Running with Scissors, 13th EAD Conference University of Dundee, 10-12 April 2019
Copyright © 2019. The copyright of each paper in this conference proceedings is the property of the author(s). Permission is granted to reproduce copies of these works for purposes relevant to the above conference, provided that the author(s), source and copyright notice are included on each copy. For other uses please contact the author(s).
BeyondAverageTools.Ontheuseof‘dumb’computationandpurposefulambiguitytoenhancethecreativeprocess PhilippaMothersilla&V.MichaelBoveaaMIT Media Lab, USA *Corresponding author e-mail: [email protected]
Abstract: In theearly phasesof thedesignprocess, embracing chance intrusions,seemingirrelevanceandambiguitycanleadtoconsideringconceptsindifferentwaysandprovokenewideas.However,thecomputationaltoolsweareincreasinglyusingin thesephasesvalueefficiencyoverserendipity; technologieswhose foundationsareanaverage.Thispaperpresentsa‘BeyondAverage’approachthatwasusedtodeveloptwotoolsthatuse‘dumb’computationandpurposefulambiguitytoenhancethecreationofnovelideas.Resultsfromstudiesusingthetoolsinadesigntaskshowthatcomputationaltoolswithamediumlevelofcontextualityandahigherlevelofinterpretabilitycanpositivelyinfluencethecreationofnewideas.Discussionsabouttheroleofcomputationintheearlyphasesofthedesignprocesssuggestthattoolswithhigherlevelsofcreativeagencycancontributetothedesigner’screativeagencyandbecomeamorenaturalpartnerintheseactivities.
Keywords:Computationaldesigntools;artificialintelligence;creativity
1.IntroductionRenowneddesignerKenyaHara(2007)writesthat“creativityistodiscoveraquestionthathasneverbeenasked”.Thisisespeciallytrueintheearlyphasesofthedesignprocess—thoseofdiscoveryanddefining—whereexploringnewinformationandconsideringitinnew,non-obviouswayshelpsdesignerstorevealnewmeaningsandassociations(Mendel,2012).Particularlyintheseearlyexplorations,usingdesigntoolsthatembracelessliteralanalogiesandallowforambiguityandserendipity(Gaver&Dunne,1999;Mothersill&Bove,2017)canprovokenewideasthatcrossovertheboundariesbetweenexistingconceptualschemas(Gero&Maher,2013).Thesecreativeleapscanhelpdesignersbreakthroughtothatmomentofinspiration(Cross,1997)whichguidesthedevelopmentofthedesigninthelatterphases.
Computationisincreasinglybeingintegratedintothetoolsusedthroughoutthecreativeprocess.Whilecurrentlybettersuitedtothemorewell-boundeddeductiveprocessofthelatterphasesofthedesignprocess(Bernal,Haymaker&Eastman,2015),ComputerAidedDesign(CAD)toolsarestartingtobeusedintheseearlier,moreabstractexplorations.Technologiessuchasgeneticalgorithmsandmachinelearningprogramsusestatisticalmathematicstorepeatedlygenerate,evaluateandoptimisedesignsolutions(Sjoberg,Beorkrem&Ellinger,2017),aswellasnavigateusthroughthe
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
2
multitudeofonlinecontentthatcaninspireournewcreations.Attheverycoreoftheseintelligenttechnologiesisanequationcalledthe‘costfunction’;theaverageoftheerrorbetweentheexpectedandactualdata,calculatedoverandoveragain.Itisfromminimisingthisaveragethatwearequicklyguidedtoconvergeonafewspecific,quantitativelybettersolutions,butisthisthebestapproachfortoolsusedintheearlier,moreabstractexplorationsofthedesignprocess?
Thesecomputationaltoolsareundoubtedlybetterthanhumansatquicklygeneratingamultitudeofdifferentdesignoptions(Steinfeld,2017),butwhenitcomestodiscoveringtheradicalinspirationneededforcreativebreakthroughsthesetechnologieshavetheirlimitations.The‘intelligent’toolsweareincreasinglyusingtofindinspirationforournewdesigns,suchasGoogleandPinterest,donotalwaysprovidethediversityofinformationandimagesthatweneedtoguideourresearchintheearlyphases;informationthathelpspromptustoquestionconceptsindifferentways,revealnewinsightsorinspireunexpectedideas(FultonSuri,2008).Artificialintelligencecanindeedhelpusfindhugeamountsofdataveryquickly,butifwearenotcarefulthesetechnologiescanalsopullusdownverycreativelyproblematic,average-driven,algorithmicrabbitholes(Carter&Nielsen,2017).
Perhapswedon’talwaysneedtheseintelligenttoolstobethat‘smart’orprovideuswithsuchoptimised,unambiguousresponses.Theambiguityprovidedbyimperfecttechnologiesandrandomnessdeliveredby‘dumb’AIscanactuallyaugmentourhumansmartness,andpotentiallyevenourcreativity(Shirado&Christakis,2017;Mothersill&Bove,2018).Thispaperexploresthisseemingparadoxandasks:howcandesigntoolsthatuse‘dumb’computationandpurposefulambiguityinfluencethecreativeprocessintheearlyphases?
