codedairenti.com

InternationalJournalofSocialRobotics(2019)11:477–494https://doi.
org/10.
1007/s12369-019-00516-zPerceivedHuman-LikenessofSocialRobots:TestingtheRaschModelasaMethodforMeasuringAnthropomorphismPeterA.
M.
Ruijten1·AntalHaans1·JaapHam1·CeesJ.
H.
Midden1Accepted:8January2019/Publishedonline:19January2019TheAuthor(s)2019AbstractAnthropomorphismisgenerallydenedastheattributionofhuman-likecharacteristicstosocialrobotsandothernon-humanobjects.
Wearguethatdifferentresearchershavedifferentinterpretationsofthisconcept,leadingtomeasuringinstrumentsthatfocusondifferentsubsetsofhuman-likecharacteristics.
Inthecurrentpaper,wediscussthesedifferentinterpretationsandexploreanewmethodformeasuringanthropomorphism,basedontheRaschmodel.
Theaimofthecurrentworkistomapanthropomorphismasarangeofhuman-likecharacteristicsonaone-dimensionalscale.
Thescale'svalidityandsensitivityweretestedbycomparingitwithtwoavailablemeasuringinstrumentsandbycomparingpeople'sresponsestodifferenttypesofagents.
Inthreestudies,weexploredwhethertheRaschmodelissuitableformeasuringanthropomorphism.
Despitesomelimitations,resultsshowedthattheRaschmodelcansuccessfullybeappliedtothemeasurementofanthropomorphism.
Implicationsforfutureworkonanthropomorphismofsocialrobotsarediscussed.
KeywordsAnthropomorphism·Measurementscale·Raschmodel1IntroductionDuringthepastdecadeswehaveseenanincreasinginterestinresearchonsocialrobots.
Theyarebeingcommercializedandbecomingavailabletothegeneralpublic.
Becauseofthis,understandingtheeffectsoftheirappearanceandbehav-ioronpeople'sinteractionswiththemisgainingimportanceaswell.
Inthenearfuture,socialrobotsmaybeprovidedmorehuman-likefeaturesandarelikelytoberepresentedassocialentitieswithfaces,engaginginsocialconversa-tionwithhumans.
Thesedevelopmentscouldmakepeopleperceivethoserobotsasmoreandmorehuman-like.
Thisperceivedhuman-likenessisanimportantdeterminantforpeople'sresponsestosocialrobots(see[1,36,38]).
More-over,human-likenessinsocialrobotshasbeenshowntopositivelyaffectpeople'sengagementwiththoserobots[4],people'sexpectationsoftherobot'snavigationbehavior[30],andtherobot'spersuasivepower[7].
Thetermanthropomor-TheresearchinthecurrentpaperwasfundedbytheNetherlandsEnergyAgency.
BPeterA.
M.
Ruijtenp.
a.
m.
ruijten@tue.
nl1EindhovenUniversityofTechnology,P.
O.
Box513,5600MBEindhoven,TheNetherlandsphismreferstotheextenttowhichpeopleperceiverobotsandothernon-humanobjectsashuman-like[2,8,9,12,20,32].
Thisconventionalandrathergeneraldescriptionofanthro-pomorphismhasleadtoavarietyofinterpretationsoftheconcept.
Asaconsequence,differentgroupsofresearchersfocusondifferentsubsetsofhuman-likecharacteristics.
1.
1SubsetsofHuman-LikeCharacteristicsAnexaminationofexistingmeasurementinstruments(e.
g.
,[2,5,32])revealedthathuman-likecharacteristicsaregener-allycategorizedintoappearances,thoughtsandemotions.
Withappearanceswerefertocharacteristicsthatreecthumanformorbehavior(i.
e.
,howanobjectorrobotlooksand/ormoves),includingbothphysicalshapesandphysicalabilities.
Withtheirmeasuringinstrumentforanthropo-morphism,Bartneckandcolleagues[2]focusedonsuchappearancesbyaskingpeopletoindicate,amongothers,towhatextentarobotlookshuman-like,lookslife-like,andshowsrealisticmovements.
Theseitemsclearlyfocusonarobot'sappearanceandtheextenttowhichthisappearanceresemblesthehumanbody.
Withthoughtswerefertocharacteristicsthatreectcogni-tivestatesandprocesses.
AccordingtoWaytzandcolleagues[32],anthropomorphismisaprocessofinductiveinference123478InternationalJournalofSocialRobotics(2019)11:477–494whichmostlikelyoccursbyattributingcognitivestatesthatareperceivedtobeuniquelyhumantootheragents(forareview,see[9]).
Hence,anthropomorphismwasmeasuredbyaskingpeopletoindicatetowhatextentanagenthascog-nitiveabilitieslikeconsciousnessandfreewill[32].
Thesecognitiveabilitiescannotbeincludedinthephysicaldesignofrobots,becausetheycanonlybeinferredfromitsbehavior.
Finally,withemotionswerefertocharacteristicsthatindicatesubjectiveconsciousexperienceswhichcanbedis-tinguishedinprimaryandsecondaryones(foranoverviewofthehierarchicalorganizationofemotions,see[26]).
Eysselandcolleagues[11]measuredanthropomorphismbyaskingpeopletoindicatetowhatextentrobotscanexperiencesuchprimaryandsecondaryemotions.
Intheirfurtherwork,Eysselandcolleagues[10,12]dif-ferentiatedbetweenpersonalitycharacteristicsthatreecthumannatureandhumanuniqueness.
Thisdistinctionwasadaptedfromearlierresearchonsocialperceptioninhumans[16,17].
Inthislineofresearch,humannaturecharacteris-ticsaredescribedascharacteristicsofthehumanspeciesthataresharedwithotheranimals(e.
g.
,innateandaffectivetraits).
Uniquelyhumancharacteristics,ontheotherhand,areconsideredtobeexclusivetohumansandnotpossessedbyanyotherspecies(e.
g.
,sociallearningandhighercognition,see[17]).
Whetherornotsuchaclear-cutdistinctionexistsbetweenhumansandotherentitiesisoutsidethescopeofthecurrentwork,butitdoesexplainwhyothershaveapproacheditasa2-dimensionalconstruct.
Theaimofthispaperistwofold.
First,weconceptualizeanthropomorphismanddiscusswhymeasurementsobtainedwithexistinginstrumentsmayhavelittlecorrespondence.
Second,weproposeamethodformeasuringanthropomor-phismbasedontheRaschmodel(seeforexample[3]),andtestitsdimensionalityandvalidityinthreestudies.
WeendwithadiscussiononthebenetsandlimitationsofusingtheRaschmodelformeasuringanthropomorphism.
1.
2ConceptualizingAnthropomorphismWearguethatanthropomorphismisasinglepredisposition,meaningthatwhateverhuman-likecharacteristicsanindi-vidualascribestoarobot—beithumannatureorhumanuniquenesscharacteristics,basicphysicalabilitiesormoraldecisionmaking,theyallstemfromthatsinglepredispo-sitiontodoso.
ThisisinlinewithWaytzandcolleagues[31],whostatedthatattributionsofhuman-likecharacteris-ticsstemfromstableindividualpredispositions.
However,Waytzandcolleagues[31]alsoreferredtoanthropomorphismastheattributionofcharacteristicsthatpeopleregardasdistinctivelyhuman,inparticularmentalcapacities.
ThisviewissharedbyZawieskaandcolleagues[37],whoarguedthatanthropomorphismonlyincludeshuman-likecharacteristicsthatrobotsdonothave.
Asaresultofthisostensiblynarrowfocus,onlymentalcapacities(i.
e.
,havingintentions,freewill,amindofyourown,con-sciousness,andexperiencingemotions)areincludedinthemeasurementofanthropomorphism(see[31]).
Morespecif-ically,theIndividualDifferencesinAnthropomorphismQuestionnaire(IDAQ)consistsofitemsthatdescribecharac-teristicsofspecicagents(e.
g.
,cars,cows,andmountains)withinoneofthreecategories(i.
e.
,technologies,animals,andnaturalentities).
Wearguethatthisoperationalizationofanthropomorphisminsufcientlymeasurestheconcept,becauseitfocusesmerelyonanarrowsubsetofhuman-likecharacteristics(mentalcapacities),anditincludesonlyveryspecicagents.
Inotherwords,webelievethataskingaper-sontowhatextentanaverageshhasfreewilldoesnotexplainorpredictthatperson'sresponsestoasocialrobot.
Wealsowonderwhethersuchascalecanbeusedforcom-paringtheperceivedhuman-likenessofdifferenttypesofagents.
Itisevidentthatthedifferentmeasuringinstrumentshavebeendevelopedwithafocusondifferentsubsetsofthecon-struct.
Consequently,thequestionarisesastowhatextentthesemeasuringinstrumentsofanthropomorphismmeasurethesameconcept,andthusultimatelytowhetherwecanfaithfullycompareresearchndings.
Weviewanthropo-morphismasaone-dimensionalconstruct,andarguethatallhuman-likecharacteristics—nomatterwhichsubsettheybelongto—areorderedaccordingtotheprobabilitywithwhichtheyareascribedtorobots.
Somehuman-likecharac-teristicsareexpectedtobemoreeasilyascribedtorobotsthanothers.
Forexample,human-likeappearancesareexpectedtobemoreeasilyascribedtosocialrobotsthanunderlyingcognitivestatesandprocesses,regardlessofanindividual'sgeneraltendencytoanthropomorphize.
Finally,wearguethattheorderingofhuman-likechar-acteristicswithrespecttothedifcultytoattributethemtorobotsissimilarforallindividuals.
Morespecically,peopleareexpectedtobemorelikelytoattributehuman-likeappear-ancestorobotsthantheyaretoattributecognitivestates.
