructthembetweentheages of18and24months.Morerealisticsolutionscouldbetworelated onesamongwhichthesecondonemightincludethefirstone.Both sharetheknowledgeofwhatthegoalis.
Thefirstpossibilityistheaccumulationofgoal-sharingvisual experienceswitha caregiver.Circumstantialevidencefor view-basedrecognitioncanbeseeninfacecellsintheinferiortemporal cortex of a monkey brain (Chapter 26 in Purves et al., 2012), which are selectively activated according to facial orientation. Appearance-based vision could be an engineering method for objectrecognitionandspatialperception.3Yoshikawaetal.(2001)
proposeda method of incrementalrecovery of the demonstra-tor’sviewusingamodularneuralnetwork.Init,thelearnercan organizespatialperceptionforview-basedimitationlearningwith thedemonstratorindifferentpositionsandorientations.Recent progressinbigdataprocessingprovidesbettersolutionstothis issue.
Thesecondisanapproachthatequalizesdifferentviewsbased on a value that can be estimated by reinforcement learning (Takahashietal.,2010).Thelearningconsequenceresemblesthe MNS functionin themonkeybrain (i.e.,regardingthedifferent actions(selfandother)asthesamegoal-orientedaction).
3Visithttp://www.cs.rutgers.edu/ elgammal/classes/cs534/lectures/appearance-based%20vision.pdfasageneralreference.
5.3. Expressions
Facialandgesturalexpressionsareaveryimportantand indis-pensablepartofartificialempathy.ApioneeringworkofWE-4RII shows very rich facial and gestural expressions, and observers evokethecorrespondingemotions(sameordifferent)(Miwaetal., 2003,2004).4Althoughtheirdesignconceptandtechnologyare
excellent,therealismofinteractionsdependsontheskillofthe designer.
We needmore realisticresearchplatforms(intwo ways)as explained by the ADR approach. One is the design of realistic robotswiththecomputationalmodelofaffectivedevelopment.The otherincludesplatformsforemotionalinteractionstudiesbetween aninfant andhis/her caregiver.For thesepurposes,TheAffetto projecthasbeeneffective(Ishiharaetal.,2011;IshiharaandAsada, 2014).“Affetto”hasachild-like(1-to2-year-old)bodysize,shape, softness,and appearance.Realizingthestructuresandfunctions requiredforahuman-likeimpressionorbehaviorsinasmallbody istechnicallychallenging.Itinvolvessynergistichardware integra-tionbasedoneffectiveandefficientspaceutilization.Inaddition tohardware design, weare tryingtoestablish methodsto uti-lize therobot as a child developmentalsimulator for cognitive developmentalroboticsand asa stimulus-presentingdevicefor psychologicalexperimentsofinterpersonalcognition.Fig.4shows anexampleof“Affetto.”
5.4. Socialbehavioranalysis
AnotheraspectofADRis(B)offeringnewmeansordatafor better understanding of the humandevelopmental process. As mentioned inSection3,robot couldbeused asreliabletoolsin psychologicalexperimentstoensurereproducibleand unbiased results.
Recently,Takahashietal.(2014)showedhowsocialinteractions withdifferentagentsaffecttheimpressionsofmentalcapabilities suchasmind-holdernessandmind-readerness.Theypreparedfive kindsofsocialagents:ahuman,anandroid(ActoroidF),a mechani-calhumanoid(infanoid),apet-likerobot(Keepon),andacomputer thatwereopponentsofthematching-pennygamepresentedinan fMRIscanner.Principalcomponentanalysis(PCA)oftheanswers fortheimpressionquestionsshowedthatthefirstandthird compo-nentscorrespondtothementalfunctionscore(mind-holderness) andtheentropy(highervaluesindicateacomplexstrategyforthe game:mind-readerness).Fig.5shows theresultof the impres-sionsintwo-dimensionalspaceintermsmind-holderness(x-axis) and mindreaderness(y-axis). Thehuman, theandroid,and the humanoidhavepositivecorrelationsbetweenmindholdernessand mindreaderness.Ontheotherhand,Keepon(thecomputer)has negativeone:high(low)mindholdernesswhilelow(high)mind readerness.
fMRI results indicated that these two aspects of social impressions correspond well to activity in two distant brain region networks. The dorso-medial cingulum net-work and the anterior–ventral temporo-parietal junction (TPJ)/posterior–superior temporal sulcus (pSTS) are activated bytheimpressionofmind-holderness(red)andmind-readerness (blue),respectively(Fig.6).Socialinteractionwithamindholder ormindreadermaydistinctlyshapetheinternalrepresentation ofoursocialbrain,whichmayinturndeterminehowwebehave towardsvariousagentsthatweencounterinoursociety.