2.ThelimitationsofaverageWhatisthebestwaytoLarissa?ThisisthequestionthatPlatoimaginedhisteacherSocratesandtheGreekgeneralMenodiscussing(Plato).SinceMenowasborninLarissa,heknewverywellhowtogettherefromprevioustravels.Aninexperiencedtravelercouldalsouseamaptomakethejourneymostefficient.Or,asatourist,hemightwishtoseethesitesalongthewayandthereforetakealessdirect,butpotentiallymoresatisfyingroute.Themoreadventuroussoulmightjustheadoutinthegeneraldirectionandletchanceguideheractionsalongthejourney.Thecoreofthisdialogueistoquestionwhatknowledgeis,butitalsorelatestoanimportantconsiderationforanyresearchintodevelopingnewcomputationaldesigntools:howshouldtheyguideus?Thisquestionhasbeenconsideredextensivelyinthefieldofcyberneticsandprovidesusefulinsightsintothechallengesforintegratingautomatedcomputationintothedesignprocess(Dubberly&Pangaro,2015).
CyberneticscomesfromtheGreekwordkybernētēs(κυβερνήτης)meaning"tosteer,navigateorgovern".Atitsmostbasic,acyberneticapproachtakesfeedbackfromasystemtounderstandhowtoreachagoalinthemostefficientway.BuildingonPlato’sanalogy,asacrowflyingoverthemountainsofAthens,wecouldnavigateourwaytoLarissausingcompassbearingsalongthemostdirectroute,modifyingourmovementstogettoourendgoal.Orappliedtothedesignprocess,computationalsystemsthatusetheseapproachescanhelpusanswerquestionssuchas“whatpossiblesolutionsfitthesegoals&constraints?”(Case,2018).
Ourcomputationaldesigntoolsareincreasinglyrelyingontheseintelligentstatistically-drivenapproachesor‘technologiesoftheaverage’.Byoptimisingtheaverageatthecoreofthecostfunctiondescribedabovetoquicklyconvergeonafewspecific,quantitativelybetter‘answers’,computationaltoolssuchasgeneticalgorithmsandmachinelearningprogramscanhelpusquicklydiagnoseamedicalcondition(Mukherjee,2017),generatethousandsofdesignsforachair(Rhodes,2016),orcreatea‘new’workofartbyanoldMasterpainter(Korsten,2016).
Beyond Average Tools
3
Whilethesetechnologiescanhelpusfindhugeamountsofcontentinsearchenginesorquicklygeneratedesignsfromsetsofdata,theefficiency-basedapproachtoanalysinginformationusedbythesesystemsmeansweareonlypresentedwiththeaverageofthismaterial.Googling‘chair’maynotbringyouimagestoinspirenewideas;youmightjustgetacollectionofpicturesthatlooksimilar.Pinterestboardsareoftenbecomingcollectionsofhomogeneouslysleekdesigns;somuchsothatdesignerssuggestthatwehavereachedthe“Pinterestsingularity”andareshunningitinanattempttonotcreateaverage-lookingdesigns(Gong,2018).
Integratingthenotionoftheaverageintothedesignprocessisnotnew(Rose,2016):fromitsoriginalapplicationtounderstandthediversityinhumansizes(leadingtotheBodyMassIndex),toitsuseinthefieldofscientificmanagement(orTaylorism)tooperationalizetheprocessesoffactoryworkers,tointegratingitintostandardizedergonomicmeasurementstodesignmass-consumableobjects(Dreyfuss&Dreyfuss,1967).Butjustasitsapplicabilitywasquestionedwhenitwasdiscoveredthatnoneofover4000pilotsmatchedallofthe10averagebodydimensionsthatcockpitswerebeingdesignedfor(Daniels,1952),perhapsweshouldbequestioningthesuitabilityoftechnologiesthatrelyonanefficiencyapproachusedintheearlyphasesofthecreativeprocess.
Incomparisontothiscurrentcomputationalapproachthatprioritisesefficiency,theearlyphasesofthedesignprocessneedalesslogicalexplorationfullofexperimentsandquestions(Schön,1983);wearetheadventurerswhoprefertherichnessofthescenicroutetoLarissa!Especiallywhendealingwiththeoftenill-formulated‘wickedproblems’thatwearedesigningfortoday(Churchman,1967),thebeginningofthedesignprocessfeelslikeaimingatashiftingtargetwhereweoftendon’tfullyunderstandtheproblem,letalonehaveadefinedgoal(Rittel,1988).Appreciatingthisflexibilityintheearlyphasesofthedesignprocessisveryimportantbecause,justas“weshapeourtoolsand,thereafter,ourtoolsshapeus”(Culkin,1967),theinspirationwecanobtaintoguideourdesignsisbeingshapedbythealgorithmsthatrulethemachinesweusetosearchfornewideas(Lynch,2016).Theargumentforintegratingtheseefficiency-basedapproachesintoourdesigntoolsisoneofconvenience(Carter&Nielsen,2017).Butcanoutsourcingourcreativetaskstotheseoverly‘user-friendly’interfacescontributetocognitiveinertia?Whilepartofthecreativeprocesscanindeedbenefitfromthecompetenceandefficiencythattheseintelligenttoolscanprovide(Steinfeld,2017),radicalbreakthroughscomeonlyfromconsideringconceptsmoreabstractly(FultonSuri,2008)andchallengingtheexistingprinciplesinourfields(Nielsen,2016).