Suchaninvariantorderingalsoentailsthatifapersonattributestheabilityofmoralreasoningtoarobot,(s)heisalsoexpectedtoattributetheabilityofseeingtothatrobot.
Anotherpersonwhodoesnotattributetheabilityofseeingtothesamerobotisnotexpectedtoattributetheabilityofmoralreasoningtoit.
Ifallhuman-likecharacteristicscanbeinvariantlyorderedacrosspeople,wecancomparepeople'sindividualpredis-positionstoanthropomorphizeandtheperceivedhuman-likenessofarobotonasinglescale.
Onemodelthatisabletomapaperson'spredispositiontoanthropomorphizeandthehuman-likecharacteristics(s)heislikelytoattributetoarobotaslocationsonasingledimension,andthusseemshighlysuitableformeasuringanthropomorphism,istheRaschmodel[3].
123InternationalJournalofSocialRobotics(2019)11:477–4944791.
3TheRaschModelTheRaschmodel(seeEq.
1)describestheoddsofacertainresponseasalogisticfunctionofpersonanditemparameters.
Inourcase,thisrelatestotheprobabilityofattributingaspecichuman-likecharacteristiciasanadditivefunctionofapersonn'sgeneralpredispositiontoanthropomorphize(θn)andthedifcultytoattributethatspecichuman-likecharacteristictoarobot(δi).
lnP(xni=1)1P(xni=1)=θnδi(1)Bothparametersinthisequation(anindividual'spredispo-sitiontoanthropomorphizeθ,andthedifcultyofattributingaspeciccharacteristictoarobotδ)areestimatedbymeansofmaximumlikelihoodestimation.
Predispositionsofper-sonsandthedifcultyofthevariousself-reportitems(i.
e.
,whetherornotarobotisthoughttopossessacertainchar-acteristic)areexpressedinlogoddunits(alsocalledlogits).
Foraspeciccharacteristicitohavea50%chancetobeattributedtoarobot,thedifcultyofthatcharacteristic(e.
g.
,δi=1)hastobematchednumericallywithanequivalentamountofapersonn'spredispositiontoanthropomorphize(θn=1).
TheRaschmodel,however,isnotonlydescriptive,butalsoprescriptive.
Itrequiresitems(orhuman-likecharacteris-tics)tobeorderedinvariantlyandtransitivelyacrosspersons,andpersonstobeorderedinvariantlyandtransitivelyacrossitems.
Thus,ifapersonattributes,forexample,fouroutof10characteristicstoasocialrobot,theRaschmodelalsoprescribeswhichfourshouldbeattributed:Theyshouldbeamongstthefourleastdifcultones.
Forthemostdifculttoattributecharacteristics,weexpecttoexclusivelyndsuchattributionsamongsttheindividualsthathaveahighpredis-positiontoanthropomorphize.
TheseformalRaschmodelexpectationscanbetestedempiricallyagainsttheobserveddata.
Providedtheresponsesareorderedfromthepersonwiththehighesttothelowestpredispositiontoanthropomorphize,themodelanticipatesthefollowingresponsestringforanaveragelydifculttoattributecharacteristic:111101010000.
Inthisresponsestring,aoneindicatesthatapersonagreedtotherobothav-ingthatcharacteristics,andazerothatapersondidnot.
Therstfourindividualsallhavehighpredisposition,soallfourarelikelytoattributethischaracteristictotherobot.
Thenextfourindividualshaveapredispositionthatapproachesnumericallythedifcultyofthischaracteristicoritem.
Asaresult,someofthemwillattributethischaracteristictotherobot,andotherswillnot.
Finallythelastfourindividualshavesuchalowpredispositiontoanthropomorphizethatallareveryunlikelytoattributethischaracteristictotherobot.
Themeansquare(MS)statisticiscommonlyusedtotestthematchbetweenmodel-predictedandobservedresponsepatterns.
TheMSint-valueistheweightedaverageofthesquaredstandardizedresiduals,inwhicheachresidualisweightedbyitsvariance[3].
Themodel-predictedresponsepattern(e.
g.
,111101010000),yieldsaMSint-valueofMS=1.
00.
ExcessivelyhighMS-valuescanbeexpectedwhentheobservedresponsestringopposestheRaschmodelpre-diction,forexamplewhenthelikelihoodofreportinganexperientialeffectincreaseswithdiminishingsusceptibility(e.
g.
,000000111111).
Incontrast,theMS-valuewouldbesmallerthan1.
00foranitemwithadeterministicresponsepattern(i.
e.
,111111000000).
Insuchacase,themodelprediction-to-datatisbetterthanwhatonewouldantici-patewithaprobabilisticmodel.
MS-valuesbelow1.
00donotreallychallengetheRaschmodelprediction,butcanbeusedtoimproveone'smeasurementinstrument.
BesidesMSintalsoMSoutt-valuesareoftenreported.
MSoutt-valuesareunweightedtstatistics,andaremoresensitivetounexpectedresponsesonrelativelyeasyordifcultitems.
Theinvarianceassumptionshouldbesufcientlymetinordertomapbothpersonsandhuman-likecharacteristicsonasinglescale(see1),andthustocompareindividualsandhuman-likecharacteristicsagainsteachotherinameaningfulway.
Forassessingitemt,MS-valuesupto1.
20arecon-sideredexcellent,andMS-valuesbelow1.
50areconsideredacceptable[35].
Theadditionaladvantageofthemodelistheinvariancebetweenitemsandpersons.
Asaresult,theassessmentofapersonspredispositiontoantropomorphizeandoftheitemswithrespecttotheirdifcultyareindependentofeachother(so-calledspecicobjectivity,see[34]).
Asresults,andincontrasttomanyothermeasurementmodels,measurementsofpersonalattributesarenotdenedbythespecicsetofitemsused(see[18]).
Itthusalsoallowsforitemsbeingdeletedand/orreplaced.
Thisenablestheuseofdifferentsetsofitems(see[33]),makingthismethodeasilyadjustableformeasuringresponsestodifferenttypesofagents(e.
g.
,animalsornaturalphenomena).
1.
4ResearchAimsThecurrentresearchwasdesignedwiththeaimtoexploreanewmethodformeasuringanthropomorphism.
Wehypoth-esizedthatanthropomorphismcanbesuccessfullymappedontoaone-dimensionalscaleandthathuman-likecharacter-isticsareorderedwithrespecttotheirlikelihoodofbeingattributedtorobotsinawaythatissimilarforallindividualsintheirencounterwithdifferenttypesofagentsindiffer-entcontexts.
Datafromthreestudieswereusedtotestthesehypotheses.
Twoofthesestudieswereoriginallydesignedwithdifferentpurposes,butwewillonlyassessresultsontheincludedmeasuringinstrumentsforanthropomorphism.
123480InternationalJournalofSocialRobotics(2019)11:477–494Intherststudy,wedevelopedandtestedtheconstructvalidityofa37-itemanthropomorphismscale.
Withconstructvaliditywerefertotherelationbetweentheorderingofitemsonthescaleaccordingtotheirdifcultyandtheirperceivedhumannatureandhumanuniqueness.
Constructvalidityishighwhenstrongrelationsbetweenthelocationsofitemsonthedimensionandtheirperceivedhumannatureandhumanuniquenessarefound.
Becauseoftheexpectedunidimen-sionality,humannatureandhumanuniquenesswerealsoexpectedtobestronglycorrelated.
Inthesecondstudy,a25-itemversionoftheanthropomor-phismscalewastestedonitsconvergentvalidity.
Withcon-vergentvaliditywerefertotheextenttowhichestimatesofthescalearerelatedtoestimatesobtainedwithtwoothermea-suringinstrumentsforanthropomorphism:thequestionnaireusedbyWaytzandcolleagues[32,referredtoastheWaytz-instrument,see"AppendixA"]andtheanthropomorphismpartoftheGodspeedquestionnairedevelopedbyBart-neckandcolleagues[2,referredtoastheGodspeed-instrument,see"AppendixB"].
Becausethefocusoftheseinstrumentsisondifferentsubsetsofhuman-likecharacter-istics,weexpectedtondmoderatecorrelationsbetweentheanthropomorphismscaleandtheWaytz-andGodspeed-instruments.
Thesecondstudywasoriginallydesignedtocomparetwodifferentrobotsontheirperceivedhuman-likeness.
Becauseofthis,wecouldusethisstudyfortestingtheinvariantorderingoftheitemsonthescaleforthosetworobots.
Inthethirdstudy,weextendedourscopefromrobotstoothertypesofagents.
A19-itemversionoftheanthropomor-phismscalewastestedonitssensitivityfordifferentiatingbetweenhumansandothertypesofagents.
Thisstudywasoriginallydesignedtoinvestigatepeople'sresponsestofourtypesofplayersinagame(i.
e.
,humans,robots,computers,andalgorithms),enablingustocompareresponsesforothertypesoftechnologiesthanrobotsaswell.
Weexpectedthatthescalewouldsuccessfullydifferentiatebetweenhumansandothertypesofagents.
Finally,weexpectedthattheanthropomorphismscalewouldshowaninvariantorderingofhuman-likecharac-teristicsonasingledimensionacrossallstudiesandallexperimentalconditions.
2Study1Inthisstudy,alistwith37human-likecharacteristicswascreated,largelybasedonearlierworkonhumannessandanthropomorphism[2,11,16,17,32],andtestedonitscon-structvalidity.
Thehuman-likecharacteristicsweremodeledasafunctionofaperson'spredispositiontoanthropo-morphizeandthedifcultytoattributethathuman-likecharacteristictoarobot.
