ProcNatlAcadSciUSA801983todenythemoreromanticviewofdiscoverythatspeaksofinsightandintuitionandthecreativemomentInanopinionheoncewroteJusticeLearnedHandintroducedintoourpatentlawtherequirementthatevide ID: 896507
Download Pdf The PPT/PDF document "ProcNatLAcadSciUSAVol80pp45694571July198..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1 Proc.NatLAcad.Sci.USAVol.80,pp.4569-4571
Proc.NatLAcad.Sci.USAVol.80,pp.4569-4571,July1983PsychologyDiscovery,invention,anddevelopment:Humancreativethinking*HERBERTA.SIMONDepartmentofPsychology,Carnegie-MellonUniversity,Pittsburgh,Pennsylvania15213ContributedbyHerbertA.Simon,August24,1982Thesubjectofthispaperisasingularlyappropriateoneforthisjournal.Weareallexpertsonthetopicofscientificdiscovery,invention,anddevelopment.MembersareelectedtotheNa-tionalAcademyofSciencesbecausetheyhavebeenadjudgedtohavedonesomecreativethinkingthatledtooneormoresignificantdiscoveries,inventions,orapplicationsofscientificideas.Butthefactthatapersonsometimesdoescreativethingsdoesnotmeanthatheunderstandsthecreativeprocess.ToimaginethatascientistcangiveafullscientificaccountofhisownthoughtprocessesisnomorereasonablethanputtingaGeigercounteronthepodiumatanAmericanPhysicalSocietymeetingandexpectingittodeliveralectureonthetheoryofradiation.Inbuildingatheoryofscientificdiscoveryorinvention,asci-entist'sbehaviorsarethedata-thephenomenaweseektoex-plain-notthedirectsourceoftheory.Muchofwhatgoesonwhenwearethinkingisinaccessibletoourconsciousawareness.Weusesuchtermsas"judge-ment,"intuition,"and"creativeinsight"tonameandlabelthosephenomenathatoccurwithoutawareness.Butlabelsarenotexplanations;contrarytotheviewofMoliere'sphysician,opium'ssleep-inducingpowerisnotexplainedbyattributingtoitadormitiveproperty.Manyscientistsandphilosophershavedoubtedwhetherascientificexplanationofdiscoveryispossibleorwhetherthereisevensomethingdescribableasamethodofdiscovery.Ein-steinhimselfisquotedassaying,"Thereisnologicalpathlead-ingto[scientific]laws.Theycanonlybereachedbyintuition,baseduponsomethinglikeanintellectuallove[Einfuehlung]oftheobjectsofexperience."IfEinsteinmeantby"logicalpath"adeductiveroutetodiscovery,hewassurelyright.Ifhemeanttosaythatthereisnotmethodinthemadnessofdiscovery,wemightquestion-andinvestigateempirically-whetherhispes-simismwasjustified.Thefactthatmanybooksandarticleshavebeenwrittenonthesubjectsofinvention,discovery,andcreativity,suggeststhatwebelievethatatleastsomesignificantaspectsoftheseprocessesareamenabletoinvestigationanddescription.Untilrecenttimes,mostofourknowledgeaboutthemhasderivedfromtheexperienceandobservationsofthoughtfulpractition-ers.Inthiscase,theGeigercounterdoesindeedhavesome-thingtosayaboutthetheoryofradiation.Butcreativepro-cessesarealsonowtheobjectsofdirectscientificinvestigation.Inthepastdozenyears,somepsychologistshavesoughttobuildandtesttheoriesabouttheprocesses,bothconsciousandun-conscious,thatgooninthehumanbrainwhendiscoveryandinventionaretakingplace.Here,Iamgoingtosayalittleaboutthispsychologicalresearch,butmainly,Iamgoingtoconsiderwhatweknowaboutthesemattersfromtheinformationthatisavailabletousmoredirectlyinourpracticeofscienceandengineering.Processesofscientificdiscoveryandapplication.Inparticular,thedevelopmentofproductsfrombasicdiscoveriestakesplaceinacomplexsocialandeconomicenvironment,inwhichbothmotivationsanddefinitionsoftheproblemdifferverymuchfromthoseintheenvironmentswherebasicdiscoveriestypicallyoccur.Ontheotherside,Iamgoingtoemphasizethecommunalitiesofprocess-andtoargue,infact,thatwhilediscoveryanddevelopmentusuallyaddressdif-ferentsubstantiveproblems,thepsychologicalprocessesofproblemsolutionarequitesimilar.Letmebeginbydefiningcreativityinapragmaticway.Fromtimetotime,humanbeingsarriveatideasthatarejudgedbytheirfellowstobebothnovelandvaluable.Thevaluesdis-cernedinthesenewideasmaybeintellectual,esthetic,prac-tical,orwhatnot.Itdoesnotmatter.Apsychologicaltheoryofcreativity(ordiscoveryorinvention)wouldaccountforthepro-cessesthatareinvolvedinbringingaboutsuchnovelandval-uableproducts.Ofcourse,weshouldnotbegthequestion:whethercreativityisaunitarythingorwhetherquitedifferentprocessesareinvolvedincreativeproductionindifferentdo-mainsisitselfanempiricalquestion,tobedecidedonthebasisofevidence.Iwillnottrytosurveythewholeliteratureoncreativity.Itdrawspartlyonhistorical,biographical,andautobiographicalaccountsofdiscovery,onsystematicsurveysofcreative(anduncreative)persons,onalimitednumberoflaboratorystudiesaimedatelicitingcreativebehaviororcomparingcreativewithnoncreativestylesofproblemsolving,andanumberofrecentattemptstosimulatecreativebehaviorwithcomputerpro-grams.Atthephenomenologicallevel,thelevelofobservableevents,thereisgreatconsistencyamongallthefindingsofthisresearch,whateveritsmethods.Icansumupthesefindingsinafewgeneralizations.Whatchieflydistinguishescreativethinkingfrommoremundaneformsare(i)willingnesstoacceptvaguelydefinedproblemstatementsandgraduallytostructurethem,(ii)con-tinuingpreoccupationwithproblemsoverconsiderableperiodsoftime,and(iii)extensivebackgroundknowledgeinrelevantandpotentiallyrelevantareas.Notalloftheseconditions-tol-eranceofambiguity,persistence,andknowledge-aresatisfiedinallcasesofdiscovery,buttheirpresencehasbeenobservedandcommentedontoomanytimestosupposethattheiras-sociationwithsuccessindiscoveryisaccidental.Noneoftheseconditionsisverysurprising;theyallhaveastrongmotivationalcomponent,andtheysatisfyoursenseofjustice-thevirtuesofpatience,persistence,anddiligencearelikelytobere-warded,eventhoughtheremaybe,asingoldmining,alargechanceelementinwhowinsthereward.Thevirtuesonlyallowyoutobuyaticketinthelottery,andsometickets,asinalllot-teries,payoffandsomedonot.Insight-basedonrecognitionHowever,somethingseemstobeleftoutofthispicture.ItseemsMuchcanbesaidaboutthedifferencesinprocessatthetwoendsofthecontinuumfrombasicdiscoveriestodevelopment*PresentedattheannualmeetingoftheNationalAcademyofSciences,Apr.26,1982,Washington,D.C.4569 Proc.Natl.Acad.Sci.USA80(1983)todenythemoreromanticviewofdiscoverythatspeaksof"insight"and"intuition"andthe"creativemoment."Inanopinionheoncewrote,JusticeLearnedHandintrod
2 ucedintoourpatentlawtherequirementthatev
ucedintoourpatentlawtherequirementthatevidenceofa"flashofge-nius"wasrequiredtomakeaninventionpatentable-therebyleavingagenerationofpatentattorneyswiththedesperateproblemofmakingsuchflashesvisible.Thereseemstobenodoubt,however,thatthefinalstepofdiscoveryisoftenasuddeneventandsometimesanunexpectedandsurprisingonetotheinventor,whocangivelittleaccountoftheprocessthatarrivedatit.Often,thefinalstepwaspre-cededbyaperiodof"incubation"duringwhichthediscovererwasostensiblypreoccupiedwithothermatters.Sucheventsmustalsoreceiveanexplanationinatheoryofcreativity.Theyseemsomewhatlessinscrutablewhenwereflectthat(i)consciousawarenessisnotaninvariablepropertyofmentalprocessesand(ii)thetheoryistoaccountfortheentireprocessandnotsimplythefinalstep.Inparticular,itistypicaloftheactofrecognitioninhumansthat,whenwerecognizeafamiliarpattern(ahumanfaceorabeetlethatisfamiliarfromourscientificstudies),weareunabletoreportreliablywhatcuesinthestimuluswereessentialorinstrumentaltotherecognition.Itisnotorious,forexample,thatthecharacteristicsrecordedintaxonomickeys(manydeterminableonlyunderthemicroscope)arenotatallthecharacteristicsthatanexperiencedtaxonomistusesforim-mediate(andoftenquitereliable)recognitionofspecieswithwhichheishighlyfamiliar.KnowledgeandpersistenceTheabilitytoachievesuddeninsightsintosituationsbyrecog-nizingfamiliarfeaturesinthemdependsonhavingstoredagreatdealofknowledge-knowledgeaboutthefamiliarpatternsthatcanberecognizedandknowledgeofthecuesforrecognition.Weknowagooddealtoday,fromresearchoncreativityinafewdisciplines,abouttheamountofknowledgethataworld-classexpertmusthave,andaboutthelengthoftimeittakestoac-quireit.Withrespecttoamountofknowledge,carefulstudyoftheabilityofchessgrandmasterstorecognizefamiliarconfigura-tionsofchesspiecesleadstotheestimatethattheyhaveatleast50,000familiar"friends"-patternsthattheywillnoteandrec-ognizeimmediatelyinapositiononachessboard.Thatnumberseemsnotunreasonablewhenwerecallthatitisofthesameorderofmagnitudeasthewordrecognitionvocabularyofacol-legegraduateforwordsinhisnativelanguage.Withrespecttothetimerequiredforaworldclassexperttoacquiretherecognitioncapabilityandotherknowledgehere-quires,studiesofbiographicaldatainthedomainsofchessplaying,musicalcomposition,andpaintingarriveataratherconsistentminimumestimateof10years.Thatis,eventhemosttalentedpersonsdonotreachworld-classperformanceintheirfieldsuntiltheyhavedevotedabout10years(ormore)ofrathersingle-mindedattentiontobecomingexpert.(Thetimemaybesomewhat,butnotagreatdeal,lessforathletes).The10-yearfigureholdsforchildprodigies(BobbieFischer,Mozart,Pi-casso)aswellasforpersonswhobegintheascenttoexpertnesslater.Knowledgeandpersistencedoseem,indeed,tobepre-requisitestohigh-levelperformance.WeakmethodsToconstructanadequatetheoryofdiscovery,onemustattendtotheconditionsunderwhichnoveltyislikelytoemerge.Prob-lemsthatcallforcreativityarepreciselyproblemsfromdo-mainsthathavenotalreadybeenwellworkedoverandinwhichsophisticated,systematicalgorithmsforsolutiondonotexist.Insuchill-structuredproblemdomains,problem-solvingmethodscannotbecloselyattunedtothecharacteristicsoftheproblemenvironment-suchtuningrequiresagreatdealtobeknownaboutthatenvironment.Ifwearegivenalinearalgebraicequa-tioninonevariabletosolve,wesimplyapplyawell-wornandoverlearnedalgorithmtosolveit.Ingenuityisrequiredonlywhensuchanalgorithmisnotknowntous.Hence,wemaypredictthatpersonstacklingproblemswhosesolutionwillhavemarksofnoveltyandrequirecreativitywilluseverygeneralmethodsthatdonotrelyonspecificknowl-edgeabouttheproblemdomain.Suchmethods,ontheotherhand,arelikelytobehighlyinefficient;allthatcommendsthemisthatnobetteronesareavailable.Researchusingcomputerprogramminglanguagestodesignproblemsolvershasuncoveredanumberofsuchmethods,whicharenowadaysusuallycalled"weakmethods."Anexampleofaweakmethodisgenerate-and-test,whichconsistssimplyinde-visingpossiblesolutionsandthentestingeachonetoseewhetheritsatisfiesthesolutionconditionsoftheproblem-essentiallytrial-and-errorsearch.Asomewhatmorepowerfulweakmethod,whichrequiresforitsapplicationonlyalittlemoreknowledgeabouttheproblemdomain,ismeans-endsanalysis.Inthismethod,apresentsituationiscomparedwithagoalsituationandoneormoredifferencesbetweenthemaredetected.Whenadifferenceisnoted,itmay(byrecognition)callforthfrommemoryanoperatorthatisrelevanttoreducingdifferencesofthiskind.Theoperatoristhenapplied,anewsituationiscre-ated,andthewholecycleisrepeatedasoftenasdesired.Mostweakmethodsrequirelargerorsmalleramountsofsearchbeforeproblemsolutionsarefound,butthesearchneednotbeblindtrial-and-error-infact,usuallycannotbe,forthesearchspacesaregenerallyfartoovasttoallowunselectivetrialanderrortobeeffective.WeakmethodsgenerallyincorporatePolya'sideaof"heuristics"-rulesofthumbthatallowsearchgeneratorstobehighlyselective,insteadofsearchingtheen-tirespace.AcomputersimulationofdiscoveryThereisnowaconsiderablebodyofevidencethatproblemsolvingofakindthatwouldberegardedascreativeifexhibitedbyhumanbeingscanbeproducedbycomputerprogramsthatuseweakmethodstosearchveryselectivelyinunfamiliarprob-lemdomainsandthatareable,withthehelpofsuchmethods,todetectinterestinglawsindataandtoinventnewinterestingconcepts.NowImustqualifythatlaststatementalittlebit.Theseprogramshavebeentestedmainlynotbysettingthemouttosearchforgenuinelynewknowledge,previouslybeyondhumanken,butbysettingthemtheproblemofrediscoveringimportantscientificlawsandconcepts,startingfromessentiallythesamesituationsthattheoriginalhumandiscoverersdid.TheyareliketheyoungGauss,youmightsay,whowas(rightly)judgedcreativebyhisteacherwhen,atatenderage,here-discoveredtheformulaforthesumofthefirstNintegers.Oneoftheseprograms,whichIshouldliketodescribebriefly,hasrediscoveredKepler'sthirdlaw,Ohm'slaw,Black'slaw
3 ofheat,Snell'slawofrefraction,theGay-Lus
ofheat,Snell'slawofrefraction,theGay-Lussaclawofgaseouschemicalreactions,someaspectsofDalton'slawofchemicalreactions,andsomeothers.Inthecourseofdoingthesethings,itreinventedtheconceptsofinertialmass,specificheat,re-fractiveindex,atomicweight,andmolecularweight(distin-guishingbetweenthelattertwo).Italsorediscoveredconser-vationlawsformomentumandheat.TheprogramwhoseexploitsIhavelistediscalledBACON,afterSirFrancisBacon,foritisadata-driveninductivema-4570Psychology:Simon Proc.Natl.Acad.Sci.USA80(1983)4571chine.Itsinputsarenumericalandnominaldata,itiscapableofperformingfactorialexperimentsonthesedata,varyingoneindependentvariableatatime,butitisprovidedwithnothe-oreticalknowledgeaboutthephenomena.Thehistoricalre-discoverytestswehaveposedtoitwerederivedprimarilyfromsituationswheretheoriginaldiscoverywas,asfarasweknow,primarilydatadriven,andwherelittlerelevanttheoryexistedbeforethediscovery.TheinitialversionsofBACONwerepro-ducedbyPatrickW.Langleyasadoctoralproject,anditsex-tensionsarejointproductsofLangley,GaryBradshaw,andmyself.BACONcarriesoutrelativelylittlesearch-itcertainlydoesnotrelyonbrute-forcetrial-and-errorexplorationofthedata.Ontheotherhand,itsmethodsarerelativelyfewandsimple,basedonahandfulofheuristics.Whatisinterestingandre-markableaboutitisthatdiscoveries(rediscoveries)ofsuchmagnitudecouldbeproducedbysuchasimplemechanism.LetmegiveaverybriefsketchofBACON'sheuristics.First,theprogramtestsforcorrelationsamongpairsofvariables.Whenasubstantialcorrelationisdiscovered(othervariablesbeingheldconstant),BACONteststheproductorratioofthevariables(asappropriate)forinvariance.Theproductorratio,whetherin-variantornot,isintroducedasanewvariable.Thus,forex-ample,BACONreachesKepler'sthirdlawbyexaminingPID,P/D2,andP2/J3.Itengagesinlittleextraneoussearchandfindsthecorrectfunctioninafewseconds.AnotherruleofthumbusedbyBACONistotryfirstlawsthataresymmetricinappropriateways.Whenexaminingca-lorimetricdata,itdiscoversthatthelawdependsontheprod-uctofthetemperature,mass,andspecificheatofoneofthesubstancesinvolved.Itthenhypothesizesdirectlythatthefinallawwillinvolvetheproductofthesamepropertiesoftheothersubstance.Athirdruleofthumbistopayattentiontonumbersthataresimplemultiplesofeachother.ThisheuristicleadsBA-CONtothechemicallawsofsimplecombinationsandtoatomicandmolecularweight.WhenBACONfindsthatthereisanin-variantrelationbetweenpropertiesoftwodifferentobjects,itpostulatesanewpropertybelongingtoeachobjecttoaccountfortherelationalinvariant.Withthisheuristic,itinventstheconceptsofinertialmass(fromexperimentsinvolvingmutualaccelerations),specificheat(fromthecalorimetricexperi-ments),andrefractiveindex.BACONisveryfarfromacomprehensivetheoryofscientificdiscovery.Initspresentform,itcannotrelateonelevelofphe-nomenatoanother(e.g.,derivethelawsofthermodynamicsfromthekineticsofgasses),cannotusetheorytoderivepos-siblelaws,doesnotinventobservationalinstruments,anddoesnotchoosedataforexamination(althoughithasfaircapacitiesforignoringirrelevantdatawithoutmuchwastedmotion).Ithasonlyamodestcapacityfordealingwithnoisydata.Evenwiththeselimitations,whatBACONdemonstratesisthatscientificdiscoveriesofsortsthathavebeenhighlyre-gardedcanbemadewiththeuseofasmallarmatoriumofheu-risticsandweakmethods.Theprocessesitusesarenotdif-ferentfromtheprocessesthathavebeenshown,inthepsychologicallaboratory,toaccountforother,moremundane,humanproblem-solvingperformances.Itshouldincreaseouroptimismthatscientificdiscoverycanbeunderstoodasanat-uralphenomenonandthatitfollowslawsalreadymadefamiliarbyotherresearchincognitivepsychology.Itisnotadomainofstrange,esoteric,humanphenomena.Scientistsneednotfearfortheirmembershipinthehumanspecies.OthercomputerdiscoveryprogramsBACONisnotthesoleswallowthatmakesthisparticularsci-entificsummer.ThereareotherprogramsthatcastlightonthediscoveryprocessbysimulatingitthatIcanonlyname.OneistheAMprogramofDouglasLenatthat,givenasetofbasicconceptsinsettheory,wentontoreinventtheintegers,theconceptofprimenumber,andtheoperationsontheintegers.Itconjectured(butdidnotprove)thefundamentaltheoremofarithmeticandGoldbach'sconjecture.AnotheristheMOL-GENprogram,ofStefikandFriedland,whichiscapableofde-signingsimpleexperimentsinmoleculargenetics.Throughtheseandothermodelingefforts,wearelearningagooddeal,todayabouttheproblem-solvingprocessesthatunderliescientificcreativity,andwearediscoveringthattheseprocessesrelyba-sicallyonweakmethodsandheuristicsearch.ConclusionAllofthisevidencebearsmainlyonthediscoveryprocessinbasicresearch.Ihavesaidnothingaboutthedevelopmentandapplicationendofthespectrum.Iwouldhypothesizethatthemajorshift,aswemovetowardthislatterend,isinthefor-mulationofthecriteriathatdefineproblemsolutions.Fromacentralconcernwithunderstandingphenomenaindepthandastrategyofabstractingfromallbutthecentrallyrelevantvari-ables,wemovetoaconcernforthewholetechnicalandsocialcontextinwhichaprincipleistobeapplied,includingunan-ticipatedconsequencesandsideeffects,aswellasquestionsofefficacyandeconomy.Manyofthedifficultiesincommuni-cationamongthegroupsatdifferentpointsalongtheresearchanddevelopmentcontinuumareassociatedwiththeseissuesofproblemformulation.Idoubtwhethertheyimplyanydiffer-encesinthebasicpsychologicalprocessesthatarerequiredforproblemsolution.1.Hayes,J.R.(1978)CognitivePsychology(Dorsey,Homewood,IL),pp.215-244:Thispresentsabriefgeneralsurveyofpsychologi-calresearchoncreativity.2.Simon,H.A.(1977)ModelsofDiscovery(D.Reidel,Dordrecht,Netherlands),Sect.5,pp.265-338:Thispresentsanelaborationofthetheorysetforthhereandcontainsnumerousreferencestotheliterature.3.Simon,H.A.(1981)TheSciencesoftheArtificial(MITPress,Cambridge,MA),2ndEd.,pp.101-123:Thisdiscussestheknowledgeofexpertsanditsacquisitionandcitesrelevantref-erences.Psychology:Simo