Computational models of creativity: A review of single- and multi-process recent approaches to demystify creative cognition

Computational

models

of

creativity:

a

review

of

single-process

and

multi-process

recent

approaches

to

demystify

creative

cognition

Vera

Mekern,

Bernhard

Hommel

and

Zsuzsika

Sjoerds

Creativityisacompellingbutheterogeneousphenomenon.As

opposedtobig-Ccreativity,whichisregardedaslimitedtothe

rarebrilliantmind,little-ccreativityisindispensableinadaptive

everydaybehavior,servingtoadjusttochanging

circumstancesandchallenges.Computationalapproaches

helpdemystifyhumancreativitybyofferinginsightsintothe

underlyingmechanismsandtheircharacteristics.Recently

proposedcomputationalmodelstocreativecognitionoften

focusoneitherdivergentorconvergentproblem-solving,but

somestarttointegratetheseprocessesintobroadercognitive

frameworks.Webrieflyreviewthestate-of-the-artinthefield

andpointouttheoreticaloverlap.Weextractbasicprinciples

thatmostexistingmodelsagreeonanddesiderataontheway

towardsacomprehensivemodel.

Address

LeidenUniversity,InstituteofPsychology,CognitivePsychologyUnit, LeidenInstituteforBrainandCognition,Netherlands

Correspondingauthor:Sjoerds,Zsuzsika(sjoerds.zs@gmail.com)

CurrentOpinioninBehavioralSciences2018,27:47–54 ThisreviewcomesfromathemedissueonCreativity EditedbyRexJungandHikaruTakeuchi

https://doi.org/10.1016/j.cobeha.2018.09.008

2352-1546/ã2018TheAuthors.PublishedbyElsevierLtd.Thisisan openaccessarticleundertheCCBY-NC-NDlicense( http://creative-commons.org/licenses/by-nc-nd/4.0/).

Introduction

Creativityisacompellingphenomenonthathasproduced admirableideasandartefacts.Adistinctionisoftenmade between big-Ccreativity, whichallows brilliant mindsto createuniqueandinventiveproducts,andlittle-ccreativity, thecognitivefunctioningthathelpseventhelessbrilliant mindadapttochangingcircumstancesandsolveeveryday problems [1,2].Because ofitsindispensability in every-dayfunctioning,little-ccreativity(henceforthcreativity)is studied widely to understand how creative cognition emerges and why it shows so much interindividual variability.

Since Guilford[3] adistinction ismadebetween diver-gent and convergent thinking in generating creative ideas. Divergent thinking produces creative ideas by exploringmultiplepotentialsolutionstoanoftenvaguely defined problem while convergent thinking serves to identifythesingle best solutionto awell-defined prob-lem.The cognitiveoperationsneededto support diver-gentandconvergentthinkinghavebeenassociatedwith possiblyantagonisticsetsofprocessesorcognitivecontrol modes,suchasflexibilityversuspersistence[4]orinsight versusanalyticprocessing[5].Yet,actualperformanceis likelytoinvolvesomedegreeofinterplaybetween diver-gent,convergent,andothercognitive(sub)processesand process-related neural networks (e.g. [6–8]), suggesting that creativity is a complex and heterogeneous phenomenon.

In this short review we consider the most recent (<3 years)computationalmodelsofaspectsofhuman creativ-ity. Computational models allow for a mechanistic approachtocognitiveprocessesinhealthyand maladap-tive cognition [9–11] and thus have the potential to demystify creative cognition. We highlight divergent andconvergentprocessesintheserecentcomputational approachestocreativecognition(seealsoTable1),tothe degreethattheycanbedistinguishedandcharacterized accordingly.Wethenbrieflyconsiderrecentissueswith dual-processaccountsinmodelingcreativity(c.f.[12,13]) and proposeaunitaryapproachthat mightofferamore parsimoniousaccounttorecognizethetrickydivisionand adaptivity between antagonistic states underlying creativity.