Mindholdernessandmindreadernessmaycorrespondto emo-tionalandcognitiveaspectsofempathythoughttheirrelationship
Fig.4. Affetto:facialexpressions(left),internalstructureoftheuppertorso(center),andanexampleofphysicaltouch(right).
Fig.5.Locationofeachopponentintwo-dimensionalspace.Thex-axisindicates “mind-holderness,”andthey-axisindicates“mind-readerness”(seetext).Thescore ofPCAcomponentsforeachopponentrepresentsthemeanvalueforallparticipants (Takahashietal.,2014).
andnetworkstructurearefarfromourcurrentunderstanding. Fur-ther,thesecondcomponentofPCAdoesnotcorrespondtoany mentalpropertyoranyspecificbrainactivity.Futureanalysisand modelingareexpectedtorevealthecorrespondingproperties.
6. Discussion
Based on our previous review (Asada, 2014b), this article focusedontheissuesofemotionalcontagion,motormimicry,and
cognitive versus emotional aspects. In the former issue, we discussed the possibility of a mechanism to connect physical and mental synchronization. To do so minimally requires an entraining system that realizes physical synchronization with theenvironment.Fetalandneonatalsimulations(Kuniyoshiand Sangawa, 2006; Mori and Kuniyoshi, 2007) are the first step, and thesesimulations shouldbe extended tointeractions with other agents. In order to support such simulations, we need neuroscientific evidence for synchronization. Recently, Hirata etal.(2014)builtahyperscanningmagnetoencephalogram(MEG) system with two MEG scanners in one shield room, one for children and another for adults, and a video display system that enables each subject to observe the opponent in approx-imately real time (with some delay), in static mode, or any other common video clips (see Fig. 7). The measurement and analysis of this hyper scanning data are expected to provide basicknowledgeaboutmother–infantsynchronization,andmodel verifications.
Studiesassessingseverelyaphasicpatients(e.g.,Varleyetal., 2001)havereportednormalToMprocessing.Thisheavilyimplies that languagecapacityis not anessential requirementfor ToM (Agnewetal.,2007)andprobablyisnotneededforempathy,either. Asmentioned inSection5.1,self-observingsystemsand hor-moneparametersinWAMOEBA(OgataandSugano,2000)were promisingfordesigningartificialemotion,andthebestactionwas sleepingtominimizeenergyconsumptionunlessexternalstimuli arose.However,animalbehavior,especiallyinhumans,is gener-atednotonlybythisfundamentalstructurenecessaryforsurvival, butalsomoreactivelybyso-calledintrinsicmotivation(Ryanand Deci,2000).Actually,fallingisaleadingcauseofaccidentalinjury anddeathinchildrenunderfive(JohandAdolph,2007). Neverthe-less,theyseemtoexploretheirenvironmentusingtheirimmature behaviors.
Fig.6. fMRIofbrainregionswhereactivitiesweremodulatedby“mind-holderness”(red)and“mind-readerness”(blue)(Takahashietal.,2014).(Forinterpretationofthe referencestocolourinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)
Fig.7.SchematicandblockdiagramsofthehyperscanningMEGsystem.(A)Standard(right)andchild-sized(left)MEGs,and(B)anaudio–visualpresentationandrecording system(Hirataetal.,2014).
Inthemachine-learninganddevelopmentalrobotics commu-nity,intrinsic motivation hasreceived increasedattention as a drivingstructureofvariousbehaviors(LopesandOudeyer,2010). The relationship between empathy and intrinsic motivation is yettobe intensivelyinvestigated. We mightconsidera certain structureofintrinsicmotivationasameanstodevelopartificial empathy.Whetherexplicitorimplicitistobeaddressedinfuture studies.
