Extending the Decision Field Theory to Model Opera

IEEETRANSACTIONSONSYSTEMS,MAN,ANDCYBERNETICS—PARTA:SYSTEMSANDHUMANS,VOL.36,NO.5,SEPTEMBER2006943ExtendingtheDecisionFieldTheorytoModelOperators’RelianceonAutomationinSupervisoryControlSituationsJiGaoandJohnD.LeeAbstract—Appropriatetrustinandrelianceonautomationarecriticalforsafeandefficientsystemoperation.Thispaperfillsanimportantresearchgapbydescribingaquantitativemodeloftrustinautomation.Weextenddecisionfieldtheory(DFT)todescribethemultiplesequentialdecisionsthatcharacterizerelianceonau-tomationinsupervisorycontrolsituations.ExtendedDFT(EDFT)representsaniterateddecisionprocessandtheevolutionofop-eratorpreferenceforautomaticandmanualcontrol.TheEDFTmodelpredictstrustandreliance,anddescribesthedynamicinteractionbetweenoperatorandautomationinaclosed-loopfashion:theproductsofearlierdecisionscantransformthenatureoflatereventsanddecisions.ThesimulationresultsshowthattheEDFTmodelcapturesseveralconsistentempiricalfindings,suchastheinertiaoftrustandthenonlinearcharacteristicsoftrustandreliance.Themodelalsodemonstratestheeffectsofdifferenttypesofautomationontrustandreliance.ItispossibletoexpandtheEDFTmodelformultioperatormultiautomationsituations.IndexTerms—Automation,decisionmaking,relianceonauto-mation,supervisorycontrol,trustinautomation.I.INTRODUCTIONAUTOMATIONcanimprovetheefficiencyandsafetyofcomplexanddangerousoperatingenvironmentsbyre-ducingthephysicalormentalburdenonhumanoperators.However,operatorsdonotalwaysrelyonautomationappro-priately,whichcanproducetheoppositeeffect,causingineffi-ciencyorunderminingsafety[1],[2].Researchexaminingtheperformanceofpeopleworkingwithautomatedsystemshasreceivedincreasingattentionastheproblemsassociatedwithunderrelianceandoverrelianceonautomationhaveresultedinseveralhigh-profiledisasters[2]–[4].Supervisorycontrolsystemsrepresentsituationsinwhichautomationandpeopleworktogethertoaccomplishtasks[5].Appropriaterelianceonautomationisanimportantfactoraffectingtheperformanceofsuchsystems,butonlyafewcomputationalmodelshavebeendevelopedtoexaminethefactorsinfluencingthedecisiontorelyonautomation[6]–[9].Thepresentstudyprovidesaquantitativemodeltodescribethefactorsthataffectrelianceondifferenttypesofautomation.ManuscriptreceivedMay26,2004;revisedDecember14,2004andApril15,2005.ThisworkwassupportedbytheNationalScienceFoundationunderGrantNo.0117494.ThispaperwasrecommendedbyAssociateEditorR.A.Hess.TheauthorsarewiththeDepartmentofMechanicalandIndustrialEngi-neering,UniversityofIowa,IowaCity,IA52246USA(e-mail:[e-mail];[e-mail]).DigitalObjectIdentifier10.1109/TSMCA.2005.855783Relianceonautomationinasupervisorycontrolsituationrepresentsdecisionmakingunderriskanduncertainty.Undersuchconditions,itisnotonlycognitivefactorsbutalsoemo-tionalfactorsthatinfluencedecisionmaking[10]–[12].Peoplereacttotheprospectofriskattwolevels:theyevaluateitcog-nitively,andreacttoitemotionally[10].Thesetworeactionshavedifferentdeterminants;specifically,cognitiveevaluationsofriskaresensitivetotheobjectivefeaturesoftheriskysituation,suchasprobabilitiesofoutcomesandassessmentsofoutcomeseverity,whereasemotionalreactionsaresensitivetothevividnesswithwhichconsequencescanbeimagined,ortorecallofpreviousexperience[10].Trustisanimportantemotionalfactorinfluencingdecisionmakingasitrelatestothedecisiontorelyonautomation[13].Trustrepresentsanaffectiveresponsetothecapabilityoftheautomation.Substantialevidencesuggeststhattrustinau-tomationisameaningfulandusefulconstructtounderstandtheoperators’relianceonautomation.Areviewofdifferentmodelsoftrustinautomationshowsthatthecurrentmodelssufferfromatradeoffbetweenquantifyingpredictionsandincludingaplausiblepsychologicalbasis[14].Conceptualandqualitativemodelsdescribethedynamicsoftrustbetweenhumansandautomation[13],[15]–[17],andtimeseriesandregressionmodelsprovideaquantitativedescriptionoftrust,butareprimarilystatisticalmodelsthatarefittothedata[1],[18]–[20].Self-confidenceinmanualcontrolcapability,anothercriticalfactorindecisionmaking[21],[22],canmediatetheeffectoftrustonreliance[18].Acomputationalmodeloftrustinautomationandself-confidencecanfacilitatetheunderstandingofthedynamicsbetweentrustandself-confidence,andhowtheseinfluencerelianceonautomation.Anotheradvantageofadoptingacomputationalmodelingapproachisthatitcangen-eratetestablepredictionsabouthumanperformanceindiffer-entsettings.Thispaperprovidesthefirststepindevelopingaquantitativemodeloftrustandreliancethatislinkedtorigorouspsycho-logicalmodelsofdecisionmaking.Specifically,weextendthedecisionfieldtheory(DFT)[23]tocharacterizetheoperators’multiplesequentialdecisionsinsupervisorycontrolsituationsinordertocreateacomputationalmodelthatpredictstrust,self-confidence,andreliance.Suchamodelcouldcomplementthedistinctionsbetweentypesandlevelsofautomationandguideautomationdesign[4].Animportantdistinctionbetweendifferenttypesofau-tomationistheinformationavailableregardingthecapabilityof1083-4427/$20.00©2006IEEE944IEEETRANSACTIONSONSYSTEMS,MAN,ANDCYBERNETICS—PARTA:SYSTEMSANDHUMANS,VOL.36,NO.5,SEPTEMBER2006Fig.1.Choosingtorelyonaut