Wehypothesizedthatitemsandper-sonscouldbemappedontoasingleone-dimensionalscale,andthattheitemswouldbeinvariantlyorderedaccordingtothedifcultywithwhichtheyareattributedtoarobot.
Additionally,forconstructvaliditypurposes,theextenttowhichthe37human-likecharacteristicswereperceivedasbeinghumannatureanduniquelyhumanwasmeasured.
Wehypothesizedthattheestimateddifcultieswithwhichthe37characteristicsareattributedtoarobotwouldberelatedtotheirperceivedhumannatureandhumanuniqueness.
Becauseoftheproposedunidimensionalityofanthropomor-phism,humannatureandhumanuniquenesswereexpectedtobestronglycorrelatedaswell.
2.
1Method2.
1.
1ParticipantsandDesignOnehundredandsixtyoneparticipantssampledthroughsocialmediaparticipatedinoneofthreegroupsinthecur-rentstudy.
Therstgroupconsistedof124participants(53malesand71females;Mage=26.
08,SDage=8.
82,Range=15to59)whoweregivenadescriptionaboutarobotandcompleted37surveyitems.
Anothergroupof20participants(9malesand9females,Mage=19.
94,SDage=1.
98,Range=18to23;twoparticipantsdidnotindicatetheirageandgender)ratedthe37human-likecharacteristicsonhumannature.
Theremaining17participants(11malesand6females,Mage=21.
12,SDage=1.
80,Range=18to24)ratedallcharacteristicsonhumanuniqueness.
Participantsinallthreegroupsparticipatedvoluntarily,gaveinformedconsent,andwerenotcompensatedforparticipation.
2.
1.
2MaterialsandProcedureAsetof37itemsdescribinghuman-likecharacteristicswasconstructed.
Forallthreegroupsofparticipants,itemswerearrangedinalphabeticalorder.
Fortherstgroupof124participants,itemswerefor-mulatedasastatementwhichcouldbeansweredwithyes(codedwitha1)orno(codedwitha0).
Theitemswerepresentedthroughanonlinesurvey.
Afterreadingashortexplanationaboutthestudy,participantswereprovidedwithashortdescriptionaboutarobot:'Therobothaseyestoper-ceivetheenvironment,hasarmsandlegstomovearoundinthisenvironment,andtodaytherobotistryingtosolveamoraldilemma'.
Thisdescriptionwasfollowedbyaninstruc-tiontonotthinkelaboratelyaboutthestatementsandtogivetheanswersthatrstcametomind.
Finally,participantsindi-catedtheirgender,ageandeducationlevel,andtheywerethankedfortheircontribution.
Thisstudytookapproximately5mintocomplete.
Thetwoothergroupsofparticipantswerenotgiventhedescriptionoftherobot,butinsteadwereaskedtoindicateto123InternationalJournalofSocialRobotics(2019)11:477–494481whatextenteachofthe37itemsonthescalewasperceivedas'typicallyhuman'(i.
e.
,humannature)or'uniquelyhuman'respectively.
Humannaturewasmeasuredwithonequestion('Towhatextentisthischaracteristictypicalforhumans')ona7-pointresponseformatrangingfrom'notatall'(codedwitha1)to'verymuch'(codedwitha7).
Humanunique-nesswasmeasuredwithonequestion('Isthischaracteristicuniqueforhumans')onadichotomousresponseformatwith'notunique'(codedwitha0)and'unique'(codedwitha1)asoptions.
Thetwoconceptswerenotfurtherintroducedorexplainedtoparticipants.
Wechoseforadichotomousscaleforhumanuniquenessbecauseacharacteristiceitheris,orisnotuniquelyhuman(i.
e.
,characteristicscannotbe'alittle'uniqueforhumans).
Boththeseevaluationstookapproxi-mately5mintocomplete.
2.
2ResultsandDiscussion2.
2.
1ModelTestTotestwhetherthedatasufcientlytthemodel,fourtestswereconducted.
First,tstatisticswereusedtotestwhetheritemsandpersonsttedtheRaschmodel(foranoverview,see[3]).
Forassessingitemt,bothintandouttwereused.
Intindicatesunexpectedobservationsonitemsthatarecloseindifcultytoaperson'spredisposition,whereasouttindi-catesunexpectedobservationsonitemsthatarerelativelyeasyordifcult[23].
Intandouttmeansquare(MS)val-ues≤1.
20areconsideredexcellent,andMS-values≤1.
50areconsideredacceptable[35].
Thesecondtestdeterminedwhethertheitemsweresufcientlyspreadovertheperceivedhuman-likenessdimension.
Thethirdonetestedthehypoth-esisthatitemsallbelongtoasingledimension.
Finally,thefourthonetestedthehypothesisthatitemswouldbeinvari-antlyorderedaccordingtothedifcultyofattributingthemtorobots.
ItemtIdeally,eachoftheitemscontributesinameaningfulwaytothemeasuringinstrument,indicatedbyasufcientitemt.
ConsideringthenotionthattheRaschmodelisstochasticandthatdatadependonprobability(andnotoncer-tainty),somemististobeexpected[27].
Anacceptableveofthe37itemshadouttMSvaluesoutsideoftheacceptableboundaries.
Thesewereitems1('experiencepain',outtMS=2.
09),26('anticipateonsurroundings',outtMS=1.
62),32('organized',outtMS=2.
74),33('estimatedistances',outtMS=2.
12),and37('avoidobjects',outtMS=1.
97).
Another7itemshadouttMSvaluesbetweenthegoodandacceptablevalues(seeTable1forestimateditemdifculties).
Itemdifcultieswereestimatedwithareliabilityof.
98,andtheaverageitemdifcultywasanchoredatM=.
00logits(SD=2.
34,Range=4.
60to4.
08).
IntMSvaluesofthe37itemsrangedfrom0.
72to1.
22(M=0.
98,SD=0.
12),andouttMSvaluesrangedfrom0.
46to2.
74(M=1.
10,SD=0.
50).
Thesendingstogetherindicatethattherewasanacceptableitemt,meaningthattherewerenotmanyobservationsthatdidnotttheoverallstructureofthedata.
PersontThepurposeofpersontmeasurementistodetectresponsepatternsthatareunlikelygiventhemodel[24].
Morespecically,persontindicateswhetherapersonrespondsasexpectedgivenhis/herindividualpredispositiontoanthro-pomorphize.
Individualpredispositionstoanthropomorphizewereestimatedwithaseparationreliabilityof.
80.
Theaver-agepredispositionwasM=.
21logits(SD=1.
16;Range=4.
13to2.
74).
Sinceitemsarerespondedtobyindividualswhocanbetiredormisreadthestatements,somemististobeexpected.
Foranacceptabletenoutof124participants(8.
1%),themodelpredictiondidnottthedataasindicatedbyat-valueoft≥1.
96.
Thesendingsindicatethattherewasanacceptablepersont,meaningthattherewerenotmanyobservationsthatdidnotttheoverallstructureofthedata.
ItemspreadAllitemsandpersonsaremappedontoasin-glescaleinFig.
1.
Ascanbeseeninthisgure,thespreadofitemssufcientlycoversthespreadofpersons.
Inotherwords,thecurrentscalewasabletoreliablymeasureindivid-ualpredispositionstoanthropomorphizeforallparticipantsinthecurrentsample.
Italsoappearedthatthetopregionofthescalecomprisedmanyitems,butnotsomanypersons.
Thus,someitemsappearedtobetoounlikelyforpersonsinthepresentsampletoattributetotherobot,andthereforedidnotcontributetotheassessmentofindividualdifferencesinpeople'spredispositiontoanthropomorphize.
Forthisrea-son,someoftheseitemswillbeomittedfromthescaleinthenextstudies.
DimensionalityNext,wetestedthehypothesisthattheitemsoftheanthropomorphismscalewouldallbelongtoonedimension.
ResultsshowedthattheRaschmodelexplained52.
8%ofthevarianceinthedata(forcomputationaldetails,see[22]).
APrincipalComponentAnalysiswasperformedonthestandardizedresiduals(i.
e.
,thedatanotexplainedbythemodel),whichcheckswhethermultipleitemssharethesameunexpectedresponsepattern(fordetails,see[22,29]).
Ifthemodelwouldtperfectly,then52.
5%oftheover-allvariancewouldbequanticationvariance,revealingaslightovert(0.
3%)tothemodel.
BecausetheRaschmodelestimatesprobabilitiesfordiscreteevents(i.
e.
,whetheraper-sonattributesacertainhuman-likecharacteristictoarobotornot),substantialquanticationvarianceistobeexpected(seealso[15]).
Theempiricalproportionofunexplainedvari-ance(i.
e.
,47.
2%)wasthushighlysimilartotheproportionofquanticationvarianceonewouldexpectwithaperfectdata-to-modelt(i.
e.
,47.
5%).
Anadditionalfactorwouldresultinanincreaseof6.
8%intheproportionofexplainedvariance.
Thesetofitemsthuslargelytappedintoasingledimensiononly.
Thesendingssupportedtheexpectedunidimensionalityofthe123482InternationalJournalofSocialRobotics(2019)11:477–494Table1Itemdifculties(δ),int-andouttmeansquares,averagevaluesofhumanuniquenessandhumannature(withthelatteradjustedtothesamescalingbysubtracting1anddividingitby7)oftheanthropomorphismscaleinStudy1Itemδ(SE)IntMSOuttMSHumanuniquenessHumannature1.
Experiencepain4.
08(.
60)1.
042.
040.
000.
542.
Unhappyaboutthedilemma3.
77(.
52)0.
910.
690.
880.
753.
Imaginative3.
52(.
47)0.
860.
510.
710.
784.
Angry3.
32(.
44)0.
870.
710.
060.
785.
Empathize2.
84(.
36)0.
730.
460.
