Artificial Creativity

8/2/2019 Artificial Creativity

1/32

ARTIFICIALCREATIVITY:

ASyntheticApproachtotheStudyofCreativeBehaviour

ROBSAUNDERS,JOHNS.GERO

TheKeyCentreofDesignComputingandCognition

DepartmentofArchitecturalandDesignScience

UniversityofSydney,NSW,2006,Australia

e-mail:{rob,john}@arch.usyd.edu.au

Abstract.Wepresentanovelapproachtothecomputationalstudyofcreativity, called Artificial Creativity. Artificial Creativity promotesthe study of the creative behaviour of individuals and societies inartificialsocietiesofagents.Itissimilartotheapproachtothattakenby Artificial Life researchers involved in developing computational

models.We presenta frameworkfor developingArtificialCreativitysystems as an adaptation of Lius dual generate-and-test model ofcreativity. An example implementation of an Artificial Creativitysystem is presented to illustrate the potential benefits of our newapproachasawayofinvestigatingtheemergentnatureofcreativityinsocieties of communicating agents. Finally, we discuss some futureresearch directions that are possible by extending the abilities ofindividualsandstudyingtheemergentbehaviourofsocieties.

1.Introduction

The aim of Artificial Creativity is to gain a better understanding of

creativity-as-it-isinthecontextofcreativity-as-it-could-be.Inotherwords,

itisthestudyofcreativityas foundinhumansocietieswithcreativityas it

maybefoundinartificialsocietiesofagentsthatmayfollowquitedifferent

socialconventions.Inthisway,thestudyofArtificialCreativityissimilarto

thestudyArtificialLife; both aresynthetic approachesto understanding a

complex, and ill-defined behavioural phenomenon, i.e. creativity and life

respectively.

The Artificial Creativity approach provides an opportunity for

researchers to study the emergence of creative behaviour in controllable

environments,affordinganumberofpossiblestudiesnotpossibleinthereal


2/32

2 A.AUTHOR,B.AUTHORANDC.AUTHOR

world. The parameters that control the behaviour of individuals can be

experimentedwithto study theaffectthat they haveon theemergence of

socialstructures.Theenvironmentthatthesocietyofagentsissituatedin

can be adjusted to study the affects that external factors have on the

creativityofindividualsandsocieties.

As with Artificial Life, one of the most interesting possibilities of

Artificial Creativity is to be able to re-run history with different starting

conditions to find out how products of creative individuals and the

structures of creative societies might have differed. For example, by re-

running an Artificial Creativity simulation with different communicationpolicies we can simulate the affect that different communication

technologiesmighthaveonthe developmentanddisseminationofcreative

ideas.

Artificial Creativity is compatible with previous approaches that have

developedcomputationalmodelsofcreativethinkingbyallowingthemtobe

deployed within the context of artificial societies as long as they can be

embedded within agents that conform to the requirements of Artificial

Creativity.The study ofthebehaviourof creativethinkingwithinartificial

societiesprovidestheopportunitytodevelopabetterunderstandingofthe

situatednessofcreativeprocesseswithinsocio-culturalsituations.AsSimon

(1981)notes,muchofthecomplexityof humanbehaviourmaycomefrom

thecomplexnatureoftheenvironmentthattheyinteractwith.

2.Creativity

Theneedtodefinethenatureofcreativityhashauntedattemptstodevelop

theoriesoftheprocessesinvolvedincreativethinking.Thedifficultyofthis

task is apparent from the number of definitions that can be foundin the

literature:Taylor(1988),forexample,givessome50definitions.Expressed

in the definitions of creativity are some widely different opinions about

what it means for a person to be creative. From reading the literature, it

seemsthatnoagreementmaybereachedondetailsofthecreativeprocess;

however,thedefinitionsprovidedcanbedividedintotwobroadcategories.

Firstly,therearedefinitionsofcreativitythatemphasisecreativethinking

and promote the view that creativity can be studied solely as a mental

phenomenon.Thesedefinitionshavebeena popularin variousapproaches

tostudyingcreativitythatdealwithindividuals,forexample,inpsychology,

cognitive science and artificial intelligence. The models of creativity

proposedbyKoestler(1964),Newelletal.(1962),andHofstadter(1979)go

intogreatdetailaboutthecognitiveprocessesinvolvedincreativethinking,

particularly theprocessesinvolvedinthegenerationofpotentiallycreative

ideas. Many of the computational models of creativity are either based


3/32

PAPERTITLE 3

directlyonthesemodels(e.g.Langleyet al.,1987;Hofstadteret al.,1995)

orarebasedonsimilarmodelsofcreativethinkingfrompsychology(e.g.

PartridgeandRowe,1994).

Definitionsof creativityinthe secondcategoryrecognisethatcreativity

goesbeyondthegenerationofnovelideasandthatsociety,astheaudience

of the creative individual, plays an important role in defining what is

creative.Creativityisthereforedefinedwithastronghonorificsensethatis

as much the result of an audiences appreciation of a work as it is the

creatorsproduction.Proponentsofthesedefinitionscontendthatcreativity

cannotoccurinavacuumandmustbestudiedinthecontextofthesocio-culturalenvironmentofthecreator(Csikszentmihalyi,1999).Thisdefinition

hasbeenpopularinfieldsthatconsiderthecreativityofmultipleindividuals

over extended periods of time, for example, in history, sociology and

anthropology(e.g.Martindale,1990).

Some researchers have attempted to combine these two views of

creativity into unified theoretical frameworks. However, the resulting

frameworks often maintain the distinction between personal and socio-

cultural notions of creativity, as in Bodens P-creativity and H-creativity

(Boden,1990)andGardnerssmall-candbig-ccreativity(Gardner,1993).

2.1.ASYSTEMSVIEWOFCREATIVITY

WhenCsikszentmihalyidevelopedhissystemsviewofcreativity,heturned

his attention away from the question What is creativity? and focussed

upontheissuessurroundingthequestionWhereiscreativity?Importantly,

Csikszentmihalyi questioned the mentalistic assumption that creative

processesareonlytobefoundinthemindofthecreativeindividual.Instead

heproposedthatprocessesessentialtocreativity,whetherpersonalorsocio-

culturallydefined, are to be found in theinteractions betweenindividuals

andthesocietythattheyaresituatedwithin.

Thesystemsviewof creativitywasdevelopedbyCsikszentmihalyias a

model of the dynamic behaviour of creative systems that include

interactions between the major components of a creative society

(Csikszentmihalyi, 1988). Csikszentmihalyi identified three importantcomponents of a creative system; firstly there is the individual, secondly

thereisacultural,orsymbolic,componentcalledthedomain,andthirdly

thereis a social, orinteractive,component called the field.A map ofthe

systemsviewofcreativityispresentedinFigure1.


4/32


SOCIETY

Field Individual

CULTURE

Domain

PERSONAL

BACKGROUND

Stimulates

Novelty

ProducesNovelty

Transmits

Information

Selects

Novelty

Figure1:Csikszentmihalyissystemsviewofcreativity(afterCsikszentmihalyi,1999).An individuals role in the systems view is to bring about some

transformation ofthe knowledge heldin the domain. The fieldis a setof

social institutions thatselects from thevariations producedby individuals

thosethatareworthpreserving.Thedomainisarepositoryofknowledgeheldbytheculturethatpreservesideasorformsselectedbythefield.

Inatypicalcycle,anindividualtakessomeinformationprovidedbythe

culture and transforms it, if the transformation is deemed valuable by

society,itwillbeincludedinthedomainofknowledgeheldbytheculture,

thusprovidinganewstartingpointforthenextcycleoftransformationand

evaluation.In Csikszentmihalyisview,creativityis nottobe foundinany

oneoftheseelements,butintheinteractionsbetweenthem.

2.1.1.LiusDualGenerate-and-TestModelofCreativity

Recognisingtheneedforaunifiedmodelofcreativityindesigncomputing,

Liu(2000)presentedasynthesisofthepersonalandsocio-culturalviewsof

creativityinasinglemodel.Liurealisedthattheexistingmodelsofpersonalcreativity complemented the socio-cultural models by providing details

abouttheinnerworkingsofthecreativeindividualmissingfromthemodels

ofthelargercreativesystem.

