`CharlesIsbell
`DeborahL.McGuinness
`MITAILab
`LoriAlperinResnick
`TechSquare
`AT&TBellLaboratories
`CambrigeMA
` MountainAvenue
`isbell@ai.mit.edu
`MurrayHill,NJ
`fdlm,resnickg@research.att.com
`Descriptionlogic-basedcon(cid:12)gurationapplications
`Theknowledgebasewascreatedbyworkingwithanex-
`havebeenusedwithinAT&Tsince toprocessover
`pertinthedomain.Thedatabaseofinstanceinforma-
`twoandahalfbilliondollarsworthoforders.Whilethis
`tionwasalsohand{compiledforthissmallapplication;
`familyofapplications[]haswidelyacknowledgedimpor-
`however,inotherapplicationsthatworkwithchanging
`tance,itisdi(cid:14)culttouseforpedagogicalpurposessince
`instanceinformation,automatictranslationroutinespe-
`thetypicalproductcon(cid:12)guredisahighlyinterconnected,
`riodicallyaccessdatabasesandthenupdatetheknowl-
`edgebase[ ].Theterminologicalknowledgebasecon-
`complicatedtechnicalpieceofequipmentliketheDACS
`IV{ . Wehavedevelopedasmaller{scalecon(cid:12)gura-
`tainsde(cid:12)nitionalinformationconcerningclassesaswell
`tionapplicationthathasanalogousreasoningprocesses
`asrules.Weworkeddirectlywithanexperttoob-
`butamoreapproachabledomain|thatofbuildinghome
`taintheserules,butinourlargerapplicationsofthis
`sort[],systembuildersbeginwithpreexistingrulespec-
`theatersystems.Thisapplicationprovidesaplatformfor
`explaininghowDescriptionLogic-basedSystems(dlss)
`i(cid:12)cationsandusearuletranslatortogenerateclas-
`work{inourcasetheclassicknowledgerepresentation
`sicrules.Rulesinthisapplicationfallintotwoclasses:
`system[ ]{andhowtheycansupportindustrialapplica-
`bothhardandfastelectricalrules(forexample,are-
`tionslikecon(cid:12)guration.
`ceivermusthaveanA/Bswitchinordertosupportsec-
`Classicisanobject{centeredrepresentationand
`ondarymainspeakers),and\rulesofthumb"(forexam-
`ple,hometheatersystemsdonothavemorethanoneTV
`reasoningtoolwithaformalfoundationindescription
`ortwoVCRs).Allproductscon(cid:12)guredbytheknowledge
`logic.Classicandmanydlssareparticularlywell
`basemustabidebythehardandfastelectricalrules,and
`suitedforapplicationsinareaslikecon(cid:12)gurationthat
`productscon(cid:12)guredfollowingour\guidance"alsofollow
`must .encoderichclassandobjectdescriptions;
`therulesofthumb.Ifclassicsupporteddefaults,the
`rulesassociatedwithguidedstereosystemswouldhave
`.provideactiveinference(suchasautomaticclassi-
`beendefaults.
`(cid:12)cationofclassesandobjectsintoageneralization
`Activeinference:Thehometheaterapplication
`hierarchy,rule(cid:12)ringandmaintenance,inheritance,
`usesclassictoprovideactiveinferenceaftertheinter-
`propagation,etc.);
`facehasguidedtheuserthroughafewsimplequestions.
` .explainthereasoningprocess;
`Weassumethatpeoplewanttobuildaudioonly,home
`.handleanincompleteandincrementallyevolving
`theateronly,orcombinationaudio/videosystemsand
`knowledgebase;and
`thattheyusuallyhaveapricerangeinmind.Thus,we
`askwhichtypeofsystemtheywant,andwhatquality
`.handleerrorsinawaythatkeepstheknowledgebase
`theyarewillingtopayfor.Withthesetwoinputs,the
`consistent,butalsoprovidesusefulinformationto
`applicationusesclassictoaskfollowupquestionsas
`theuser.
`appropriateandtoproduceacomplete(abstract)de-
`Wewillprovidesomeexamplesinourdomainthatillus-
`scriptionofaconsistentproduct.
`trateeachoftheseareas.
`Forexample,imaginethattheuserchoosesacombi-
`Classandobjectdescriptions:Asinanyapplica-
`nationsystem,butdoesnotspecifyaparticularlevelof
`tion,weneedadomainontologyinwhichtowork.Our
`quality.classicdeducesthatthetargetsystemmust
`hometheaterapplicationcontainsaknowledgebasein-
`havemainspeakers,aVCR,andaTV.Sinceallstereo
`cludingaconcepttaxonomyandinstancedescriptions.