410.
736.
Happy2.
71(.
35)0.
880.
860.
120.
697.
Chosethedilemma2.
71(.
35)1.
031.
130.
820.
758.
Satised2.
29(.
30)0.
790.
580.
120.
549.
Responsible2.
29(.
30)0.
860.
640.
240.
7210.
Freewill2.
20(.
30)1.
141.
140.
180.
6311.
Understandothers'emotions1.
88(.
27)0.
860.
620.
240.
6612.
Ambitious1.
68(.
26)0.
950.
720.
760.
8113.
Understandsthedilemma1.
21(.
23)1.
040.
920.
940.
6714.
Recognizeothers'emotions0.
55(.
21)0.
890.
850.
060.
6015.
Intentionnottoharmothers0.
47(.
21)1.
061.
030.
590.
6516.
Thinkaboutthedilemma0.
42(.
21)0.
940.
850.
880.
8117.
Self-conscious0.
42(.
21)0.
990.
950.
290.
7318.
Jump0.
26(.
20)1.
221.
240.
000.
3419.
Deliberateactions0.
15(.
20)1.
111.
380.
060.
5820.
Talk0.
19(.
20)0.
951.
040.
240.
7821.
Solveriddles0.
47(.
20)1.
010.
990.
470.
6722.
Recognizevoices0.
72(.
21)1.
111.
460.
000.
4823.
Understandlanguage0.
89(.
21)0.
860.
750.
240.
7524.
Rational0.
93(.
21)1.
051.
360.
710.
7625.
Seedepth0.
97(.
21)1.
071.
210.
000.
5526.
Anticipateonsurroundings1.
45(.
23)1.
091.
620.
120.
5827.
Consciousaboutsurroundings1.
66(.
24)0.
720.
540.
000.
5028.
Detectcolor1.
83(.
24)0.
920.
870.
000.
5429.
Purposeful1.
96(.
25)0.
921.
500.
180.
5830.
Calculate2.
16(.
26)0.
940.
780.
350.
6731.
See2.
54(.
29)1.
130.
980.
000.
3532.
Organized2.
73(.
31)1.
172.
740.
180.
5233.
Estimatedistances2.
73(.
31)1.
122.
120.
060.
4834.
Pickupobjects3.
05(.
34)1.
020.
860.
060.
5435.
Walk3.
61(.
42)1.
021.
490.
000.
4836.
Detectobjects3.
02(.
48)1.
070.
990.
000.
4937.
Avoidobjects4.
60(.
61)0.
991.
970.
000.
39scale,showingthatindividualdifferencesinpredispositionstoanthropomorphizecanbeassessedonasinglescaleofequaladditiveunits.
Inotherwords,allhuman-likecharac-teristicsincludedinthescaleweresuccessfullymappedontoasingledimension,rangingfromlowtohighonperceivedhuman-likeness.
InvariantorderingTotestthehypothesisthatitemsonthescalewouldbeinvariantlyorderedaccordingtothedif-cultyofattributingthemtorobots,thesamplewassplitinhalfanditemdifcultieswereestimatedtwice:onceforparticipantswithevenandonceforparticipantswithoddidenticationnumbers.
Consistentwiththehypothesisofperson-independentitemdifculties,thetwoestimatesofthe37itemswerehighlysimilar,r=.
97,p.
20),allowingdataofbothexperimentalconditionstobecombinedintoasinglesamplefortheanalyses.
Allpartic-ipantsparticipatedvoluntarily,gaveinformedconsent,andwerenotcompensatedforparticipation.
3.
1.
2MaterialsandProcedureParticipantsperformedthestudyonline.
Onthewelcomepage,theycouldchoosetocompletethestudyinDutchorinEnglish,afterwhichtheywereprovidedinformationabouttheprocedureofthestudyintheirpreferredlanguage.
Next,theywatchedashort(about1min)videoofoneofthetworobots,dependingontheexperimentalconditiontheywerein.
Therobotwithhuman-likephysicalfeatureswasrunningaroundandpouringwaterinacup(thevideocanbefoundathttps://goo.
gl/npyDfG),andtherobotwithhuman-likecog-nitivefeaturesappearedtobecomeangryatapersonwholeftdirtontheoor(thevideocanbefoundathttps://goo.
gl/i2Sqgg).
Afterparticipantswatchedthevideoofoneofthetworobots,theycompletedthethreemeasuringinstrumentsfor123InternationalJournalofSocialRobotics(2019)11:477–494485Table2Itemdifculties(δ),int-andouttmeansquaresoftheanthropomorphismscaleinStudy2Itemδ(SE)IntMSOuttMS13.
Understandsthedilemma5.
76(.
76)1.
305.
212.
Unhappyaboutthedilemma5.
76(.
76)0.
710.
109.
Responsible4.
43(.
46)0.
760.
225.
Empathize4.
22(.
43)1.
151.
644.
Angry3.
88(.
39)1.
002.
0112.
Ambitious3.
05(.
31)1.
030.
7311.
Understandothers'emotions2.
62(.
28)0.
790.
8117.
Self-conscious2.
46(.
27)1.
040.
9714.
Recognizeothers'emotions1.
29(.
22)1.
030.
9727.
Consciousaboutsurroundings0.
23(.
21)0.
930.
7719.
Deliberateactions0.
58(.
21)1.
081.
0721.
Solveriddles0.
63(.
21)1.
201.
1723.
Understandlanguage0.
72(.
22)1.
040.
8931.
See1.
01(.
22)0.
940.
9225.
Seedepth1.
01(.
22)1.
015.
9720.
Talk1.
27(.
24)0.
960.
9529.
Purposeful1.
39(.
24)0.
940.
6826.
Anticipateonsurroundings1.
69(.
26)0.
920.
7218.
Jump2.
40(.
31)1.
010.
9230.
Calculate2.
84(.
36)0.
920.
8136.
Detectobjects3.
28(.
41)0.
820.
3133.
Estimatedistances3.
67(.
48)0.
780.
5122.
Recognizevoices3.
67(.
48)1.
060.
9035.
Walk3.
92(.
52)1.
120.
8334.
Pickupobjects5.
16(.
82)1.
590.
70anthropomorphism.
Therstonewasa25-itemversionoftheanthropomorphismscalethatwasdevelopedandtestedinStudy1andadjustedforthecurrentstudy(seeTable2fortheitemsinthisstudy).
Someofthemostdifcultitems,aswellastheeasiestone,weredeletedbecausetheywereexpectedtocontributelittletoestimationsofpeople'spredispositiontoanthropomorphize(i.
e.
,items1,3,6,7,8,10,and37inTable1).
ThreeitemsweredeletedbecausetheconstructvaliditytestinStudy1showedthattheydidnotsufcientlyrelatetoanthropomorphism(i.
e.
,items24,28,and32,inTable1).
Item15inTable1wasdeletedbecauseitwasphrasedasadoublenegation.
Itemswereformulatedasastatementwhichcouldbeansweredwithyes(codedwitha1)orno(codedwitha0).
ThesecondquestionnairewastheGodspeed-instrument(see"AppendixB"),whichconsistedof5items(α=.
71)witha5-pointresponseformat.
Fivedummyitemswereincludedtomakethegoalofthisquestionnaireandofthisstudylessobvious.
Participants'averagedresponsesacrossthevetar-getitemswereusedintheanalyses.
ThethirdquestionnairewastheWaytz-instrument(see"AppendixA"),whichcon-sistedof6items(α=.
78)witha5-pointresponseformat.
Participants'averagedresponsesacrossthesixitemswereusedintheanalyses.
Aftercompletingthesequestionnaires,participantscom-pletedseveralquestionnairesthatwererelatedtotheoriginalpurposeofthestudy.
Thesequestionnairesmeasuredcon-ceptssuchasdesireforcontrol,needtobelong,trust,andpredictability.
Dataonthesequestionnaireswillnotbeusedinouranalysis.
Finally,participantsindicatedtheirageandgender,weredebriefedandthankedfortheirparticipation.
Thestudytookapproximately10mintocomplete.
3.
2ResultsandDiscussion3.
2.
1ModelTestInthissection,thehypothesesthatitemsandpersonscanbemappedontoasingleone-dimensionalscale,andthatitemsareinvariantlyorderedaccordingtothedifcultywithwhichtheyareascribedtoarobotaretested.
ThesectionhasthesamestructureasinStudy1.
ItemtAsinStudy1,mostitemsttedthemodelsufcientlywithintandouttMSvalues≤1.
50(seeTable2foresti-mateditemdifculties),exceptforitems13('understands123486InternationalJournalofSocialRobotics(2019)11:477–4941234567Study2-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Study1(a)12345678Cognive-8-7-6-5-4-3-2-10-8-7-6-5-4-3-2-1012345678Physical(b)Fig.
3IteminvarianceplotoftheitemdifcultiesofaStudies1and2,andbrobotswithmostlyphysicalandcognitivehuman-likefeatures.
Eachnumberrepresentsanitem,correspondingwiththenumbersinTable2.
Redlinesindicate95%condenceintervals.
(Colorgureonline)moraldilemmas',outtMS=5.
21),5('empathize',outtMS=1.
64),4('angry',outtMS=2.
01),25('seedepth',outtMS=5.
97),and34('pickupobjects',intMS=1.
59).
Itemdifcultieswereestimatedwithareliabilityofα=.
98.
TheaverageitemdifcultywasanchoredatM=.
00logits(SD=3.
14,Range=5.
16to5.
76).
IntMSvaluesofthe25itemsrangedfrom0.
71to1.
59(M=1.
00,SD=0.
18).
OuttMSvaluesofthe25itemsrangedfrom0.
10to5.
97(M=1.
23,SD=1.
35).
