Explanatory Pluralism And The Co-Evolution Of Theories In Science …mechanism.ucsd.edu/~bill/teaching/w17/phil151/mccauley... · 2017-01-12 · Explanatory Pluralism And The Co-Evolution

ExplanatoryPluralismAndTheCo-EvolutionOfTheoriesInScience*

RobertN.McCauley1.IntroductionOverthepastdecadeorsoPatriciaandPaulChurchlandhavemademajorcontributionstophilosophicaltreatmentsofintertheoreticreductioninscience.Thehistoricimportanceofthisissueinthephilosophyofscienceispatentandso,therefore,istheimportanceoftheChurchlands’contributions.Theirinsistenceonthecentralityofthisissuetodiscussionsinthephilosophyofmindmay,however,beevenmorepraiseworthyinanerawhenmanyinthatfield(evenamongthosewhoclaimthemantleofnaturalism)makerepeateddeclarationsaboutthestatusofthepertinentsciencesandthemind-bodyproblemgenerallyinwhatoftenappearstobeblitheignoranceofboththosesciencesandtherelevantliteratureinthephilosophyofsciencesince1975.Inarecent,jointpapertheChurchlands(1990)discussandlargelydefusefivewell-wornobjections(concerningqualia,intentionality,complexity,freedom,andmultipleinstantiation)tothereductionofpsychologytoneurobiology.MyconcernswiththatputativereductionandwiththeChurchlands’accountoftheoverallprocessareofaverydifferentsort.Twomodelshavetraditionallydominateddiscussionsofintertheoreticrelations.Afterbrieflysurveyingthecontrastsbetweenthem,section2examineshowtheChurchlands’accountoftheserelationsintermsofacontinuumofintertheoreticcommensurabilitycapturesthosemodels’respectiveadvantagesinasingleproposal.ThatsectionendsbyexamininghowPatriciaChurchland’ssubsequentdiscussionsoftheco-evolutionoftheoriesenhancesthisaccountbyexploringsomeofitsunderlyingdynamics.Inshort,theco-evolutionoftheoriesconcernscross-scientificinteractionsthatchangethepositionofaparticularintertheoreticrelationshipontheChurchlands’continuum.Insection3IlocatesomerevealingequivocationsintheChurchlands’discussionsof“theco-evolutionoftheories”bydistinguishingthreepossibleinterpretationsofthatnotionthatwindtheirwaysthroughtheChurchlands’workandthroughNeurophilosophyinparticular.WiththeaidofadistinctionconcerninglevelsofanalysisthatIhavedevelopedelsewhere,Iargue,ineffect,thattheChurchlands’accountoftheco-evolutionoftheoriesandtheirmodelofintertheoreticreductionobscurecriticaldistinctionsbetweenthreequitedifferenttypesofintertheoreticrelations.Section4positionsthesethreetypeswithinamorefine-grainedaccountofintertheoreticrelationsthatwillofferabasisforevaluatingtheirrelativemeritsasanalysesoftheinterfaceofpsychologyandneuroscience.Oneofthesethree,thepictureofco-evolutionmodeledonthedynamicsofscientificrevolutions,hasattractedthemostattention.ThisinterpretationhasencouragedtherecurringeliminativistinclinationsconcerningfolkpsychologyforwhichtheChurchlandsarerenown,but,ofthethree,itisalsotheinterpretationthatisleastplausibleasananalysisoftherelationsbetweenpsychologyandneuroscience.Psychology(folkorotherwise)maywellundergosubstantialrevision,andfuturescientificprogressmaywellleadtotheeliminationofsomepsychologicaltheories,buttheChurchlandshaveofferedanunhelpfullyoversimplifiedaccountoftheintertheoreticdynamicsinquestion.

ExplanatoryPluralismAndTheCo-EvolutionOfTheoriesInScience p,2

Insection5Ishallsupportandelaborateuponanotheroftheinterpretationsofco-evolutionthatemergesfromNeurophilosophyby,amongotherthings,examiningacase(concerningtheconnectionistnetwork,NETtalk)thattheChurchlandsandtheircollaboratorshavehighlighted.Thisthirdinterpretationrecognizesnotmerelythevalueofintegratingscientificdisciplinesbutofpreservingapluralityofsemi-autonomousexplanatoryperspectives.AlthoughtheChurchlandsnowoftenseemtofavorthisthirdinterpretationtoo,someoftheircommentscontinuetoconflatethethreedistincttypesofintertheoreticrelations.2.ThreePhilosophicalModelsofIntertheoreticRelationsinScienceUntilthelate1970s(atleast)twomodelsofintertheoreticrelationsinsciencedominatedphilosophers’attentions.Thefirst,ageneralpurposemodelofintertheoreticrelations,wasdeeplyrootedinlogicalempiricism;thesecond,ineffectamodeloftheorychange,emergedlargelyinreactiontothefirst(Bechtel1986).Ishallbrieflydiscusstheminorder.AlthoughErnestNagel’sTheStructureofScience(1961)containsthemosttime-honoredtreatmentoftheoryreduction,RobertCausey’sUnityofScience(1977)probablyprovidesthemostcomprehensivediscussionofthetopic.Theirgeneralapproachtotheoryreductionproceedswithintheconstellationofcommitmentsthatcharacterizelogicalempiricism,includingtheassumptionsthatasatisfactoryaccountofscientificrationalityrequiresheedtojustificatoryconsiderationsonly,thatscientifictheoriesarebestunderstoodascomplexpropositionalstructuresandbestrepresentedviaformalreconstructions,thatscientificexplanationresultsfromthedeductionofexplanandafromscientificlaws,thatscientificprogressresultsfromthesubsumptionofreigningtheoriesbytheoriesofevengreatergenerality,andthatscienceultimatelyenjoysanunderlyingunityoftheoryandontology.Thismodelconceivestheoryreductionasaspecialcaseofdeductive-nomologicalexplanation.Itisaspecialcasebecausetheexplanandumisnotastatementdescribingsomeeventbutratheralawofthereducedtheory.Inordertocarryoutsuchreductions,thepremisesinthemostcomplexcasesofheterogeneousreductiveexplanationsmustinclude(1)atleastonelawfromthereducingtheory:(2)statementsindicatingthesatisfactionoftherequisiteinitialconditionsspecifiedinthatlaw;(3)bridgelawswhichsystematicallyrelate–withinaparticulardomaindelineatedbyappropriateboundaryconditions–thetermsfromthepertinentlaw(s)ofthereducingtheorytothosefromthelawofthereducedtheory;(4)statementsindicatingthesatisfactionofthoseboundaryconditions(underwhichtheeventsdescribedinthelawofthereducingtheoryrealizetheeventsdescribedinthelawofthereducedtheorythatistobeexplained).Suchpremisespermitastraightforwarddeductionofthelawofthereducedtheory.Becausetheboundaryconditionsincludedinthebridgelawsarecastintermsofpredicatescharacteristicofthereducingtheory,thereductionreflectsanasymmetrybetweenthetwotheories.Thereducingtheoryexplainsthereducedtheory,finally,becausethereducingtheoryencompassesawiderarrayofeventswithinitsexplanatorypurview.Thissetofevents,presumably,includesalloftheeventsthereducedtheoryexplainsandmore,sothattheprinciplesofthereducingtheoryarebothmoregeneralandmorefundamental.Themostpopularshowcaseillustrationisthereductionofthelawsofclassicalthermodynamicstotheprinciplesofstatisticalmechanics.


