Understanding control at the knowledge level

《Understanding control at the knowledge level》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、UnderstandingControlattheKnowledgeLevelB.ChandrasekaranLaboratoryforAIResearchTheOhioStateUniversityColumbus,OH43210Email:chandra@cis.ohio-state.eduAbstractfoodisinthekitchen.Thatthecatwouldnatu-Whatisitthatunifiesthecontroltaskrallygotothekitchenundertheseco

2、nditionsinallitsmanifestations,fromthether-seemsreasonabletothehostandpresumablytomostattotheoperatorofanucleartheguest.Whenpeopletalkthisway,theyarepowerplant?Atthesametime,howdonotassertingthattheneuralstuffofthecatisweexplainthevarietyofthesolutionssomekin

3、dofalogicalinferencemachinework-thatweseeforthetask?IproposeaingonPredicateCalculusexpressions.ItisKnowledgeLevelanalysisofthetasksimplyausefulwayofsettingupamodelofwhichleadstoatask-structureforcon-agentsandusingthemodelforexplainingtheirtrol.Differencesinav

4、ailabilityofbehavior.Theattributionsofthegoalandknowledge,thedegreeofcompilationknowledgecanbechangedonthebasisoffur-intheknowledgetomapfromobserva-therempiricalevidence,buttherevisedmodeltionstoactions,andpropertiesrequiredwouldstillbeinthesamelanguageofgoal

5、sandofthesolutionstogetherdeterminetheknowledge.differencesinthesolutionarchitec-TheKnowledgeLevelstillneedsarepresenta-tures.Iendbydiscussinganumberoftion,butthisisarepresentationthatisnotpos-heuristicsthatariseoutoftheKnowl-itedintheagent,butoneinwhichoutsi

6、derstalkedgeLevelanalysisthatcanhelpintheabouttheagent.Newellthoughtthatlogicwasdesignofsystemstocontrolthephysi-anappropriaterepresentationforthispurpose,calworld.leavingopenthepossibilityofotherlanguagesalsobeingappropriateinsomecircumstances.WhatistheKnowl

7、edgeLevel?NewellusedthephraseSymbolLevel,torefertotherepresentationallanguagesactuallyusedBynowmostofusinAIknowaboutthetoimplementartificialdecisionagents(orex-KnowledgeLevelproposalofNewell[Newell,plaintheimplementationofnaturalagents).81].Itisawayofexplaini

8、ngandpredictingtheLogic-basedlanguages,LispandFORTRAN,behaviorofadecision-makingagentwithoutneuralnetdescriptions,andevenBrookssub-committingoneselftoadescriptionofthesumptionarchitecture