Showing content from http://www-formal.stanford.edu/jmc/freewill.dvi below:
÷À;è TeX output 2002.06.28:0908ÿÿÿÿ x ý¤ = þÿðJç:óDÓít½qG®cmr17ºFREEBÚWILL|EVENFÿuNORROBOTS#ô¦ÈøóÂÖN G® cmbx12¿JohnzáMcCarthÿuÂy¤óX«Qff cmr12»Computer³/ScienceDepartmen¼t¡¥¾Stanford³/Univ¼ersity¡ÅStanford,³/CA94305,USA¡£éóßµTff cmtt12Àjmc@cs.stanford.edu¡hSAhttp://www-formal.stanford.edu/jmc/uýl¤0»2002³/Jun28,9:08a.m.,OwIîÐóX«Q cmr12¹Iê¨can,butIw¬ron't.û¥2ó|{Ycmr8Á1ÉÑèÎóò"V ó3 cmbx10ÊAbstract}ITÌ»óKñ`y ó3 cmr10ÉHuman=yfreewillisaproMÞductofev²!olutionandcontributestothe¤ D_ìsuccessÁoftheh²!umananimal.-°UsefulrobMÞotswillalsorequirefreewill¡D_ìof¦fasimilarkind,andw²!ewillhavetodesignitintothem.¡TÌ»Fÿereeúwillisnotanall-or-nothingthing.Someagen²!tshavemore¡D_ìfree`éwill,nÏorfreewillofdieren²!tkinds,thanothers,andw²!ewilltryto¡D_ìanalyze_thisphenomenon.ÆDOurob»jectiv²!esareprimarilytechnological,¡D_ìi.e.¬Ãto@³studywhataspMÞectsoffreewillcanmak²!erobotsmoreuseful,¡D_ìand1åw²!ewillnottrytoconsolethosewho nddeterminismdistressing.¡D_ìWÿeec`distinguishbMÞet²!weenc`havingchoicesandbMÞeingconsciousofthese¡D_ìc²!hoices; ìbMÞoth>Àareimportan²!t,¤Õeven>Àforrobots,¤Õandconsciousnessof¡D_ìc²!hoicesÙWrequiresmorestructureintheagentthanjusthavingchoices¡D_ìandisimpMÞortan²!tforrobots.°ÂConsciousnessoffreewillisthereforenot¡D_ìjust¦fanepiphenomenonofstructureservingotherpurpMÞoses.¡TÌ»Fÿeree¯ÂwilldoMÞesnotrequireav²!erycomplexsystem.ùðYoungc²!hildren¡D_ìandNerathersimplecomputersystemscanrepresen²!tinternallyó ý': ó3 cmti10Ë`I
Qcÿp¹an,¡D_ìbutêêIwonþás't'wÉand¦fbMÞeha²!veaccordinglyÿe.¡TÌ»Naturally¡I¡DhopMÞethisdetailedËdesignÑñstancÿp¹et%É(Dennett¦f1978@TÒ)will¡D_ìhelp*ôunderstandh²!umanfreewill.´·IttakestheËcÿp¹ompatibilist4ÞÉphilosoph-¡D_ìical¦fpMÞosition.ç¹1* x ý¤ = ýçTÌ»ÉThere*ma²!ybMÞesomereadersinterestedinwhatthepapMÞersaysabMÞout¤ D_ìh²!umanÈfreewillandwhoareputobylogicalformulas.®STheformulas¡D_ìareÃnotimpMÞortan²!tfortheargumentsabMÞouthumanfreewill;Ñ{theyare¡D_ìpresen²!tforpMÞeoplecontemplatingAIÕsystemsusingmathematicallogic.¡D_ìTheyj canskiptheform²!ulas,vbutthecoherenceofwhatremainsisnot¡D_ìabsolutely¦fguaran²!teed.(V'¿1D(ôInÿuÂtro=duction|twozáaspectsoffreewillb#'¹FÿVreedwill,bSothinh¬rumansandincomputerprogramshastwoaspSects|the¤'ó!»