人工智能原理-北京大学-2--PartIBasicsChapter2IntelligentAg-(2

SchoolofElectronicandComputerEngineeringPekingUniversityWangWenminArtificialIntelligenceArtificialIntelligence2ArtificialIntelligencePart1.BasicsPart2.SearchingPart3.ReasoningPart4.PlanningPart5.LearningContents:ArtificialIntelligence3Part1.Basics1.Introduction2.IntelligentAgentsContents:ArtificialIntelligence4Objectives教学目的2.IntelligentAgentsOverviewseveralapproachesforAI.纵览AI的各种研究途径。Discussthenatureofintelligentagents,thediversityofenvironments,andtheresultingmenagerieofagenttypes.讨论智能体的性质、环境的多样性、以及由此产生的各种类型的智能体。ArtificialIntelligence52.IntelligentAgentsContents:2.1.ApproachesforArtificialIntelligence2.2.RationalAgents2.3.TaskEnvironments2.4.IntelligentAgentStructure2.5.CategoryofIntelligentAgentsArtificialIntelligence62.1.ApproachesforArtificialIntelligenceContents:2.1.1.CyberneticsandBrainSimulation2.1.2.Symbolicvs.Sub-symbolic2.1.3.Logic-basedvs.Anti-logic2.1.4.Symbolismvs.Connectionism2.1.5.StatisticalApproach2.1.6.IntelligentAgentParadigmArtificialIntelligence::Basics::Agent7In1940sand1950s,anumberofresearchersexploredtheconnectionbetweenneurology,informationtheory,andcybernetics.1940年代至1950年代，许多研究者探索神经学、信息论和控制论之间的关系。Someofthembuiltmachinesthatusedelectronicnetworkstoexhibitrudimentaryintelligence.他们当中有些人采用电子网络打造机器来展现初步的智能。ManyoftheseresearchersgatheredformeetingsoftheTeleologicalSocietyatPrincetonUniversityandtheRatioClubinEngland.许多研究者聚集在普林斯都大学和英国Ratio俱乐部，召开了目的论学会的会议。By1960,thisapproachwaslargelyabandoned,althoughelementsofitwouldberevivedinthe1980s.到了1960年，这种途径基本上被抛弃了，尽管有些要素于1980年代复活。Overview概述2.1.1.CyberneticsandBrainSimulationArtificialIntelligence::Basics::Agent8HerbertSimonandAllenNewellstudiedhumanproblem-solvingskillsandattemptedtoformalizethem.赫伯特·西蒙和艾伦·纽厄尔研究了人类问题求解技能，并且试图对其形式化。Theirworklaidthefoundationsofartificialintelligence,aswellascognitivescience,operationsresearchandmanagementscience.他们的工作奠定了人工智能、以及认知科学、运筹学和管理学的基础。Theirresearchteamusedtheresultsofpsychologicalexperimentstodevelopprogramsthatsimulatedthetechniquesthatpeopleusedtosolveproblems.他们的团队采用了心理学实验结果开发程序，仿真人们解决问题的技巧。Soar,acognitivearchitecture,wasoriginallycreatedatCMUinthemiddle1980s,nowmaintainedatUniversityofMichigan.Soar，一种认知架构，是以CMU为核心于1980年代中期开发，如今由密歇根大学维护。Overview概述2.1.1.CyberneticsandBrainSimulationArtificialIntelligence92.1.ApproachesforArtificialIntelligenceContents:2.1.1.CyberneticsandBrainSimulation2.1.2.Symbolicvs.Sub-symbolic2.1.3.Logic-basedvs.Anti-logic2.1.4.Symbolismvs.Connectionism2.1.5.StatisticalApproach2.1.6.IntelligentAgentParadigmArtificialIntelligence::Basics::Agent10SymbolicAIisbasedonhigh-level“symbolic”(human-readable)representationsofproblems,logicandsearch.符号AI是基于人类易懂的高级“符号”来表现问题、逻辑和搜索。Theapproachisbasedontheassumptionthatmanyaspectsofintelligencecanbeachievedbythemanipulationofsymbols.该方式是基于这样一种假设：智能的许多方面能够通过符号操作来获得。SymbolicAI符号AI2.1.2.Symbolicvs.Sub-symbolicThemostsuccessfulformofsymbolicAIisexpertsystems,itprocessestherulestomakedeductionsandtodeterminewhatadditionalinformationitneeds,i.e.whatquestionstoask,usinghuman-readablesymbols.符号AI最成功的形式是专家系统，它对规则进行操作来进行推断和确定需要什么附加信息，即采用人类易懂的符号询问一些问题。SymbolicArtificialIntelligence::Basics::Agent11Bythe1980s,manyresearchersbelievedthatsymbolicsystemswouldneverbeabletoimitatealltheprocessesofhumancognition,especiallyperception,robotics,learningandpatternrecognition.到了1980年代，许多研究者已确信，符号系统将永远无法模仿人类认知的全部过程，尤其是感知、机器人技术、学习和模式识别。Sub-symbolicAI亚符号AI2.1.2.Symbolicvs.Sub-symbolicsub-symbolicAnumberofresearchersbegantolookinto“sub-symbolic”approaches,basedonneuralnetworks,statistics,numericaloptimization,etc.一些研究者开始关注“亚符号”方式，以神经网络、统计学、数值优化等为基础。ArtificialIntelligence::Basics::Agent12Examples:Symbolicvs.Sub-symbolicAI符号与亚符号AI2.1.2.Symbolicvs.Sub-symbolicSymbolicApple符号化苹果ApplestructurekindoriginbodyfruitstemappletreeshapesizecolortasteredgreenhandroundappleApple0.850.630.240.350.365.304.10Sub-symbolicApple亚符号化苹果ArtificialIntelligence132.1.ApproachesforArtificialIntelligenceContents:2.1.1.CyberneticsandBrainSimulation2.1.2.Symbolicvs.Sub-symbolic2.1.3.Logic-basedvs.Anti-logic2.1.4.Symbolismvs.Connectionism2.1.5.StatisticalApproach2.1.6.IntelligentAgentParadigmArtificialIntelligence::Basics::Agent14UnlikeNewellandSimon,JohnMcCarthyfeltthatmachinesdidnotneedtosimulatehumanthought,butshouldinsteadtrytofindtheessenceofabstractreasoningandproblemsolving,regardlessofwhetherpeopleusedthesamealgorithms.与纽厄尔和西蒙不同，麦卡锡觉得机器无需仿真人类的思考，反倒应该试图去发现抽象推理和问题求解的本质，不管人们是否使用同样的算法。HisAIlaboratoryatStanford(SAIL)focusedonusingformallogictosolveawidevarietyofproblems,includingknowledgerepresentation,planningandlearning.他的斯坦福大学AI实验室（SAIL）专注于用使用形式逻辑来解决各种问题，包括知识表征、规划和学习。LogicwasalsothefocusoftheworkinEurope,whichledtothedevelopmentoftheprogramminglanguagePrologandthescienceoflogicprogramming.逻辑也被欧洲的研究工作所专注，导致编程语言Prolog和逻辑编程科学的发展。Logic-based基于逻辑2.1.3.Logic-basedvs.Anti-logicArtificialIntelligence::Basics::Agent15ResearchersatMIT(suchasMarvinMinsky)foundthatsolvingdifficultproblemsinvisionandnaturallanguageprocessingrequiredad-hocsolutions.MIT的研究者（比如明斯