PersontIndividualpredispositionstoanthropomorphizewereestimatedwithareliabilityofα=.
78.
Theaveragepre-dispositionwasM=.
33logits(SD=1.
59,Range=5.
70to5.
47).
Forareasonableeightoutof131participants(6.
1%),themodelpredictiondidnottthedataasindicatedbyat-valueoft≥1.
96.
DimensionalityTheRaschmodelexplained63.
7%ofthevarianceinthedata.
Ifthemodelwouldtperfectly,then63.
5%oftheoverallvariancewouldbequanticationvari-ance.
Theempiricalproportionofunexplainedvariance(i.
e.
,36.
3%)wasthushighlysimilartotheproportionofquan-ticationvarianceexpectedwithaperfectdata-to-modelt(i.
e.
,36.
5%).
Anadditionalfactorwouldresultinanincreaseofatrivial3.
6%intheproportionofexplainedvariance.
Thesetofitemsthuslargelytappedintoasinglefactoronly.
InvariantorderingTheorderingoftheitemdifcultieswashighlysimilartothatobtainedinStudy1,asindicatedbyastrongpositivecorrelationbetweentheitemdifcultiesesti-matedinStudies1and2(r=.
88,p<.
001,seeFig.
3afortheinvarianceplot).
Thisresultsupportstheexpectationthattheprobabilitywithwhichthevarioushuman-likechar-acteristicsareascribedtorobotsislargelyindependentoftheindividual'spredispositiontodoso.
Inotherwords,thescaleshowedanorderingofhuman-likecharacteristicsthatissimilarfordifferentindividualsindifferentsamples.
Toexplorewhethertheexpectedinvarianceofitemdif-cultiesalsoholdsacrossthetwodifferentrobotsthatwereevaluated,thesamplewassplitinhalfwithrespecttotherobotthatwasevaluated.
Consistentwiththehypothesisofrobot-independentitemdifculties,thetwosetsofestimates(onefortherobotwithmostlyphysicalhuman-likefeatures,theotherfortherobotwithmostlycognitivehuman-likefea-tures)werehighlysimilar(r=.
97,p<.
001,seeFig.
3bfortheinvarianceplot).
Thisndingagainsupportstheexpec-tationthathuman-likecharacteristicsareinvariantlyorderedwithrespecttothedifcultytoattributethemtorobots.
3.
2.
2ConvergentValidityTotesttowhatextentestimatesobtainedwiththeanthropo-morphismscaleconvergedwiththetwocommonlyusedmea-suringinstrumentsforanthropomorphism,thethreescaleswerecompared.
Resultsindicatedalow,butstatisticallysignicantcorrelationbetweenourscaleandtheGodspeed-instrument(r=.
22,p=.
01).
Aftercorrectingformeasure-menterrorattenuation,thecorrelationremainedratherlow(r=.
30,forcomputationaldetails,see[6]).
Inaddition,amoderateandstatisticallysignicantcorrelationwasfoundbetweenourscaleandtheWaytz-instrument(r=.
46,p<.
001).
Aftercorrectingformeasurementerrorattenuation,thiscorrelationremainedrathermoderate(r=.
59).
3.
3ConclusionsInthisstudy,anadjusted25-itemversionoftheanthropo-morphismscalewascomparedwithtwoavailablemeasuringinstrumentsforanthropomorphismtotestforconvergentvalidity.
Resultsshowedthat,asinStudy1,people'sresponsessufcientlyttedtheRaschmodel,indicatedby123InternationalJournalofSocialRobotics(2019)11:477–494487anacceptabledata-to-modelt.
Thisresultsupportedtheexpectedinvariantorderingofhuman-likecharacteristicswithrespecttotheiritemdifculty.
Additionally,thescalecorrelatedwithexistingmeasuresofanthropomorphism,butnottotheextenttowhichwecouldclaimconvergence.
Thisindicatesthatourunderstandingofanthropomorphismisstilllimited.
Wewillelaboratemoreonthisinthegeneraldiscus-sion.
Nodifferencesbetweentherobotswerefoundonanyofthemeasuringinstrumentsforanthropomorphism.
Theorder-ingofitemsonouranthropomorphismscalewasalsohighlysimilarforbothrobots,indicatingthatthetworobotsinthisstudywereperceivedasequallyhuman-like.
Aninterestingquestionthusishowwellthescaledifferentiatesbetweenothertypesofagents.
Inthenextstudy,people'spredisposi-tionstoanthropomorphizehumans,robots,computers,andalgorithmswillbecompared.
4Study3Thecurrentstudywasoriginallydesignedtoinvestigatepeo-ple'sresponsestodifferenttypesofplayersinasocialgame:theUltimatumGame(see[13]).
Dataofthisstudywillbeusedtoexploretheanthropomorphismscale'ssensitivityfordifferentiatingbetweenvarioustypesoftechnologies.
Wecomparedpeople'sresponsestohumans,robots,computers,andalgorithms.
Wehypothesizedthatthescalewouldsuc-cessfullydifferentiatehumansfromrobots.
Inaddition,weexploredwhetheradifferenceexistsbetweentheattributionofhuman-likecharacteristicstocomputersandalgorithms.
4.
1Method4.
1.
1ParticipantsandDesignTwohundredandtwoparticipants(89malesand113females;Mage=34.
69,SDage=11.
12,Range=18to76)wererecruitedviaAmazonMechanicalTurk(MTurk)topartici-pateinanonlineexperiment.
Theywererandomlyassignedtooneoffourexperimentalconditions(Agenttype:humanvs.
robotvs.
computervs.
algorithm)ofabetween-subjectsdesign.
Participantswerepaid$1fortheirparticipation.
4.
1.
2MaterialsandProcedureParticipantsperformedthestudyonline.
Tocreateasocialinteractionthatprovidesopportunitiestoanthropomorphizetheagents,participantsplayedtheUltimatumGame(see[13]).
Inthisgame,twoplayersdivideasumofcredits.
Therstplayer(theagent)proposesacertaindivisionandthesec-ondplayer(theparticipant)decideswhether(s)heacceptsorrejectstheoffer.
Participantsinthealgorithmconditionweretoldthattherewerenootherplayersavailable,andthattheywouldbeconnectedtoanalgorithmthatwould'ran-domlygenerateoffers'duringthegame.
Participantsintheotherthreeconditionsweretoldtobeplayingthegamewithhumans,robots,orcomputers.
Duringthegame,participantswereshownpicturesoftheotherplayers.
AnexampleofeachoftheagentsisprovidedinFig.
4.
AfterplayingtheUltimatumGame,participantscom-pletedanadjusted19-itemversionoftheanthropomorphismscale.
Themaindependentvariableinthisstudywaspar-ticipants'behaviorintheultimatumgame,andtheanthro-pomorphismscalewasincludedtoexplorethepossiblemediating/moderatingroleofanthropomorphism.
Sincethelatterwasnotthemaingoal,andduetotimeconstraints,fewerindicatorswereusedinthisstudy.
Itemsforthisadjustedver-sionwereselectedfromtheoriginalsetof37itemsinsuchawaythattheywouldstillcoverawiderangeoftheanthro-pomorphismcontinuum(seeTable3fortheitemsinthisstudy).
Itemswereformulatedasastatementwhichcouldbeansweredwithyes(codedwitha1)orno(codedwitha0).
Aftercompletingtheanthropomorphismquestionnaire,participantsindicatedtheirageandgender,weredebriefed,thankedfortheirparticipation,andpaidthroughtheMTurksystem.
Theexperimenttookapproximately6mintocom-plete.
Fig.
4Examplesofpicturesusedintheahuman,brobot,ccomputer,anddalgorithmgroupsinStudy3123488InternationalJournalofSocialRobotics(2019)11:477–4944.
2ResultsandDiscussion4.
2.
1ModelTestInthissection,thehypothesesthatitemsandpersonscanbemappedontoasingleone-dimensionalscaleandthatitemsareinvariantlyorderedaccordingtothedifcultywithwhichtheyareascribedtoagentsaretested.
Thesectionhasthesamestructureasinstudies1and2.
ItemtMostitemsttedthemodelsufcientlywithintandouttMSvalues≤1.
50(seeTable3forestimateditemdifculties),exceptforitems12('ambitious',outtMS=1.
59),33('estimatedistances',outtMS=1.
97),and30('calculate',outtMS=5.
59).
Itemdifcultieswereestimatedwithareliabilityofα=.
98.
TheaverageitemdifcultywasanchoredatM=.
00logits(SD=1.
83,Range=4.
15to3.
05).
IntMSvaluesofthe19itemsrangedfrom0.
76to1.
29(M=1.
00,SD=0.
17).
OuttMSvaluesofthe19itemsrangedfrom0.
46to5.
59(M=1.
14,SD=1.
12).
PersontIndividualpredispositionstoanthropomorphizewereestimatedwithareliabilityofα=.
88.
TheaveragepredispositionwasM=.
15logits(SD=3.
28,Range=5.
79to5.
31).
Forareasonabletenoutof202participants(5.
0%),themodelpredictiondidnottthedataasindicatedbyat-valueoft≥1.
96.
DimensionalityTheRaschmodelexplained52.
4%ofthevariance.
Ifthemodelwouldtperfectly,then52.
1%oftheoverallvariancewouldbequanticationvariance.
Thepro-portionofunexplainedvariance(i.
e.
,47.
6%)washighlysim-ilartotheproportionofquanticationvarianceexpectedwithaperfectdata-to-modelt(i.
e.
,47.
9%).
Anadditionalfactorwouldresultinanincreaseof8.
3%oftheexplainedvariance.
Next,theorderingofitemsinthecurrentstudywascomparedwiththatofStudy1.