Withregardtothecognitiveandemotionalaspects,Decetyand Lamm(2006)proposedamodelinwhichbottom-up(i.e.,direct matchingbetweenperceptionandaction)andtop-down(i.e., reg-ulation,contextualappraisal,andcontrol)informationprocesses arefundamentallyintertwinedinthegenerationandmodulationof empathy.Bottom-upprocessesaccountfordirectemotionsharing, which areautomaticallyactivated(unlessinhibited)by percep-tualinput.Executivefunctionsimplementedintheprefrontaland cingulatecorticesserve toregulateboth cognitionandemotion through selective attention and self-regulation. This metacog-nitivelevel is continuouslyupdated bybottom-up information, and in return controlsthe lower level by providing top-down feedback.Top-downregulation,throughexecutivefunctions, mod-ulateslowerlevelsandaddsflexibility,makingtheindividualless dependentonexternalcues.Themetacognitivefeedbackloopalso playsacrucialroleintakingintoaccountone’sownmental compe-tenceinordertoreact(ornot)totheaffectivestatesofothers.This modelshouldbesupplementedbytop-downprocessesthatare notclassicallyassociatedwithexecutivefunctionanditsassociated neuralstructures,inparticularthoseinthemedialanddorsolateral prefrontalcortices.
Thissystemhasnotbeenimplementedinafunctionalartificial system,butitseemsvaluabletoattempttodoso.Ifimplemented, contextualfactorssuchasearlyexperienceswithprimary care-givers(attachment),currentmoodstates,andotherenvironmental
contingenciescouldbeconsideredsincethesesfactorsare capa-bleofmodulatingempathy(Gonzalez-Liencresaetal.,2013).This wouldrequireasystematicdefinitionforempathy,sympathy,and personaldistress(severalpapersarguetherelationshipbetween empathyandpersonaldistress(e.g.,Shamay-Tsooryetal.,2009; Bushetal.,2000;Lammetal.,2007)becausetheyseemclosely related.
Acknowledgements
This research was supported by Grants-in-Aid for Scientific Research (Research Project Number: 24000012). The author expresseshisappreciationfor constructivediscussions withDr. MasakiOgino(KansaiUniversity),Dr.YukieNagai(Osaka Univer-sity),HisashiIshihara(OsakaUniversity),Dr.HideyukiTakahashi (OsakaUniversity),Dr.AiKawakami(TamagawaUniversity),Dr. MatthiasRolf(OsakaUniversity),andotherprojectmembers.
References
Agnew,Z.K.,Bhakoo,K.K.,Puri,B.K.,2007.Thehumanmirrorsystem:amotor reso-nancetheoryofmind-reading.BrainRes.Rev.54,286–293.
Amodio,D.M.,Frith,C.D.,2006.Meetingofminds:themedialfrontalcortexand socialcognition.Nat.Rev.Neurosci.7,268–277.
Asada,M.,2011.Cancognitivedevelopmentalroboticscauseaparadigmshift?In: Krichmar,J.L.,Wagatsuma,H.(Eds.),NeuromorphicandBrain-BasedRobots. CambridgeUniversityPress,pp.251–273.
Asada,M.,2014a.Affectivedevelopmentalroboticshowcanwedesignthe devel-opmentofartificialempathy?In:Proc. ofthe9thACM/IEEE International ConferenceonHuman–RobotInteractionWorkshoponHRI:ABridgeBetween RoboticsandNeuroscience(USB-ROM).
Asada,M.,2014b.Towardsartificialempathy.Int.J.Soc.Robot,Publishedonline:29 November,http://dx.doi.org/10.1007/s12369-014-0253-z
Asada,M.,Hosoda,K.,Kuniyoshi,Y.,Ishiguro,H.,Inui,T.,Yoshikawa,Y.,Ogino,M., Yoshida,C.,2009.Cognitivedevelopmentalrobotics:asurvey.IEEETrans.Auton. Ment.Dev.1(1),12–34.
Auton.Syst.37,185–193.
Asada,M.,Nagai,Y.,Ishihara,H.,2012.Whynotartificialsympathy?In:Proc.ofthe InternationalConferenceonSocialRobotics,pp.278–287.
Bush,G.,Luu,P.,Posner,M.I.,2000.Cognitiveandemotionalinfluencesinanterior cingulatecortex.TrendsCogn.Sci.4,215–222.
Chartrand,T.L.,Bargh,J.A.,1999.Thechameleoneffect:theperception–behaviorlink andsocialinteraction.J.Pers.Soc.Psychol.76(6),839–910.
Chen,Q.,Panksepp,J.B.,Lahvis,G.P.,2009.Empathyismoderatedbygenetic back-groundinmice.PLoSONE4,e4387.
Craig,A.B.,2003.Interception:thesenseofthephysiologicalconditionofthebody. Curr.Opin.Neurobiol.13,500–505.