3.AlternativestotheaverageItisoftenintheearlyphasesofthedesignprocess—thoseofdiscoveryanddefining—thatcreativeleapscanleadtoradicalbreakthroughs(Cross,1997).Activitiesinthesephasesinclude‘gatheringdisparateinformation’,‘generatinghypotheses’and‘identifyingnoveldirections’(Mothersill&Bove,2018);activitieswhereawidevarietyofinformationisexploredandconsideredinnon-obviouswaystohopefullyrevealnewmeaningsandassociations.Theseactivitiesinvolvetheoftenserendipitouscreativechallengesthathumansareverygoodat:consideringdifferentcontexts,embracingambiguityandusinganalogytofindnewinterpretationsandassociations(Bernaletal.,2015).
TheseelementsofthecreativeprocesswerechampionedbycreativityresearchersEdwarddeBonoandWilliamGordon.DeBonodevelopedthepracticeoflateralthinking,whichutilisedthefactthatthehumanmindisveryefficientatrecognisingpatterns;ifwearepresentedwithinformationwhichdoesnotimmediatelyseemrelevant,wenaturallytryto‘makesense’ofit.Lateralthinkingwelcomeschanceintrusions,irrelevance,andambiguityinordertoprovokedifferentpatternsandcreatenewideas(Bono,1970).ThisstrategywasalsoembracedbyGordoninthepracticeof
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
4
synectics—literallymeaning‘thejoiningtogetherofdifferentandapparentlyirrelevantelements’—where‘perfect’ideasarerejectedinfavourofthenon-rationalitythatcangeneratemoreevocativemetaphorsandseedsofinspiration(Gordon,1961).
Whencomparedtothecertaintyofferedtousthroughthetechnologiesoftheaveragedescribedabove,theearlyphasesofthedesignprocessoftenfollowalesslogicalandpredictablepath(Mitchell,1993)andsopotentiallyrequiredifferentapproaches.Purposelyintegratingnoiseintotheverypredictableandcontrollablesystemswearesofamiliarwith,suchasthroughambiguityandchanceintrusions,can“createamarginoferrorinwhichcreativeinterpretationandmisinterpretationmightthrive”(Bernes,2017).Ifweareopentoexploringthesemomentsofcreativereinterpretation,wemightdiscoverentirelynewapproachestoadesignproblemandinvent“waysofthinkingwhichhaven'tyetbeeninvented”(Nielsen,2016).
Ifambiguityandopennesstochanceinterventionsareimportantaspectsoftheearlyphasesofthedesignprocessthatcanhelpusdiscovernewideas,thenwebelievetheyshouldalsobeintegratedintothetoolsweuseinthosedesignactivities.Incontrasttothedriveforquantification,optimisationand‘intelligence’incurrenttechnologies(Sjobergetal.,2017),weareexploringhowthemoreserendipitousprinciplesofcreativity—thoseofseemingirrelevanceandambiguity—canbeusedasanapproachforcreatingnewcomputationaltools.Thefollowingsectionsdescribethe‘BeyondAverage’approachwehavetakentodeveloptwocomputationaldesigntoolsandtheevaluationscarriedouttounderstandhowtheycanbeusedtogeneratenewideas.
4.A‘BeyondAverage’approach Buildingontheseserendipitousprinciplesofcreativitypresentintheearlyphasesofthedesignprocess,weproposethefollowingdesignspacedimensionstoguidethedevelopmentofcomputationaltoolsthatcancontributetotheactivitieswherenewideasarediscovered:
Contextuality Thisdimensionassessestheamountofcontextualinformation—orseemingirrelevance—thatthetoolusestoguidethecollection,generationandreviewingofinspirationalinformationanddesignoutputs.Thisdimensioncanalsorelatetothe‘smartness’ofthetool.Atoolwithahighcontextualityintegratesalotofadvancedcomputationsuchasthemachinelearninganalysisofextensivedatasetstocalculateacontextually‘optimised’andrelevantresponse,e.g.asusedinasearchenginesuchasGoogle.Incontrast,atoolwithlowcontextualityisonethatusesmuchsimpleralgorithmssuchasrandomness,hencedoesn’tgeneraterecommendationslearnedfromprevioususesandcanoftenprovideseeminglyirrelevantresponses.
InterpretabilityThisdimensiondetermineshowdirectorambiguoustheinformationorcreativeguidanceprovidedbythetoolis;isitaprescriptionoraprovocation?Thisdimensioncanalsorelatetotheagencythattheuserhaswhenusingthetool.ExamplesoftoolswithlowinterpretabilityaresearchengineslikeGooglewhereauserentersaspecificrequestandthetoolreturnsverydirectlyrelatedinformationthatrequireslittleadditionalinterpretation;theuserisveryactiveinchoosingaspecificconcepttoexplorebutmorepassivewheninterpretingtheinformation.AnexampleofatoolwithahigherlevelofinterpretabilityisEnoandSchmidt’s(1975)ObliqueStrategiescarddeckthatdoesnotrequiretheusertochooseaninitialconceptbutreliesontheiractiveperceptionandimaginationto‘makesense’ofthemoreambiguousinformation.