Resultsshowedamoder-atebutsignicantcorrelationbetweenthetwoestimates(r=.
58,p<.
01,seeFig.
5fortheinvarianceplot).
Despitethesignicantcorrelationbetweentheitemdif-culties,manyoftheestimateddifcultiesappearedoutsideofthe95%condenceinterval,indicatingsubstantialdiffer-encesbetweenthetwostudies.
Someoftheseitems(i.
e.
,'pickupobjects',or'walk')hadloweritemdifcultiesinStudy1thaninStudy3,andsome(i.
e.
,'calculate',or'understandslanguage')hadloweritemdifcultiesinStudy3thaninStudy1.
Somehuman-likecharacteristicswerethusmoreeasyormoredifculttoattributetoaspecictypeofagentthanothercharacteristics.
Morespecically,theexpectedinvari-antorderingofhuman-likecharacteristicswithrespecttotheirprobabilityofbeingascribedtonon-humanagentswasnotsupportedwhenotheragentsthanrobotswereevaluated.
Inthefollowingsections,wewillinvestigatetheinvariantorderingoftheitemsinStudy3inmoredetail.
4.
2.
2InvariantOrderingofItemsToexplorewhichhuman-likecharacteristicsdifferedintheirprobabilityofbeingattributedtohumans,robots,computers,andalgorithms,itemdifcultieswereestimatedseparatelyTable3Itemdifculties(δ),int-andouttmeansquaresoftheanthropomorphismscaleinStudy3Itemδ(SE)IntMSOuttMS14.
Recognizeothers'emotions3.
05(.
36)1.
291.
032.
Unhappyaboutthedilemma2.
36(.
32)1.
100.
704.
Angry2.
17(.
31)0.
830.
6011.
Understandothers'emotions1.
90(.
29)0.
930.
5313.
Understandsthedilemma1.
51(.
27)0.
990.
7927.
Consciousaboutsurroundings1.
04(.
25)0.
780.
4618.
Jump1.
04(.
25)0.
860.
7335.
Walk0.
27(.
22)0.
870.
6417.
Self-conscious0.
17(.
22)0.
760.
5426.
Anticipateonsurroundings0.
12(.
22)0.
960.
6434.
Pickupobjects0.
02(.
22)0.
920.
7112.
Ambitious0.
30(.
21)1.
091.
5920.
Talk0.
30(.
21)0.
990.
7733.
Estimatedistances0.
95(.
21)1.
051.
9736.
Detectobjects1.
20(.
21)0.
800.
7121.
Solveriddles1.
54(.
21)1.
341.
1729.
Purposeful2.
12(.
22)1.
171.
1423.
Understandlanguage3.
04(.
24)1.
031.
3130.
Calculate4.
15(.
29)1.
255.
59123InternationalJournalofSocialRobotics(2019)11:477–494489Fig.
5IteminvarianceplotoftheitemdifcultiesinStudies1and3.
Eachnumberrepresentsanitem.
Redlinesindicate95%condenceintervals.
(Colorgureonline)12345Study3-5-4-3-2-10-5-4-3-2-1012345Study1foreachofthesefourplayertypes.
Althoughallcorrelationsbetweenthefoursetsofitemdifcultiesweresignicant(seeTable4),someofthecharacteristicswereclearlydifferentintheirprobabilityofbeingascribedtospecicagents.
Figure6displaystheiteminvarianceplotsbetweeneachoftheagenttypes.
AscanbeseeninthisFigure,threeofthe19items(i.
e.
,items18'jump',34,'pickupobjects',and35'walk')wereconsistentlylesslikelytobeascribedtoalgorithmsandcomputersthantorobotsandhumans.
Thisshouldnotcomeasasurprise,ascomputersandalgorithmslackthemorphologythataccommodatessuchphysicalactiv-ities.
Thisndingrevealedthatacomparisonbetweentheagenttypesisunfaironspecicattributes.
Usingacomputer'sabilitytoperformphysicalfeaturesasameasurementofitshuman-likenessissimilartojudgingpeoplewhoareboundtoawheelchairaslesshumanthantheirablecounterpartsfornotbeingabletojump.
4.
2.
3SensitivityinDifferentiatingBetweenAgentsTotesttheextenttowhichtheanthropomorphismscalediffer-entiateshumansfromotherplayertypes,aone-wayAnalysisofVariance(ANOVA)wasconductedwithagenttypeasindependentvariableandtheindividualpredispositiontoanthropomorphizeasdependentvariable.
Resultsindicatedastatisticallysignicanteffectofagenttype,F(3,201)=42.
04,p<.
001,η2=.
39.
Morespecically,anthropomorphismwashighestforhumans(M=3.
50,SD=2.
35),followedbyrobots(M=0.
63,SD=2.
81),algorithms(M=1.
53,SD=2.
82)andcomputers(M=1.
59,SD=2.
29),seeFig.
7a.
Pairwisecomparisons(LSD)showedastatisticallysig-nicantdifferencebetweenhumansandallotheragents,t(198)=11.
03,p<.
001,d=1.
89.
Post-hoccompar-isonsusingBonferronicorrectionindicatedthatdifferencesbetweenanytwogroupsotherthanhumanwerenotsigni-cant.
Interestingly,afterremovingthethreemorphology-relateditems(i.
e.
,items18,34,and35)fromthescale,thedif-ferencesinanthropomorphismbetweennon-humanplayertypesbecamesmaller,withestimationsbeinghighestforhumans(M=3.
48,SS=2.
34),followedbyrobots(M=0.
83,SD=2.
90),algorithms(M=1.
32,SD=2.
83)andcomputers(M=1.
35,SD=2.
37),seeFig.
7b.
Thiscouldbecausedbythenatureoftheexperiment,becauseallplayersbehavedintheexactsamewayduringtheUltima-tumGame.
Whenthoseplayersweresubsequentlycomparedwithrespecttocharacteristicsthattheywereequallylikelytopossessbasedonthebehaviorthattheyshowed,itshouldnotcomeasasurprisethatnodifferencesbetweenthemwerefound.
Table4CorrelationsbetweenitemdifcultiesinthefourexperimentalconditionsinStudy3HumanRobotComputerAlgorithmr=.
74p<.
001r=.
68p<.
01r=.
88p<.
001Computerr=.
59p<.
01r=.
50p=.
03Robotr=.
73p<.
001123490InternationalJournalofSocialRobotics(2019)11:477–4941234567Computer-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Algorithm(a)algorithmvscomputer1234567Robot-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Algorithm(b)algorithmvsrobot1234567Human-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Algorithm(c)algorithmvshuman1234567Robot-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Computer(d)computervsrobot1234567Human-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Computer(e)computervshuman1234567Human-7-6-5-4-3-2-10-7-6-5-4-3-2-101234567Robot(f)robotvshumanFig.
6IteminvarianceplotsoftheitemdifcultiesofeachofthefourexperimentalconditionsinStudy3:algorithm,computer,robot,andhuman.
Eachnumberrepresentsanitem.
Redlinesrepresent95%condenceintervals.
(Colorgureonline)4.
3ConclusionsInthecurrentstudy,anadjusted19-itemversionoftheanthropomorphismscalewasusedtoinvestigatethescale'ssensitivity.
Resultsshowedthat,asinStudies1and2,peo-ple'sresponsessufcientlyttedtheRaschmodel,indicatedbyanacceptabledata-to-modelt.
Thisresultsupportedtheexpectedinvariantorderingofthehuman-likecharacteristicswithrespecttotheiritemdifculty.
Anadequatelevelofsensitivitywasfound.
Thescalewasabletodifferentiatehumansfromdifferenttypesoftechno-logicalplayers,butitdidnotdifferentiatethoseplayersfrom123InternationalJournalofSocialRobotics(2019)11:477–494491-3-2-1012345Person-measurealgorithmcomputerrobothumanExperimentalcondion(a)-3-2-1012345Person-measurealgorithmcomputerrobothumanExperimentalcondion(b)Fig.
7VisualizationofaveragedpersonmeasuresonthescaleabeforeandbafterremovingthebiaseditemsinStudy3.
Whiskersrepresent95%condenceintervalseachother.
Presumablytheconceptualdifferencesbetweenthoseplayertypesweretoosmalltobedetectedbythecur-rentversionofthescale.
Additionally,thescale'ssensitivitydroppedwhenthreemorphology-relateditemswereremovedfromtheanalysis.
5GeneralDiscussionThecurrentresearchwasdesignedtoexplorewhetheranthro-pomorphismcanbesuccessfullymeasuredusingtheRaschmodel,whethertheconceptcanbemappedontoaone-dimensionalscale,andwhetherhuman-likecharacteristicsareorderedinawaythatissimilarforallindividualsintheirencounterwithrobotsandothertypesofagentsindif-ferentcontexts.
Wearguedthathuman-likecharacteristicscanbeorderedaccordingtotheprobabilitywithwhichtheyareascribedtorobots,andthatthisorderingofhuman-likecharacteristicsontherangeofperceivedhuman-likenessissimilarforallindividualsintheirencounterwithdifferenttypesofagents.
Wedevelopedasetofhuman-likecharac-teristicsanduseddatafromthreestudiestotestthescale'spsychometricqualities.
Thesestudieshaddesignswithdif-feringcontextsandexperimentalconditions,andpeople'spredispositionstoanthropomorphizewerecomparedindif-ferentsampleswithdifferenttypesofagents.
Intherststudy,wehypothesizedthatitemsandpersonscouldbemappedontoasingleone-dimensionalscale,andthatitemswouldbeinvariantlyorderedaccordingtothedif-cultywithwhichtheyareattributedtoarobot.
Additionally,theestimateddifcultieswithwhichthe37characteristicsareattributedtoarobotwereexpectedtoberelatedtotheirperceivedhumannatureandhumanuniqueness.