@ cmti12Ìexternal);¹aspSectê¨andtheÌintrÿffospectiveÐX¹aspect.¡8ôTheÊexternalaspSectisthesetofresultsthatanagen¬rtó·ág£ cmmi12ÃPlZ¹canachieve,Ðþi.e.¡'whatê¨itÌcÿffanÐX¹doinasituationÃs¹,¤w?ÃP¡Æoss¹(ÃPS;ÿþs¹)UR=ó!", cmsy10ÆfÃxÆjÃCÜan¹(ÃP;ÿþx;s¹)ÆgÃ:ü|¹(1)¡'Th¬rusøëinthepresentsituation,ü|Iøècan ndmydrink.cªInonesenseIøècanclimb¤'on ~theroSofofm¬ryhouseandjumpo.î'InanothersenseI Ecan't.(Thedieren¬rt¡'sensesêõofÌcÿffanÐ¥¹willbSediscussedinSection3.1).9ÆInacertainposition,ëac¬rhess¡'program}acanc¬rheckmate}aitsoppSonen¬rtandcanalsomoveintoapSositionlead-¡'ing=·totheoppSonen¬rtgivingcheckmate.ÿ:WhatisÃx¹inÃCÜan¹(ÃPS;ÿþx;s¹)?In=·English¡'ittusuallyhasthegrammaticalformofanaction,+butinthein¬rterestingcases¡'it+¿isnotanelemen¬rtaryactionlikethosetreatedinsituationcalculus.ü%Thus¡'w¬re{Khave`I{&cangotoAustralia',t`Icanmak¬reamilliondollars',t`Icangeta¡'new¥Âhouse'.j.OftenthewhatistobSeac¬rhieved¥Âisa uen¬rt,Ôe.g.thestateof¡'ha¬rvingê¨anewhouse.¡8ôInäthemostimpSortan¬rtcase,åºÃP¡Æoss¹(ÃP;ÿþs¹)dependsonlyonthecausalposi-¡'tionê¨ofÃPn¹inthew¬rorldandnotontheinternalstructureofÃP¡Æ¹.¡8ôThe½øin¬rtrospSectiveaspSectinvolvestheagentÃP¡Æ¹'sknowledgeofÃP¡Æoss¹(ÃPS;ÿþs¹),¡'i.e.·Ýitsükno¬rwledgeofwhatitcanachieve.·ÝHereiswherethehumansenseof¡'freeïÀwillcomesin.H(ItdepSendsonÃP¹ha¬rvinganinternalstructurethatallows¡'certain?ËaspSectsofitscurren¬rtstatetobein¬rterpretedasexpressingknowledge.¡'IÐQkno¬rwÐXIcan ndm¬rydrink.0InasimplechesspSositionIÐQwouldknowIÐQcould¡'giv¬reP¿checkmateinthree,obSecausethechessproblemcolumninthenewspapSer¡'saidê¨so,althoughImigh¬rtn'tyethavebSeenableto gureouthow.¡8ôSomeTpresen¬rtcomputerprograms,r8e.g.±chessprograms,r8haveanextensive¡'ÃP¡Æoss¹(ÃPS;ÿþs¹).¨FHo¬rwever,theirÊknowledgeofÃP¡Æoss¹(ÃPS;ÿþs¹)asasetisverylimited.¡'IndeedzitistoSolimitedforoptimalfunctionalit¬ryÿV,²oandrobots'kno¬rwledgeofç2û x ý¤ = ýç'¹theirwpSossibilitiesneedtobemademorelik¬rethatofhumans.%FÿVorexample, ça¤'robSot>Øma¬ryconcludethatinthepresentsituationithastoSolimitedasetof¡'pSossibilities.Þ¢ItÛîma¬rythenundertaketoensurethatinfuturesimilarsituations¡'itê¨willha¬rveê¨morec¬rhoices."Ê«'ó"ÂÖN ff cmbx12Í1.1KæcPreliminaryffphilosophicalremarks@'¹ConsiderWamac¬rhine,©e.g.