Recent

computational

approaches

to

creativity

Modelsofdivergentcreativity

Divergentthinkinghasbeenrelatedtoassociative think-ing [13], and can be modeled as spreading activity in neural networks.Three recentpublications used a net-work science approach to study how individual differ-ences in creative associative thinking might arise from structural differences in semantic networks [14,15,16
]. Findingssuggestedthatthesemanticnetworksofhighly creativeindividualsshowedmoresmall-worldproperties, which allows for faster search over a wider network of

$_{ThisworkwassupportedbytheEuropeanResearchCouncil(ERCAdvancedGrant)undertheEuropeanUnion’sHorizon2020researchand}

associations,increasingtheprobabilityofreturningnovel associations[15].Kenettetal.[16
]alsofoundthat break-ingassociations in asimulated semanticnetworkled to largerpartsofthenetworkbreakingapartinlowcreative individuals,whilenetworks in high creativeindividuals remained fairly intact. This network science computa-tionalapproachthussuggeststhatstructural characteris-ticsofsemanticnetworksinfluencetheextentof diver-gentthinking.

Another recent approach implementedacomputational modelofapopulartasktostudydivergentcreativity,the AlternativeUsesTask(AUT[3]).IntheAUT, individu-als produce as many as possible alternative uses for a commonobject(e.g. towel,brick)withinlimitedtime.In the model, performance on this task relies on object replacementand objectcomposition (OROC).The sys-temwasmodeledwithinatheoreticalframework called CreaCogs[17,18
,19]).CreaCogs-OROCorganizes mem-ory into three layers: first, a subsymbolic level where feature spaces (e.g. shape, color, affordance) of objects arerepresentedinadistributedfashion;second,alevelof conceptsgroundedinthesubsymboliclevel;andthird,a problem template level representing known problems and solutions encoded over concepts and relations between them (Figure 1). Each level is grounded in thesubordinateleveltobeabletouse,say,featuresfrom related concepts to find objects with features that can replaceacueobjectintheAUT,orviceversa.Themore feature spaces are considered, the more divergent the search for a replacement use can become, making the divergenceofsearchintheAUT-dependentonthesize and number of feature spaces in the CreaCogs-OROC knowledgebase — apossible source of interindividual differences.SimulationsoftheAUTinCreaCogs-OROC

showthatthesystemcanproduceanswerscomparableto findingsin humans[20].

Theoretically,CreaCogs-OROCcanbeusedtoconstruct insight problems [19,21]by takingasimple problem withan existingsolutionandreplacingor (de)composingobjects usedinthesolutiontochangetheproblemtoacreative problem. The authors suggest an example problem in whichtheparticipant shouldfindhowto buildaseesaw fromasurfingboard andabucketto decidewhoof two peopleisheavier.Althoughinsightproblemconstructionin CreaCogshas notyet been simulated, thecreative (de) compositionofobjectsandobjectreplacementto re-repre-sentabalancingscaleisreminiscentofprocessesin model-ingtheAUT.Themorefeaturesorobjectsareconsidered in constructing insight problems, the more divergent a searchforthesolutionmighthavetobecome.Thecreative problem-solving(orproblem-generating)approachinthe CreaCogsframework thusseemsto lenditself tomodel divergentbehaviorinmultiplecreativityparadigms.

Modelsofconvergentcreativity

Whiletheabovementionedsetofmodelsfocusedonthe spread of search,or divergentcognition,similar models are used to study convergent, more targeted search. Another prototype system within the CreaCogs frame-work (comRAT)simulates performanceontheRemote AssociatesTest(RAT[22]),aconvergent-creativitytask in which three verbal concepts are presented and a solutionword that canbe combined with either one is soughtfor(e.g.market,glue,man!super).ComRATwas developedasanRATsolver(comRAT-C[17,19])anda semantic RAT problem generator (comRAT-G [23]). ComRAT-Ccomprises aknowledge base of word pairs modeled in CreaCogs’ concept level. Activation of an

Table1

Summaryofrecentcomputationalmodelsappliedtocreativecognition

Authors Modeledcreativityprocess Descriptioncomputationalapproach Benedeketal.[14];

Kenettetal.[15]; Kenettetal.[16
]