Damasio,A.,Carvalho,G.B.,2013.Thenatureoffeelings:evolutionaryand neurobi-ologicalorigins.Nat.Rev.Neurosci.14,143–152.
deWaal,F.B.,2008.Puttingthealtruismbackintoaltruism,Theevolutionof empa-thy.Annu.Rev.Psychol.59,279–300.
Decety,J.,Lamm,C.,2006.Humanempathythroughthelensofsocialneuroscience. ScientificWorldJournal6,1146–1163.
Edgar,J.L.,Paul,E.S.,Harris,L.,Penturn,S.,Nicol,C.J.,2012.Noevidenceforemotional empathyinchickensobservingfamiliaradultconspecifics.PLOSONE7(2),1–6. Fana,Y.,Duncana,N.W.,deGreckc,M.,Northoffa,G.,2011.Isthereacore neu-ralnetworkinempathy?anfMRIbasedquantitativemeta-analysis.Neurosci. Biobehav.Rev.35,903–911.
Gallese,V.,Fadiga,L.,Fogassi,L.,Rizzolatti,G.,1996.Actionrecognitioninthe pre-motorcortex.Brain119(2),593–609.
Gergely,G.,Watson,J.S.,1999.Earlysocio-emotionaldevelopment:contingency perceptionandthesocial-biofeedbackmodel.In:Rochat,P.(Ed.),EarlySocial Cognition:UnderstandingOthersintheFirstMonthsofLife.LawrenceErlbaum, Mahwah,NJ,pp.101–136.
Goetz,J.L.,Keltner,D.,Simon-Thomas,E.,2010.Compassion:anevolutionary anal-ysisandempiricalreview.Psychol.Bull.136,351–374.
Gonzalez-Liencresa,C.,Shamay-Tsooryc,S.G.,Brünea,M.,2013.Towardsa neu-roscienceofempathy:ontogeny,phylogeny,brainmechanisms,contextand psychopathology.Neurosci.Biobehav.Rev.37,1537–1548.
Grezes,J.,Armony,J.,Rowe,J.,Passingham,R.,2003.Activationsrelatedto“mirror” and“canonical”neuroninthehumanbrain:anfMRIstudy.Neuroimage18, 928–937.
Harrison,N.A.,Singer,T.,Rotshtein,P.,Dolan,R.J.,Critchley,H.D.,2006.Pupillary contagion,Centralmechanismsengagedinsadnessprocessing.Soc.Cogn.Affect. Neurosci.1,5–17.
Hatfield,E.,Cacioppo,J.T.,Rapson,R.L.,2000.Emotionalcontagion.Curr.Dir.Psychol. Sci.2,96–99.
Hatfield,E.,Rapson,R.L.,Le,Y.-C.L.,2009.Emotionalcontagionandempathy.In: Decety,J.,Ickes,W.(Eds.),TheSocialNeuroscienceofEmpathy.MITPress,pp. 19–30(Chapter2).
Hirata,M.,Ikeda,T.,Kikuchi,M.,Kimura,T.,Hiraishi,H.,Yoshimura,Y.,Asada,M., 2014.Hyperscanningmegforunderstandingmother–childcerebral interac-tions.Front.Hum.Neurosci.8(118).
Inui,T.,2013.Embodiedcognitionandautismspectrumdisorder.Jpn.J.Occup.Ther. 47(9),984–987(inJapanese).
Ishihara,H.,Asada,M.,2014.Fivekeycharacteristicsforintimatehuman–robot interaction:developmentofuppertorsoforachildrobot‘Affetto’.Adv.Robot. (inpress).
Ishihara,H.,Yoshikawa,Y.,Asada,M.,2011.Realisticchildrobot“Affetto”for under-standingthecaregiver–childattachmentrelationshipthatguidesthechild development.In:IEEEInternationalConferenceonDevelopmentandLearning, andEpigeneticRobotics(ICDL-EpiRob2011)(CD-ROM).
Joh,A.S.,Adolph,K.E.,2007.Learningfromfalling.ChildDev.77(1),89–102. Kawakami,A.,Furukawa,K.,Katahira,K.,Kamiyama,K.,Okanoya,K.,2013a.Relations
betweenmusicalstructuresandperceivedandfeltemotion.MusicPercept.30 (4),407–417.
Kawakami,A.,Furukawa,K.,Katahira,K.,Okanoya,K.,2013b.Sadmusicinduces pleasantemotion.Front.Psychol.4(311),1–15.