Beyond Average Tools
5
Thesedimensionscreateaframingthroughwhichtoconsiderhowcomputationaldesigntoolscaninfluencethecreationofnewideasintheearlyphasesofthedesignprocess.Figure1showsourproposedpositioningofthe‘BeyondAverage’tools(describedinthenextsection)onthedesignspacedimensions,withGoogleincludedasabenchmarkofcurrenttools.
Figure1.Existingand‘BeyondAverage’toolsproposedmappingontodesignspacedimensions
5.‘BeyondAverage’designtools5.1.design(human)designcreativeprompttooldesign(human)designisacomputationalcreativeprompttoolthatprovokesnewassociationsbetweenconceptsinauser’sproject(http://reframe.media.mit.edu).Usingtextfromadesigner’sownnotesandreadings,design(human)designpresentsarandomisedprompt,helpingtojuxtaposeconceptsinnewways(Figure2).ThistoolwasdevelopedinresponsetofindingsfromfieldresearchatdesignconsultancyIDEO;thattoolsoffering‘structuredserendipitousinspiration’couldhelpprovokenewinterpretationsandideas(Mothersill&Bove,2017).
AsshowninFigure1,weproposethatthedesign(human)designtoolhasmediuminterpretabilityand,atitssimpleststate,alow-to-mediumlevelofcontextuality.Ifthetextcorpusismodifiedtoincludeinformationonlyrelatedtoacertaintopicorpersonaldataset,thelevelofcontextualitybecomesmedium-to-high.
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
6
Figure2.Screenshotfromdesign(human)designcreativeprompttool
5.2.LookingSidewaysinspirationexplorationtoolLookingSideways(http://sideways.media.mit.edu)isanonlineexplorationtoolthatseekstoprovokeunexpectedinspirationandcreatenewassociationsbyprovidinguserswithaselectionofsemi-randomlychosen,looselyrelated,diverseonlinesourcesfromart,design,historyandliteratureforeverysearchquery(Figure3).
AsshowninFigure1,weproposethattheLookingSidewaystoolhasalowerlevelofinterpretabilitythanthedesign(human)designtoolduetotheuser’smoreactiveengagementwithit.Atitsmostsimplestate,ithasamediumlevelofcontextuality,howeverifthedatabasesthatthetoolissearchingarecustomisedtoacertaintopicorpersonal‘creativewateringholes’,thelevelofcontextualitycanbecomequitehigh.
Figure3.ScreenshotfromLookingSidewaysinspirationexplorationtool
Beyond Average Tools
7
6.EvaluationmethodologyToevaluatethecreativepotentialofthesetools,wecarriedoutstudieswithbothprofessionalandstudentdesigners.18participants(10men,8women)tookpartinanobservedstudywheretheywereaskedtogeneratecreativeresponsestooneoftwothemes(“automatedsystems(inthehome,work,cityetc.)thatwetrust”and“thefutureofwellness(inthehome,work,cityetc.)thatisintegrated”)usingtheBeyondAveragetoolstoprovideinspiration.Thetextcorpusthatthedesign(human)designtooldrewfromwascustomizedforeachthemeusingwordsfromrelevantWikipediapagesandarticles.Theresultspages(includingimages,news,shoppingetc.)fromGoogle’ssearchenginewasusedasacontroltool.Theparticipantshad10minutestouseeachtooltoexplorethethemesandgenerateideasbasedontheinspirationtheyprovided,notingdownanyideasorsketchesusingpenandpaper.Aslearningfromprevioustoolswasinevitable,theorderofthetoolswasrandomisedacrossparticipants.Finally,participantscompletedasurveythataskedquestionsrelatedtothepotentialofeachtooltoprovideunexpectedcreativity(https://bit.ly/2FkvMEU).
Shah&VargasHernandez’s(2003)metricsformeasuringideationeffectiveness—novelty,variety,quality,quantity—aswellasmetricsrelatingtodeBono’s(1970)analysisoflateralthinking—whetherideasareofimmediateusefulness,areasforfurtherexplorationornewapproachestoproblem,andiftheyareverticallyorlaterallyrelated—wereintegratedintoquestionsthatparticipantsratedona5pointLikertscale.Overallcommentsabouthowthetoolsinfluencedtheparticipants’generationofnewideas,howthetoolscouldintegrateintotheircreativepracticeandanysuggestionsformodificationswerealsocollected.
7.FindingsWhilewedidcollectnumericaldataaboutthecreativitymetricsanddesignspacedimensionsdescribedabove,weacknowledgethatitishardtodrawgeneralisablequantitativefindingsfromthesetypesofsubjective,noteasilyrepeatablecreativeinterventions,especiallywithourrelativelysmallsamplesize.Therefore,herewewillpresentgeneraltrendsindicatedbythequantitativedataandextendtheanalysisoftheseinsightswiththequalitativefeedbackalsocollected.Asbothofthethemestestedprovidedsimilarresponses(mostratingswerewithinoneLikertpoint),wehavecombinedthedataintoasingleaverageusedintheresultsbelow.Theredidappeartobesomeeffectsduetothedifferentorderofthetoolsshowntotheparticipants,butthosewillbediscussedfurtherinthenextsection.