Inthesec-ondstudy,wetestedtheextenttowhichestimatesmadewiththreedifferentmeasuringinstrumentsforanthropomorphismwouldberelated.
Inthethirdstudy,wehypothesizedthattheanthropomorphismscalewouldsuccessfullydifferenti-atehumansfromothertypesofagents.
Allhypotheseswere(atleastpartially)conrmed,andthenextsectionsdescribeinmoredetailthedimensionalityofanthropomorphism,implicationsforfurtherresearch,andthecomparisonofthethreedifferentmeasuringinstruments.
5.
1DimensionalityofAnthropomorphismAcrossstudies,aninvariantorderingofhuman-likecharac-teristicswasfound,indicatingthatthisorderingwassimilarfordifferentpeopleintheirencounterwithdifferenttypesofagentsindifferentcontexts.
Thisndingwassupportedbydimensionalityteststhatconsistentlyindicatedthatthedatacouldberepresentedinaone-dimensionalstructure.
Morespecically,ineachofthestudies,anadditionalfactorwouldresultinonlyasmallincreaseintheproportionofexplainedvariance.
Thisresultsupportedthehypothesisthatanthropomorphismcanberepresentedasaone-dimensionalconstruct.
Testsofconstructvalidityshowedsignicantcorrelationsbetweenitemdifcultiesandtheirperceivedhumannatureandhumanuniqueness,supportingtheexpectationthatthescalemeasuresanthropomorphism.
Together,thesendingsindicatethathuman-likecharacteristicsareorderedinsuchawaythattheyrangefromlowtohighonasingledimension.
5.
2ImplicationsforFurtherResearchFindingsonsomeofthecharacteristicshaveimportantimplicationsforfutureresearchandthusneedsomefurther123492InternationalJournalofSocialRobotics(2019)11:477–494consideration.
Forexample,inStudy1theitem'Experiencepain'wasratedasextremelylowinhumanuniquenessandasmediuminhumannature,butitappearedtobethemostdifcultonetoascribetoarobot.
Onepossibleexplanationforthisunexpectedndingcouldbethatanervoussystemisnecessaryforexperiencingpain,whichisnotuniqueforhumans,butissomethingthatrobotsclearlydonothave.
Thisresultalsoraisestheissueofphysicalversuscogni-tivecapacities.
Forhumansandotherorganisms,theabilitytomovearoundinthe(physical)environmentisagiven,whereasforcertaintechnologicalartifactsthismaynotbesoobvious.
Likewise,futurearticialintelligencemaycre-ateagentswithhighmentalcapacitieswithoutabody,whichmakesiteasierforthemtosolveamoraldilemmathantopickupanobject.
Thisissuemaybecomeapparentinthenearfuture,andmoreresearchisneededtofurtherinvesti-gatethis.
AlargebenetoftheRaschapproachistheabilitytoselectitemsforaspecicdesign,andassuchcancopewithsuchtechnologicaldevelopments.
Anotherimportantndingwasthatwhenmorphology-relatedcharacteristics(i.
e.
,theitemsaboutarobotbeingabletojump,walk,andpickupobjects)weredisregardedinStudy3,thedifferencesonanthropomorphismbetweenrobots,computers,andalgorithmsdecreased.
Futureresearchcanbedesignedtoexplorewhetherwecancreateameasureforanthropomorphismthatisuniversalforvarioustypesofagents(e.
g.
,humans,animals,technologies,anddeities).
5.
3ComparisonofMeasuringInstrumentsTheanthropomorphismscalewascomparedwithtwoavail-ablemeasuringinstrumentsforanthropomorphismtotestforconvergentvalidity.
TheseavailableinstrumentsweretheGodspeed-andWaytz-instruments.
Wetestedwhethermeasurementsobtainedwiththethreeinstrumentswouldberelated.
Thishypothesiswasconrmedbysignicantcor-relationsbetweenthescaleandbothotherinstruments,butthesecorrelationswereratherlow.
Althoughthesecorrelationsweresmallerthanexpected,thisshouldnotcomeasasurprise,giventhenatureofthedif-ferentinstruments.
Theywerealldevelopedwithdifferentviewsontheconceptanthropomorphism.
TheGodspeed-instrumentfocusesonmostlyappearance-relatedfeatures,whereastheWaytz-instrumentfocusesmostlyoncognition-relatedfeatures.
Thelowcorrelationalsoclearlyshowsthatwestillhavelimitedunderstandingofwhatanthropomorphismentails,andthuswhatindicatorsarebestusedinitsmeasurement.
Itisonlywhenmultipledifferentmeasuringinstrumentscon-versethatwecanclaimtofullygraspwhattheindicatorsofanthropomorphismare.
WebelievethattheRaschmodelisapromisingtooltouncoversuchindicatorsandthuswhatanthropomorphismtrulyentails.
5.
4LimitationsandFutureResearchInallstudies,theitemsoftheanthropomorphismscalewereorderedalphabetically,whichmayhaveinuencedpeople'sresponsestocertainitemsbecauseofordereffects.
Theitemsonthescaleshouldbeorderedrandomlyinfuturestudiestopreventtheoccurrenceofsuchorderingeffects.
AllstudieswereperformedwithmainlyDutchandAmer-icanparticipants,sotheirculturalbackgroundsandexperi-enceswithtechnologycouldhavebeenquitesimilar.
Earlierworkhasshownthatpeople'stendenciestoattributehuman-likenesstonon-humanscanberelatedtoreligion[14],andthatpeoplefromdifferentcountries(suchasindividualisticversuscollectivisticones)responddifferentlytocomputers[19]andevaluaterobotsdifferently[28].
Itwouldbeinter-estingtoinvestigatewhetherculturaldifferencesinuencepeople'spredispositionstoanthropomorphize.
Inaddition,weusedsocialmediaforsamplingpartici-pantsinstudies1and2,andthusreliedondatagatheredfromexperimenters'acquaintances.
Thismayhaveledustoendupwithhomogeneousgroupsofparticipants,whichcouldpartlyexplainthehighconsistencybetweenthosestudies.
Forincreasingourunderstandingoftheconceptanthropomor-phism,itisimportanttoalsocollectdatafrommorediversegroupsandcheckwhetherthehighconsistencyprevails.
InStudy2,eachstatementwaspresentedasageneralstate-mentabout'arobot',andnotspecicallyaimedattherobotthatparticipantswatchedintheshortmovieclip.
Thiscouldexplainwhynodifferencesinanthropomorphismwerefoundbetweenthetworobots.
Futureresearchthatisdesignedtoinvestigateevaluationsofdifferentrobotsshouldthereforenotusethescaleinageneralway,butratherphraseeachquestiontobespecicallyaboutacertainrobot(orothernon-humanobject).
Wedidnottestwhetherpeoplehadpreviousknowledgeaboutorexperiencewithrobots.
Thisexperiencecouldinu-encepeople'sresponsestothoserobots(seee.
g.
,[21,25])andshouldthereforebetakenintoaccountinfuturestudies.
TheRaschmodeloffersapromisingapproachtoinvestigatehowsuchexperiencesaffectaperson'spredispositiontoanthropo-morphize,thedifcultyofascribingaspeciccharacteristicstoarobot,orboth.
5.
5ConclusionsDespitesomelimitations,wehaveshownthatanthropomor-phismcanbemeasuredonaone-dimensionalscale,andthatitemsareorderedinvariantlywhenthisscaleisappliedtorobots.
TheRaschmodelthusprovidesareliablewayofmea-suringanthropomorphism.
Becauseoftheinvariantorderingofthehuman-likecharacteristics,thescaleprovidesoppor-tunitiesforcomparingvarioustypesofagentsandrobotswitheachotheracrossstudies.
Themethodalsoprovidesthe123InternationalJournalofSocialRobotics(2019)11:477–494493possibilitytoselectitemsbasedonthecontextofthestudy,makingitaversatiletoolformeasuringanthropomorphism.
FindingsofStudy3showedthatdifferenttypesofagents(excepthumans)weredifculttodistinguish,presumablybecausethebehaviorofthesedifferentagentswasidentical.
Itisthereforeaninterestingquestionhowthemethodperformswheninteractionswithdifferenttypesofagentsareevalu-ated.
Ultimately,ascalecanbedevelopedthatcontributestoourunderstandingofwhatmakespeopleattributehuman-likecharacteristicstosocialrobotsandothernon-humans.
Thisunderstandingcanhelpdesignerstocreaterobotsassocialentitiesthatareacceptedasmembersofoursociety.
OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.
0InternationalLicense(http://creativecommons.
org/licenses/by/4.
0/),whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinktotheCreativeCommonslicense,andindicateifchangesweremade.
AWaytz-InstrumentItemsoftheanthropomorphismquestionnaireadaptedfrom[32],answeredona5-pointora7-pointresponseformat.
"Towhatextentdoestherobothavethoughtsofitsown""Towhatextentdoestherobothaveintentions""Towhatextentdoestherobothaveafreewill"Towhatextentdoestherobothaveaconsciousness""Towhatextentdoestherobothavedesires""Towhatextentdoestherobothavevaluesandnorms""Towhatextentdoestherobotexperienceemotions"BGodspeed-InstrumentTheitemsoftheanthropomorphismpartoftheGodspeedquestionnaire(adaptedfrom[2]),answeredona5-pointora7-pointresponseformat.
"Fake-Natural""Machinelike-Humanlike""Unconscious-Conscious""Articial-Lifelike""Movingrigidly-Movingelegantly"References1.
AirentiG(2015)Thecognitivebasesofanthropomorphism:fromrelatednesstoempathy.
IntJSocRobot7(1):117–1272.
