Upgrade to remove ads
Upgrade to remove ads
Unformatted text preview:

~`ORGLibraryNftwestemUnnrersityChapterQualitativePhysics:Past,Present,andFutureKennethD.ForbusQualitativeReasoningGroupDepartmentofComputerScienceUniversityofIllinoisatUrbana,Champaign1IntroductionQualitativephysicsisconcernedwithrepresentingandreasoningaboutthephysicalworld.Thegoal ofqualitativephysicsistocaptureboththecommon-senseknowledgeofthepersononthestreetandthetacitknowledgeunderlyingthequantitativeknowledgeusedbyengineersandscientists.Theareaisnowalittleovertenyearsold,which,atleastmeasuredinthespanofAI,isalongtime.Soitmakessensetostepbackandtrytosystematizetheworkinthefieldanddescribethecurrentstateoftheart.I'llstartbydescribingwhatqualitativephysicsis,whyoneshouldbedoingit,andwhereitcamefrom.ThenI'llsketchthecurrentstateof theart,atleastthepart thatisnowfairlystable.ThenI'lldescribewhatIthinkliesaroundthecomer,includingsomepointerstorecentworkandsomeinterac-tionsbetweenqualitativephysicsandotherfields.Finally,I'lldescribesomeopenproblems, eachofwhichwillprobablyrequirequiteafewinspiredPh.D.thesestocrack.Qualitativephysicsisgrowingrapidly,andthusanysurveyislikelytobe-comequicklydated.Forexample,severalproblemswhichweredescribedasvirginterritorywhenthismaterialwaspresentedatAAAI-86havenowbeenat240::.Forbus-leastpartiallyexplored.Nevertheless,Ithinkthegeneralframeworkforunder-standingtheareathatwaspresentedthenremainssound,andsoIhavere-mainedfaithfultothatorganization.2WhyQualitativePhysics?Consider whatweneedtoknowaboutthephysicalworldtomakecoffee.Weknowthattopourcoffeefromthepotintoacuprequireshavingthecup underthespoutofthekettle,andthatifwepourtoomuchin,therewillbe amessonthefloor.Weknowallthiswithoutknowingthemyriadequationsandnumeri-calparametersrequiredbytraditionalphysicstomodelthissituation.Supposeweweregoingtobuilda householdrobotthat,amongotherdu-ties,madecoffee.Wemightstartbyusingtraditionalphysicstomodelthesit-uation.Immediatelyseveralproblemsarise.Therearefewformalaxiomatictheoriesofphysics.Theformalaspectsofphysics,theequations,donotbythemselvesdescribewhentheyareapplicable.What,forexample,istheequa-tion forthecup? Thereisn'tone,perse,butrathervarious aspectsofthecuppotentiallyparticipateinseveraldifferentequationsdescribing"whathappens"intheworld.Manyeverydayphysicalphenomena,suchasboiling,are noteasilydescribedbyasingleequation.Andevenwhenequationsexist,peoplewhoknownothingaboutthem canoftenreasonfluentlyaboutthephenomena.Soequationscannot benecessaryforperformingsuchreasoning.Butsupposeforamomentthatwehadsuchasetofequations.Couldweusethem?Realisticequationsrarelypermitclosed-form,analyticsolutions.Evenwhentheydo, thehighcomputationalcomplexityofsymbolicalgebraicmeansit'snotthe sortofcomputationyouwantgoingoninsidearoboten-gagedinreal-timeactivity.Analternaterouteisnumericalsimulation.Byplugginginnumericalvalues,wecouldgenerateaveryprecisedescriptionofwhatwillhappen.Butsuchsimulationsrequireimmensecomputationalre-sources.Worsevet,itassumestheexistenceofa completesetofaccuratevaluesforallinputparameters.Typicallywejustdon'thavesuchaccuratein-formation,thusforcingustosearchaspaceofparameterscorrespondingtotherangesthevariousinputparametersmaytake.Thisincreasestheamountofcomputationevenmore,makingnumericalsimulationinfeasible.Evenifnumericalsimulationweretechnologicallyfeasible,bysayshirt-pocketsupercomputers,orbyallowingroughapproximations,itstillwouldbeinsufficient