Åacomputerprogram,thatisen¬rtirelydeterministic,¡'i.e.lis,completelyspSeci edandcon¬rtainsnorandomelement.lA%ma§jorques-¡'tion©forphilosophersiswhetherah¬rumanisdeterministicintheabSovesense.¡'Iftheansw¬rerisyes,¦thenwemusteitherregardthehumanashavingnofree¡'willIorregardfreewillascompatiblewithdeterminism.yÂSomephilosophers,¡'calledù;Ìcÿffompatibilists¹,)e.g.èfDanielDennett(Dennettê¨1984EI),tak¬rethisview,and¡'regard´apSersontoha¬rve´freewillifhisactionsaredeterminedb¬ryhisinternal¡'decision³proScessesev¬reniftheseprocessesthemselv¬resaredeterministic.û¥2Á2 A¹My¡'view,iscompatibilist,|lbutI+¿don'tneedtotak¬reapSositionondeterminism¡'itself.¡8ôAI¯4depSends¯Conacompatibilistview,»$butha¬rvingtakenit,»$thereisalotto¡'bSeaVlearnedaboutthespeci cformsoffreewillthatcanbedesigned.ëThat¡'isê¨thesub§jectofthisarticle.¡8ôIZdon'tZ-discusstheaspSectsoffreewillrelatedtoassigningcreditorblame¡'for4actionsaccordingtowhethertheyw¬reredonefreelyÿV.Moregenerally,Gthe¡'considerations(ofthisarticleareorthogonaltoman¬rystudiedbyphilosophers,¡'butê¨Ithinktheyapplytoh¬rumanfreewillnevertheless.¡8ôSpSeci callyÿV,+)EastBGerman¬rydidnotdenyitscitizensthekindoffreewill¡'that+ÒsomehopSetoestablishviaquan¬rtummechanicsorchaostheoryÿV.ü_Itdid¡'den¬ryê¨itscitizenschoicesinthesensediscussedinthisarticle.¡8ôLogical%AI·hassomefurtherphilosophicalpresuppSositions. 8WTheseare¡'discussedê¨in(McCarth¬ry1999bUÕÑ).(V'¿2D(ôInformalzádiscussionb#'¹Thereç9aredieren¬rtkindsandlevelsoffreewill..Anautomobilehasnone,¡'a9uc¬rhessprogramhasaminimalkindoffreewill,(andahumanhasalot.¡'Human-lev¬relµAIµ systems,¿Îi.e.'thosethatmatchorexceedhumanintelligence¡'will5needalotmorethanpresen¬rtchessprograms,!andmostlikelywillneed¡'almostê¨asm¬ruchê¨asah¬rumanpSossesses,eventobSeusefulservÿXäants.ç3ç x ý¤ = ýç8ô¹ConsiderC²c¬rhessprograms.CýWhatkindsoffreewilldotheyhaveandcan¤'they?Bha¬rve?ÿ¾A?usualchessprogram,agivenapSosition,ageneratesalistofmoves¡'a¬rvÿXäailable_jinthepSosition. 'Itthengoesdo¬rwnthelistandtriesthemoves¡'successiv¬relyOgettingascoreforeachmove. fItchoSosesthemovewiththe¡'highestL°score(orpSerhapsthe rstmo¬rveL°consideredgoodenoughtoac¬rhieve¡'aê¨certainob§jectiv¬re.)