Liuproposeda dualgenerate-and-testmodelofcreativityasasynthesis

ofSimonetalsmodelofcreativethinkingandCsikszentmihalyissystems

view. Asits name suggests, thedual generate-and-test model of creativity

encapsulatestwo generate-and-testloops:oneat theleveloftheindividual

and the other at the level of society. The generate-and-test loop at the

individual level, illustrated in Figure 2(a), provides a model of creative


5/32

PAPERTITLE 5

thinking, incorporatingproblemfinding, solutiongeneration andcreativity

evaluation.Thesocio-culturalgenerate-and-testloopmodelstheinteractions

among the elements of Csikszentmihalyis systems view of creativity, as

illustrated in Figure 2(b). In particular, it captures the role that the field

playsasasocio-culturalcreativitytest.Thecombineddualgenerate-and-test

modelofcreativityisillustratedinFigure2(c).

GenerateProblem

Finding

no

yes

yes

no

Field

Individual

Domain

Personalcreativesolution

generationandrecognition

Social-culturalrecognitionofcreativeideasbyagroupofauthorisedpeople

Sourceof

initialdataand

knowledgein

thedomain

Creativity

Test

GenerateProblem

Finding

no

Individual

Personalgenerate-and-test

Test

Domain

Knowledge

Creative

Product

yes

no

yesField

Individual

Domain

Socio-culturaltest

Problems&

SolutionsSocio-culturalgenerate

(a)

(b)

(c)

Figure2:Liu'sDualGenerate-and-TestModelofCreativeDesign:(a)thepersonalgenerate-and-testmodel,(b)thesocio-culturalgenerate-and-testmodel,(c)thecombineddual

generate-and-testmodel.

Lius model unifies Simon et als and Csikszentmihalyis models of

creativity to form a computational model of creativity that shows how

personalandsocio-culturalviewsofcreativitycanbemodelledinasingle

system.ComparedtoBodensmodelofcreativity,thedualgenerate-and-test

model of creativity models both the P-creativity and H-creativity of

individuals using the generate-and-test loops at different levels.Using the

languageofGardnerwe may say thatwhatdistinguishessmall-c creativity

frombig-ccreativityisthatbig-ccreativityaffectschangestothedomain

whereassmall-ccreativitydoesnot.Lius dual generate-and-test model shows that it is possible to cast

Csikszentmihalyis systems model in computational terms and thereby

providesuswithausefulbasisforaframeworkfordevelopingmodelsof

Artificial Creativity. Before developing Lius model further, we will

examinetherequirementsofacomputationalmodelofArtificialCreativity.


6/32


3.ArtificialCreativity

The Artificial Creativity approach that we propose here is based on

Langtons approach todevelopingcomputational modelsof ArtificialLife

(Langton, 1989). The essential requirements of a computational model of

ArtificialCreativityare:

Themodelcontainsasocietyofagentssituatedinaculturalenvironment.

Thereisnoagentthatcandirectthebehaviourofalloftheotheragents.

Therearenorulesthatdictateglobalbehaviour.

Agentsinteractwithotheragentstoexchangeartefactsandevaluations.

Agentsinteractwiththeenvironmenttoaccessculturalsymbols.

Agentsevaluatethecreativityofartefactsandotheragents.

Many of the requirements of a computational model of Artificial

Creativity are similar to the requirements of a computational model of

Artificial Life. Although some of the details are different, both types of

modelsconsistofa populationofagents,andbothrequirethatthereareno

rulesoragentsthatcandictateglobalbehaviour.

AnadditionalrequirementofArtificialCreativityagentsnotfoundinthe

requirementsofArtificialLifeisthattheagentsinanArtificialCreativitymodelmustbeabletomakeindependentvaluejudgementsandadapttheir

behaviour accordingly.Morespecifically,agentsinan ArtificialCreativity

systemmustbe abletomakeevaluativejudgementsaboutthe creativityof

agentsandproductsinordertoimplementthepersonalandsocio-cultural

creativitytestsfoundinLiusmodel.

Toillustratetheapproach,considerhowonewouldmodelasocietyof

artists.First,wewoulddefinearepertoireofbehavioursfordifferentartistic

agentsandcreatelotsoftheseagents.Wewouldthenstartasimulationrun

by specifying some initial social configuration of the agents within a

simulatedculturalenvironment.Fromthispoint onwardsthe behaviour of

the system would depend entirely on the interactions between different

agents and the interactions between the agents and their culturalenvironment.Importantly,therewouldbenosingleagentthatcouldenforce

a definition of creativity by controlling the behaviour of all of the other

agents. In addition, there would be no rules in the agents or in the

environmentthatwoulddefineaglobaldefinitionofcreativity.Thenotions

ofwhomandwhatarecreativeheldbythesocietywouldemergefromthe

multiplenotionsofcreativityheldbytheindividualagents.


7/32

PAPERTITLE 7

3.1. THEIMPORTANCEOFEMERGENCE

The requirements of Artificial Creativity are designed to model the

emergenceofphenomenainsocietiesofagentsconsistentwithcreativityin

human society. Emergence is an important feature of Artificial Creativity

systems,wherephenomenaatacertainlevelarisefrominteractionsatlower

levels.

Inphysicalsystems,temperatureandpressureareexamplesofemergent

phenomena. Temperature and pressure are emergent properties of large

ensemblesof molecules andare dueto interactionsat themolecularlevel.Anindividualmoleculepossessesneithertemperaturenorpressure;theyare

propertiesthatonlyemergewhenmany moleculesarebrought together.In

Artificial Life, the stable patterns in cellular automata, and the flocking

behaviourofsimulatedbirdsareexamplesofemergentphenomena.

InArtificialCreativity,thesocio-culturalevaluationsofwhomandwhat

are creative are emergent phenomena; no individual can dictate the

collectiveevaluationsofwhomandwhatarecreative,theycanonlytryto

influence other individuals by exposing them to their products and their

personal evaluations. The emergence of macro-level creativity from the

interactionsofindividualsatthemicro-levelisillustratedinFigure3.

Figure3:Abehaviour-basedapproachtothestudyemergenceofcreativebehaviouratthe

levelofsocietybymodellingthebehaviourofindividuals(afterLangton,1989).

In Bodens terms we might be tempted to say that H-creativity is

emergent whereas P-creativityis notbecausetheprocessesthatimplementP-creativity test are fixed. However, in the Artificial Creativity system

describedlatertheinteractionbetweenagentsandthecontinuallearningof

the agents through exposure to new artefacts mean that what an agent

considerstobeP-creativeisanemergentpropertyofthewholesystem.An

individualembeddedwithinanArtificialCreativitysystemisaffectedbyits

socio-cultural context such that it will not produce the same P-creative

productsas itwouldin isolation.Hence,bothH-creativityandP-creativity

mustbeconsideredemergentpropertiesofcreativesystems.


8/32


4. AFrameworkforArtificialCreativity

Thissectionpresentsaframeworkfordevelopingcomputationalmodelsof

Artificial Creativity. The framework is presented by adapting Lius dual

generate-and-test model to meet the requirements of Artificial Creativity

listedabove.

4.1. ADAPTINGLIUSMODELTOARTIFICIALCREATIVITY

A critical aspect of Lius model that must be addressed to developcomputational models ofartificial creativityis the definition of thesocio-

cultural creativity test. A literal implementation of Lius model would

produce a separate process that would model the socio-cultural creativity

test.This is a viable solutionformodelling some aspects ofcreativity,as

demonstratedbythecomputationalmodeldevelopedbyGaboratostudythe

memeticspreadof innovationsthrough asimulatedcultureGabora(1997).