Tounderstandiftheparticipantshadasimilarexperienceusingthetoolsasweexpected,Figure4showstheparticipantsratingsofhowcontextualandambiguoustheyconsideredresponsesgeneratedbythetools(Lowcontextuality/interpretability=1;highcontextuality/interpretability=5).Theparticipantsgenerallyagreedwithourhypothesisforwherethesetoolssitwithinthedesignspacedimensions:Googlewasconsideredtogiveverydirect,highlycontextualresponses,design(human)designwasconsideredtohavethemostinterpretabilityandmediumcontextuality,andLookingSidewayswasconsideredtogivemediumlyambiguousandcontextualresponses(slightlylowerthanourexpectation,likelyduetotechnicallimitationswiththeprototype).
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
8
Figure4.Existingand‘BeyondAverage’toolsmappingontodesignspacedimensionsbyparticipantscomparedwithproposedmapping
Reviewingthedatamappedagainstthedesignspacedimensionsindividuallyrevealssomelargertrendsabouthowthelevelsofcontextualityandinterpretabilityaffectcreativeoutput.Figures5and6showtheratingsforeachofthetoolsforthemetricsdescribedabovemappedalongthedesignspacedimensions.Lineshavebeenaddedbetweenthediscretedatapointstoindicatetrendsinhowthecreativitymetricsmightvaryasadesigntoolincludesmoreorlesscontextualityandinterpretability.Quantityofideasisnotincludedasalltoolsgeneratedsimilarresults(1-2ideas),probablyduetotheshorttimeallowedforthetask.
7.1.TheinfluenceofcontextualityonthecreativeprocessFigure5showsthatGoogle—thetoolwiththehighestcontextuality—hadthelowestratingsformostofthemetrics(between2.33and3.83).Despiteparticipants’familiaritywithusingGoogletogatheralargequantityofinformationonatheme,itshighcontextualitymeantthisknowledgewassituatedintermsofwhatotherpeoplehavedoneandthoughtbefore;the“generallyaccepted‘norm’answers”.Whilethishelpedsomeparticipantsidentifycommonfeaturesortrends,itledotherstofeeltherewas“toomuchpriminginthewrongdirection.”ThehighcontextualityofGooglewasconsideredbeneficialwhentheparticipanthasalready“honedinonsomethingnarrow”andis“thinkingaboutframingtheirenquiry”,butwas“notusefulfordeeplyassessingwhere[their]ideasweresituated”andthereforenottherighttoolforcomingupwithnewideas.
Beyond Average Tools
9
Figure5.Mapofcreativitymetricsagainstthelevelofcontextualityineachofthetoolsstudied
Incontrast,thedesign(human)designtool(mediumcontextuality)wasratedhighestforallmetrics(between3.17and4.67).Thelowerlevelofcontextualitywasfoundhelpfulinliberatingtheparticipantsfromtheirownpreconceptions.Beingprimedwithtextrelatedtothetwothemesallowedthetooltoeasilyprovidemanysimplebutdifferent“relativelystablestartingpoints”fromwhichideascouldbeconstructed.However,duetotheformatofthetool,someparticipantsfeltthatthepromptsoftenfellintomoreproject-basedtasksratherthangeneralinspiringconcepts,limitingtheirboundariesofthought.Anotherparticipantalsocommentedthatwhile“arbitrarinesscanbeverypowerfulforlateralthinking…sometimesitcanfeelforcedordifficulttodrawconnections”andthat“knowingwhentoskipandwhentoponder”aseeminglyirrelevantconnectionrequiresconsideration,andpotentiallyguidance.
HelpingtoseelinksbetweenideaswasoneofthefeaturesthatparticipantslikedintheLookingSidewaysexplorationtool;addingalevelofcontextualitytoseeminglyunconnectedconcepts.Thisabilitytovisuallymaphowrandomconceptsintersect“providednicetangents”toopenuptheirexistingideadomain.Asparticipantscontrolledthecontextoftheexplorationbyenteringtheirownsearchterms“someconnectiontothegoalisthere”whichguidedoneparticipant“intoaheadspacethatiscomfortableandthatIfeelauthoritativein,butisnewterritory.”Despitethisfeedback,participantsstillratedthetoolasfairlylowcontextualityanditdidnotscoreashighlyasthedesign(human)designtoolintermsofcreativity(between2.56and4.11).Ingeneral,participantslikedthatthesearchresultswerenotdefinedbypopularitysuchasonGoogle,butduetolimitationsinthenumberofcontentsourcesinthecurrentprototype,therewasn’talargeenoughamountofinformationavailabletoexploreaconceptdeeply—asGoogleprovides—orconsidermanynewperspectives—asthedesign(human)designtoolprovides.
Overall,itappearsthattoolswhichprovidemorehighlycontextualresponses,i.e.Google,aregoodforexploringanarrowsubjectoncedesignparameters(orsearchterms)areknownbutthefocusedrangeofsimilarinformationlimitstheabilitytogeneratenewideasorconnections.Toolsthathavealowercontextuality—design(human)designandLookingSideways—canprovidetangentiallyassociatedresponsesthatpromptparticipantstoreconsiderhowconceptscouldbeinterpretedandconnected,providingthemwithinteresting“startingpoints”fornewideastoexplorefurther.