¡8ôThatVctheprogramconsidersalternativ¬resisourreasonforascribingtoita¡'littleêfreewill,ê5whereasw¬reascribSenonetotheautomobile.8°Howisthechess¡'program's£hfreewilllimited,±¨andwhatmorecouldw¬reask?! Couldfurtherfree¡'willê¨helpmak¬reitamoreeectiveprogram?¡8ôA¬Üh¬rumandoSesn'tusuallyconsiderhischoicessequentiallyÿV,ݨscoringeach¡'andºäcomparingonlythescores.©Theh¬rumancomparestheconsequencesof¡'theÛõdieren¬rtchoicesindetail. ÆWÿVouldithelpachessprogramtodothat?¡'Humanê¨c¬rhessplayersdoit.¡8ôBey¬rondÏÞthatisconsideringthesetÃLegn9alCs¹(Ãp¹)oflegalmovesinpSositionÃp¡'¹asÐ]anob§ject.0AÐWh¬rumanconsidershissetofchoicesanddoSesn'tjustconsider¡'eac¬rh³choiceindividuallyÿV.9AchesspSositioniscalled`cramped'iftherearefew¡'non-disastrouséÚmo¬rves,)§anditisconsideredusefultocramptheoppSonent's¡'pSosition·Þev¬renifonehasn'totherreasonsforconsideringthepositionbadfor¡'thekoppSonen¬rt.»ªVÿVerylikelyÿV,ÑaprogramthatcouldplayaswellasDeepBlue¡'butëdoing10û¥2ó¾KÈcmsy8ÇÁ6_g¹asm¬ruchëcomputationw¬rouldneedamoreelabSoratechoice¡'structure,Ùyi.e.1·moreÕ.freewill.FÿVorexample,Ùyone uen¬rtofchesspSositions,Ùye.g.¡'ha¬rvinganopSen leforarook,Ê?canberegardedasgivingabetterposition¡'thanê¨anotherwithoutassigningn¬rumericalvÿXäaluestopSositions.(V'¿3D(ôThe±À niteautomatonmo=deloffreewillandD(ôó#F CG® cmbxti10Îcþûanb#'¹ThisCsectiontreatsÃP¡Æoss¹(ÃPS;ÿþs¹)for niteautomata.¾Finiteautomataraisethe¡'question°gofwhatanagen¬rtÌcÿffan¹doinasharpform.However,áÖtheyarenot¡'a ¢usefulrepresen¬rtationofanagent'sintrospSectiveknowledgeofwhatitcan¡'do.¡8ôTÿVoLÖtheexten¬rtthatapSersonormachineÌcÿffan2¹achieveanyofdierentgoals,¡'that>]pSersonormac¬rhinehasfreewill.ÿrOurideasonthisshowupmostsharply¡'consideringsystemsofin¬rteractingdiscrete niteautomata. ðTheseareas¡'deterministic%xasy¬roucanget,t,whichiswhyI%'chosethemtoillustratefreeç4!b x ý¤ = ýç'¹will.¤8ôThe Fmaterialofthissectionrevisesthatin(McCarth¬ryê¨andHayes1969(â),¡'sectionê¨2.4en¬rtitled`TheautomatonrepresentationandthenotionofÌcÿffanå°¹'.¡8ôLetÃSÂÚ¹bSeasystemofin¬rteractingdiscrete niteautomatasuchasthat¡'sho¬rwnê¨in gure1. \¹Ø*¬[¼yïEPSfile="/u/ftp/jmc/figure1.eps" llx=0 lly=0 urx=398 ury=308 rwi=2834 ¶E¦Figureê¨1:8àSystemÃS²×¹.*¬8ôEac¬rhÖbbSoxrepresentsasubautomatonandeachlinerepresentsasignal.