Whenthereducingtheoryoperatesatalowerlevelofanalysisthanthereducedtheory,theaddedgeneralityofitsprinciplesisadirectfunctionofthisfact.Thesearecasesofmicroreductionswherealowerleveltheoryanditsontologyreduceahigherleveltheoryanditsontology(OppenheimandPutnam1958,Causey1977).Microreductionistsholdthatifwecanexhaustivelydescribeandpredictupperlevel(ormacro)entities,properties,andprinciplesintermsoflowerlevel(ormicro)entities,properties,andprinciples,thenwecanreducetheformertothelatterandreplace,atleastinprinciple,theupperleveltheory.Virtuallyalldiscussionsofintertheoreticrelationspresupposethisarrangementamong(andwithin)thesciencesintermsoflevelsofanalysis.(See,forexample,ChurchlandandSejnowski1992,pp.10-11.)Numerousconsiderationscontributetothedepictionofthearchitectureofscienceasalayerededificeofanalyticallevels(Wimsatt1976).Ideally,movingtowardlowerlevelsinvolvesmovingtowardthestudyofincreasinglysimplesystemsandentitiesthatareubiquitous,enduring,andsmall.Conversely,movingfromlowertohigherlevelsciencesinvolvesmovingtowardstudiesoflarger,rarersystemsofgreatercomplexityand(often)lessstabilityandwhosehistoryislessancient.Becausethealtitudeofalevelofanalysisisdirectlyproportionaltothecomplexityofthesystemsittreats,higherlevelsciencesdealwithincreasinglyrestrictedrangesofeventshavingtodowithincreasinglyorganizedphysicalsystems.1Asasimplematteroffact,oftenmorethanoneconfigurationoflowerlevelentitiescanrealizevarioushigherlevelkinds(especiallywhenfunctionallycharacterized).Theresultingmultipleinstantiationshighlightboththeimportanceandthecomplexityoftheboundaryconditionsinthebridgelawsofheterogeneousmicroreductions.Criticsofthemicroreductionistprogram(e.g.,Fodor1975)seethatcomplexityassufficientgroundsforquestioningtheprogram’sfeasibilityinthecaseofthespecialsciences,whilemoresympatheticparticipantsinthesediscussionssuchasRobertRichardson(1979)andtheChurchlands(1990)suggestthatwhenscientiststraceoutsuchconnectionsbetweenhigherandlowerlevelentitiesinspecificdomainstheyvindicatetheoverallstrategy,whilerecognizingthedomainspecificityofitsresults.Reductionistsdifferamongthemselvesastothepreciseconnectionsbetweenentitiesatdifferentlevelsthatarerequiredforsuccessfulreductiveexplanation.Theyallagree,however,thatthetheorieswhicharepartiestothereductionshouldmapontooneanotherwellenoughtosupportsystematicconnections,usuallycontingentidentities,betweensome,ifnotall,oftheentitiesthatpopulatethem.Thetestoftheresultingcontingentidentitiesismet,ultimately,bytheexplanatorysuccessesthereductionsaccomplish(McCauley1981;Enc1983).Feyerabend(1962)andKuhn(1970)arethemostprominentproponentsofthesecondmajoraccountofintertheoreticrelations.Theyforgedtheirearlydiscussionslargelyinresponsetoboththelogicalempiricistprogramanditsreductionistblueprintforscientificprogress.Feyerabendemphasizedhowscrutinyofmanyoftheshowcaseillustrationsofintertheoreticreductionsrevealedthefailureofthesecasestoconformtothelogicalempiricists’model.Kuhndiscussednumerousexamplesinthehistoryofsciencewheresuccessivetheorieswerenotevenremotelyplausiblecandidatesforthesortofsmoothtransitionsthestandardreductivemodelenvisions.Instead,Kuhnproposedthatprogressinscienceconsistsofextendedintervalsofrelativetheoreticalstabilitypunctuatedbyperiodicrevolutionaryupheavals.Bothholdthatthecasesinquestioninvolveconflictsbetweenincommensurabletheories.Althoughthesubsequentliteratureisrifewithassessmentsofthisclaim(Thagard1992offersthemostsuggestiveofrecenttreatments),thecriticalpointfornowisthat,whateverincommensurabilityamountsto,itstandsinstarkoppositiontoanymodelofintertheoretic