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
10
7.2.TheinfluenceofinterpretabilityonthecreativeprocessMappingthesameresultsontotheinterpretabilityaxis,Figure6showsacleartrendtowardsgreatercreativitywithhigherlevelsofinterpretability.ForGoogle(lowinterpretability)participantsarereliedupontocomeupwithinterestingsearchterms,hencetheresponsescanonlybe“ascreativeasyourownmindessentiallyallowsyoutobe.”ThisimprovedwithhigherlevelsofinterpretabilityintheLookingSidewaystoolasitsabilitytoconnectrandomuser-definedconceptsprovidedfresh,unexpectedinputthat“encouragedmomentumandoutgrowth”and“awaytoriffoutfromwhereIalreadyam”.Presentingtheresponsesinamorevisual,unorganisedmanneralsoallowedfortheparticipantsto“makeamess”,inspiringlessliteralconnectionsandmorevariedinterpretationsbecausetheycanfindtheirownsenseinthecontent.
Thetoolthatprovidedthemostvariedandnewconnectionswasthedesign(human)designtool(highinterpretability).Participantsfoundthatwhentheyallowedthemselvestoletgoofcontrollingthetoolandconsidertheoftenambiguousresponsesinamoreflexibleway,therandomjuxtapositionsofconceptschallengedthemtotakeon“amorenon-structuralthinking”thatprompted“newandverydifferentpointsofviewsonmyideas”;afeelingthatseveralparticipantsdescribedasbeingrareincomparisontoothercompuationaldesigntoolstoday.However,whilemanyparticipantsenjoyedthepossibilitytoquicklyiteratethroughahighnumberofambiguouspromptsasithelpedthemgetintoadifferentmindset,afewconsideredthejuxtapositionofeventwooftheoftenverybroadconceptsrequiredalotoftimetothinkdeeplyaboutthepotentialconnectionsbetweenthem.
Overall,thereseemstobeacleartrendthathigherlevelsofambiguityintheresponsesprovidedbythetools—somethingwecouldalsodescribeasahigherlevelofcreativeagencyonthemachine’spart—allowedformorevarietyofinterpretationswithintheinformationpresentedandthereforeagreaterpossibilityfornewconnectionsandideastobemade.
Figure6.Mapofcreativitymetricsagainstthelevelofinterpretabilityineachofthetoolsstudied
7.3.TherolesoftheBeyondAveragetoolsinthedesignprocessFromtheresultsdiscussedabove,wesuggestthatcomputationaltoolswithamediumlevelofcontextualityandamedium-to-highlevelofinterpretabilitycanpositivelyinfluencecreativityintheearlyphasesofthedesignprocess.Thelateralresponsestosearchqueriesandsomewhatrandom
Beyond Average Tools
11
provocationsenabledbyhigherlevelsofinterpretabilityallowparticipantstohavesomeagencyoverthedirectionofexplorationsbutalsobeprovokedtorethinkhowsomethingseeminglyirrelevantcouldbecontextual;responsesthatmakejustenoughsenseandprovideahighpotentialcontextualityforparticipantstogeneraterelevantbutnovelideas.
Figure7showsthisquadrantofthedesignspacedimensionswasalsoratedthemostdesirableforinspiringnewideas,supportedbythedesign(human)designtoolbeingratedfavouritebymostparticipants(11outof18).However,oneparticipantcommentedthatdesiringtoolsinthisquadrantofthedesignspaceseemedlikeaparadox.ThisrelatestohowparticipantsfeltGoogle—andthegeneraltrendforefficientsearchtools—hadconditionedthemtothinkinalogicalwayandusingtheBeyondAveragetoolshelpedthemembracemoreambiguous,non-deterministicapproaches.
Figure7.Proposedmappingofexistingand‘BeyondAverage’toolsontodesignspacedimensionscomparedwithdesiredamountofcontextualityandambiguityratedbyparticipants
Theeffectofthesedifferentapproacheswasnoticeablethroughtheordereffectsthatemerged.WhentheBeyondAveragetoolsweretestedfirst,participantsstartedtoconsiderhowtheycoulduseGooglemorecreatively,withmixedsuccessduetoitsmoreefficiency-orientedsearchapproach.
Thefactthatthesetoolscaninfluenceeachotherisanexcitingfinding.Whilesomeparticipantsdiddistinguishthetoolsforseparatedesignactivities,e.g.design(human)designforbrainstormingandLookingSidewaysasamappingtooltodocumenttheircreativeprocess,mostthoughttheywouldbeusefulasasuite.Usingamediumlycontextualisedversionofthedesign(human)designtoolwasconsideredausefulcreative‘icebreaker’forseedinginterestingnewdirectionsforfurtherexploration,followedbytheLookingSidewaystooltosuggestlateralconnectionsbetweenconceptsandGoogletogathermorefocusedinformationtofurtherframetheirideas.IntegratinginformationrelatedtokeyconceptsexploredinGoogleandtheLookingSidewaystoolbackintoamorecontextualisedversionofthedesign(human)designtoolwassuggestedasawaytofurthergeneratenovelbutmorefocusedideasrelatedtotheparticipant’semergingthemesanddesignparameters.
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
12
Thisimaginedroleofthetoolsinthedesignprocessindicatesasomewhatcyclicalneedforhighlevelsofcontextualityandinterpretabilityinexplorationandideationactivities.Whenusingcomputationaltoolswithveryhighlevelsofcontextuality,e.g.Google,thecreativeagencyisdeterminedbythehuman;thesearchtermsaredeterminedbythedesigner,oftenthroughsomenon-computationalmeanssuchasbrainstorming.Whenthecomputationaltoolcanhavecreativeagencyaswell,e.g.throughusinghigherlevelsofinterpretabilityasthedesign(human)designandLookingSidewaystoolsdo,thecomputercancontributetothedesigner’screativeagencyandbecomemoreofanaturalpartnertoguidetheearlyphasesofthedesignprocess.