Kawakami,A.,Furukawa,K.,Katahira,K.,Okanoya,K.,2014.Musicevokesvicarious emotionsinlisteners.Front.Psychol.5(431).
Kuhl,P.,Andruski,J.,Chistovich,I.,Chistovich,L.,Kozhevnikova,E.,Ryskina,V., Stol-yarova,E.,Sundberg,U.,Lacerda,F.,1997.Cross-languageanalysisofphonetic unitsinlanguageaddressedtoinfants.Science277,684–686.
Kuniyoshi,Y.,Sangawa,S.,2006.Earlymotordevelopmentfrompartiallyordered neural-bodydynamics: experimentswithacortico-spinal-musculo-skeletal model.Biol.Cybern.95,589–605.
Kuriyama,T.,Shibuya,T.,Harada,T.,Kuniyoshi,Y.,2010.Learninginteractionrules throughcompressionofsensori-motorcausalityspace.In:Proc.oftheTenth InternationalConferenceonEpigeneticRobotics(EpiRob10),pp.57–64. Lakin,J.L.,Jefferis,V.E.,Cheng,C.M.,Chartrand,T.L.,2003.Thechameleoneffectas
socialglue:evidencefortheevolutionarysignificanceofnonconsciousmimicry. J.NonverbalBehav.27(3),145–162.
Lamm,C.,Batson,C.D.,Decety,J.,2007.Theneuralsubstrateofhumanempathy: effectsofperspective-takingandcognitiveappraisal.J.Cogn.Neurosci.19(1), 42–58.
Leite,I.,Martinho,C.,Paiva,A.,2013.Socialrobotsforlong-terminteraction:asurvey. Int.J.Soc.Robot.5,291–308.
subliminalsignalsoffear.Neuroimage24,235–243.
Lopes,M.,Oudeyer,P.-Y.,2010.Guesteditorialactivelearningandintrinsically moti-vatedexplorationinrobots:advancesandchallenges.IEEETrans.Auton.Ment. Dev.2(2),65–69.
Lungarella,M.,Metta,G.,Pfeifer,R.,Sandini,G.,2003.Developmentalrobotics:a survey.Connect.Sci.15(4),151–190.
Meltzoff,A.N.,2007.The‘likeme’frameworkforrecognizingandbecomingan intentionalagent.ActaPsychol.(Amst.),26–43.
Miwa,H.,Itoh,K.,Matsumoto,M.,Zecca,M.,Takanobu,H.,Roccella,S.,Carrozza, M.C.,Dario,P.,Takanishi,A.,2004.Effectiveemotionalexpressionswith emo-tionexpressionhumanoidrobotWE-4RII.In:Proc.ofthe2004IEEE/RSJIntl. ConferenceonIntelligentRobotandSystems,pp.2203–2208.
Miwa,H.,Okuchi,T.,Itoh,K.,Takanobu,H.,Takanishi,A.,2003.Anewmental modelforhumanoidrobotsforhumanfriendlycommunication– introduc-tionoflearningsystem,moodvectorandsecondorderequationsofemotion. In:Proc.ofthe2003IEEEInternationalConferenceonRobotics&Automation, pp.3588–3593.
Moll,H.,Tomasello,M.,2006.Level1perspective-takingat24monthsofage.Brit.J. Dev.Psychol.24,603–613.
Mori,H.,Kuniyoshi,Y.,2007.Acognitivedevelopmentalscenariooftransitional motorprimitivesacquisition.In:Proceedingsofthe7thInternational Confer-enceonEpigeneticRobotics,pp.165–172.
Nagai,Y.,Kawai,Y.,Asada,M.,2011.Emergenceofmirrorneuronsystem:immature visionleadstoself-othercorrespondence.In:IEEEInternationalConferenceon DevelopmentandLearning,andEpigeneticRobotics(ICDL-EpiRob2011) (CD-ROM).
Nagai,Y.,Rohlfing,K.J.,2009.Computationalanalysisofmotionesetoward scaffold-ingrobotactionlearning.IEEETrans.Auton.Ment.Dev.1(1),44–54. Neisser,U.(Ed.),1993.ThePerceivedSelf:EcologicalandInterpersonalSourcesof
SelfKnowledge.CambridgeUniversityPress.