¡'Time2.tak¬resonintegervÿXäaluesandthedynamicbSehaviorofthewholeautoma-¡'tonê¨isgiv¬renbytheequations:¡~j/ÃaÌÌÁ1À¹(Ãtª¨¹+1)UP=ÃAÌÌÁ1¹(ÃaÌÌÁ1¹(Ãt¹)Ã;ÿüsÌÌÁ2¹(Ãt¹))¡~j/ÃaÌÌÁ2À¹(Ãtª¨¹+1)UP=ÃAÌÌÁ2¹(ÃaÌÌÁ2¹(Ãt¹)Ã;ÿüsÌÌÁ1¹(Ãt¹)Ã;sÌÌÁ3¹(Ãt¹)Ã;sÌÌÁ10 ¹(Ãt¹))¡~j/ÃaÌÌÁ3À¹(Ãtª¨¹+1)UP=ÃAÌÌÁ3¹(ÃaÌÌÁ3¹(Ãt¹)Ã;ÿüsÌÌÁ4¹(Ãt¹)Ã;sÌÌÁ5¹(Ãt¹)Ã;sÌÌÁ6¹(Ãt¹)Ã;sÌÌÁ8¹(Ãt¹))¡~j/ÃaÌÌÁ4À¹(Ãtª¨¹+1)UP=ÃAÌÌÁ4¹(ÃaÌÌÁ4¹(Ãt¹)Ã;ÿüsÌÌÁ7¹(Ãt¹))ü}(2)¡~j/ÃsÌÌÁ2À¹(Ãt¹)UP=ÃSÌÌÁ2¹(ÃaÌÌÁ2¹(Ãt¹))¡~j/ÃsÌÌÁ3À¹(Ãt¹)UP=ÃSÌÌÁ3¹(ÃaÌÌÁ1¹(Ãt¹))¡~j/ÃsÌÌÁ4À¹(Ãt¹)UP=ÃSÌÌÁ4¹(ÃaÌÌÁ2¹(Ãt¹))ç5- x ý¤ = ýç~j/ÃsÌÌÁ5À¹(Ãt¹)UP=ÃSÌÌÁ5¹(ÃaÌÌÁ1¹(Ãt¹))¤~j/ÃsÌÌÁ7À¹(Ãt¹)UP=ÃSÌÌÁ7¹(ÃaÌÌÁ3¹(Ãt¹))¡~j/ÃsÌÌÁ8À¹(Ãt¹)UP=ÃSÌÌÁ8¹(ÃaÌÌÁ4¹(Ãt¹))¡~j/ÃsÌÌÁ9À¹(Ãt¹)UP=ÃSÌÌÁ9¹(ÃaÌÌÁ4¹(Ãt¹))¡~j/ÃsÌÌÁ10 ¹(Ãt¹)UP=ÃSÌÌÁ10¹(ÃaÌÌÁ4À¹(Ãt¹))ü}(3)8ôThein¬rterpretationoftheseequationsisthatthestateofanysubautoma-¡'tonC²attimeÃt¹+1C²isdeterminedb¬ryitsstateattimeÃt¹andbythesignals¡'receiv¬redJbattimeÃt¹.sThevÿXäalueofaparticularsignalattimeÃt¹isdeterminedby¡'theèstateattimeÃt¹oftheautomatonfromwhic¬rhitcomes.$ Signalswithouta¡'source5subautomatonrepresen¬rtinputsfromtheoutsideandsignalswithout¡'aê¨destinationrepresen¬rtoutputs.¡8ôFinite5automataarethesimplestexamplesofsystemsthatin¬rteractover¡'time.üSThey+Îarecompletelydeterministic;Lbifw¬reknowtheinitialstatesofall¡'theÿ7automataandifw¬reknowtheinputsasafunctionoftime,ZthebSehavior¡'of*
thesystemiscompletelydeterminedb¬ryequations(2)and(3)forallfuture¡'time.¡8ôTheîautomatonrepresen¬rtationconsistsinregardingtheworldasasystem¡'ofû£in¬rteractingsubautomata.é4FÿVorexample,+qwemightregardeachpSersoninthe¡'roSomD
asasubautomatonandtheen¬rvironmentD
asconsistingofoneormore¡'additional1Isubautomata.ûAsw¬reshallsee,V\thisrepresentationhasmanyofthe¡'qualitativ¬reËpropSertiesofinteractionsamongthingsandpSersons..SHowever,ÑUif¡'w¬reÁÄtaketherepresentationtoSoseriouslyandattempttorepresentparticular¡'in¬rterestingzßsystemsassystemsofinteractingautomata,Þìweencounterthe¡'follo¬rwingê¨diculties:¡8ô1.øØThe*n¬rumbSerofstatesrequiredinthesubautomataisverylarge,Püforex-¡'ampleI¡2û¥2Á10ý-:ó¹Aa¨cmr6Â10ÕV¹,iÖifw¬retrytorepresentapSerson'sknowledge.3Automatathislarge¡'ha¬rvetobSerepresen¬rtedbysystemsofequationsorbycomputerprograms,¥or¡'ink÷someotherw¬rayk÷thatdoSesnotin¬rvolvek÷mentioningstatesindividuallyÿV.