7.4.FutureresearchTheseresultshavehighlightedexcitingopportunitiesforustopursue.Modificationstothetoolsinclude:automatingthecustomisationofthetextcorpusinthedesign(human)designtooltogeneratemorecontextuallyspecificprovocations,expandingthenumberofcontentsourcesintheLookingSidewaystool,andfixingseveraluserinteractionissues.ExtendingtheLookingSidewaystool,wearealsodevelopingtheDesignDaydreamstableandpost-itnote;alow-techaugmentedrealitytoolthatcanprojectthedigitalcontentexploredontoobjectsintherealworld(Figure8).
Figure8.DesignDaydreamsaugmentedrealityviewers(aspartofalargeraugmenteddraftingtable)
Acknowledgingthatobservedstudiesarelimitedwheninvestigatingthedesignprocess,wearealsocarryingoutlongerunobservedstudiestofurtheranalysethetools.Intheselessstructuredstudies,weimaginetheremightbeagreaterhesitancytoembracetheserendipitouslogicoftheBeyondAveragetools,especiallyinreal-worldprojectswhenproductivitydemandsarehigher.Weaimtoinvestigatethisapparentlimitationofthetools’effectivenessbyexploringhowtheresponsesprovidedcanbetherightbalanceofdisruptiverandomnessandefficientrelevance.Throughunderstandinghowtobetterframethebenefitsthesetoolscanprovidewithindifferentdesignactivities,weaimtostimulatepurposefulmomentsofunexpectedcreativereinterpretationfordesigners,aswellasslowlybroadentheirattitudesaboutthedifferentwayscomputationaltoolscanguideustobe‘productive’inthecreativeprocess.
Beyond Average Tools
13
8.ConclusionIntheearlyphasesofthedesignprocess,embracingchanceintrusions,seemingirrelevanceandambiguitycanleadtoconsideringconceptsindifferentwaysandprovokenewideas.However,thecomputationaltoolsweareincreasinglyusinginthesephasesvalueefficiencyoverserendipity;technologieswhosefoundationsareanaverage.Thispaperexploredhowdevelopingcomputationaldesigntoolsthatembraceseemingirrelevanceandambiguitycouldinfluencethecreativeprocessintheearlyphases.
The‘BeyondAverage’approachdefinedtwodesignspacedimensions:contextuality—how‘smart’responsesfromthetoolwere—andinterpretability—howambiguoustheresponseswere.Situatedatdifferentpositionsalongthesedimensionsaretwotoolsdevelopedbytheauthors:thedesign(human)designcreativeprompttoolandtheLookingSidewaysexplorationtool.Resultsfromstudiesusingthesetoolstoprovideinspirationtoparticipantsastheyattemptedtogeneratenewideasaroundatheme(withGoogleasacontrol)showedthatcomputationaltoolswithamediumlevelofcontextualityandahigherlevelofinterpretabilitycanpositivelyinfluencethecreationofnewideas.
Imaginingthesetoolsusedasasuiteintheirdesignprocess,participantssuggestedjumpingbetweenthetoolswhentheyneededdifferentlevelsofcontextualityandinterpretability;usingtheveryambiguousdesign(human)designtooltoprovokenewseedsofideasthattheycandeeplyexploreinthemoresituatedGooglesearchengineandLookingSidewaystool.Extendingthisdiscussiontoconsidertheroleofcomputationintheearlyphasesofthedesignprocess,wesuggestthattoolswithhigherlevelsofcreativeagency—thosewithhighlevelsofbothcontextualityandinterpretability—cancontributetothedesigner’screativeagencyandbecomeamorenaturalpartnerintheseactivities.
ReferencesBernal,M.,Haymaker,J.R.,&Eastman,C.(2015).Ontheroleofcomputationalsupportfordesigners
inaction.DesignStudies,41,163-182.
Bernes,J.(2017,May)"ThePoetryofFeedback."e-fluxJournal#82.AccessedJune01,2018.https://www.e-flux.com/journal/82/127862/the-poetry-of-feedback/
Bono,E.D.(1970).Lateralthinking.Atextbookofcreativity.London.
Carter,S.,&Nielsen,M.(2017).Usingartificialintelligencetoaugmenthumanintelligence.Distill,2(12),e9.
Case,N.(2018,February)."HowToBecomeACentaur."JournalofDesignandScience.AccessedJune01,2018.https://jods.pubpub.org/pub/issue3-case
Churchman,C.W.(1967)"Guesteditorial:Wickedproblems."B141-B142.
Cross,N.(1997).Creativityindesign:analyzingandmodelingthecreativeleap.Leonardo,311-317.
Culkin,J.M.(1967).“Aschoolman'sguidetoMarshallMcLuhan.”SaturdayReview,Incorporated,1967.
Daniels,G.S.(1952).Theaverageman?(No.TN-WCRD-53-7).AIRFORCEAEROSPACEMEDICALRESEARCHLABWRIGHT-PATTERSONAFBOH.