relationsthatrequiresneatmappingsbetweentheories’principlesandontologiescapableofsupportingstrictdeductive-nomologicalexplanations.Thehistoryofscienceprovidesampleevidencethatwheresuchincompatibilityissufficientlyseverethetheoryanditsontologythatareeventuallydeemeddeficientundergoelimination.Stahl’ssystemofchemistryisthepreferredillustration,butDarwin’stheoryofinheritancecouldservejustaswell.Theunmistakablesensethatbothofthesemodelsofintertheoreticrelationsdescribesomeactualcasesfairlyaccuratelyandthattheyeachcaptureimportantinsightsabouttheissuesatstake,theirprofoundconflictsnotwithstanding,couldinducepuzzlement.AnaccountofintertheoreticrelationsintermsofacontinuumofcommensurabilitythatPaulChurchland(1979)initiallysketchedandwhichtheChurchlandshavesubsequentlydeveloped(P.S.Churchland1986,pp.281f.;ChurchlandandChurchland1990)substantiallyresolvesthatperplexitybyreconcilingthoseconflictsandallottingtoeachmodelameasureofdescriptiveforce.TheChurchlandspointoutthat,infact,differentcasesofintertheoreticrelationsvaryconsiderablywithrespecttothecommensurabilityofthetheoriesinvolved.So,theyproposethatsuchcasesfallalongacontinuumofrelativeintertheoreticcommensurability,where,ineffect,thetwomodelssketchedaboveconstitutethatcontinuum’send-points.Oneendofthecontinuumrepresentscaseswhereintertheoreticmappingisextremelyloworevenabsent.Thesearecasesofradicalincommensurabilitywhererevolutionaryscienceandthecompleteeliminationofinferiortheoriesensue.Whatevervaguenessmaysurroundthenotionof‘incommensurability,’theChurchlandsareclearlyconfidentthatthedevelopmentswhichbroughtabouttheeliminationofthebodilyhumours,theluminiferousether,caloricfluid,andthelike,involvesufficientlydrasticchangestojustifythesortofextremedeparturesfromthetraditionalmodelofreductionthatKuhnandFeyerabendadvocated.Attheotherendofthiscontinuum,wherethemappingofonetheoryonanotherisnearlyexhaustiveandtheformertheory’sontologyiscomposedfromtheentitiesthelattertheorycountenances,themostrigorousmodelsoftheoryreductionmostnearlyapply(e.g.,Causey1972).Theconstraintsproponentshaveimposedontheoryreductionaresodemandingthatitisafairquestionwhetheranyactualscientificcasequalifies.TheChurchlandshaveurgedconsiderablerelaxationoftheconditionsnecessaryforintertheoreticreduction.Insteadofconformitytotherigorouslogicalandontologicalconstraintstraditionalmodelsimpose,PaulChurchland(1979;seetooHooker1981;Bickle1992)suggeststhatthereducingtheoryneedonlypreservean“equipotentimage”ofthereducedtheory’smostcentralexplanatoryprinciples.Thereductioninvolvesanimage,sincethereducingtheoryneednotduplicateeveryfeatureofthereducedtheory’sprinciples,butonlyenoughoftheirsalientonestosuggesttheirgeneralcharacterandtoindicatetheirsystematicimport(seeSchaffner1967).Thatimageisequipotent,though,sincethereducingtheory’sprincipleswillpossessalloftheexplanatoryandpredictivepowerofthereducedtheory’sprinciples–andmore.Fromthestandpointoftraditionalmodels,Churchlandproposesaformofapproximatereduction,whichfallswellshortofthelogicalempiricists’standards,butwhichalsosuggestshowtruetheories(e.g.,themechanicsofrelativity)cancorrectandevenapproximatelyreducetheoriesthatarefalse(e.g.,classicalmechanics).Switchingtothemetaphorofimageryisappropriate,since,asWilliamWimsatt(1976,p.218)notedoveradecadeago,ifthestandardmodelsofreductionallegethatafalsetheoryfollowsfromatrueone,theputativedeductionhadbetterinvolveanequivocationsomewhere!InrecentyearstheChurchlandshaveeachenlargedonthiscontinuummodel.Forexample,withinhisneurocomputationalprogramPaulChurchlandhasadvancedaprototypeactivationmodelof


explanatoryunderstandingthat,presumably,includestheunderstandingthatarisesfromreductiveexplanations.Churchlandholdsthattheneurocomputationalbasisofexplanatoryunderstandingresidesintheactivationofaprototypevectorwithinaneuralnetworkinresponsetoimpingingcircumstances.Adistributedrepresentationoftheprototypeintheneuralnetworkconstitutesthebrain’scurrentbeststabatdetectinganunderlyingpatternintheblooming,buzzingconfusion.ForChurchlandexplanatoryunderstandingisanarrayofinputsleadingtotheactivationofoneoftheseexistingprototypesasopposedtoanother.Churchlandinsiststhattheactivationofaprototypevectorincreases,ratherthandiminishes,availableinformation.Itinvolvesa“speculativegain”ininformation(1989,p.212).Thus,contrarytoanti-reductionistcaricature,thisaccountofexplanatoryunderstandingimpliesthatreductiveexplanationsamplifyourknowledge.Theoriginalityoftheinsightsareductiveexplanationoffersdependsuponthenovelapplicationofexistingcognitiveresources,i.e.,ofanindividual’srepertoireofprototypevectors.Consequently,reductiveexplanationinvolvesneitherthegenerationofnewschemesnorthedestructionofoldones.Theapproximatecharacterofintertheoreticreductionsisafunctionofthis“conceptualredeployment”onwhichtheyturn(1989,p.237).Inconceptualredeploymentadevelopedconceptualframeworkfromonedomainisenlistedforunderstandinganother.Inshort,successfulreductiveexplanationrestsonananalogicalinferencebyvirtueofwhichwedeemanimageofatheoryequipotenttotheoriginal.Havingestablishedtheinitialapplicabilityofanexisting,alternativeprototypevector,itinevitablyundergoesareshapingasaconsequenceofexposuretothenewlyadoptedtrainingset.Thisreshapingofactivationspaceistheneurocomputationalprocessthatdrivestheremainingco-evolutionofthereductivelyrelatedtheories.Inherdiscussionsoftheco-evolutionoftheories,PatriciaChurchlandhasintroducedadynamicelementintothecontinuummodel.Shesuggeststhatthepositionoftwotheories’relationsonthiscontinuumcanchangeovertimeastheyeachundergoadjustmentsinthelightofoneanother’sprogress.Thesuggestionthatscientifictheoriesco-evolvearisesfromananalogywiththeco-evolutionofspeciesandfromthepictureofthesciencesbrieflyoutlinedabove.Ontheco-evolutionarypicturethesciencesexertselectionpressuresononeanotherinvirtueofageneralconcernforsupplyingasmuchcoherenceaspossibleamongourexplanatoryschemes.Ifthevarioussciencesarearrangedintiersofanalyticallevels,theneachwillstandatvaryingdistancesfromtheothersinthisstructure.Typically,proximityisacentralconsiderationinassessingtheforceofselectionpressures.Thus,thepivotalrelationshipsarethosebetweenascienceandthosesciencesatimmediatelyadjacentlevels.Forexample,thepresumptionisthattheneurosciencesbelowandthesocio-culturalsciencesabovearemorelikelytoinfluencepsychologythanarethephysicalsciences,sincetheyarelocatedbelowtheneurosciencesand,therefore,atanevengreaterdistance.Itisthisprocessoftheco-evolutionoftheoriesandtheChurchlands’accountofitthatwilldominatetheremainderofthispaper.IshallattendtoitsimplicationsfortherelationshipofcognitivepsychologytothesortofneurocomputationalmodelingthattheChurchlandsendorse.3.ThreeWaysTheoriesMightCo-evolvePatriciaChurchland’sNeurophilosophy(1986)containsthemostextensivediscussionofreductionintermsoftheco-evolutionoftheoriesavailable.2Churchlandfocusesontherelationbetween