Colton (2000) applied a similar socio-culturalcreativity test to assess the

increaseincreativityduetotheco-operationofagentssearchingaspaceof

mathematical possibilities using different search heuristics. However,

implementing a single function, or agent, that model a socio-cultural

creativitytestwouldviolateoneoftherequirementsforArtificialCreativity

outlinedpreviously,i.e.thatnoruleoragentshoulddirectglobalbehaviour.Liudoesnotgointodetailsaboutthedefinitionofthisfunctionbutit

appears that he considers this function to be outside the scope of

computationalmodelsandsomethingthatcanonlybeimplementedbysome

form of interaction with human society. Many computational models

developed reinforce this view by concentrating on the constrained

generationofnovelideasintheircomputationalmodelsandrelyingonusers

toevaluatethecreativeworthofideas.Forexample,seeClancey(1997)for

adiscussionofthesocialsituatednessofHaroldCohensAARON.

To computationally model the behaviour of creative societies, it is

necessary to define a socio-cultural creativity test without violating the

requirementsofArtificialCreativity.Thekeytosolvingthisproblemisto

realisethatthepersonalcreativitytestinsideeachindividualcanbeusedto

developasocio-culturaltestforcreativity.Thesocio-culturalcreativitytest

can be modelled by permitting the communication of artefacts and

evaluations of personal creativity between individuals. An illustration of

twoindividualscommunicatingcreativityevaluationsisillustratedinFigure

4.


9/32

PAPERTITLE 9

IndividualB

IndividualA

Problem

Finding

Creativity

TestGenerate

Creativity

TestGenerateProblem

Finding

no

no

yes

To

Domainyes

yes

Figure4:Thecommunicationofevaluationsbetweenindividualsanditsintegrationintothe

individualgenerate-and-testcycle.

In the interaction illustrated in Figure 4, Agent A communicates an

artefactthatitconsiderstobecreative,i.e.thatpassesitpersonalcreativity

test, to Agent B. Agent B evaluates the artefact according to its own

personal creativity test and sendsits evaluation back to Agent A. In thisway,Agent Bcan affectthe generationof future artefactsbyAgentA by

rewardingAgentAwhenitgeneratesartefactsthatAgentBconsiderstobe

creative. More subtly, Agent A can affect the personal creativity test of

AgentBbyexposingittoartefactsthatAgentAconsiderstobecreative,

becausetheevaluationofcreativityinvolvesanevaluationofnovelty,Agent

AaffectsachangeinAgentBsnotionofcreativitybyreducingthenovelty

of the type of artefacts that it communicates. By exposing Agent B to

artefactsthatAgentAconsiderstobecreative,becausetheyarenoveland

yetunderstandable,itcanaltertheevaluationofcreativitymadebyAgent

B.

Agent-centric evaluations of creativity permit the emergence of socio-

cultural definitions of creativity as the collective function of many

individualevaluations.Withoutagent-centricevaluationsofinterestingness

thecollectionofagentswouldsimplyrepresentparallelsearchesofthesame

designspace.Toimplementthesocio-culturalcreativitytestasacollective

functionofindividualcreativitytestsa communicationpolicyisneeded.A

simplecommunicationpolicywouldbeforagentstocommunicateaproduct

whentheirevaluationofthatproductisgreaterthansomefixedthreshold.

More complex communication policies might incorporate more strategic

knowledgeaboutwhentocommunicateandwhotocommunicatewith.


10/32


To complete the implementation of the field as a collection of

individuals, the individuals must be given the ability to interact with the

domain according to some domain interaction policy. A simple domain

interactionpolicywouldfollowthecommunicationpolicyaboveandallow

agents to addproductsof thegenerative process if thepersonalcreativity

evaluationisgreaterthanadomaininteractionthreshold.Thisapproachis

illustratedinFigure4.However,toensuresomelevelofsocialagreement

before the addition of products to the domain, a slightly more complex

domain interaction policy ensures thatno individual is allowed to submit

theirownworktothedomain.Thus,atleastoneotheragentmustfindanindividualsworkcreativebeforeitisenteredintothedomain.

Makingtheseamendmentsto Lius dualgenerate-and-testresultsin the

modelofsocio-culturalcreativityillustratedinFigure5.

Field Domain

Individual

Individual Individual

Individual

Individual

Individual

Individual Individual

Figure5:TheArtificialCreativitymodelofsocio-culturalcreativity.

5. TheDigitalClockworkMuseProject

In The Clockwork Muse Martindale (1990) presented an extensive

investigation into the role that an individuals search for noveltyplays inliterature,music,visualartsandarchitecture.He concludedthatthesearch

fornoveltyexertsasignificantforceonthedevelopmentofstyles.

Martindale illustrated the influence of the search for novelty by

individuals in a thought experiment where he introduced The Law of

Novelty.TheLawofNoveltyforbidstherepetitionofwordordeedand

punishesoffendersbyostracisingthem.MartindalearguedthatTheLawof

Noveltywasmerelyamagnificationoftherealityincreativefields.

Someof theconsequencesof thesearchfornoveltyare thatindividuals

thatdonotinnovateappropriatelywillbeignoredinthelongrunandthat


11/32

PAPERTITLE 11

the complexity of any one style will increase over time to support the

increasing need for novelty. In this section, we present a computational

model of the Law of Novelty developed using our Artificial Creativity

approachusingcuriousdesignagentsthatsearchfornovelty(Saunders&

Gero,2001b).

Ourmodelconsistsof multiplecuriousdesign agentswithina single

field conducting searches for interesting and potentially creative genetic

artworks.Eachagentisequippedwithanevolutionaryartsystemtoallow

itto generategeneticartworksandcancommunicatewithoneotheragent,

chosenatrandom,oneachtimestep.Individualsthatproduceartworksthatare considered creative by other agents are rewarded with creativity

credit.

5.1. THEINDIVIDUAL:ACURIOUSDESIGNAGENT

This subsection describes the important components of a curious design

agent and the interactive evolutionary system that it interacts with. The

agentsintheDigitalClockworkMuseProjecthavebeendevelopedusinga

model of curiosity that wehaveappliedto several domainscan befound

elsewhere (Gero and Saunders, 2000; Saunders and Gero, 2001a; 2001b;

2001c).Themodelofcuriosityprovidestheessentialabilityforagentsto

evaluate thecreativity of artefacts andtakeappropriate action, i.e.evolvenew artefacts, communicate with otherindividuals in the field,or add an

artefacttothedomain.

5.1.1. InteractiveEvolution

Every agent in The Digital Clockwork Muse uses an interactive

evolutionaryartsystem,similartotheonesdevisedbyDawkins,Sims,Todd

and Latham, and others (Dawkins, 1987; Sims, 1991; Todd and Latham,

1992) to generate genetic artworks. Interactive evolutionary art systems

useastandardevolutionarysystem,e.g.ageneticalgorithm,toevolvesmall

populationsofartworksthatarepresentedtoahumanuserforevaluation.In

our system, agents take the place of human users and interact with the

evolutionaryartsystemstosearchfornovelgeneticartworks.Theflowofinformationbetweenanagentanditsevolutionaryartsystemisillustratedin

Figure6.


12/32


IndividualB

Creativity

TestGenerateProblem

Finding

no yes

img img img

EvolutionarySystem

img img img

img img img

SelectsInteresting

Images

EvolvesPopulation

ofImages

Figure6:Acuriousdesignagentandaninteractiveevolutionaryartsystem.

5.1.2. GeneticArtworksKarlSimsisbestknownforhisworkdevelopingoneofthefirstinteractive

evolutionaryartsystemsforcomplextwo-dimensionalbitmapimages(Sims,

1991). Using a process similar to Genetic Programming, Sims devised an

evolutionary art system that produced artworks by evolving symbolic

functiontrees.

An example genetic artwork of the type evolved by the agents in this

project is shown in Figure 3. This genetic artwork was evolved over the

InternetaspartoftheInternationalInteractiveGeneticArt(IIGA)project

(WitbrockandReilly,1999). Theevolutionarysystemsusedinthisproject

weredevelopedusingsourcecodefromtheIIGAproject.


