我与小娜（08）：人工智能的伟大胜利
我与小娜（08）：人工智能的伟大胜利
小娜知道，1月27日，英国《自然》刊登重要文章，题为“MasteringthegameofGowithdeepneuralnetworksandtreesearch”，标志着世界人工智能（AI）历史性的伟大胜利。为什么？小娜知道，ThegameofGo（机器人）haslongbeenviewedasthemostchallengingofclassicgamesforartificialintelligenceowingtoitsenormoussearchspaceandthedifficultyofevaluatingboardpositionsandmoves.HereweintroduceanewapproachtocomputerGothatuses‘valuenetworks’toevaluateboardpositionsand‘policynetworks’toselectmoves.Thesedeepneuralnetworksaretrainedbyanovelcombinationofsupervisedlearningfromhumanexpertgames,andreinforcementlearningfromgamesofself-play.Withoutanylookaheadsearch,theneuralnetworksplayGoatthelevelofstate-of-the-artMonteCarlotreesearchprogramsthatsimulatethousandsofrandomgamesofself-play.WealsointroduceanewsearchalgorithmthatcombinesMonteCarlosimulationwithvalueandpolicynetworks.Usingthissearchalgorithm,ourprogramAlphaGoachieveda99.8%winningrateagainstotherGoprograms,anddefeatedthehumanEuropeanGochampionby5gamesto0.Thisisthefirsttimethatacomputerprogramhasdefeatedahumanprofessionalplayerinthefull-sizedgameofGo,afeatpreviouslythoughttobeatleastadecadeaway.
这是《自然》杂志对此文内容的摘要。具体说明，且听下回分解。
袁萌 2月10日