Divergent Networkscienceapproach;Percolationanalysis

OlteÛeanuandFalomir[20]; OlteÛeanu[19];

Olte_Ûeanu[21];

OlteÛeanu,Falomir,andFreksa[18
]

Divergent Prototypesystem(OROC)inCreaCogstheoreticalframework

Olte_ÛeanuandFalomir[17]; Olte_Ûeanu,Falomir,andFreksa[18
]; OlteÛeanu,Schultheis,andDyer[23]

Convergent Prototypesystem(comRAT)inCreaCogstheoreticalframework

Schatz,Jones,andLaird[24] Convergent Semanticmemorymodelincognitivearchitecture(Soar)

Kajı´cetal.[25
] Convergent Spikingneuronmodel

Augelloetal.[32
] Divergentandconvergent* Cognitivearchitecture(MicroPsi/Psi) Wiggins[28];

WigginsandBhattacharya[29]

Divergentandconvergent Cognitivearchitecture (IDyOT)

RAT problemactivatesall wordsrelatedto thequeried word pair, modeling an associative search of the full knowledgebasetoenableconvergenceupononeanswer found in three word pairs [17,19].It returns an answer afterfindingawordthatwasassociatedwitheachofthe cue words. Sometimes, the system was also able to convergeupon(alternative)answerswhenonlytwoword pairsintheknowledgebasesharedaword,indicatingthat thelearned associations structureprovides arobust sys-temto solveRAT problems[17].

As shown recently[24], spreading activation starting at threeRATcuewordscanalsoleadtothecorrectsolution bystronglyincreasingtheactivationlevelofonewordof theknowledgebase[24].Freerecall,implementedinthis computationalapproach,explainedsuccessbetterthana cued-retrieval approach in which the system was only allowedtoreturnananswerthatmatchedallthree prob-lemwords.Thisperformanceincreasethuscomesdespite thelackofadeliberatequerytomatchthesolutionword to word pairsof the threeRAT cue words [24]. Appar-ently, a winner-takes-all approach in (divergently)

spreadingactivityoveranassociativenetworkissufficient to simulateconvergentsearchprocess intheRAT. Thiswasalsoobservedinabiologicallyplausiblespiking neuron model of RAT performance [25
]: a selection networkofneuronsactivatesoneofthreeRATcuewords atatime,andrandomlyswitchesbetweencues.Activation spreadsfromthecuetoallneuronsthatrepresentthecue wordin adistributed fashion.Activationofassociated words representedinoverlappingneuralnetworksisfedbackinto awinner-takes-allresponsenetworkwhichconvergeson themostactivatedwordandrespondswhenauto-inhibitory responseprocesseshavedecayed(Figure2).Performance ofthismodelmatchedhumanperformancewell[25
]and, again,thesuccessofthismodelmightindicatethat spread-ing activation and a winner-takes-all approach could explainRATperformance.

Moreintegrativecomputationalcreativityaccounts

While the models described above claim to focus on divergentorconvergentcreativityspecifically,bothkinds of models rely on associative search processes or

Figure1 Problem template level - structural

PT1

PT2

C3

C2

C1

Current Opinion in Behavioral Sciences

spreadingofactivation.Thisraisesthequestionwhether more integrative models accounting for both divergent and convergent thinking are feasible, an idea that is already apparent in CreaCogs. Some computational approacheshaveindeedtriedtomodelcreativity inthe broaderscopeofcognition.

Oneapproachtostudycreativethinkinginabroader cogni-tive architecture is Information Dynamics of Thinking (IDyOT[26–29].Thismodelisbasedontheideaof predic-tive coding, according to which the brain is constantly occupiedwiththeefficientprocessingofsensory informa-tionbyminimizingentropyand unexpectedness. Predic-tionsareproducedbygeneratorsthatcompeteforattention inaglobalworkspace,implementingBaars’Global Work-spaceTheory[26,28–30].Predictionsaremadefrom mem-ory, a multiplelayer hierarchy including distributed and conceptualrepresentationsaswellasanintermediatelayer of conceptual spacesrepresenting concepts and relations betweenconceptsgeometrically,asdescribedby Ga¨rden-fors[31].AccordingtoWiggins[28],spontaneouscreativity arisesasaby-product,fromthebrain‘freewheeling’asit continuespredictioninabsenceofrelevantsensoryinput.A novelidea mightbepredicted fromfinding anunvisited point in oneof the conceptual spaces[28,29]. So far, no explicitaccounthowthis‘freewheeling’leadstomoreorless divergentideaswassuggested,butthedetailedandbroad approachmightofferarichsimulationofcreativecognition. Interestingly, one recent model by Augello et al. [32
] doesexplicitlymodeldivergentandconvergentprocesses