13/32

PAPERTITLE 13

Figure3:Anexampleofageneticartworkinteractivelyevolvedbyahumanuser.(Fromthe

archiveofevolvedgeneticartworksinInteractiveGeneticArtIII.)

5.1.3. Sensing

A 32x32-pixel image of each genetic artwork is analysed by a curiousdesign agent to determine its novelty. Although this is a low-resolution

imageitisstilllargeenoughtoallowcomplexartworkstobeevolved.To

sense the image, a relatively simple combination of a Laplacian edge-

detectorandafixedintensitythresholdfunctionwereusedtotransforma

geneticartworkintoabinaryimage,asshowninFigure7.

(a) (b)

Figure7:Theimageprocessingappliedtogeneticartworkstoextracttheedgestructureof

theimages,(a)theoriginalimage,and(b)thebinaryimageproducedbytheimageprocessing

tofindthemostprominentedges.


14/32


5.1.4. Novelty

Each agent is equipped with a neural network to learn the categories of

images as it explores the space of possible genetic artworks. A self-

organising map, or SOM, (Kohonen, 1995) is used to categorise each

artworkthatanagentencountersintoacategoryrepresentedbyoneofthe

networksneurons.Ateachpresentationofanartworktheprocessedbinary

image is converted into a vector consisting of 1024 values. As an agent

exploresthespaceofpossibilitiesitlearnsamapoftypicalartworksforthe

region of the genetic art space it currently occupies. By comparing new

artworks against this map, the agent can detect novel, and potentiallyinteresting,artworks.

Themapthattheneuralnetworkproducesprovidesaformofshort-term

memory for the agent to compare new artworks with previously created

ones.Thelargerthenetwork,themoreneuronstheagenthas,andthemore

categoriesofartworksitcanrememberandrecallforcomparison.

Figure 8 shows the neighbourhoods thathave formed for similar input

patterns,e.g.aroundE2andA6,aswellasthemixingofthesepatternsin

theintermediateareas,e.g.aroundD4.Themixingofrepresentationsinthis

wayprovidesanagentwiththeabilitytogeneralisefrompastexperiences

andhencepredictaspectsofunseenartefacts.Thisisanimportantability

forcuriousdesignagentsbecauseitallowsthemtodeterminethenoveltyof

newartefactswithoutsamplingallofthedesignspace.

A B C D E F

1

2

3

4

5

6

Figure8:Theprototypesrepresentedbythe36neuronsofaself-organisingmaphavingjust

categorisedtheinputshowninFigure4batlocationE2.

Novelty(N)iscalculatedasthecategorisationerrorofanagentsSOM

asitattemptstoidentifyasuitablecategoryforanartwork.Noveltyvalues,

i.e.thevaluesofoutputbythebestmatchingneuronoftheneuralnetwork


15/32

PAPERTITLE 15

dependonthesizeof image,inthiscasethesevaluesareintherangeN=0

and N=32, with N=0 being an exact match and N=32 being a complete

mismatch. Effectively this measures the distance of the closest category

prototypetotheinputpattern.

TheEuclideandistancebetweentheclosestcategoryprototypeandanew

inputpatternisarathercrudemeasureofnovelty,andmoresophisticated

measureshavebeendevelopedbyseveralresearchersincludingtheauthors

(Kohonen, 1993; Marsland et al., 2000; Saunders and Gero, 2001c),

however, for the purposes of this demonstration system the measure of

novelty provided by the categorisation error is sufficient andcomputationallyinexpensive.

Novelty is used as the sole criterion to evaluate evolved artworks for

interestingness.Assuchwedefinetheinterestingnessofanartworkbased

on the degree to which it could not have been predicted from previous

experience. Thisis similar to BodensnotionofP-novelty(Boden,1990).

Ourdefinition of interestingnessbasedonnoveltyalonelackstheexplicit

requirement for usefulness needed to model P-creativity as defined by

Boden but,we argue that because interesting artworks are actionable,i.e.

theypromotecuriousaction,theusefulnessofanartworkisitspotentialto

leadtootherinterestingartworksandistherefore,withintheconfinesofthis

simplesystem,relatedtoitsnovelty.

5.1.5. Interestingness

Interest in an artwork is calculated using an approximationto the Wundt

curve, a well-known arousal response curve developed from studies of

animals and humans to exposed to arousal producing stimuli, including

novelty(Berlyne,1971).TheWundtcurveissketchedinFigure9.Berlyne

(1971) refers to the Wundt curve as a hedonic function, to indicate its

relationship to the pleasure/pain response that is often associated with

arousingstimuli.


16/32


HEDO

NIC

VALUE

NOVELTYNx

Hx

Reward

Punish

0

1

-1

n1 n2

Figure9.Thehedonicfunctionusedtocalculateinterest.Thehedonicfunctionisshownasa

solidline,therewardandpunishmentcurvesareshowndashed.

In our model the hedonic function is calculated as the sum of two

sigmoidal functions whereas the Wundt curve is calculatedas the sum of

cumulative-Gaussianfunctions. Themost important feature ofthe hedonic

function used in this research that it shares in common with the Wundt

curveis that it is the sum of two non-linear functions. Ineither case the

functionsaresummedtoproduceaninvertedUshapedcurve,assketched

in Figure 9. The sigmoidal function labelled Reward represents the

intrinsicrewardgiventotheagentforfindinganarousal-inducingstimulus

over a fairly lowthreshold, n1.The second function, labelled Punish,is

the amount of punishment that the agent receives for finding an arousal-

inducingstimulusoverahigherthreshold,n2.Byalteringthethresholdsfor

the reward and punishment sigmoid curves this peak can be positioned

anywherealongthenoveltyaxis.

The agents in the Digital Clockwork Muse use the above hedonic

function to calculate the level of interest that they have in a particular

artworkbaseduponthenoveltydetectedbytheself-organisingmap.Figure9illustratestheuseof thehedoniccurvewithanexamplenoveltyvalueNx

thatismappedtoitscorrespondinghedonicvalueHx.

5.1.6. Curiosity

Throughacombinationoftheneuralnetworkandthehedonicfunctionthe

agentsdisplayaformofcuriousbehaviour.Givenasetofnewartworks

an agent will favour those that are imperfectly represented by the self-

organisingmap,indicatingtheneedforsomelearning,butarenotsonovel


17/32

PAPERTITLE 17

as to fall beyond the peak of the hedonic function. Thus the agent is

motivated to choose artworks it has a good chance of improving its

representation of by favouring similar-yet-different artworks at each time

step (Berlyne, 1971). In other words, the agent shows little interest in

artworks that are either too similar or too different to its previous

experiences(Schmidhuber,1991)

Anagentsinterestinanartworkdeterminestheartworksactionability.

If an artwork is the most interesting at a given moment without being

interestingenoughtobeconsideredcreativethentheartworkisselectedas

thestartingpointforfurthersearchbutnotsenttoanyotheragents.

5.2. THEFIELD:ACOMMUNITYOFINTEREST

Todefine a fieldin anArtificial Creativitysystemwe need todefine the

communication mechanisms and policies used by agents to exchange

artworks and evaluations. For this project we have chosen to use the

simplestimplementationspossible.

5.2.1. Communication

Iftheinterestingnessofanartworkbreachesathresholdvaluethatmarksthe

lowerboundoftherangeofpotentiallycreativeartworksthentheartworkis

senttootheragentsforpeerreview.Artworksareexchangedasmessagesthatencodethesymbolicdescriptions

of the artworks. Receiving agents must then express the genetic

representationtorecovertheartworkandthenevaluateit.Havingexpressed

areceivedartworkanagentevaluatesitaccordingto itspersonalcreativity

testbasedonitsownexperiences.Theexperiencesofareceivingagentare

likely to be different than those of the sender and this can lead to very

different evaluationsof thesame artwork.An artwork thatwasinteresting

foritscreatormaybeboringtoasecondagentbecauseitistoofamiliaror

uninterestingtoathirdbecauseitisnotfamiliarenough.