`systemsshouldhavesomekindofpreampli(cid:12)erandam-
`pli(cid:12)er,andthechoiceofanunspeci(cid:12)edcombinationsys-
` DACSIV{ isadigitalcross{connectsystemthat
`temdoesnotimplyanythingaboutthem,theuseris
`processesdigitizedsignalsforsomeUSstandardtransmission
`askedtodecidewhethershewantsseperateampli(cid:12)er
`rates.Classicisfreelyavailableforacademicpurposes,and
`componentsorareceiver(whichincludesanampli(cid:12)er,
`preampli(cid:12)erandatuner).Becausethereisnopricingin-
`commerciallyavailableforotherpurposes.Ithasbeendis-
`formation,classicisunabletoinfermorecomponents.
`tributedtooveruniversitiesandisinuseinmanyinternal
`projectswithinAT&T.
`Theinformationthatclassicisgivenandhasdeduced
`
`Page 1 of 6
`
`FORD 1014
`
`
`
`Figure :TheclassicHomeEntertainmentCon(cid:12)gurationSystem.Thewindowhasfourmajorareas.The
`(cid:12)rst(upperleft)isacomponentspane,withiconsrepresentingallthekindsofcomponentsthatclassicknowsabout.
`Clickingononeoftheseiconsaddsacomponentofthattypetothestereosystemunderconstruction.Thesecond
`area(upperright)isthelivingroompane,whereallthecomponentsthathavebeenaddedtothestereosystem
`(eitherdirectlybytheuserorbyclassic'sinferences)arepresentedgraphically.Therearealsoiconsrepresenting
`thestereosystemitselfaswellasimportantconceptsofwhichitisaninstance(e.g.HighQualitySystem).
`Clickinganiconallowstheusertoadd,removeordisplayinformationabouttheobjectthaticonrepresents.An
`iconwithasmallsquareinitsupperright-handcornerisrequired.Thethirdarea(lowerleft)isthedisplaypane.
`Informationaboutcomponents,pricinginformationandexplanationsaredisplayedhere.The(cid:12)nalarea(lowerright)
`istheerrorpane.Iconsrepresentingobjectsthathavecausederrorsaredisplayedhere.Clickingononeofthese
`iconsallowstheusertodisplayorexplaininformationaboutthatobject.
`
`Page 2 of 6
`
`FORD 1014
`
`
`
`Figure:Selectedinformationaboutthestereo
`systembeingcreated.Thusfar,allwereallyknow
`aboutthesystemisthatitisacombinationaudioand
`hometheatersystemandthatitisfollowingourexpert's
`rulesofthumb.
`ispresentedgraphically(seeFigure ).
`classiccalculatesthedeductiveclosureoftheinfor-
`mationprovided.Theusercanviewthecompletedin-
`formationonanycomponentjustbyclickingonitsicon.
`Theusercanalsoviewpropertiesofthewholesystem
`byclickingonaspecialstereosystemicon.Becauseshe
`haschosenanunspeci(cid:12)edstereosystem,ifsheclickson
`it,shewillnotseemuchthatshedoesnotalreadyknow:
`thisisaguidedcombinationsystem(seeFigure) .
`Thischanges,however,oncesheaddssomeinforma-
`tion.Forexample,decidingthatsheiswillingtospend
`atleast$ onherdreamsystemnotonlymakesthe
`salespersonhappy,butallowsclassictodeducethat
`shewantsahigh{qualitycombinationsystem.classic
`theninfersthathersystemmusthaveinadditionto
`itsmainspeakers,apairofsurroundspeakers,acenter
`speakerandasubwoofer.Asisoftenthecase,providing
`informationaboutthesystemasawholeimpliesprop-
`ertiesofindividualcomponents.Clickingontheiconfor
`thepreampli(cid:12)er,forexample,revealsthat,amongother
`things,itmusthavealistpriceofatleast$ (see
`Figure ).
`Explanation:classiccanjustifyallofitsbeliefs[].
`Notonlycantheuserviewanypieceofinformation,she
`canalsohaveanydeductionexplained.
`Inourexam-
`ple,iftheuseraskshowthepreampli(cid:12)eracquiredits
`pricerestriction,shelearnsthatarule(cid:12)redthatsays
`thathigh{qualitysystemsmusthavehigh{qualitycom-
`ponents,whichforpreampli(cid:12)ersenforcesaminimum
`priceof$ (seeFigure).Theexplanationfacility
`canalsoanswerotherquestionssuchaswhyoneobject
`doesordoesnot\subsume"(isorisnotmoregeneral
`than)anotherobject,whyarule(cid:12)redonanobject,or
`whyanerroroccurred.