neuroscienceandpsychology,butherdiscussionclearlyaspirestomoralsthataregeneral.Hercommentsatvariouspointsseemtosupportthreedifferentco-evolutionaryscenarios,thoughtwoofthemare,quiteclearly,closelyrelated.ThethreearedistinguishedbythelocationsontheChurchlands’continuumtowhichtheypredictco-evolvingtheorieswillincline.OnsomeoccasionsChurchlandsuggeststhatpsychologyandtheneuroscienceswillco-evolveinthedirectionofapproximatereduction.Shestates,forexample,that“theco-evolutionarydevelopmentofneuroscienceandpsychologymeansthatestablishingpointsofreductivecontactismoreorlessinevitable....Theheartofthematteristhatifthereistheoreticalgiveandtake,thenthetwoscienceswillknitthemselvesintooneanother”(1986,p.374).Themetaphoroftwosciencesknittedintooneanotherimpliesanintegrationthatistight,orderly,anddetailed.AlthoughChurchland,presumably,doesnotthinkthatthatintegrationwillsatisfythetraditionalmicroreductionists’stringentdemandsonintertheoreticmapping,talkofknittingtwosciencesintooneanother,theon-goingpursuitofaunifiedmodelofreduction(ChurchlandandChurchland1990),andanewinterestinestablishingpsycho-physicalidentitiesechocommitmentsoftraditionalmicroreductionism,wherethesortofreductivecontactinquestionledtotalkofan“inprinciplereplaceability”ofthereducedtheoryinwhichthelowerleveltheoryenjoysbothexplanatoryandmetaphysicalpriority.Morerecently,theChurchlandshavebeenclearaboutthefutilityofattemptstoreplaceupperleveltheories,buttheystillgenerallysubscribetotheexplanatoryandmetaphysicalpriorityofthelowerleveltheory–especiallyinthecaseofpsychologyandneuroscience.(See,forexample,P.S.Churchland1986,pp.277,294,and382.)Certainly,aco-evolutionaryaccountofintertheoreticrelationshasnoproblemtranslatingthegeneralmicroreductiveimpulse.Withinthisframeworkitamountstotheclaimthattheselectionpressuresthatthescienceatthelevelofanalysisbelowthatofthetheoryinquestionexertswillhaveanoverwhelminglygreatereffectonthattheory’seventualshapeandfatethanwillthesciencesabove(seesection5).OntheChurchlands’account,suchintertheoreticintegrationwouldenabletheneurosciencestosupplyanequipotentimageofpsychologicalprinciples.PaulChurchland’sspeculationsabouttheneuralrepresentationofthesensoryqualiaassociatedwithcolorvisionmightconstituteanappropriateillustration.Thefitbetweenourcommonsensenotionsaboutourexperiencesofcolorsandthesystemofneuralrepresentationheproposesisquiteneat(1989,p.102-08).HereafterIshallrefertothissenseoftheco-evolutionoftheoriesas“co-evolutionM,”i.e.,co-evolutioninthedirectionofapproximatemicroreduction.IntheChurchlands’jointdiscussion(1990,chapter6.1),whereitplaysbothapredictiveandnormativerole,thisnotionofreductionreceivesconsiderableattention.TheChurchlandsclearlyholdthat“itisreasonabletoexpect,andtoworktoward,areductionofallpsychologicalphenomenatoneurobiologicalandneurocomputationalphenomena”(1990,p.249).Co-evolutionMisnottheonlyaccountofco-evolutioninNeurophilosophy,for,astheChurchlandshavesubsequentlyasserted,inthecaseofpsychologyandneuroscience,“thereareconflictingindications”aboutthedirectioninwhichconjecturesatthesetwolevelsofanalysiswilllikelyco-evolve(1990,p.253).Ifintegrationisthefateofpsychologyandneuroscience,PatriciaChurchlandrepeatedlyhintsthatthiswillonlyoccurafterpsychology’sinitialdemolitionandsubsequentreconstructioninaccordwiththemandatesoftheneurosciences.Sheclaims,forexample,that“...thepossibilitythatpsychologicalcategorieswillnotmaponetooneontoneurobiologicalcategories...doesnotlooklikeanobstacletoreductionsomuchasitpredictsafragmentationandreconfigurationofthepsychologicalcategories”(1986,p.365).Withthissecondview,aswiththefirst,noquestionarisesaboutwheretheblamelies,ifthetheoriesofpsychologyandneurosciencefailtomapontooneanotherneatly.(SeeWimsatt1976.)Atleastfortheshortterm,Churchland


seemstoexpectthatthisintertheoreticrelationwillmigrateinjusttheoppositedirectiononthecontinuumofintertheoreticcommensurabilityfromwhatco-evolutionMpredicts,i.e.,towardagrowingincommensurabilitythatpredictsafragmentationofpsychologicalcategories.Ifthe“fragmentationandreconfiguration”ofpsychologicalcategoriesinvolvedonlytheelaborationoradjustment(oreventheinprinciplereplaceability)ofpsychologicaltheoriesbydiscoveriesintheneurosciences,co-evolutionMmightsuffice.Onthissecondview,though,thisprocesscanleadtotheeventualeradicationofmajorpartsofpsychology.So,forexample,Churchlandremarksthat“thereisatendencytoassumethatthecapacitiesatthecognitivelevelarewelldefined...inthecaseofmemoryandlearning,however,thecategorialdefinitionisfarfromoptimal,andrememberingstandstogothewayofimpetus”(1986,p.373,emphasisadded3).HereChurchlandanticipatesthatjustasthenewphysicsofGalileoandhissuccessorsoustedthelatemedievaltheoryofimpetus,sotooshalladvancesinneurosciencedisposeofpsychologists’speculationsaboutmemory.This,then,isco-evolutionS(co-evolutionproducingtheeliminationsoftheoriescharacteristicofscientificrevolutions)inwhichthetheoreticalperspectivesoftwoneighboringsciencesaresodisparatethateventuallythetheoreticalcommitmentsofonemustgo--inthefaceoftheother’ssuccess.Co-evolutionSunderliesthepositionforwhichtheChurchlands’advocacyhasbeenfamous,viz.,eliminativematerialism.4Theyhavecontendedthatprogressintheneuroscienceswillprobablybringabouttheeliminationoffolkpsychologyaswellasanyotherpsychologicaltheoriesthatinvolvecommitmentstothepropositionalattitudes(presumably,includingmuchofmainstreamcognitiveandsocialpsychology).Justasscientistsbanishedphlogistonandcaloricfluid,sotoowillthepropositionalattitudesbeexpelledasneuroscienceprogresses.Thepsychologicalconjecturesinquestion(will)failtomatchthedescriptive,explanatory,andpredictivesuccessesoftheirneuroscientificcompetitors.Moreover,theirsubstantialdissimilaritiestothoseallegedcompetitorsprecludeanysortofreconciliation.Consequently,numeroustheoreticalnotionsinpsychologystandtogothewayofimpetus.ThisisthepredictedresultwhentheChurchlandsemphasize,amongthose“conflictingindications,”theuncongenialrelationsbetweenpsychologyandneuroscience.Revisingtheirextremeeliminativism,theChurchlandssometimesseemtointendthesetwointerpretationstoaddressdifferentstagesintheco-evolutionaryprocess(asIsuggestedabove):first,thedemolitionofmuchcurrentpsychologyviaco-evolutionSfollowedbythereconstructionofaneuroscientificallyinspiredpsychologyviaco-evolutionM.Thecrucialpointfornowisthatthesetwointerpretationsofco-evolutionholdthattherelationshipbetweentwotheorieswill,overtime,shiftinonedirection(asopposedtotheother)ontheChurchlands’continuum.Anobviousquestionarises,though.Ifeitherdirectionispossible,thenwhatarethevariablesthatdeterminethedirectionofanyshift?(Thisquestionpressestherevisedversionofeliminativismnolessthantheoriginal.)TheChurchlandshavenotaddressedthisquestiondirectly,becausetheyhaverecognizedthatthecomplexitiesoftheintertheoreticrelationsinquestionandoftherelationshipofpsychologyandneuroscience,inparticular,requiremore.Enterthethirdinterpretation.OneofPatriciaChurchland’sextendedcommentsabouthergeneralmodelofreduction(1986,pp.296-7)isespeciallyrevealing,sinceitreflectsatvariouspointstheinfluenceofallthreeinterpretations.