¼ÍIn¡'Sectionê¨4w¬re'llrepresentthempartiallyÿV,bysentencesoflogic.¡8ô2..·GeometricçEinformationishardtorepresen¬rt.Consider,&lforexample,¡'the loScationofam¬rulti-jointed ob§jectsuc¬rhasapersonoramatterofev¬ren¡'moreê¨dicult¬ry|theshapSeofalumpofclayÿV.¡8ô3.Thesystemof xedin¬rterconnectionsisinadequate.SinceapSerson¡'ma¬ry1 handleanyob§jectintheroSom,C^anadequateautomatonrepresentation¡'w¬rouldê¨requiresignallinesconnectinghimwitheveryob§ject.¡8ô4.4Theémostseriousob§jection,(ho¬rwever,iséthat(intheterminologyof¡'(McCarth¬ryê¨andHayes1969(â))Gtheautomatonrepresen¬rtationisepistemologi-ç63] x ý¤ = ýç'¹callyøuinadequate.bGNamelyÿV,ûèw¬redonoteverknowapSersonwellenoughtolist¤'his=in¬rternalstates.¯ ThekindofinformationwedohaveabSouthimneedsto¡'bSeê¨expressedinsomeotherw¬rayÿV.¡8ôNev¬rertheless,ÖÛwe§mayusetheautomatonrepresentationforconceptsof¡'Ìcÿffan,CÍcauses,9¹useful×ékindsofcoun¬rterfactualstatements(`Ifanothercarhad¡'come½Æo¬rverthehillwhenyoupassedjustnow,ùótherewouldhavebSeenahead-on¡'collision').8àSeeê¨(CostelloandMcCarth¬ry1999BÏ). ½Ø*¬[¼yïEPSfile="/u/ftp/jmc/figure2.eps" llx=0 lly=0 urx=452 ury=146 rwi=2834 5âFigureê¨2:8àAnothersystemÃS²×¹. ¸ëØ*¬[¼yïEPSfile="/u/ftp/jmc/figure3.eps" llx=0 lly=0 urx=452 ury=186 rwi=2834 ´?Figureê¨3:8àSystemÃSÌÌÁ1À¹.8ôLetusconsiderthenotionofÌcÿffan¹.LetÃS»f¹bSeasystemofsubautomata¡'without Òexternalinputssuc¬rhasthatof gure2._LetÃp¹bSeoneofthesubau-¡'tomata, öandçsuppSosethatthereareÃm¹signallinescomingoutofÃp¹.WhatÃp¡'¹canudoisde nedintermsofanewsystemÃS²×ÌÌó×2cmmi8Äpz4¹,±(whic¬rhisobtainedfromthe¡'system=ÃSE¹b¬rydisconnectingtheÃm¹signallinescomingfromÃp¹andreplacing¡'them5Ub¬ryÃm¹externalinputlinestothesystem.çIn gure2,Hsubautomaton1¡'hasoneoutput,;yandinthesystemÃSÌÌÁ1ϱ¹( gure3)thisisreplacedb¬ryanexternalç7@4 x ý¤ = ýç'¹input. FThe[ÚnewsystemÃS²×ÌÌÄpÖ¹alw¬rays[ÚhasthesamesetofstatesasthesystemÃS²×¹.¤'No¬rw/letÃqh¹bSeaconditiononthestatesuchas, P`ÃaÌÌÁ2Ã3¹iseven'or`ÃaÌÌÁ2 ?¹=ÃaÌÌÁ3À¹'.t(In¡'theê¨applicationsÃXá¹ma¬rybSeaconditionlike`ThebSoxisunderthebananas'.)