`Inferencescanalsohavetem-
` classicactuallyinfersfarmoreabouttheobject;how-
`ever,classicisabletoprune\uninteresting"displayinfor-
`mation.Forexample,inthisapplicationcomponentshave
`informationassociatedwiththemthatdescribehowthey
`shouldbedisplayed.Displayingtheseimplementationdetails
`wouldonlybeconfusing.
`
`Figure :Informationpropagation.Afteraddinga
`pricerestrictiontothestereosystem,classicdeduces
`thatitisahigh{qualitysystem.Sinceitisalsoguidedby
`ourexpert'srulesofthumb,thispropagatespricingin-
`formationtoallofthecomponents,includingthepream-
`pli(cid:12)er.
`platesassociatedwiththemthatmaybeusedtopresent
`explanationsintermsthatareacceptabletotheuser.
`Incompleteandevolvingknowledgebases:As
`wasthecasewithspecifyingaminimumpriceforthe
`stereosystem,classicallowscontinualre(cid:12)nementof
`(andchangesto)itsknowledgebase.Inadditiontospec-
`ifyingpriceinformation,ausermightalso\instantiate"
`acomponentdescriptionbychoosingaparticularmake
`andmodel.Sincethesystemmustbeconsistent,thein-
`terfacewillonlygeneratechoicesthatappeartosatisfy
`allconstraintsthatclassichasderivedaboutthecom-
`ponent(byusingthespeci(cid:12)cationofthecomponentas
`aquerytothedatabaseofindividuals).
`Itisworthstressingthataddinginformationtothe
`systemortoanyofitscomponentscanhavemanyim-
`plications.Forexample,choosingaparticulartunerof-
`tenallowsclassictoinstantiatethepreampli(cid:12)er(see
`Figure).
`Errors:Althoughclassicandtheapplicationmin-
`imizetheplaceswhereausercanmakeanerror,errors
`canstilloccur.Theapplicationdoesnotaskclassicto
`precalculateallpossibleconsequencesofagivenchoice.
`Theusercouldmakeachoicewhichwouldcausearule
`to(cid:12)rethatwouldthencauseapropagationofsomein-
`consistentinformation.Inourexample,theuseralready
`hasaTVand,asitturnsout,systemsunderourguid-
`ancemayonlyhaveatmostoneTV.So,attemptingto
`addanotheronegeneratesanerror.classicdoesnot
`allowtheknowledgebasetobeinaninconsistentstate,
`soitwillrollbacktheknowledgebasetotheprevious
`consistentstate,meanwhilesavingcopiesofalltheindi-
`vidualsthatledtotheerror,intheirinconsistentstates.
`Iftheuserasksclassicforanexplanationoftheerror,
`classiccanaccesstheinconsistentstateinformationto
`generateanexplanation(seeFigure).
`Additionalfunctionality:
`Inadditiontoinstanti-
`
`Page 3 of 6
`
`FORD 1014
`
`
`
`Figure:Explainingnewinformation.classicis
`capableofexplainingallofitsinferences.Above,the
`preampli(cid:12)er'spricerestrictionistheresultofaparticular
`ruleabouthigh-qualitystereosystems.
`atingcomponentsandaddingnewcomponents,theuser
`mayalsodeletearequirementonthesystem.
`Inthis
`case,anydeductionsthatweremadeasaresultofthis
`requirementareremovedfromthespeci(cid:12)cation.Ifthe
`userisnotfamiliarwithdi(cid:11)erenttypesofstereoequip-
`ment,shemaywishtotrustourexpert,andbuilda
`systemstartingwithoneoftheexamplesystems,where
`allthecomponentsareknowntoworkwelltogether.She
`canthenre(cid:12)nethissystemaccordingtoherneeds,re-
`questingalternativemakesandmodelstotheonescho-
`sen,andaddingandremovingcomponents.Whenthe
`userhas(cid:12)nishedre(cid:12)ningthesystemtohersatisfaction,
`shecanasktheapplicationtocompletethespeci(cid:12)ca-
`tionforher.Theapplicationwillthenchooseconsistent
`makesandmodelsforallthecomponentsshehasleft
`unspeci(cid:12)ed.Shecanthenviewapartslist,afterwhich
`shemightwanttoshiptheordero(cid:11)tothefactory(see
`Figure).