Anagentmayfindareceivedartworkmoreinterestingthanitsowncurrent

artworks,inwhichcaseitcanusethereceivedartworkasthestartingpoint

foranewsearchofthegeneticartspace.An advantage of passing the genetic representations of artworks between

agents,ratherthantheartworksthemselves,isthatifareceivingagentfinds

an artwork interesting it can usethegenetic representationto evolvenew

artworks without having to reverse engineer an artwork first. This is a

computationally efficient approach to distributing artworksbut it removes

thepossibilityofmemeticevolutionofartworksthroughtheintroductionof

errorsduringthe imitationprocess(Dawkins, 1976).Tosafeguard against

plagiarismandtherebystopapopularartworkbeingcopiedbyallmembers

of a population unaltered, an agent is not allowed to pass on a received


18/32


artwork as its own in the same cycle; it must perform at least one

evolutionarygenerationfirst.

Before using an artwork received from elsewhere an agent must pay the

creatoroftheinterestingartworksomecredit,proportionaltotheinterestthe

receiving agent has in the artwork. The amount of credit accumulated

throughoutalifetimeisusedtoassesshowcreativeaparticularindividual

is.

5.3. THEDOMAIN:AREPOSITORYFORCREATIVEARTWORKS

Adomainismaintainedbythecollectiveactionsofagentsinitsassociated

field. We have implemented the minimal domain interaction policy that

ensuressomeformofsocialagreementwithinafieldbeforeanartworkcan

beaddedtoadomain.Agentscannotaddtheirownartworkstothedomain;

they can only add artworks that they receive from others. Toqualifyfor

addition to the domain an artwork must be of particularly high

interestingnessforthereceivingagent,mostlikelyhigherthanthatrequired

for an artwork to be considered worthy of communicating to another

memberof isfield.If so, the artworkis addedtothe domainwith alabel

indicatingtheagentthatcreatedit.

Futuregenerationsofgeneticartistsbegintheirsearchwithartworksthat

havebeenaddedtothedomain,however,thedynamicnatureofthesocio-cultural evaluation process means that artworks that were considered

creativearelikelytobe nolongerconsideredcreativebecausetheyaretoo

familiarto thefield.Therefore,thedomaindoesnotprovideinstantaccess

tocreativeworks,butratherastoreoffamiliarstartingpointsfromwhich

newcreativeartworkscanbeproduced.Therealadvantageofstartingwith

artworks stored in the domain is that they are already familiar to other

membersofthefield.Theresultofashortsearchfornovelartworksstarting

withexamplesfromthedomainislikelytobenewartworksthataresimilar-

yet-differentwithrespecttothedomain,makingthemidealcandidatesfor

beingcreative.

ResearchersofArtificialCreativitycanalsousetherecordskeptinthe

domain as a means to trace the development of artistic stylesconsideredcreativeovertime.

6. ExperimentsinArtificialCreativity

The following experiments were conducted with the aim of confirming

Martindales predictionsfor Artificial Creative systemsand to investigate

otherinterestingemergentbehaviour.


19/32

PAPERTITLE 19

6.1. THELAWOFNOVELTY

Weinvestigatedthe effectsof thesearchfor novelty, byproducingagents

withdifferenthedonicfunctions.Theaimwastoshowthatagentsarenot

recognised as creative when they fail to innovateinappropriately. Agents

can innovateinappropriately either byproducing boringimagesthat are

too similar to images previously experienced by other agents, or by

producing radical images that are too different for other agents to

appreciate.

We have simulated both types of inappropriate innovation in a singlesimulation.Forthisexperimentwecreatedagroupofagentsmostofwhom,

agents0-9,sharedthesamehedonicfunction,i.e.thesamepreferencefor

average novelty (N=11). Two of the agents have quite different novelty

preferences. One, agent 10, has a preference for low amounts of novelty

(N=3)andtheother,agent11,hasapreferenceforhighamountsofnovelty

(N=19).Agentswithalowernoveltypreferencetendtoinnovateataslower

rate than those with a higher hedonic preference. The results of the

simulationarepresentedinTable1.

TABLE1.Theattributedcreativityforagroupofagentswithdifferentpreferencesfor

novelty.

Agent

ID

Preferred

Novelty

Attributed

Creativity

0 N=11 5.43

1 N=11 4.49

2 N=11 4.50

3 N=11 3.60

4 N=11 4.48

5 N=11 1.82

6 N=11 6.32

7 N=11 8.93

8 N=11 10.72

9 N=11 5.39

10 N=3 0.0

11 N=19 0.0

Theresultsshowtheagentswiththesamepreferencefornoveltytobe

somewhat creative according to their peers, with an average attributed

creativity of 5.57. However, neither agent 10 nor agent 11 received any

credit for their artworks.Consequently noneof the artworksproducedby

theseagentsweresavedinthedomainforfuturegenerations.Whenthese

agentsexpirednothingremainedinthesystemoftheirefforts.


20/32


The results show that while an agent must innovate to be considered

creative, it must do so at a pace that matches other agents to achieve

recognition. The agent with a preference for high levels of novelty and

hence rapid innovation was justas unsuccessful in gainingrecognition as

theagentwithalownoveltythresholdthatinnovatedtooslowly.

6.2. THEEMERGENCEOFCLIQUES

Wehavealsoinvestigatedthebehaviourofgroupsof agentswithdifferent

hedonicfunctions.Todothiswecreatedagroupof10agents,halfofthemhadahedonicfunctionthatfavourednoveltyN=6andtheotherfiveagents

favoured novelty values close to N=15. Figure 9 shows the payments of

creativity credit between the agents in recognition of interesting artworks

sentbytheagents.

2 8 2

21 3 4 5 6 7 8 9

1

2 1 1 3

4 5 2 5

2 23 3

1 16 3 5

3 4 5 1

3 5 1 4

4 3 2 4

1 4 4 4

0

0

1

2

3

4

56

7

8

9

Sender

Rec

eiver

Figure9:Amatrixshowingthetotalnumberofmessagescarryingcreditforbeingcreative

betweentheagentsofthesimulation.

Two areas of frequent communication can be seen in the matrix of

payment messages shown inFigure 9. The agents with the same hedonic

functionfrequentlysendcreditforinterestingartworksamongstthemselves

butrarelysendthemtoagentswithadifferenthedonicfunction.Therearea

largenumberofcreditmessagesbetweenagents0-4andagents5-9,butonly

onepaymentbetweenthetwogroupsagent4creditsagent5forasingle

interestingartwork.

The result of putting collections of agents with different hedonic

functionsinthesamegroupappearstobetheformationofcliques:groups

ofagentsthatcommunicatecreditfrequentlyamongstthemselvesbutrarely

acknowledgethecreativityofagentsoutsidetheclique.Asaconsequenceof

the lack of communication between the groups the style of artworks

producedbythetwocliquesalsoremainsdistinct.


21/32

PAPERTITLE 21

Communicationbetweencliquesisrarebutitisanimportantaspectof

creativesocialbehaviour.Communicationbetweencliquesoccurswhentwo

individuals in the different cliques explore design subspaces that are

perceptuallysimilar.Eachoftheindividualsisthenabletoappreciatethe

others work because they have constructed appropriate perceptual

categories. The transfer of artworksfroma sourceto a destinationclique

will introducenew variablesinto thecreativeprocessesof thedestination

clique,thetwocliquescanthenexploreindifferentdirections,justastwo

individuals do when they share artworks. Cliques can therefore act as

super-artists,exploringadesignspaceasacollectiveandcommunicatinginterestingartworksbetweencliques.

Figure10is ascreenshotoftherunningsimulationthathasformedtwo

cliques. To help visualise the emergent cliques, the distances between

agentsare shortenedforagentsthatcommunicatefrequently.Thedifferent

stylesofthetwogroupscanalsobeseen,withagents0-4producingsmooth

radialimageswithlowafractaldimension(~1.4)andagents5-9producing

fracturedimageswith clearlydefinededgesandahigherfractaldimension

(~1.7).