together.Theircomputationalpainterisdesignedwithin theMicroPsicognitivearchitecture[33],andthepainter replacesimagefeatureswithacreativealternative(e.g.a face area with a flower, similar to the human painter Arcimboldo).Augello etal. refer to theFourQuadrants Model,whichmodelstheinteractionbetweenconvergent anddivergentprocessesandrecognizesthatbothofthese processescanbeimplicitorexplicit(Figure3[34]). Long-term memory consists of distributed representations in whicheach neuron represents thecentroid of apattern clusterrepresentingan imageor imagedetail [35].The working memory module clusters more abstract repre-sentationsofimagedetails,enablingfeaturesubstitution inonedomain(e.g.faces)withfeaturesofanotherdomain (e.g. flowers). Cognitive control over these processes allowsfor moreorlessdivergentsearchandcomesfrom aresolutionnetwork.Thisnetworkcontrolsforexample thepermitteddistancesbetweencentroidsintheworking memorytoallowforbroaderornarrowersearch,andthe interaction between convergent analytic and tacit pro-cessesindecidingwhetherareplacementissuccessfulor not(Figure 3).

Dual-process

modeling

Neurocognitiveresultshavebeen takenasevidence for distinct neural mechanisms underlyingdivergent, flexi-ble,orunfocused thinkingontheonehand and conver-gent,persistent,orfocusedthinkingontheother,bewith referenceto frontaland striatal dopaminergicpathways [36,37], dopamine receptors families [38],or brain net-works(e.g.[39–41]).However,dual-processaccountsare

Figure2 cue1 cue2 cue3 bias WTA noise resetsignal primarycue

WTA response inhibition

˜ A response cue selection responsenetwork Groupofneurons

net Networkofneurons

Gatingneurons

Flowof information

Inhibitory connection

Current Opinion in Behavioral Sciences

increasinglycriticized,oftenbecausetheprocessesand/or theoutcomesofthetwohypotheticalpathwaysare diffi-cultto distinguishorbecausebothtypesofprocessesor outcomes rely on some sort of interaction between the two pathways [5,12,13,42–45]. The model by Augello et al. [32
] is rather exceptional in trying to integrate divergentandconvergentprocessesincreativecognition (butseeCLARION[46–49]).However,theissues men-tionedabovehaveledtoacallforanevenmore integra-tive perspectiveofcreativecognition[12,13,50
].

Towards

a

unitary

account

What basicingredients should suchan evenmore inte-grative perspective ofcreative cognitionhave?We con-siderthreeingredientsessential.First,mostofthemodels that wehave discussed agreeon theimportanceof dis-tributed representations of objectsor conceptsto facili-tate replacement or composition, as for instance sug-gested explicitly in CreaCogs. Such representations

alsofacilitateaspectsofcreativitythatgobeyond diver-gent and convergent thinking, such as the creation of metaphors(creatingaconnectionbetweentwoseemingly unrelatedconceptsthathoweversharesomefeatures)or replacement by second-best solutions (‘plan b’) if best solutions are impossible or not feasible. Second, the degreeofflexibilitythatmostaspectsofcreativebehavior requirecallfor thecontextualization of representations. Even distributedrepresentations of objects or concepts arerelativelystaticwithoutameanstoweighthepossible featuresaccordingtosituationalrequirementsorpresent goals.Itisthiscontextualizationthatfacilitatesthe crea-tionofmetaphors andthebreakingofoverlearned asso-ciations between concepts. Third, modelsneed to take into account individual differences more. Most models certainly allowfor theconsiderationof suchdifferences butmakingthemanexplicitgoalofmodeldevelopment would drastically increase our insight into the mecha-nismsunderlyingthesubstantialindividualdifferencesin creativitythatcanbeobservedinreallife.Itisimportant to distinguish between trait-like differences between individuals (resulting from genetic predisposition and/ oroverlearning)thataredifficultorimpossibleto elimi-natethroughinterventions,andstate-likedifferencesthat eventhesameindividualcanshowindifferentsituations or underdifferentgoals.