Dreyfuss,H.,&Dreyfuss,H.(1967).Themeasureofman:humanfactorsindesign.NewYork:WhitneyLibraryofDesign.
Dubberly,H.,&Pangaro,P.(2015).Cyberneticsanddesign:Conversationsforaction.Cybernetics&HumanKnowing,22(2-3),73-82.
Eno,B.,&Schmidt,P.(1975).Obliquestrategies:Overonehundredworthwhiledilemmas.
PHILIPPA MOTHERSILL & V. MICHAEL BOVE
14
FultonSuri,J.(2008).Informingourintuition:Designresearchforradicalinnovation.RotmanMagazine,52-57.
Gaver,W.,&Dunne,A.(1999,May).Projectedrealities:conceptualdesignforculturaleffect.InProceedingsoftheSIGCHIconferenceonHumanFactorsinComputingSystems(pp.600-607).ACM.
Gero,J.S.,&Maher,M.L.(Eds.).(2013).Modelingcreativityandknowledge-basedcreativedesign.PsychologyPress.
Gong,S.(2018,Nov22).DesigneratIDEOCoLabCambridge.Personalinterviewwithauthor.
Gordon,W.J.(1961).Synectics.HarperandRowPublishers.
Hara,K.(2007).Designingdesign.LarsMullerPublishers.
Korsten,B.(2016)TheNextRembrandtproject(acollaborationbetweenMicrosoft,ING,DelftUniversityofTechnologyandtheMauritshuisandRembrandthuisartmuseums).AccessedNovember22,2018.https://www.nextrembrandt.com/
Lynch,M.P.(2016).TheInternetofus:Knowingmoreandunderstandinglessintheageofbigdata.WWNorton&Company.
Mendel,J.(2012).Ataxonomyofmodelsusedinthedesignprocess.interactions,19(1),81-85.
Mitchell,W.J.(1993).Acomputationalviewofdesigncreativity.InGero,J.S.,&Maher,M.L.(Eds.).Modelingcreativityandknowledge-basedcreativedesign,(pp.25-42),PsychologyPress.
Mothersill,P.&Bove,V.M(2017)Humans,MachinesandtheDesignProcess.ExploringtheRoleofComputationintheEarlyPhasesofCreation,TheDesignJournal,20:sup1,S3899-S3913,DOI:10.1080/14606925.2017.1352892
Mothersill,P.&Bove,V.M.(2018)AnOntologyofComputationalToolsforDesignActivities,ProceedingsofDRS2018InternationalConference,25–28June2018,Limerick,Ireland,p1261-1277,DOI:10.21606/dma.2018.456
Mukherjee,S.(2017,April).“A.I.versusM.D:Whathappenswhendiagnosisisautomated?”TheNewYorker.AssessedJune01,2018.https://www.newyorker.com/magazine/2017/04/03/ai-versus-md
Nielsen,M.(2016,November)"ThoughtasaTechnology."TheConversation.AccessedJune01,2018.http://cognitivemedium.com/tat/index.html
Plato,Meno,97
Rhodes,M.(2016,October).“So.AlgorithmsAreDesigningChairsNow.”Wired.AccessedJune01,2018.https://www.wired.com/2016/10/elbo-chair-autodesk-algorithm/
Rittel,H.W.J.(1988).Thereasoningofdesigners.IGP
Rose,T.(2016)Theendofaverage:Howtosucceedinaworldthatvaluessameness.PenguinUK.
Schön,D.A.(1983).Thereflectivepractitioner:Howprofessionalsthinkinaction(Vol.5126).Basicbooks.
Shah,J.J.,Smith,S.M.,&Vargas-Hernandez,N.(2003).Metricsformeasuringideationeffectiveness.Designstudies,24(2),111-134.
Shirado,H.,&Christakis,N.A.(2017).Locallynoisyautonomousagentsimproveglobalhumancoordinationinnetworkexperiments.Nature,545(7654),370.
Sjoberg,C.,Beorkrem,C.,Ellinger,J.(2017,November).EmergentSyntax:MachineLearningfortheCurationofDesignSolutionSpace.InACADIA2017:DISCIPLINES&DISRUPTION(Proceedingsofthe37thAnnualConferenceoftheAssociationforComputerAidedDesigninArchitecture),pp.552-561
Steinfeld,K.(2017,November).DreamsMayCome.InACADIA2017:DISCIPLINES&DISRUPTION(Proceedingsofthe37thAnnualConferenceoftheAssociationforComputerAidedDesigninArchitecture),pp.590-599
Beyond Average Tools
15
AbouttheAuthors:
PhilippaMothersillisaPhDstudentintheMITMediaLabObject-BasedMediagroup,whereshedrawsfromdesigntheory,cognitivepsychologyandcomputersciencetodevelopnewcomputationaldesigntoolsforearlystagesinthecreativeprocess.
V.MichaelBoveistheheadoftheObject-BasedMediagroup.Heistheauthororco-authorofover100journalorconferencepapersondigitaltelevisionsystems,videoprocessinghardware/softwaredesign,multimedia,scenemodeling,userinterfaces,visualdisplaytechnologies,andoptics.
Acknowledgements:wewouldliketothankMaxLeverandJosieKufortheirtechnicaldevelopmentassistanceandtheKennedyMemorialTrustforprovidingfinancialsupportforaportionofthework.