`Discussion:Wefeelthatdescriptionlogic-based
`technologyisparticularlywellmatchedtothisstyleof
`con(cid:12)gurationproblemforseveralreasons.First,theap-
`plicationisfairlylogical(notheuristic)sowewouldei-
`therhavetoimplementthelogicinaprogramminglan-
`guageorstartwithatoollikeclassicthatincorporates
`aformallogic.Second,thisdomainisnaturallyhierar-
`chicalandruleinformationisappropriateatmanydif-
`ferentlevelsofthetaxonomy.dlsssupporthierarchical
`rulesinsteadofusingamoretraditional,(cid:13)atrule-based
`approach.Thismaysimplifyknowledgeengineeringand
`maintenance[].classicrulescanbesimplerbecause
`theyonlyneedtocontaincontentappropriatetoacer-
`tainlevelofconceptinthehierarchy,andtheydonot
`needtocontainanycontrolinformation.Finally,the
`applicationnaturallyincorporatesmanydi(cid:11)erenttypes
`ofinference;afewofwhichinclude:inheritance,prop-
`agation,boundsconstraints,andrules.Thesecanbe
`encodeddirectlyindlssinsteadofneedingtobepara-
`phrasedintorules.Possiblymoreimportantly,explana-
`tionsofthereasoningprocessmaybeintermsofthe
`naturallyoccurringinferences.
`Thishometheatersystemisasimpleexampleofa
`
`Figure:Moreinformationpropagation.Choosing
`aparticulartunercausesclassictodeduceamatching
`preampli(cid:12)er.Noticethatinstantiatedcomponentsare
`distinguishedfromnon-instantiatedonesinthatinstan-
`tiatedcomponentshave\real"picturesassociatedwith
`themwhileuninstanitatedcomponentsonlyhaveicons.
`Figure:Errors.AddinganotherTV,whilehaving
`acertainappeal,isnotallowedbythesystembecause
`guidedstereosystemscanhaveatmostoneTV.The
`o(cid:11)endingobjectisplacedintheerrorpanewhereitis
`availabletobeinspectedandexplained.
`
`Page 4 of 6
`
`FORD 1014
`
`
`
`familyofapplicationswhereadescriptionlogic-based
`platformisusedtoimplementstandardcon(cid:12)guration
`tasksandprovidethebasisforadditionalfunctionalities.
`Thedeployedapplicationsbuiltonthisdesignhavepro-
`videdmanyadvantagesincludingdecreasedorderpro-
`cessingintervals(facilitatinghypotheticalcon(cid:12)guration
`evaluations,whichwerepreviouslyinfeasible),reduc-
`tionsinpersonnelrequiredtomaintainproductinforma-
`tion,accurateandup{to{datepricingforsalesquotes,
`eliminationofduplicationindatabases,andidenti(cid:12)ca-
`tionofincompatibleknowledge.
`Acknowledgements
`Wewishtothanktheentireclassicgroup,particu-
`larlyPeterPatel-SchneiderandRonBrachman,fortheir
`insightfulcommentsonthiswork.Wealsowishto
`thankthePROSE/QUESTARteam,particularlyJon
`Wright,HarryMoore,JayBerman,CharlieFoster,and
`PatSaleh,forcontinuingfeedbackonapplicationsneeds.
`References
`[ ]R.J.Brachman,D.L.McGuinness,P.F.Patel-
`Schneider,L.A.Resnick,andA.Borgida.Livingwith
`classic:WhenandHowtoUseakl-one-LikeLan-
`guage.
`InPrinciplesofSemanticNetworks:Explo-
`rationsintherepresentationofknowledge,J.Sowa,ed-
`itor,Morgan-Kaufmann,pp. {, .
`[]D.L.McGuinnessandA.Borgida.ExplainingSub-
`sumptioninDescriptionLogics.InProc.IJCAI,Mon-
`treal,August .
`[ ]R.J.Brachman,P.G.Selfridge,L.G.Terveen,B.Alt-
`man,A.Borgida,F.Halper,T.Kirk,A.Lazar,D.L.
`McGuinness,andL.A.Resnick.
`IntegratedSupport
`forDataArchaelogy.inInternationalJournalofIntel-
`ligentandCooperativeInformationSystems,(), ,
`pp. { .
`[]J.R.Wright,D.L.McGuinness,C.Foster,and
`G.T.Vesonder.ConceptualModelingusingKnowledge
`Representation:Con(cid:12)guratorApplications.
`InProc.
`Arti(cid:12)cialIntelligenceinDistributedInformationNet-
`works,IJCAI- ,Montreal, .
`[]Wright,J.R.,Weixelbaum,E.S.,Brown,K.,Vesonder,
`G.T.,Palmer,S.R.,Berman,J.I.,Moore,H.H.,A
`knowledge-basedcon(cid:12)guratorthatsupportssales,en-
`gineering,andmanufacturingatAT&TNetworkSys-
`tems.InProceedingsoftheInnovativeApplicationsof
`Arti(cid:12)cialIntelligenceConference,pp. { , .
`
`Page 5 of 6
`
`FORD 1014
`
`
`
`Figure:Acompletedsystem.classicguaranteesthatthesystemaboveiswhollyconsistentandwillwork
`together.
`
`Page 6 of 6
`
`FORD 1014