All three basic ingredients are available from a recent attemptto provide aunitary alternativeto conventional dual-routetheorizing inaction control.As suggestedby HommelandWiers[50
],bothpossibleandactualevents, includingconceptsandideas,arerepresentedby distrib-uted networks of the features that characterize these events (so-calledevent files)—an assumptionadopted from the theoryof eventcoding (TEC [51]).While all featuresof agivenevent aremaintainedin thesystem, thecurrentcontribution of featurestorepresentingthis eventincontextofthepresentsituationisweightedbyits relevance for the present intention and task goal [52]. This implies, among other things,that event fileswith more task-relevant features compete more strongly for selectionforfurtherprocessing.Thefeature-based repre-sentationprincipleofeventfilesallowsforfeature-based selection,competition,andspreadofactivation,aswellas for task-specific selectivity of representation — which coversourfirsttworequirementsfor aunitaryapproach. However, to account for individual differences and to allowformodelingbothdivergentandconvergentwaysof concept-selection,anotherprincipleisrequired.As Hom-melandWiers[50
]havesuggested,thedegreeof com-petitionbetweenmultipleeventfilescanberegulatedto make iteither weak,as required for divergentthinking and other associative reasoning patterns, or strong, as required for focused in-depth convergent thinking. The idea isthat competitionand selectionis regulated bythepresentmetacontrolstate,whichreflectsa partic-ularbalancebetweenextremepersistence,characterized

Figure3 Implicit Processing Divergent Processes Convergent Processes Explicit Triggering Training Primi ng P reparatio n Insight Habit Explicit Monitoring Work ing mem ory & s e ria l proc e s sing imitatio ns Narr owed At tent ion Explicit Processing Reflective (REF) Exploratory (EXP) Tacit (TAC) Analytic (AN)

Current Opinion in Behavioral Sciences

TheFourQuadrantsModel,implementingfour(exploratory,analytic, reflective,tacit)strategies,recognizesthecrosstalkbetweenimplicit andexplicitprocessesontheonehandanddivergentandconvergent processesontheother.Augelloetal.[32
]mapthefourresulting processesontotheircreativepaintertoindicatetheimportanceof interactionsbetweentheprocessesincreativecognition.Highlighted areonlytheinteractionsinvolvedinthecreativeprocessproposedin thesystembyAugelloetal.[32
].

bystrongcompetitionandtop–downfocus,andextreme flexibility,characterized by weak competitionand top– downfocus[53].Theparticularmetacontrolbiasonthis persistence-flexibility dimension has been claimed to showsystematicinter-individualandintra-individual var-iability;some people tend to bemore persistent where otherstendtobemoreflexible,butthesamepersonmay alsosometimestendtobemorepersistentandsometimes more flexible, depending on situational demands ( Fig-ure4,leftversusrightpanel).Moreover,wehaverecently suggestedthatpeoplemaydifferintheiradaptivity,which referstotheeasewithwhichtheykeepandreadjusttheir balancebetweenpersistenceandflexibilityinthecaseof changing environmental demands (Figure 4, topversus bottompanel).

Adopting these three key principles holds promise for developingaunitary computationalapproach of human creativity that avoids previous problems with dual-pro-cessmodels(asdiscussed inRefs.[12,13]).Asuccessful unitary approach is likely to help further demystifying creativecognitionbycomingtogripswiththeunderlying mechanisms.

Conflict

of

interest

statement

Acknowledgement

ThisresearchwasfundedbyanAdvancedGrantoftheEuropeanResearch Council(ERC-2015-AdG-694722)toBH.

