Artificial emergence: visions and nightmares The visions of Drexler and others of molecular nanotechnology will be realised sometime. The vision is appealing — trillions of tiny autonomous machines creating useful products or cleaning up after us, using t

EngineeringEmergence

FionaPolack

DepartmentofComputerScience

UniversityofYorkYork,YO105DD,UK

ona@cs.york.ac.uk

April7,2005

Arti cialemergence:visionsandnightmares

ThevisionsofDrexlerandothersofmolecularnanotechnologywillberealisedsometime.Thevisionisappealing—trillionsoftinyautonomousmachinescreatingusefulproductsorcleaningupafterus,usingthecollectivee ectsofsomeminimalbuilt-infunctionality.However,thenightmarescenariosareeverpresent:apollutionbusterthatgoesrogueandbuststheplanetmayseemalittleextreme,butahostofmedicalnanitesthataccidentallyblockanarteryasaside-e ectofclearingitisnotsofarfetched.Conventionally,wemightseeengineeringasthewaytocontrolriskintechnologicaladvances;wearguethat,inordertoexploittheemergentpropertiesof(arti cial)complexsystemssafely,weneedtobeabletoengineerthesesystems.

Engineeringandassurance

Engineeringisaquality-enhancingactivity.Thegoalistoproducerobustsystems,providingthenecessaryassuranceoffunctionalityandsafety.Sidee ectsincludebeingabletode nethesafeoperatingconditionsforsystems,andbuildinginsaferesponsesifthesystemmovesoutsideitsintendedoperationalenvelope.Thereisconsiderableexpertise,atYorkandelsewhere,inconstructingandcertifyingconventionalcriticalsystems.Atypicalapproachtakesassumptionsandevidence(factsandstatistics)andusesthesetoconstructanargumentthatagivensystemwilloperatesafelywithinstatedenvironmentalparameters.Assuranceisconventionallyconstructedusingevidence.Forexample,con denceinasystem’sreliabilitymightbeenhancedbyusingmaterialswitharecordedhistoryofperformance,orbyusingtechniques(andpeople)thathaveproducedreliablesystemsbefore.

Weareseekingtoestablishengineeringprinciplesforemergentsystems,suchthatwecouldconstructassurancearguments.ThisworkisconductedinpartundertheauspicesoftheTUNAproject1.

Akeyproblemisdecomposition—aconventionalincremental-componentdevelopmentdoesnotpreserveemergence,or,moreaccurately,makesnoguaranteeofemergenceintheendsystem.Weneedtobeabletoarguethevalidityofeachpartofthesystem,aloneandincombination.Wecannotrelyonananiteassemblywherethenanitesdonotsurvivelongenoughtodoanything,orwherethenanitesintheenvironmentinwhichassemblyisrequireddonothavetheresourcestoperformtheassembly.EmergenceandArchitecture

Aninitialobservationisthatemergencedependscruciallyonrepresentation.Totakeantrivialexample,thecellsofacellularautomaton(CA)are nitestateautomata;theupdateruledeterminesthenextstateofeachcellfromthecurrentstatesofthecellandsomecollectionofothercells,referredtoasitsneighbours.Suchasystemcanbeprogrammedveryeasily,andisutterlyuninteresting;thevalueofthestate uctuates,andatsomepointmaystop uctuating.TheCAdisplaysemergence(prettypatterns)onlywhenthecellsandtheirneighbourlinessaretranslatedontosomerepresentation.Changetherepresentation,andtheemergentstructureschangeorvanish[3].

Wecannotde netheautomatonandtherepresentationusingasinglesetofconcepts.Automataarestate-and-operationsystems;arepresentationisavisualisationbuiltupfromsomediscretisedmodel

TheoryUnderpinningNaniteAseemblers,EPSRCgrantEP/C516966/1,isa

two-yearfeasibilitystudyattheUniversitiesofYork,KentandSurrey.TheparticipantsareS.Stepney,S.A.Schneider,P.H.Welch,J.C.P.Woodcock,A.L.C.Cavalvanti,H.TreharneandF.A.C.Polack1http://www.cs.york.ac.uk/nature/tuna.