Figure10:Ascreenshotofthesimulationclearlyshowingthetwocliques.Thesquares

representagents.Theimagesshowthecurrentlyselectedgeneticartworkforeachagent.The

numberaboveeachsquareshowstheagentsattributedcreativity.Thedarklinesbetween

agentsindicatethecommunicationofcredit.


22/32


7. FutureResearch

ArtificialCreativitysimulationspermittheinvestigationofmanyinteresting

aspectsofcreativityincomputationalmodels.Thissubsectionwillexplore

someoftheseaspectsandhowcomputationalmodelscanbedevelopedto

gainabetterunderstandingofthem.

7.1. COMPUTATIONALLYSTUDYINGHISTORICALCREATIVITY

One of the most interesting possibilities supported by the currentlyimplemented system is the possibility to study H-creativity in artificial

societies.Runningsimilarsimulationsto thosedocumentedaboveforlong

periodsoftimeprovidesanopportunitytostudytherevisionofcreativity

over many generations. As noted by Boden (1990) the attribution of H-

creativityto individuals andtheir productsis oftenrevisedover time. We

canexpecttofindsimilarrevisionsinArtificialCreativitysimulations:we

would expect to observesimilarre-evaluations,e.g.the recognitionof the

creativityofindividualsbylatergenerationsthatwasnotrecognisedbytheir

peers.

7.2. THESITUATEDNESSOFARTIFICIALCREATIVITY

Thestudyofsituatednesshasbecomeapopulartopicofresearchinrecent

years in design computing (Gero& Reffat, 1997; Gero, 1998) anddesign

cognition(Suwa&Tversky,1997;Suwa,GeroandPurcell,1999).Artificial

Creativityprovidesnewopportunitiestostudysituatednesscomputationally.

Importantly, the computational modelling of the individual, field and

domainina singlesystempermitsthestudyofboththeinteractivenotions

of situatedness, and the cultural notions of situatedness. Interactive

situatedness has been studied computationally before, e.g. in models of

reflection-in-action(GeroandSaunders,2000).

The cultural situatedness of individuals is crucial to design as a

profession concerned with the socio-cultural change, yet it has been

neglectedincomputationalstudiesbecauseofthe difficultiesof situatinga

computational model in real world cultures. Some examples of programsthathaveinteractedwithcreativefieldsintherealworldareprogramsthat

havebeeninvolvedintheinventionofnewproducts,mostnotablediscovery

systemslikeEURISKO(Boden,1990).However,allofthesesystemshave

requiredhumaninterventionwhentomediateinteractionswiththerelevant

fields and domains. Artificial Creativity allows us to study socio-cultural

situatedness as an emergent property of communicating agents that are

individuallysituatedintheirhistoryofexperiences.


23/32

PAPERTITLE 23

7.3. CREATIVEINDIVIDUALS

ArtificialCreativityprovidesuswithopportunitytostudythecharacteristics

ofcreativeindividuals.Severalimportantcharacteristicsof eachindividual

can be identified and adjusted by altering parameters. We have already

investigatedthebehaviourofagentswithdifferenthedonicfunctionsandwe

could extend these studies by adjusting different parameters or using

functions other than the Wundt curve. Other parameters that we could

experiment with include those that control the learning and the sensors,

effectors and interaction policies ofagents.For example,by changingtheparametersthatcontrolthesizeoftheSOMscurrentlyusedintheagentswe

can effectively experiment with individuals with longer and shorter

memories.

7.3.1. TheLifecycleofCreativeIndividuals

Thereissomedebateoverwhethercreativeindividualsdotheirbestwork

whentheyareyoungorwhethertheycontinuetobejustascreativeinlater

life (Simonton). The study of Artificial Creativity provides us with

opportunity to study the lifecycles of creative individuals and the

mechanismsbywhichthecreativepotentialofindividuals,withrespectto

society, increases or decreases as the agent develops. Adjusting the

parameters discussed above we can experiment with agents that havedifferent abilities and observe how their creative potential changes over

theirlifespan.Forexample,byusinganeuralnetworkthatlearnsquickest

whenyoungwecanstudyhowthecreativityofanagentisaffectedasthe

abilitytolearnistradedoffagainsttheknowledgegained.

7.3.2. TheEvolutionofCreativeIndividuals

18th Century theories of creativity suggested that it was an inherited

characteristic based on the observation that creative ability often ran in

families,implyingthatthegeneticmakeupofanindividualdeterminedtheir

creative ability. These theories have been superseded by more balanced

explanations of the roles of nature and nurture in the development of

creativeindividuals,butthisdoesnotmeanthatwecannotinvestigatetheevolutionof creativeindividualsthroughtheacquisitionofgenetictraitsin

Artificial Creativity simulations. To do so, we simply need to define a

genotypethat determines the values for the characteristics of individuals,

e.g.sizeandtypeofneuralnetwork,andselectindividualsforreproduction

based on their creative status. Interesting observations would include

whether or notthe genetic make-up ofcreativeindividualsconvergeupon

certainparametersoverseveralruns.Ifconvergencesofthissortcouldbe

found then we should be able to identify something like creative

personalitytraitsinindividuals.


24/32


7.3.3. TheDiversificationofCreativeIndividuals

ArtificialCreativityisaboutmodellingcreativesocietiesandwhilewehave

restrictedourselvestomodellingproducersthereareseveralotherimportant

players in these societies that do not directly produce artefacts but are

involvedinthedistributionandevaluationofthem.

Tomodelmorecomplexsocialrelationsweshouldimplementmodelsof

otherindividuals,e.g.consumers,distributors,critics,etc.Eachwouldhave

their own role to play in simulations of creative societies and would

influence each others behaviour; consumers would evaluate interesting

works and pay credits to producers of interesting works, distributorscommunicatetheworksofproducerswidely,andcriticscommunicatetheir

evaluationsofworkswidely.TheserelationshipsareillustratedinFigure11.

Usingdifferenttypesofindividuals,wecanbegintomodelsomemore

of the complexity discussed by Csikszentmihalyi (1999). We could also

model more complexrelationshipsbetweenagents,such as those between

designerandclient,byallowingcommunicationofrequirements.Asavery

simpleexample,aclientcouldcommunicateanexampleworkandinstructa

producer tomake something like itandexpect the new workto still be

interesting, i.e. similar-yet-different to the example. Importantly, the

personalcreativitytestsplayacriticalroleindeterminingthebehaviourof

allofthedifferenttypesofindividualsandthesocio-culturalcreativitytest

isstillandemergentpropertyofthewholesociety.

ProducerandConsumer

Producer,CriticandConsumers

Producer,DistributorandConsumers

Producer

Consumer

Distributor

Critic

Design

Evaluation

Figure11:Threedifferenttypesofindividualswithdifferentabilitiesandtheirrolesinthecommunicationofdesignsandevaluationsincreativedesignsocieties.

7.4. DOMAINSOFCREATIVITY

Domainsplayacrucialroleincreativesocietiesbydefiningofthetypesof

interactions that can take place between individuals within a field and

betweendifferentfields.Therearemanypossiblescenariosforstudyingthe

relationship between domainsandthe creativityof individuals and fields.


25/32

PAPERTITLE 25

The following discussion provides only a very brief outline of the

possibilities.

7.4.1. MultipleDomains

People that move from one field to another, or contribute to multiple

domains,are often considered some of the most creative participants.We

can investigate the mechanisms involved in the creativity of these

individuals by modelling multiple fields and their associated domains.

Fields are defined by the communications of the individuals within them,

creating separate fields means creating groups of individuals that arespecialised in the type of information that they communicate. Effectively

thiscreatesabarriertoentryfornon-membersofafield.Individualsthat

transferfromonedomaintoanothermustfindsomewaytoovercomethis

barrier. One possible isfor agent toapproach the design space ofa field

whilewithinanotherfield,in asimilarwaythathasalreadybeenobserved

incliques.Theagentmaythenbeallowedentrancetoafieldbecausethe

agent producesartefacts thatare appreciatedby themembers of thefield.

ThisprocessisillustratedinFigure12(a).