References

and

recommended

reading

ofspecialinterest

1. CsikszentmihalyM:Reflectionsonthefield.RoeperRev1998, 21:80-81.

2. CsikszentmihalyM:Implicationsofasystemsperspectivefor thestudyofcreativity.In HandbookofCreativity.Editedby SternbergRJ. CambridgeUniversityPress;1998:313-335.

3. GuilfordJP:Creativity:yesterday,todayandtomorrow.JCreat Behav1967,1:3-14.

4. NijstadBA,DeDreuCKW,RietzschelEF,BaasM:Thedual pathwaytocreativitymodel:creativeideationasafunctionof flexibilityandpersistence.EurRevSocPsychol2010,21:34-77.

5. KouniosJ,BeemanM:Thecognitiveneuroscienceofinsight. AnnuRevPsychol2014,65:71-93.

6. DietrichA:Thecognitiveneuroscienceofcreativity.Psychon BullRev2004,11:1011-1026.

7. EysenckHJ:Creativityandpersonality:suggestionsfora theory.PsycholInq1993,4:147-178.

8. HeilmanKM:Possiblebrainmechanismsofcreativity.ArchClin Neurpsychol2016,31:285-296.

9. ForstmannBU,WagenmakersEJ(Eds):AnIntroductionto Model-BasedCognitiveNeuroscience.Springer;2015.

10. SjoerdsZ,denOudenHEM:Computationelepsychiatrie:een toekomstvoorwiskundigemodellenindeclassificatieen behandelingvanpsychopathologie?Neuropraxis2015, 19:141-152.

11. MaiaTV,HuysQJM,FrankMJ:Theory-basedcomputational psychiatry.BiolPsychiatry2017,82:382-384.

12. BarrN:Intuition,reason,andcreativity:anintegrative dual-processperspective.In TheNewReflectionisminCognitive Psychology:WhyReasonMatters.EditedbyPennycookG. Routledge;2018:93-118.

13. SowdenPT,PringleA,GaboraL:Theshiftingsandsofcreative thinking:connectionstodual-processtheory.ThinkReason 2015,21:40-60.

14. BenedekM,KenettYN,UmdaschK,AnakiD,FaustM, NeubauerAC:Howsemanticmemorystructureand

intelligencecontributetocreativethought:anetworkscience approach.ThinkReason2017,23:158-183.

15. KenettYN,AusterweilJL:Examiningsearchprocessesinlow andhighcreativeindividualswithrandomwalks.In Proceedingsofthe38thAnnualMeetingoftheCognitiveScience Society.2016:313-318.

16.

KenettFlexibilityYN,ofLevythoughtO,KenettinhighDY,creativeStanleyindividualsHE,FaustM,representedHavlinS: bypercolationanalysis.ProcNatlAcadSciUSA2018,115 :867-872http://dx.doi.org/10.1073/pnas.1717362115.

These authors use a network science approach to model how the structureofassociativesemantic networkscan influence flexibilityin Figure4 persistent trait bias flexible trait bias f f p p high

adaptivity Demanded bias

Trait bias Current state bias

low adaptivity f f p p

Current Opinion in Behavioral Sciences

divergentsearchincreativecognition.Specifically,individualswhohave semanticnetworkswithsmall-worldcharacteristicsshowmoreflexibility insearchbecausetheirnetworkisappearsmorerobusttoremovalof linksbetweenwordsinthispercolationanalysis.

17. OlteÛeanuA-M,FalomirZ:comRAT-C:acomputational compoundRemoteAssociatesTestsolverbasedonlanguage dataanditscomparisontohumanperformance.Pattern RecognitLett2015,67:81-90.

18.

OltethatÛeanucananswerA-M,FalomirhumanZ,creativityFreksaC:tests:Artificialanapproachcognitivesystemsandtwo casestudies.IEEETransCognDevSyst2018,10:469-475.

AshortreviewoftheCreaCogstheoreticalframework,whichwas devel-opedtoaccountforhumandataoncreativitytasksbysimulating per-formanceonthesetasks.Theauthorsdiscusstwoprototypesystemsto simulateconvergent(comRAT-C)anddivergentcreativity(OROC)and suggest two extra creativitytaskswhich can be modeledwithin the CreaCogsframework.