Ogata,T.,Sugano,S.,2000.Emotionalcommunicationbetweenhumansandthe autonomousrobotWAMOEBA-2(WasedaAmoeba)whichhastheemotion model.JSMEInt.J.Ser.C:Mech.Syst.Mach.Elem.Manuf.43(3),568–574. Pessoa,L.,2013.TheCognitive-EmotionalBrain.MITPress.
Premack,D.,Woodruff,G.,1978.Doesthechimpanzeehaveatheoryofmind?Behav. BrainSci.,515–526.
Purves,D.,Augustine,G.A.,Fitzpatrick,D.,Hall,W.C.,LaMantia,A.-S.,McNamara, J.O.,White,L.E.(Eds.),2012.Neuroscience.,5thedition.SinauerAssociates,Inc. Riek,L.D.,Robinson,P.,2009.Affective-centereddesignforinteractiverobots.In: ProceedingsoftheAISBSymposiumonNewFrontiersinHuman–Robot Inter-action,pp.102–108.
Rizzolatti,G.,2005.Themirrorneuronsystemanditsfunctioninhumans.Anat. Embryol.201,419–421.
Rizzolatti,G.,Sinigaglia,C.,2008.MirrorsintheBrain–HowourMindsShareActions andEmotions(F.Anderson,Trans.).OxfordUniversityPress.
Rochat,P.,2001.TheInfant’sWorld.HarvardUniversityPress(Chapter4). Russell,J.A.,1980.Acircumplexmodelofaffect.J.Pers.Soc.Psychol.39,1161–1178. Ryan,R.M.,Deci,E.L.,2000.Intrinsicandextrinsicmotivations:classicdefinitions
andnewdirections.Contemp.Educ.Psychol.25(1),54–67.
Schraw,G.,1998.Promotinggeneralmetacognitiveawareness.J.Instr.Sci.26, 113–125.
Shamay-Tsoory,S.G.,Aharon-Peretz,J.,Perry,D.,2009.Twosystemsforempathy: adoubledissociationbetweenemotionalandcognitiveempathyininferior frontalgyrusversusventromedialprefrontallesions.Brain132,617–627. Singer,T.,Lamm,C.,2009.Thesocialneuroscienceofempathy.In:Miller,M.B.,
King-stone,A.(Eds.),TheYearinCognitiveNeuroscience2009.NewYorkAcademy ofSciences,pp.81–96(Chapter5).
Smith,A.,2006.Cognitiveempathyandemotionalempathyinhumanbehaviorand evolution.Psychol.Rec.56,3–21.
Sommerville,J.A.,Woodward,A.L.,Needham,A.,2005.Actionexperiencealters 3-month-oldinfants’perceptionofothers’actions.Cognition96,B1–11. Sonnby-Borgstrom,M.,2002.Automaticmimicryreactionsasrelatedtodifferences
inemotionalempathy.Scand.J.Psychol.43,433–443.
Sperry,R.W.,1952.Neurologyandthemind-brainproblem.Am.Sci.40,291–312. Takahashi,H.,Teradad,K.,Moritaa,T.,Suzukie,S.,Hajib,T.,Kozimag,H.,Yoshikawah,
M.,Matsumotoi,Y.,Omorib,T.,Asadaa,M.,Naito,E.,2014.Different impres-sionsofotheragentsobtainedthroughsocialinteractionuniquelymodulate dorsalandventralpathwayactivitiesinthesocialhumanbrain.Cortex58,289– 300.
Takahashi,Y.,Tamura,Y.,Asada,M.,Negrello,M.,2010.Emulationandbehavior understandingthroughsharedvalues.Robot.Auton.Syst.58(7),855–865. Uithol,S.,vanRooij,I.,Bekkering,H.,Haselager,P.,2011.Understandingmotor
resonance.Soc.Neurosci.6(4),388–397.
Varley,R.,Siegal,M.,Want,S.,2001.Severeimpairmentingrammardoesnot pre-cludetheoryofmind.Neurocase7(6),489–493.
Watanabe,A.,Ogino,M.,Asada,M.,2007.Mappingfacialexpressiontointernalstates basedonintuitiveparenting.J.Robot.Mechatron.19(3),315–323.
Yoshikawa,Y.,Asada,M.,Hosoda,K.,2001.Developmentalapproachtospatial per-ceptionforimitationlearning:incrementaldemonstrator’sviewrecoveryby modularneuralnetwork.In:Proc.ofthe2ndIEEE/RASInternationalConference onHumanoidRobot,pp.107–114.