(a) (b)

Figure12:Bridgingfieldsanddomains.In(a)anagenttraversesadesignspacegetsclose

enoughtoanotherthedesignspaceofanotherfieldthatitcansuccessfullycommunicatewith

itsmembersandberewarded.In(b)anagentstraddlestwofieldsandreceivescreditfrom

membersofbothasittransfersartefactsbetweendomains.Solidarrowsrepresentthetransfer

ofartefacts,dashedlinesrepresentthetransferofevaluationsorcredit.

7.4.2. StaticDomains

Wecandefinestaticfieldsbydefiningthedesignspacetobeexploredbya

field as a fixed property of its domain. Communication between fields

exploring different domains will be limited to products that lie in the

intersection ofdomaindesign spaces.Bysettingup suchenvironmentswe

can study the spread of ideas through related fields as a consequence of

interests shared by individualsdifferent domains. Agents that are able to

locatethemselvesattheintersectionofmultipledomainshavethepotential

to benefit by addressing a larger audience and by transferring creative

artefacts from one domain to another. An individual situated at the

intersectionoftwodomainsisillustratedinFigure12(b).


26/32


7.4.3. DynamicDomains

We can alsodefine dynamicfields byallowingindividuals to modify the

designspaceofthedomain,orbybasingthedesignspaceofadomainon

the products of another field. Allowing individuals to modify the design

spaceof a domain wouldallow a designspacetoshiftinthe directionof

interesting works communicated from other fields. We would be able to

study a sort of social curiosity as the field moved collectively in the

directionofnewandinterestingpossibilities.

D3

D1 D2

Figure13:Themovementofdesignspacesasaconsequenceof'collectivecuriosity'.

7.4.4. HierarchiesofDomains

Basingthedomainofonefieldonthedomainofanother,wouldallowthe

investigation of the effects of technological innovation in one domainopeningupnewcreativepossibilitiesinanother.Theexpansionofdesign

spaces in this way is illustrated in Figure 14 wherethe expansion of the

design space for domain D2 opens up new possibilities and expands the

designspaceofdomainD1.

D2

D1 D1

D2

Figure14:Theexpansionofadesignspaceasaconsequenceofchangesinasubordinate

domainsdesignspace.

7.4.5. EmergentDomains

Oneway to study theemergence of newdomains would betoimplement

permit the subdivision of a domain when subgroups of its field are


27/32

PAPERTITLE 27

sufficiently different that they no longer share a common understanding.

This process of domain subdivision is reminiscent of speciation in the

naturalworld.Speciationisaprocesswherebyagroupoforganismsbecome

differentiatedastheyadapttotheirenvironment.Thedefinitionofaspecies

isagroupoforganismsthathavebecomesowelladaptedtoeachotherthat

theycanonlyreproducewithothermembersofthesamegroup.

Similarly, a field can be defined as a group of individuals that have

becomesospecialisedthattheyarelimitedtocommunicatingcreativeideas

toothermembersofthefield.Thedivisionofafieldintotwonewfieldsis

illustratedinFigure16.Thenewfieldsemergeasthecommunicationlinksbetween two subgroups weaken until they constitute two separate fields.

Thealreadyobservedformationofcliquescanbeseenasaprecursorofthe

emergenceofnewfieldsanddomains.

(a) (b) (c) (d)

Figure15:Thedivisionofafieldintotwonewfields.

Speciation is an important topic of research in Artificial Life and by

adapting some of the mechanisms studied there to model the splitting of

domainswemaybeabletostudytheemergenceofnewdesignschoolsand

possiblydesigndisciplinesinartificialsocieties.

7.5. CREATIVESOCIETIES

Primarily,ArtificialCreativityisconcernedwiththestudyof theemergent

behaviour of creative societies. The followingstudies are concernedwith

the possibilities for experimenting with social conventions, e.g.

communicationprotocols,toinvestigatetheiraffectsoncreativity.

7.5.1. CommunicationModels

ArtificialCreativitysimulationsprovidethenecessarymechanismsforbothdirect and indirect communication. Direct communication is supported by

thetransmission ofproductsandevaluationsas messages betweenagents.

Indirect communication is supported by the storage and retrieval of

symbolic representationsin a domain.Usingthis framework wecan study

the affects that different communication mechanisms have on the

developmentofcreativesocieties.

Oneinterestingpossibilitywouldbetosimulatetheintroductionofnew

communication technologies and see their affect on the creativity of


28/32


individualsandsocieties.Forexample,wecanaskthequestion:Whatare

affects of the introduction of global communication technologies like the

telephone,e-mailandtheWorldWideWeboncreativefieldslikedesign?

Eachtechnologyopensupnewpossibilitiesforthecommunicationofideas

orthecommunicationofdifferenttypesofartefacts.Wewouldanswerthis

questionbysimulatingsometypicalcommunicationanddomaininteraction

policiesofindividualsbeforeandaftertheintroductionofglobaldirectand

indirect communication and observe the behavioural changes of the

simulatedsociety.

7.5.2. TheEconomicsofCreativity

Creativefieldsareconstructedthroughthesocialinteractionsofindividuals

determined by their communication policies. The communication policy

implementedin allof thecreativeagentsin thecurrentsystemimplements

communication of creditfor producing interesting artworks but the credit

hasnoworthinsidethesimulation,itisimportantonlytoobserverswhocan

monitorthe creativityof individualsbywatching theircreditratings.This

doesntreflectthestatusofbeingconsideredcreativeintherealworld.For

instance, being considered creative imbues a status that can be used to

attract financial resources. Also, being creative costs, in both time and

money,andthisforcesanindividualtoconsiderthebenefitsofsearchingfor

radicallycreativesolutionsagainstthecostincurred.

7.5.3. ThePoliticsofCreativity

Givingdifferentcommunicationabilitiestodifferenttypesofagentsmakes

theselectionofwhichagentstocommunicatedwithanimportantfactorin

self-promotion of creative individuals. As Csikszentmihalyi (1999) points

out,convincingotherpeoplethatyouvehadacreativeideaisoftenharder

thanhavingtheideainthefirstplace.Insocietieswithinunequalstatusof

individuals,thequestionofwhichindividualstocommunicatewithbecomes

importantforindividualsseekingtherecognitionfrompeers.

Agentsinfluencethe behaviour ofotheragentsbycommunicatingtheir

artworks and their evaluations of those artworks. The current simulation

modelsaperfectegalitariansocietywherealloftheagentsarecreatedequalandnoagenthasmoreinfluencethananyoftheothers.Wearenotlimited

to modelling such utopian societies inArtificial Creativity andwith some

smalladjustmentswecanmodelsocietieswheresomearemoreequalthan

others.

By weighting evaluations according to the creativity of the individual

sendingit,alreadycreativeagentscanpromotethecreativestatureofother

agentsmore quicklythanuncreative agents.Recognitionbyan established

creative agent would have a significant impact on the status of young

agentstryingtogainrecognition.


29/32

PAPERTITLE 29

8. Conclusions

ThecomputationalworkpresentedinthispaperhasillustratedtheArtificial

Creativityapproachtodevelopingmodelsofcreativesocieties.Byadapting

Liusdualgenerate-and-testmodelofcreativitywehaveproducedamodel

of creative societies that can be used to study socio-cultural creative

behaviour as an emergent propertyarising fromthe creative behaviour of

individuals. The implemented system models the evolution of notions of

creativitywithinanartificialsocietyovertimeasindividualscomeandgo,

thefieldchangesincomposition,andthedomainisaltered.The emergence of social behaviour, e.g. The Law of Novelty, and

dynamicsocialstructures,e.g.cliques;suggestthattheArtificialCreativity

approach to developing models of creative societies may contribute new

insightsintothenatureofcreativedesigninsocio-culturalsituations.Figure

14 illustrates the different levels at which creativity may be studied as a

pyramid ofemergentproperties.Eachlevelrepresentsa differentaspectof

creativity that is emergent from the ones below it. Layers that represent

processes are coloured darker than those that represent products, process

andproductlayersalternateupthepyramid;propertiesofcreativeproducts

are a consequence of the processes that created them and the behaviour

higher-levelprocessesareaffectedbytheproductsthattheyoperateupon.