19. Olte_ÛeanuA-M:ACognitiveSystemsFrameworkforCreative ProblemSolving.2016.

20. Olte_ÛeanuAM,FalomirZ:Objectreplacementandobject compositioninacreativecognitivesystem.Towardsa computationalsolveroftheAlternativeUsesTest.CognSyst Res2016,39:15-32.

21. Olte_ÛeanuAM:Towardsanapproachforthecomputationally assistedcreationofinsightproblemsinthepracticalobject domain.In CEURWorkshopProceedings.2016:1-13.

22. MednickSA:Theassociativebasisofthecreativeprocess. PsycholRev1962,69:220-232.

23. OlteÛeanuA-M,SchultheisH,DyerJB:Computationally constructingarepositoryofcompoundremoteassociates testitemsinAmericanEnglishwithcomRAT-G.BehavRes Methods2017:1-10 http://dx.doi.org/10.3758/s13428-017-0965-8.

24. SchatzJ,JonesSJ,LairdJE:Anarchitectureapproachto modelingtheremoteassociatestest.InProceedingsofthe16th InternationalConferenceonCognitiveModelling(ICCM).2018.

25.

spikingKajicI,GosmannneuronmodelJ,StewartofwordTC,associationsWennekersT,forEliasmiththeRemoteC:A AssociatesTest.FrontPsychol2017,8.

Aspikingneuronmodelthatoffersabiologicallyplausiblecomputational accountofassociativesearchasabasisforconvergentcreativity.The authorsshowthatthismodelcansolveRemoteAssociatesTest pro-blemsandthatdataiscomparabletohumandataonthistest. 26. VanderVeldeF,ForthJ,NazarethDS,WigginsGA:Linkingneural

andsymbolicrepresentationandprocessingofconceptual structures.FrontPsychol2017,8.

27. WigginsGA:Themind’schorus:creativitybefore consciousness.CognitComput2012,4:306-319.

28. WigginsGA:Creativity,information,andconsciousness:the informationdynamicsofthinking.PhysLifeRev2018http://dx. doi.org/10.1016/j.plrev.2018.05.001.

29. WigginsGA,BhattacharyaJ:Mindthegap:anattempttobridge computationalandneuroscientificapproachestostudy creativity.FrontHumNeurosci2014,8.

30. BaarsBJ:ACognitiveTheoryofConsciousness.Cambridge UniversityPress;1988.

31. Ga¨rdenforsP:ConceptualSpaces:TheGeometryofThought.MIT Press;2000.

32.

AugellocreationA,byInfantinoacognitiveI,LietoarchitectureA,PilatoG,integratingRizzoR,VellacomputationalF:Artwork creativityanddualprocessapproaches.BiolInspiredCogn Archit2016,15:74-86.

Thispublicationdescribesacreativecomputationalpainterthatcreates paintings inthestyleofArcimboldo.Thepainterisintegratedintoan existingcognitivearchitecture(MicroPsi)inwhichtheauthorsrecognize theinteractionbetweendivergentandconvergentcreativeprocesses. Throughconceptualabstractionofdistributedrepresentationsofimages orimagedetailsthepaintercanreplacepartsofapaintingcomingfrom onedomain(i.e.faces)withalternativesfromanother(i.e.flowers).

33. BachJ:TheMicroPsiagentarchitecture.In Proceedingsof ICCM-5,InternationalConferenceonCognitiveModeling. 2003:15-20.

34. TubbR,DixonS:Afourstrategymodelofcreativeparameter spaceinteraction.In In ProceedingsoftheFifthInternational ConferenceonComputationalCreativityICCC-2014.Editedby ColtonS,VenturaD,Lavra9cN,CookM.ProceedingsoftheFifth InternationalConferenceonComputationalCreativityICCC-2014 2014:16-22.

35. AugelloA,InfantinoI,PilatoG,RizzoR,VellaF:Introducinga creativeprocessonacognitivearchitecture.BiolInspiredCogn Archit2013,6:131-139.