...somemisgivingsmaylingeraboutthepossibilityofreductionshoulditbeassumedthatareductivestrategymeansanexclusivelybottom-upstrategy...


Thesemisgivingsarereallyjustbugbears,andtheyhavenoplaceinmyframeworkforreduction. ...ifthereductionissmooth,itsreductiongivesit[thereducedtheory]–anditsphenomena–afirmerplaceinthelargerscheme...Ifthereductioninvolvesamajorcorrection,thecorrected,reducedtheorycontinuestoplayaroleinpredictionandexplanation...Onlyifonetheoryiseliminatedbyanotherdoesitfallbythewayside. ...coevolution...iscertaintobemoreproductivethananisolatedbottom-upstrategy.

Thesecondparagraphtracespointsonthecontinuum.Italludesinitiallytoco-evolutionM–itsfinalsentencetoco-evolutionS.Itisthefirstandthirdparagraphs,though,whereshadowsofathirdinterpretationappear.Closelyrelatedtoco-evolutionMisco-evolutionP(co-evolutionasexplanatorypluralism).Theirmanysimilaritiesnotwithstanding,itisworthteasingthemapart.Asafirstpass,whereco-evolutionManticipatesincreasingintertheoreticintegrationlargelyguidedbyandwithadefaultpreferenceforthelowerlevel,co-evolutionPconstruestheprocessaspreservingadiversesetofpartiallyintegratedyetsemi-autonomousexplanatoryperspectives–wherethatnon-negligiblemeasureofanalyticalindependencerestsateachanalyticallevelontheexplanatorysuccessandtheepistemicintegrityofthetheoriesandonthesuggestivenessoftheempiricalfindings.Co-evolutionM,ineffect,holdsthatselectionpressuresareexertedexclusivelyfromthebottomup,whereasco-evolutionPattendstotheconstraintsimposedbytheneedsanddemandsoftheoriesoperatingathigherlevels.Theseapparentlysmalldifferencesarebutthefringeskirmishesofsomeofthemostbasicepistemologicalandmetaphysicalbattlesinthephilosophyofscience.Spacelimitationsprecludeextensivedevelopment,butbroadly,iftheyarenotpersuadedbyco-evolutionS,physicalistspreferco-evolutionM,sinceitsuggestsascienceunifiedinboththeoryandontologythataccordsprioritytothelower(i.e.,physical)levels.Morepragmaticallymindedphilosophersoptforco-evolutionP,foregoingassurancesofandworriesaboutaunifiedscienceandmetaphysicalpurityinfavorofenhancedexplanatoryresources.FornearlyadecadenowtheChurchlandshavebeennegotiatingtheirinterestsinunifiedscienceandmetaphysicalpurityontheonehandwiththeirinterestsinenhancedexplanatoryresourcesandinternalismontheother.(SeeMcCauley1993andnote10below.)Therelaxationoftheireliminativismandtheiremergingpreferenceforco-evolutionPindicatetheinfluenceofpragmaticcurrentsintheirthought.Co-evolutionPisprominentinNeurophilosophyandevenmoresosince.5PatriciaChurchlandclaimsthat“...thehistoryofsciencerevealsthatco-evolutionoftheorieshastypicallybeenmutuallyenriching,”that“[r]esearchinfluencesgoupanddownandalloverthemap,”that“co-evolutiontypicallyis...interactive...andinvolvesonetheory’sbeingsusceptibletocorrectionandreconceptualizationatthebehestofthecohorttheory,”andthat“psychologyandneuroscienceshouldeachbevulnerabletodisconfirmationandrevisionatanylevelbythediscoveriesoftheother”(1986,pp.363,368,373,and376).Figure6.2.1seemsthemostplausibleinterpretationoftherelationshipbetweenthesethreenotionsofco-evolutionandtheearliercontinuummodel;itroughlyindicatestheregionsofthatcontinuumwherethecasescoveredbythethreetypesofco-evolutionendup.(SeeChurchlandandChurchland1990,p.252.)


Figure6.2.1.Threenotionsofco-evolutionsituatedontheChurchlands’continuum.

Section4willsuggestthatthepictureinFigure6.2.1ofintertheoreticrelationsandofco-evolution,inparticular,isoversimplifiedtothepointofdistortion.Theintertheoreticdynamicsofscientificrevolutionsarequitedifferentfromthoseofapproximatemicroreductionandexplanatorypluralism.Crucially,co-evolutionPisincompatiblewithco-evolutionS.Themutualintertheoreticenrichmentco-evolutionPenvisionswillnotarise,ifneuroscienceisradicallyreconfiguring(letaloneeliminating)psychology.Neitherthehistoryofsciencenorpragmaticaccountsofscientificpracticeoffermuchreasontothinkthatco-evolutionSprovideseitheranaccuratedescriptionorausefulnormfortherelationshipbetweenpsychologyandneuroscienceorforanysuchrelationshipbetweentheoriesinsciencesoperatingatdifferentanalyticallevels.Thedifferencesbetweenco-evolutionP’sandco-evolutionMarealsoimportant.Atstakeisthequestionoftherelativepriorityofneuroscientific(lowerlevel)andpsychological(upperlevel)contributionstothescienceofthemind/brain.Thistopicwilldominatesection5.Incriticizingco-evolutionSandcurtailingco-evolutionM,theaimofthenexttwosectionsis,ultimately,toendorseanddevelopthenotionofexplanatorypluralism.4.ExploringExplanatoryPluralism:DebunkingCo-evolutionsEnlistingadistinctionWimsatt(1976)introducedbetweenintralevelandinterlevelcontexts,Ihavepreviouslydevelopedamodelofintertheoreticrelationsthatdiscloseswhyweshouldnotexpectadvancesinneurosciencetoeliminatemuchpsychologydirectly(McCauley1986).Moregenerally,itsuggeststhatco-evolutionSdoesnotveryhappilymodeltheco-evolvingrelationsoftheoriesatdifferentlevels.Thesortsofunequivocaleliminationsoftheoriesandontologiesthatco-evolutionScountenancesariseinintralevelcontextsinvolvingconsiderableincommensurability.Thesecontextsconcernchangeswithinaparticularscienceovertime.Theyincludetheclassiccasesthatphilosophersgroupundertherubricof“scientificrevolutions”–impetus,phlogiston,caloricfluid,andthelike.Withinaparticularlevelofanalysissomenewlyproposedtheoryprovessuperiortoitsimmediatepredecessorwithwhichitissubstantiallydiscontinuous.Whenthescientificcommunityoptsforthisnewtheory,mosttracesofitspredecessorrapidlydisappear.Sincetheyofferincompatibleaccountsofmanyofthesamephenomena,thenewtheoryexplainstheoldtheoryaway.Bycontrast,intralevelsituationswherethemappingsbetweentheoriesarereasonablygoodfallneartheotherendoftheChurchlands’continuum.Herethenewtheoryexplainsitspredecessorwhichitalsotypicallycorrects.Scientistsregardtheearliertheory’sdomainasaspecialcasetowhichthenewtheoryappliesandforwhichtheoldtheorycontinuestosufficeasauseful