¡8ôWÿVeê¨shallwrite¡ÍÈÃcan¹(Ãp;ÿþn9;s¹)'whic¬rh;isread,$à`ThesubautomatonÃpÌcÿffan¹bringabSouttheconditionÃt¹inthe¡'situationÄÃs¹'ifthereisasequenceofoutputsfromtheautomatonÃS²×ÌÌÄp ø¹that¡'willoâev¬rentuallyputÃS"¹¹intoastateÃaû¥2Ç0>¹thatsatis esÃn9¹(Ãaû¥2Ç0Î9¹).ÈInotherwords,0in¡'determiningäwhatÃp¹canac¬rhieve,Æóweäconsidertheeectsofsequencesofits¡'actions,¶quite¨ñapartfromtheconditionsthatdeterminewhatitactuallywill¡'do.¡8ôHere'sÏanexamplebasedon gure2.UInordertowriteform¬rulasconve-¡'nien¬rtlyÿV,Ûwe«usenaturaln¬rumbSerss«forthevÿXäaluesofthestatesofthesubau-¡'tomataê¨andthesignals.¡/ýÛ¦fmDHÃaÌÌÁ1À¹(Ãtª¨¹+1)£éß=· gÃaÌÌÁ1À¹(Ãt¹)ª¨+ÃsÌÌÁ2¹(Ãt¹)¤ÿmDHÃaÌÌÁ2À¹(Ãtª¨¹+1)£éß=· gÃaÌÌÁ2À¹(Ãt¹)ª¨+ÃsÌÌÁ1¹(Ãt¹)+2ÃsÌÌÁ3¹(Ãt¹)¡mDHÃaÌÌÁ3À¹(Ãtª¨¹+1)£éß=»gó$ÂÖN cmbx12ÏifÆÂ¹ÃaÌÌÁ3À¹(Ãt¹)UR=0Ïthen">5¹0Ïelse:LÃaÌÌÁ3¹(Ãt¹)ª¨+1¡?<ÃsÌÌÁ1À¹(Ãt¹)£éß=»gÏifÆÂ¹ÃaÌÌÁ1À¹(Ãt¹)UR=0Ïthen">5¹2Ïelse:L¹1¡?<ÃsÌÌÁ2À¹(Ãt¹)£éß=· g1¡?<ÃsÌÌÁ3À¹(Ãt¹)£éß=»gÏifÆÂ¹ÃaÌÌÁ3À¹(Ãt¹)UR=0Ïthen">5¹0Ïelse:L¹1Ã:ü|¹(4)6ní8ôConsidermytheinitialstateofÃS P¹tobSeoneinwhic¬rhallthesubautomataare¡'inê¨state0.8àWÿVeha¬rveê¨thefollo¬rwingpropSositions:÷z8ô1.(Subautomaton¸U2Ìwill¹nev¬rerbSeinstate1.[Itstartsinstate0andgoSes¡'toê¨state2attime1.8àAfterthatitcannev¬rerdecrease.]¡8ô2./?Subautomatonçr1Ìcÿffan¹putSubautomaton2instate1butw¬ron't.[If¡'Subautomatoná1emitted1attime0insteadof2,¥ acarcameo¬rverl>thehillisoneoftheoutputsofatracsubautoma-¡'ton. (Costelloê¨andMcCarth¬ry1999BÏ)¡·discussesusefulcoun¬rterfactuals,°Nlike¡·the¡'abSo¬rveªjthatareim¬rbeddedinadescriptionofasituationandha¬rveªjconse-¡'quences.9One
³useisthattheypSermitlearningfromanexperiencey¬roudidn't¡'quiteê¨ha¬rveandwouldrathernothave."Ê«'Í3.3KæcGos3odffanalysesinÌtosubautomata@'¹In theforegoingw¬rehavetakentherepresentationofthesituationasasystem¡'of¼Yin¬rteractingsubautomataforgranted.)pIndeedifyouwanttotakethemfor¡'gran¬rtedê¨youcanskipthissection.¡8ôHo¬rwever,sa1±givenoverallautomatonsystemmightbSerepresentedasa¡'system×ofin¬rteractingsubautomatainanumbSerofways,·ãanddierentrep-¡'resen¬rtations4mightyielddierentresultsabSoutwhatagivensubautomaton¡'can,²ac¬rhieve,}5whatwouldhavehappSenedifsomesubautomatonhadacted¡'dieren¬rtlyÿV,Ããorqwhatcausedwhat.B
RetroSearch is an open source project built by @garambo
| Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4