36. CoolsR,IvryRB,D’EspositoM:Thehumanstriatumis necessaryforrespondingtochangesinstimulusrelevance.J CognNeurosci2006,18:1973-1983.

37. CoolsR:Dopaminergiccontrolofthestriatumforhigh-level cognition.CurrOpinNeurobiol2011,21:402-407.

38. DurstewitzD,SeamansJK:Thedual-statetheoryofprefrontal cortexdopaminefunctionwithrelevanceto catechol-O-methyltransferasegenotypesandschizophrenia.Biol Psychiatry2008,64:739-749.

39. BootN,BaasM,vanGaalS,CoolsR,DeDreuCKW:Creative cognitionanddopaminergicmodulationoffronto-striatal networks:integrativereviewandresearchagenda.Neurosci BiobehavRev2017,78:13-23.

40. LiuZ,ZhangJ,XieX,RollsET,SunJ,ZhangK,JiaoZ,ChenQ, ZhangJ,QiuJetal.:Neuralandgeneticdeterminantsof creativity.Neuroimage2018,174:164-176.

41. ShiL,SunJ,XiaY,RenZ,ChenQ,WeiD,YangW,QiuJ: Large-scalebrainnetworkconnectivityunderlyingcreativityin resting-stateandtaskfMRI:cooperationbetweendefault networkandfrontal-parietalnetwork.BiolPsychol2018, 135:102-111.

42. SubramaniamK,KouniosJ,ParrishTB,Jung-BeemanM:Abrain mechanismforfacilitationofinsightbypositiveaffect.JCogn Neurosci2009,21:415-432.

43. Jung-BeemanM,BowdenEM,HabermanJ,FrymiareJL, Arambel-LiuS,GreenblattR,ReberPJ,KouniosJ:Neuralactivity whenpeoplesolveverbalproblemswithinsight.PLoSBiol 2004,2:e97.

44. KleibeukerSW,Ce´dricP,KoolschijnMP,JollesDD,DeDreuCKW, CroneEA,StruzikZR,BenedekM:Theneuralcodingofcreative ideagenerationacrossadolescenceandearlyadulthood. FrontHumNeurosci2013,7:905http://dx.doi.org/10.3389/ fnhum.2013.00905.

45. BenedekM,JaukE,SommerM,ArendasyM,NeubauerAC: Intelligence,creativity,andcognitivecontrol:thecommonand differentialinvolvementofexecutivefunctionsinintelligence andcreativity.Intelligence2014,46:73-83.

46. SunR,WilsonN,LynchM:Emotion:aunifiedmechanistic interpretationfromacognitivearchitecture.CognitComput 2016,8:1-14.

47. SunR:TheCLARIONcognitivearchitecture:extending cognitivemodelingtosocialsimulation.In Cognitionand Multi-AgentInteraction.EditedbySunR. CambridgeUniversityPress; 2006:79-99.

48. SunR:Themotivationalandmetacognitivecontrolin CLARION.In IntegratedModelsofCognitiveSystems.Editedby GrayWD. OxfordUniversityPress;2007:63-75.

49. He´lieS,SunR:Incubation,insight,andcreativeproblem solving:aunifiedtheoryandaconnectionistmodel.Psychol Rev2010,117:994-1024.

50.

Hommelactioncontrol.B,WiersTrendsRW:TowardsCognScia2017,unitary21:940-949.approachtohuman

contextualizationofcognitiverepresentationsandthebiasingofcognitive controltowardsflexibility(whichshouldpromotedivergentprocesses) andpersistence(whichshouldpromoteconvergentprocesses). 51. HommelB,Mu¨sselerJ,AscherslebenG,PrinzW:TheTheoryof

EventCoding(TEC):aframeworkforperceptionandaction planning.BehavBrainSci2001,24:849-878.

52. MemelinkJ,HommelB:Intentionalweighting:abasicprinciple incognitivecontrol.PsycholRes2013,77:249-259.

53. HommelB:Betweenpersistenceandflexibility.In Advancesin MotivationScience.EditedbyElliotAJ. Elsevier;2015:33-67.