forourrobot.First,westillneedto interpret theoutput ofthesimulation.Alistofnumericalstateparametersisnotthemostperspicuousrepresentationofanevent.Second,anyrunof anumericalsimulatorprovidesaspecificsetofpredictionsaboutwhatthesystem beingsimulatedwilldo.Thiswill suffice forsometasks,but notforall.Oftenwewanttocharacterizethepossibilitiesthatmightoccur,withsomeguaranteeofcompleteness.Forin-Chapter7QualitativePhysics241stance,afault-treeanalysisofapowerplantthatcapturedonlyasmallfractionofthefailuremodesofthesystemwouldbeinappropriate.Withnumericalsimulationsitisoftenhardtotellwhenonehas capturedallofthepossiblebe-haviors.iInmanysituationsoneneedsarapidandroughestimateofwhatispossible,ratherthanaveryprecisepredictionbased onmanyunsupportedas-sumptions.Arobotpouringcoffeeshouldbecognizantofthepossibilityofoverflow,andnotspenditstimecalculatingjusthowbig theresultingpuddlemightbe.These problemsare notspecifictomakingcoffee;theyholdmoregener-allywheneveronetriestoreasonaboutthephysicalworld.Tosummarize,theseproblemsare:Themoaehngproblem:Howdoesonemapfromreal-worldobjectsto theabstractionsofone'sphysics?Theresolcalonproblem:Carryingoutnumericalsimulationsrequiresmoredetailthanisoftenavailable.Reasoningtechniquesthatcanexploitlowresolution,partialinformationarerequiredforcommonsensereasoning.3.Thenarrownessproblem:Traditionalsimulationprovidespreciseanswersgivenaparticular setofassumptions.Manyreasoningproblemsrequireknowingalternativepossibilities,ratherthanasingleprojection.Atfirsttheseproblemsmayseemsurprising.Physics,oneofthecrowningsuccessesofthescientificmethod,hasbeencarried.onforhundredsofyears.Butconsider:Physicistsalreadyhavecommonsensetheoriesoftheworld.Theirzoalistocreatemodelscapableofmorepreciseexplanations.Withfewexceptions,thefocusof formalizationlieswithbuildingnewmodelsthathavesignificantlybetterpredictiveandexplanatorypowerthanourimplicitcom-monsensemodels.Qualitativephysicsarisesfromtheneedtoshareourintui-tionsaboutthephysicalworldwithourmachines.Therearemanypotentialapplicationsofqualitativephysics.Asarguedelsewhere[GentnerandStevens,1983:deKleerandBrown,1984;deKleer,1984j,thetacitknowledgeofengineersandscientistsrestsonthissharedframework.Ifwearetobuildprogramsthatcapturethisexpertise,wemustun-derstandthefoundationqualitativephysicsprovides.Wewillreturntothispointafterbrieflysummarizingtheessenceofqualitativephysics.It issaidthatiftheangularincrementinthesimulation oftheaerodynamicpropeniesoftheBostonJohnHancockbuildinghadbeenhalved,thefactthatthebuilding'swindowswouldtendtopopoutinhighwinds couldhavebeenpredicted.Instead,itwasdiscoveredempirically.242Forbus2.1TheEssenceThekeytoqualitativephysicsistofindwaystorepresentcontinuousproper-tiesoftheworld bydiscretesystemsofsymbols.Onecan alwaysquantizesomethingcontinuous,but


View Full Document
Download Qualitative Physics - Past, Present, and Future
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Qualitative Physics - Past, Present, and Future and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Qualitative Physics - Past, Present, and Future 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?