calculatingheuristic.Althoughcorrectedandincorporatedasaspecialcaseintoamoregeneraltheory,Newton’slawsofmotionworkwellformostpracticalpurposes.Anewtheorydisruptssciencelesstotheextentitpreserves(ratherthanoverthrows)thecherishedinsightsandconceptualapparatusofitspredecessors.Itmayrequirereinterpretationofestablishednotions(“planets,”“genes,”“grammaracquisition,”etc.),butchangesareevolutionaryonlywhentheypreserveafairmeasureofintensionalandextensionaloverlapwiththeirpredecessors.Whensucceedingtheoriesinsomesciencearelargelycontinuous,noonespeaksofelimination.Thechangeisevolutionary,notrevolutionary.Consequently,thenewtheoryisperfectlycapableofprovidinganequipotentimageoftheold.Thesearethecaseswherethenewtheoryoverwhelminglyinheritstheevidencefortheold.Revolutionaryorevolutionary,progresswithinsomescienceeliminatesfeaturesofearliertheorieseventually.Inrevolutionarysettingsthechangesareabruptandtheeliminationis(relatively)immediate.Inevolutionarycontextsincompatibilitiesaccrueovertime.Althoughthetransitionfromonetheorytoitsimmediatesuccessormaybemoreorlesssmooth,overaseriesofsuchtransitionsalltracesofancestraltheoriesmaycompletelydisappear.Considerthefateof“naturalmotions”fromAristotelianthroughNewtonianmechanics(McCauley1986,pp.192-93).Similarly,overthepasthundredyearsthe“memorytrace”hasundergoneconsiderableevolutionarytransformation.Sometheoristswouldarguethatthereinterpretationshavebeensosubstantialthattheoriginalnotion(andwhatitallegedlyreferredto)hasvirtuallyvanished.6Interlevelrelationsconcerntheoriesatdifferent(typicallyneighboring)levelsofanalysisataparticularpointintime(incontrasttointralevelcasesconcernedwithsuccessivetheoriesatthesamelevelofanalysis).TheChurchlands’continuummapsontointerlevelcasestoo.Whensciencesatadjoininglevelsenjoysubstantialintertheoreticmapping(insituationsapproximatingclassicmicroreductions)theyheavilyconstrainoneanother’sform–otherwise,whywouldanyonehaveattemptedtocharacterizetheirrelationsintermsofdeductivelogicandidentitystatements?Thisistheeffectoftheknittingoftwosciencesintooneanotherthatco-evolutionMenvisions.Awellintegratedlowerleveltheoryhasresourcessufficienttoreproducetheexplanatoryandpredictiveaccomplishmentsofthecorrespondingupperleveltheory,however,thisoftencomesatconsiderablecomputationalexpense.AstheChurchlandshaveemphasized,thisdoesnotdisgracethehigherleveltheorynorleadtotheevaporationofthephenomenaitseekstoexplain.Whenconsideringinterlevelcaseswithrelativelyunproblematicintertheoreticrelations,theChurchlands,likethetraditionalreductionistsbeforethem7,havefocusedexclusivelyontheirresemblancestotheintralevelsettingsdescribedabove.(See,forexample,P.S.Churchland1986,p.294.)Afterall,heretootheelaborationsoftheupperleveltheory’scentralconceptsthatthelowerlevel,reducingtheoryoffersoftencorrectthelessfine-grained,upperleveltheory’spronouncements.However,becausethetheoriesaretightlyknit,theupperleveltheorystillprovidesausefulandefficientapproximationofthelowerleveltheory’sresults.Thissoundsquitelikethecasesofscientificevolutiondescribedabove.Beneaththeseresemblances,though,liesmallbutrevealingdifferences.First,unliketheevolutionaryintralevelcases,thereducedtheoryininterlevelsituationsdoesnotstandinneedoftechnicalcorrectionineverycase.Forafewsituationsatleast,itsresultswillconformpreciselywiththoseofthelowerleveltheory,becauseforthesecasesitadequatelysummarizestheeffectsofallrelevantlowerlevelvariables.8Thiscontrastswiththeinescapable,ifoftennegligible(froma