Eachlevelinfluencesthelayersbelowthembyprovidingasituationfortheproducts andprocesses. Forexample, emergent social structures influence

the communication of products and the products communicated influence

thebehaviourofindividualsthatsendandreceivethem.


30/32


Individual

Processes

Individual

Products

Individual

Behaviours

Communicated

Products

Social

Behaviour

CulturalProducts

Emergence Influence

Figure16:Apyramidofcreativity.

Thefoundationsofthecreativepyramidaretheprocessesinternaltothecreative agent that allows it to generate-and-test ideas. The result of

executing these processes is the creative products. Traditionally,

computational research has concentrated on thesetwo levels byencoding

processesthoughtto be important increativityina pieceof software and

getting experts to examine the results of running those processes to

determine whether the processesare creative. Intraditionalcomputational

models,thehigherlevelsofthepyramidarenotmodelledinthesoftware

andareprovidedbypeople.

ArtificialCreativitysuggestsa differentapproach;insteadofevaluating

the products of a piece of software to determine its creativity, it focuses

uponthebehavioursofagentsandartificialsocieties.ArtificialCreativityis

concerned with modelling the creative behaviours of individuals, e.g.curiosity,andstudyingtheemergentsocialbehaviourswhenindividualsare

puttogether.BecauseindividualsinanArtificialCreativitysimulationmust

beabletoevaluatethecreativityofcommunicatedproductsandhenceother

individuals,thedetailsoftheproductsofindividualsbecomelessimportant.

More important in the study of Artificial Creativity are the socio-cultural

structuresthatemergeasaconsequenceof thecommunicationofproducts

andevaluations.

The Artificial Creativity approach permits the computational study of

highestlevelsofcreativityillustratedinFigure14withouthavingtodevelop


31/32

PAPERTITLE 31

agents that can integrate, and achieve creative status, in human society.

ArtificialCreativity simulations permit theexperimentationwith creativity

in artificial societies thatwouldbeimpossible in therealworld,allowing

thestudyofcreativity-as-it-isinthecontextofcreativity-as-it-could-be.

Acknowledgements

Tocome

References

Berlyne,D.E.:1971,AestheticsandPsychobiology,Appleton-Century-Crofts,NewYork.

Boden.M.A.:1990,TheCreativeMind:MythsandMechanisms.Cardinal,London.

Clancey, W. J.: 1997, Situated Cognition: On Human Knowledge and ComputerRepresentations,CambridgeUniversityPress,Cambridge,England.

Colton,S.,Bundy,A.andWalsh,T.:2000,AgentBasedCooperativeTheoryFormationinPure Mathematics, in G. Wiggins (ed.), Proceedings of AISB 2000 Symposium onCreativeandCulturalAspectsandApplicationsofAIandCognitiveScience ,pp.1118,Birmingham,UK.

Csikszentmihalyi,M.:1988,Society,culture,andperson:asystemsviewofcreativity,inR.J.Sternberg(ed.),TheNatureofCreativity,CambridgeUniversityPress,Cambridge,UK,

pp.325339.Csikszentmihalyi, M.: 1999, Implications of a Systems Perspective for the Study of

Creativity,inR.J.Sternberg(ed.),HandbookofCreativity,CambridgeUniversityPress,Cambridge,UK,pp.313335.

Dawkins,R.:1976,TheSelfishGene,OxfordUniversityPress,Oxford.

Dawkins,R.:1987,TheBlindWatchmaker,NortonandCo.,London.

Gabora,L.: 1997,The origin andevolutionof cultureand creativity.Journal ofMemetics:

EvolutionaryModelsofInformationTransmission1(1):128.

Gardner,H.:1993,CreatingMinds,BasicBooks,NewYork.

Gero, J. S.: 1998, Conceptual designingas a sequence of situatedacts,in I. Smith (ed.),ArtificialIntelligenceinStructuralEngineering,Springer,Berlin,pp.165177.

Gero,J.S.andReffat,R.:1997,Multiplerepresentationsforsituatedagent-basedlearning,inB.VarmaandX.Yao(eds.),ICCIMA'97,GriffithsUniversity,Australiapp.8185.

Hofstadter, D.: 1979,Godel, Escher,Bach: AnEternalGolden Braid, BasicBooks, New

York.Hofstadter, D.and theFluid AnalogiesResearchGroup(1995,FluidConcepts&CreativeAnalogies:ComputerModelsof theFundamentalMechanismsofThought,BasicBooks,NewYork.

Koestler,A.:1964,TheActofCreation,Macmillan,NewYork.

Kohonen,T.:1993,Self-OrganizationandAssociativeMemory,3rded.,Springer,Berlin.

Kohonen,T.:1995,Self-OrganizingMaps,SpringerVerlag,Berlin.

Langley,P., Simon,H.A.,Bradshaw,G.L.,and Zytkow,J.M.:1987,ScientificDiscovery:ComputationalExplorationsoftheCreativeProcesses,MITPress,Cambridge,MA.

Langton,C.G.:1989,Artificiallife,ArtificialLife,AddisonWesley,pp.147.


32/32


Liu,Y.T.:2000,Creativityornovelty?DesignStudies21(3):261276.

Marsland,S.,Nehmzow,U.andShapiro,J.:2000,Arealtimenoveltydetectorforamobilerobot, inProceedings of EUREL European Advanced Robotics Systems Conference,Salford,England.

Martindale,C.:1990,TheClockworkMuse,BasicBooks,NewYork.

Newell, A., Shaw, J. C.and Simon,H. A.: 1962, The process ofcreative thinking,inH.Gruber, G. Terrell and M. Wertheimer (eds.),ContemporaryApproaches to CreativeThinking,AthertonPress,NewYork.

Partridge,D.Rowe,J.:1994,ComputersandCreativity,IntellectBooks,Oxford.

Saunders,R.andGero,J.S.: 2001a,DesigningforInterestandNovelty:MotivatingDesign

Agents,inProceedingsofCAADFutures2001,Eindhoven.Saunders,R. andGero,J. S.:2001b, ACuriousDesignAgent: AComputationalModelof

Novelty-Seeking Behaviour in Design,in Proceedings of the Sixth Conference onComputer Aided Architectural Design Research in Asia (CAADRIA 2001), Th eUniversityofSydney,Australia.

Saunders,R.andGero,J.S.:2001c,TheDigitalClockworkMuse:AComputationalModelofAestheticEvolution,inG.Wiggins(ed.),ProceedingsoftheAISB'01SymposiumonAIandCreativityinArtsandScience,SSAISB,York,UK.

Schmidhuber, J.: 1991, A possibility for implementing curiosity and boredom in model-buildingneuralcontrollers, inJ.A.MeyerandS.W.Wilson(eds.),ProceedingsoftheInternational Conference on Simulation of Adaptive Behaviour: From Animals to

Animats,MITPress/BradfordBooks,pp.222227.

Simon,H.A.:1981,TheSciencesoftheArtificial,MITPress,Cambridge,MA.

Sims, K.:1991,Artificialevolutionfor computergraphics.ComputerGraphics25(4):319-328.

Suwa, M.,and Tversky, B.:1997, Whatdo architectsand studentsperceivein their design

sketches?Aprotocolanalysis,DesignStudies,18,385403.

Taylor,C. W.:1988, Various approachesto the definitionof creativity,inR. J.Sternberg(ed.),TheNatureofCreativity,CambridgeUniversityPress,Cambridge,UK,pp.99124.

Todd,S.andLatham,W.:1992,EvolutionaryArtandComputers.AcademicPress,London.

Witbrock, M. and Reilly, S. N.: 1999, Evolving genetic art, in P. J. Bentley, (ed.),EvolutionaryDesignbyComputers,MorganKaufman,SanFrancisco,CA,pp.173.

Date post:	05-Apr-2018
Category:	Documents
Upload:	walesa-danto
View:	224 times
Download:	0 times

Artificial Creativity

Documents