BartneckC,KulicD,CroftE,ZoghbiS(2009)Measurementinstrumentsfortheanthropomorphism,animacy,likeability,per-ceivedintelligence,andperceivedsafetyofrobots.
IntJSocRobot1(1):71–813.
BondTG,FoxCM(2013)ApplyingtheRaschmodel:fundamentalmeasurementinthehumansciences.
PsychologyPress,Portland4.
BruceA,NourbakhshI,SimmonsR(2002)Theroleofexpressive-nessandattentioninhuman–robotinteraction.
In:ProceedingsofIEEEinternationalconferenceon2002roboticsandautomation.
ICRA'02,vol4.
IEEE,pp4138–41425.
CarpinellaCM,WymanAB,PerezMA,StroessnerSJ(2017)Theroboticsocialattributesscale(rosas):developmentandvalidation.
In:Proceedingsofthe2017ACM/IEEEinternationalconferenceonhuman–robotinteraction.
ACM,pp254–2626.
CharlesEP(2005)Thecorrectionforattenuationduetomea-surementerror:clarifyingconceptsandcreatingcondencesets.
PsycholMethods10(2):206–2267.
ChidambaramV,ChiangYH,MutluB(2012)Designingpersua-siverobots:howrobotsmightpersuadepeopleusingvocalandnonverbalcues.
In:ProceedingsoftheseventhannualACM/IEEEinternationalconferenceonhuman–robotinteraction.
ACM,pp293–3008.
DuffyBR(2003)Anthropomorphismandthesocialrobot.
RobotAutonSyst42(3):177–1909.
EpleyN,WaytzA,CacioppoJT(2007)Onseeinghuman:athree-factortheoryofanthropomorphism.
PsycholRev114(4):864–88610.
EysselF,ReichN(2013)Lonelinessmakestheheartgrowfonder(ofrobots):ontheeffectsoflonelinessonpsychologicalanthropomorphism.
In:8thACM/IEEEinternationalconferenceonhuman–robotinteraction(HRI),2013.
IEEE,pp121–12211.
EysselF,HegelF,HorstmannG,WagnerC(2010)Anthropomor-phicinferencesfromemotionalnonverbalcues:acasestudy.
In:RO-MAN,2010IEEE.
IEEE,pp646–65112.
EysselF,KuchenbrandtD,BobingerS(2011)Effectsofanticipatedhuman–robotinteractionandpredictabilityofrobotbehavioronperceptionsofanthropomorphism.
In:Proceedingsofthe6thinter-nationalconferenceonhuman–robotinteraction.
ACM,pp61–6813.
GüthW,SchmittbergerR,SchwarzeB(1982)Anexperimentalanalysisofultimatumbargaining.
JEconBehavOrgan3(4):367–38814.
GuthrieS(1993)Facesintheclouds:anewtheoryofreligion.
OxfordUniversityPress,Oxford15.
HaansA,KaiserFG,BouwhuisDG,IJsselsteijnWA(2012)Individualdifferencesintherubber-handillusion:predictingself-reportsofpeople'spersonalexperiences.
ActaPsychol141(2):169–17716.
HaslamN,BainP,DougeL,LeeM,BastianB(2005)Morehumanthanyou:attributinghumannesstoselfandothers.
JPersonalSocPsychol89(6):937–95017.
HaslamN,LoughnanS,KashimaY,BainP(2008)Attributinganddenyinghumannesstoothers.
EurRevSocPsychol19(1):55–8518.
KaiserFG,MertenM,WetzelE(2018)HowdoweknowwearemeasuringenvironmentalattitudeSpecicobjectivityasthefor-malvalidationcriterionformeasuresoflatentattributes.
JEnvironPsychol55:139–14619.
KatagiriY,NassC,TakeuchiY(2001)Cross-culturalstudiesofthecomputersaresocialactorsparadigm:thecaseofreciprocity.
Usabilityevaluationandinterfacedesign:cognitiveengineering,intelligentagents,andvirtualreality,pp1558–156220.
KennedyJS(1992)Thenewanthropomorphism.
CambridgeUni-versityPress,Cambridge21.
LemaignanS,FinkJ,DillenbourgP,BraboszczC(2014)Thecog-nitivecorrelatesofanthropomorphism.
In:Proceedingsofthe2014ACM/IEEEinternationalconferenceonhuman–robotinteraction22.
LinacreJM(2003)Datavariance:explained,modeledandempir-ical.
RaschMeasTrans17(3):942–943123494InternationalJournalofSocialRobotics(2019)11:477–49423.
LinacreJM(2006)Mistdiagnosis:intouttmean-squarestandardized.
www.
winsteps.
com/winman/diagnosingmist.
Accessed1Oct201824.
MeijerRR,SijtsmaK(2001)Methodologyreview:evaluatingper-sont.
ApplPsycholMeas25(2):107–13525.
RuijtenPAM,CuijpersRH(2017)Dynamicperceptionsofhuman-likenesswhileinteractingwithasocialrobot.
In:Proceedingsofthecompanionofthe2017ACM/IEEEinternationalconferenceonhuman–robotinteraction.
ACM,pp273–27426.
ShaverP,SchwartzJ,KirsonD,O'ConnorC(1987)Emotionknowledge:furtherexplorationofaprototypeapproach.
JPersonSocPsychol52(6):1061–108627.
ShawF(1991)Fitsabout"mist".
RaschMeasTrans5(1):13228.
ShibataT,WadaK,IkedaY,SabanovicS(2009)Cross-culturalstudiesonsubjectiveevaluationofasealrobot.
AdvRobot23(4):443–45829.
SmithEVJr(2002)Understandingraschmeasurement:detectingandevaluatingtheimpactofmultidimenstionalityusingitemtstatisticsandprincipalcomponentanalysisofresiduals.
JApplMeas3(2):205–23130.
SyrdalDS,DautenhahnK,WaltersML,KoayKL(2008)Sharingspaceswithrobotsinahomescenario-anthropomorphicattribu-tionsandtheireffectonproxemicexpectationsandevaluationsinaliveHRItrial.
In:AAAIfallsymposium:AIinEldercare:newsolutionstooldproblems,pp116–12331.
WaytzA,CacioppoJ,EpleyN(2010a)WhoseeshumanThestabilityandimportanceofindividualdifferencesinanthropomor-phism.
PerspectPsycholSci5(3):219–23232.
WaytzA,MorewedgeCK,EpleyN,MonteleoneG,GaoJH,CacioppoJT(2010b)Makingsensebymakingsentient:effectancemotivationincreasesanthropomorphism.
JPersonSocPsychol99(3):410–43533.
WrightBD(1977)Solvingmeasurementproblemswiththeraschmodel.
JEducMeas14(2):97–11634.
WrightBD,LinacreJM(1987)Dichotomousraschmodelderivedfromspecicobjectivity.
RaschMeasTrans1(1):5–635.
WrightBD,LinacreJM,GustafsonJE,Martin-LofP(1994)Rea-sonablemean-squaretvalues.
RaschMeasTrans8(3)36.
YoungJE,SungJ,VoidaA,SharlinE,IgarashiT,ChristensenHI,GrinterRE(2011)Evaluatinghuman–robotinteraction.
IntJSocRobot3(1):53–6737.
ZawieskaK,DuffyBR,StronskaA(2012)Understandinganthro-pomorphisationinsocialrobotics.
PomiaryAutomatykaRobotyka16:78–8238.
ZotowskiJ,StrasserE,BartneckC(2014)Dimensionsofanthro-pomorphism:fromhumannesstohumanlikeness.
In:Proceedingsofthe2014ACM/IEEEinternationalconferenceonhuman–robotinteraction.
ACM,pp66–73Publisher'sNoteSpringerNatureremainsneutralwithregardtojuris-dictionalclaimsinpublishedmapsandinstitutionalafliations.
PeterA.
M.
RuijtenisAssistantProfessoroftheHuman-TechnologyInteractionGroupatEindhovenUniversityofTechnology.
HehasaBachelorinElectricalEngineeringandaMasterinHuman-TechnologyInteraction.
HisresearchinterestsareSocialRobots,Anthropomor-phism,BehaviorChange,andapplicationsofthisinHuman-InspiredMachines.
AntalHaansisassistantprofessorinenvironmentalpsychologyintheHuman-TechnologyInteractiongroupatEindhovenUniversityofTechnology.
Hisresearchfocusesontheinterplaybetweenhumansandtheirsurroundings—includingbuiltandmediaenvironments—inexplaininghumanexperienceandperformance.
Researchinterestsincludeenvironmentalperception,presenceinVR,mediatedsocialtouch,psychologicalmeasurement,and(smart)urbanlightingappli-cations.
JaapHamisassociateprofessoratEindhovenUniversityofTech-nology.
Hestudiestheory-drivenandproblem-drivenquestionsfocus-ingonhowtechnologycanchangehumanbehaviorandthinking.
HeacquiredresearchgrantsandpublishedpapersinjournalsandconferenceproceedingsonPersuasiveTechnology,PersuasiveSocialRobotics,AmbientPersuasion,andVRforchangingsustainabilitybehaviorandhealthbehavior.
CeesJ.
H.
Middenisem.
professoratEindhovenUniversityoftechnol-ogyspecializingintheinteractionbetweenhumansandtechnologicalsystems.
HereceivedadegreeinPsychologyattheUniversityofLei-den.
Hisresearchinterestsconcernthesocialandcognitivefactorsofhuman-technologyinteractionsasthesebecomeapparentinthecon-sumptionanduseofproductsandsystems.
Hepublishedonpersua-sivetechnologyandbehaviorchange,socialrobotics,environmentalconsumerbehaviorandtheperceptionandcommunicationoftechno-logicalrisks.
123