practicalstandpoint),divergenceofthecalculationsofsometheoryanditssuccessor,suchasclassicalmechanicsandthemechanicsofrelativity.(SeeChurchlandandChurchland1990,p.251.)Ininterlevelcasescorrectionscanarisebecausetheupperleveltheoryisinsufficientlyfine-grainedtohandlecertainproblems.Bycontrast,inintralevelcasescorrectionsalwaysarisebecausetheearliertheoryiswrong–byalittleinevolutionarycases,byalotinrevolutionaryones.Itfollowsthattheupperleveltheoryisnotalwaysamerecalculatingheuristic(asthereplacedpredecessorisincasesofscientificevolution).Moreover,theupperleveltheory’sheuristicadvantagesinwellintegratedinterlevelcontextsaretypicallyenormous,comparedwithintralevelcases.Thedivergenceofcomputationaleffortbetweentheclassicalandstatisticalsolutionsforsimpleproblemsaboutgases(aninterlevelcase)dwarfsthatbetweenclassicalmechanicsandthemechanicsofrelativityforsimpleproblemsaboutmotion(anintralevelcase).Ofapiecewiththisobservation,theChurchlandsquiteaccuratelydescribethequantumcalculationsofvariouschemicalproperties(anotherinterlevelcase)as“daunting”(1990,p.251).Finally,theupperleveltheorylaysoutregularitiesaboutasubsetofthephenomenathatthelowerleveltheoryencompassesbutforwhichithasneithertheresourcesnorthemotivationtohighlight.Thatisthepriceofthelowerleveltheory’sgeneralityandfinergrain.Iftheseconsiderationsarenotcompelling,scrutinyofinterlevelcircumstancesthatsupportrelativelylittleintertheoreticmappingrevealsfarmoreimportantgroundsforstressingthedistinctionbetweeninterlevelandintralevelsettings.Heretwosciencesatadjacentlevelsaddresssomecommonexplanandaunderdifferentdescriptions,buttheirexplanatorystoriesarelargely(thoughnotwholly)incompatible.OntheChurchlands’view,thisisjusttherelationshipbetweenneuroscienceandmostoffolkpsychology,andifrememberingistogothewayofimpetus,therelationshipbetweenneuroscienceandsomeimportantpartsofscientificpsychologyaswell.Ifalloftheseintertheoreticrelationsshouldreceiveaunifiedtreatment,astraditionalreductionists,theChurchlands(e.g.,ChurchlandandSejnowski1990,p.229),andFigure6.2.1suggest,thenitisperfectlyreasonabletoexpecteliminationinthoseinterlevelsituationsinvolvingsignificantincommensurability.Theproblem,though,isthatneitherthehistoryofscience,norcurrentscientificpractice,northescientificresearchtheChurchlandschampion,noraconcernforexplanatorypluralismoffersmuchreasontoexpecttheoryeliminationinsuchsettings.Incommensurabilityininterlevelcontextsneitherrequirestheeliminationoftheoriesonprincipledgroundsnorresultsinsucheliminationsinfact.Admittedly,intheearlystagesofascience’shistoryitisnotalwayseasytodistinguishlevelsofanalysisand,consequently,todistinguishwhatwouldcountasaninterlevel,asopposedtoanintralevel,elimination.Crucially,though,thehistoryofscienceandespeciallythehistoryoflatenineteenthandtwentiethcenturyscienceoffernoexamplesoflarge-scaleinterleveltheoryelimination(particularlyofthewholesalevarietystandardeliminativismandco-evolutionSenvision)oncetheupperlevelscienceachievessufficienthistoricalmomentumtoenjoytheaccoutrementsofotherrecognizedsciences(suchascharacteristicresearchtechniquesandinstruments,journals,universitydepartments,professionalsocieties,andfundingagencies).Thereasonissimpleenough.Maturesciencesarelargelydefinedbytheirtheoriesand,moregenerally,bytheirresearchtraditions(Laudan1977),hence,eliminationofanupperleveltheorybyalowerleveltheorymayrisktheeliminationoftheupperlevelscientificenterprise!(Presumably,thisiswhyNagelalwaysspokeofthereductionofascience,ratherthanofatheory,whenaddressinginterlevelcases.)Amotiveforundertakinginterlevelinvestigation(especiallywhentheintertheoreticconnectionsarenotplentiful)istoexploreonescience’ssuccessfulproblemsolvingstrategiesasameansofinspiringresearch,provokingdiscoveries,andsolvingrecalcitrantproblemsatanotherlevel.


(BechtelandRichardson1993focusinparticularontheproblemofunderstandingtheoperationofmechanisms.)Monitoringdevelopmentsintheoriesatneighboringlevelsisoftenafruitfulheuristicofdiscovery.Thestrategy’sfruitfulnessdependspreciselyonthetwosciencesmaintainingameasureofindependencefromoneanother.Thisisthemarkofexplanatorypluralismandco-evolutionP.Apaucityofinterlevelconnectionsonlyenhancesthe(relative)integrityandautonomyoftheupperlevelscience.AsWimsattnotes“ininterlevelreduction,themoredifficultthetranslationbecomes,themoreirreplaceabletheupperleveltheoryis!Itbecomestheonlypracticalwayofhandlingtheregularitiesitdescribes”(1976,p.222).Thetheoriesatthetwolevelspossessdifferentconceptualandexplanatoryresources,whichunderscoredifferentfeaturesoftheircommonexplanandum.Theyprovidemultipleexplanatoryperspectivesthatshouldbejudgedonthebasisoftheirempiricalsuccess--notonhopesabouttheirputativepromiseforthetheoretical(orontological)unificationofscience.Forthepragmaticallyinclined,explanatorysuccessisbothsufficientlyvaluableandrarethatitwouldbeimprudenttoencouragetheeliminationofanypotentiallypromisingavenueofresearch.AsChurchlandandSejnowskiremark,“theco-evolutionaryadviceregardingmethodologicalefficiencyis‘letmanyflowersbloom’”(1992,p.13).TheChurchlandshavearguedfamously,though,thatfolkpsychologyisbarren(P.S.Churchland1986,pp.288-312andP.M.Churchland1989,pp.2-11).Thoseargumentshaveprovokedanentireliteratureinresponse(seeGreenwood1991andChristensenandTurner1993).IamsympatheticwiththeChurchlands’arguments,atleastwhentheywieldthemagainstpositionsinthephilosophyofmindthatdenytheexplanatorygoalsandtheconjecturalandfalliblecharacteroffolkpsychology.Thatfolkpsychologyoffersexplanationsandthatitisconjecturalandfalliblearebothcorrect.Thatisjustnotthewholestory,though.Thepivotalquestionforapragmatistiswhetherfolkpsychologycancontributetotheprogressofourknowledge,or,better,whetherfolkpsychologycontainsresourcesthatmayaidsubsequent,moresystematicpsychologicaltheorizing.Attributiontheory,thetheoryofcognitivedissonance,andotherproposalswithinsocialpsychologyemployasrichversionsofthepropositionalattitudesasdoesfolkpsychology(BechtelandAbrahamsen1993).Moreover,asDennett(1987)hasemphasized,employingtheintentionalstanceaidstheorizingaboutoperativesubsystemsinsub-personalcognitivepsychology.9Thesearejusttwofrontswherepsychologicalscienceseemstobesimultaneouslyemployingand,eversogradually,transformingfamiliarfolkpsychologicalnotions.Arguably,then,theChurchlandsmayhaveunderestimatedthepossiblecontributionoftheresourcesoffolkpsychology,becausetheyhavebeeninsufficientlyattentivetotheirroleinsocialpsychologicalandcognitivetheorizing(McCauley1987,1989).Indeed,theysometimesdisregardthepsychologicalaltogether.10(See,however,note14below.)Isuspectthatsuchneglectisbornofinsistingonaunifiedaccountofintertheoreticrelationsandofentertainingimagesofco-evolutionS,inparticular.TheChurchlandsarecorrecttoemphasizethesalientroleoftheoryeliminationinscientificprogress,buttheseeliminationsareintralevelprocessesandmostunivocallyso(1)whenthelevelsinquestionconcernscientificpursuitsaswellestablishedasneuroscienceandpsychologyand(2)whenthoselevelsareconstruedasthickly,i.e.,asinclusively,asthedistinctionbetweenthosetwosciencesimplies.ThetheoriesandcharacteristicontologiesinformingStahl’saccountofcombustionandYoung’saccountofthepropagationoflightwerereplacedbytheories(withnewontologies)thatoperatedatthesamelevelsofanalysisandthatwereidentified,bothnowandthen,ascontinuationsoftheresearchtraditionsassociatedwiththoselevels.Eliminationinscienceisprincipallyanintralevelprocess.


Thatisnottoassertthatinterlevelconsiderationsplaynorole.Evenwithlevelsofanalysissothicklyconstrued,Idonotmeantodenythatscientists’decisionsatlevelsaboveandbelowinfluencetheoreticaldevelopmentsatagivenlevel.NordoIwishtodenythatatthattargetedlevelsuchdevelopmentscaninvolveeliminations.Rather,thecriticalpointisthattheseinfluencesarereliablymediatedbydevelopmentsintheconceptualapparatusandresearchpracticesthatareassociatedwiththeresearchtraditionofthetargetedlevel.(SeeBechtelandRichardson1993,especiallychapter8andBechtel,1996.)Ifitisconstruedasanexplanatoryconstruct,then,IagreewiththeChurchlandsthatmuchoffolkpsychologymaywellundergosubstantialrevisionand,perhaps,eveneliminationeventually.11WhatIamsuggesting,though,is:(1)thatthosechangeswilloccurprimarilyasaresultofprogresswithinsocialandcognitivepsychology,i.e.,thattheywillariseastheconsequenceofintralevelprocesseswithinthepsychologicallevelofanalysis;(2)that,invirtueoftheroleofintentionalattributionsinthetheoriesofsocialandcognitivepsychology,thisdisplacementwillprobablybequitegradual,i.e.,that,sofar,thechangesareprovingevolutionary,notrevolutionary;(3)thattheoreticaldevelopmentswithinthosesub-disciplinesofpsychologywillmediatewhateverco-evolutionaryinfluenceneurosciencehasinthisoutcome.MappingtheChurchlands’continuumontotheintralevel-interleveldistinctionyieldsthearrangementinFigure6.2.2.Itreadilyaccommodatesco-evolutionMandco-evolutionP,butco-evolutionSfindsnoobvioushome.Thepointisthattheinteractionofpsychologyandneuroscience,likeallco-evolutionarysituations,isacaseofinterlevelrelations.Inshort,co-evolutionSembodiesacategorymistake.Itconflatesthedynamicsoftheco-evolutionoftheoriesatdifferentlevelsofanalysiswiththoseofscientificrevolutions,whichareintralevelprocesses.12

Figure6.2.2.Mappingdegreesofintertheoreticcontinuity(theChurchlands’continuum)ontointralevelandinterlevelcontexts.

WhatfollowsonthisrevisedpictureinFigure6.2.2aboutinterlevelcasesthatreflectsubstantialincommensurability?Infact,Ithinksuchcasesareextremelyrare,especiallyifthesciencesin


questionarewellestablished,sincepartofbecomingawell-establishedscienceispreciselytopossesstheoriesthatrecognizablycoherewithatleastsomefeaturesoftheoriesatcontiguouslevels.Arguably,thedistinctionsbetweenlevelsofanalysisalreadypresumetheextremeimprobabilityofsuchradicalincompatibilitybetweentheoriesoperatingatadjoiningscientificlevels.(Ofcourse,notallexplanatorytheoriesarescientifictheories.)Ifanalysesdivergeinnearlyallrespects,thenitmaynolongerbeclearthattheyshareacommonexplanandum–temptingsomeresearcherstoadoptobscurantiststrategiesofmetaphysicalextravagance.13Theproblemssurroundingco-evolutionSnotwithstanding,inelaboratingWimsatt’smetaphoroftheco-evolutionoftheoriestheChurchlandshavefundamentallyreinvigoratedthestudyofchangeininterlevelrelationsovertime(arguablyinitiatedinSchaffner1967).Aswiththissection,thenextwillsaymoreaboutco-evolutionPbyopeningwithmoreaboutwhatitisnot.5.ExploringExplanatoryPluralism:BeyondCo-evolutionmThedemandinscienceforcoherenceoftheoriesatadjacentlevelsofanalysisisanadditionalmotive,beyondthepromiseofnewdiscoveries,forprobingpossibleinterlevelconnections.Themotiveistoascertainwhetherornotresearchatnearbylevelscohereswithandsupportsscientists’findings,andifitdoesnot,toexplorepossibleadjustmentstoincreasetheprobabilityofsuchmutualsupport.Thiscan,amongotherthings,clarifyrespectsinwhichthetwosciencesshareacommonexplanandum.Inthelongtermscientists’concernforcoherenceamongtheirresultsinevitablytendstoencouragebetterintertheoreticmappingininterlevelsettings.Forgingsuchconnectionsproducesnewdiscoveriesintherespectivesciences.Onestrategy,thoughcertainlynottheonlyone,istoadvancehypotheticalidentitiesbetweentheoreticalontologiesinordertopoweranengineofdiscovery.TherelationshipbetweenMendeliangeneticsandbiochemicalgeneticsoverthefirsthalfofthiscenturyisanespeciallyaptillustrationoftworelatedresearchprogramsatneighboringlevelsofanalysisaidingoneanotherthroughtheinvestigationofaseriesofproposalsaboutwhichstructureswere,infact,thegenes.Scientists’twoprimarymotivesforinquiriesintoresearchatneighboringlevels,then,arefinallyoneandthesame.Thismightseemtosuggestthatco-evolutionMpredominates;however,anumberofcountervailingconsiderations(someofwhicharebrieflyexaminedinthissection)favoranexplanatorypluralismwherethesciencesmaintainsomeindependenceoftheory,method,andpractice.So,evenapproximatemicroreductionneednotbeinevitable.Twoissuesespeciallydistinguishco-evolutionMandco-evolutionP.Thefirstconcernstherelativemetaphysical,epistemic,and/orexplanatorypriorityofupperandlowerleveltheoriesintheco-evolutionaryprocess.Thesecondconcernsthegroundsofferedforanydisparateassignmentsofthesepriorities.Thedefaultassumptionadoptedinananalysisofco-evolutionMthataccordswiththetraditionalmicroreductionisticrationaleforphysicalismattributescomprehensiveprioritytolowerlevels.Classicalmicroreductionwouldforecastaco-evolutionaryprocesswheretheoverwhelmingmajorityoftheselectionpressuresareexertedfromthebottomup.Theupperleveltheorymaycontributeintheprocessofdiscovery,providinganinitialvocabularyandproblemsforresearch,butsoonerorlateritmustconformtothelowerleveltheory’sexpectations.Herethegroundsforthispriorityrestnotmerelyonthetheoreticalmaturityandsuperiorprecisionlowerleveltheoriestypicallyenjoy(withwhichpragmatismhasnocomplaint)butalsoonpresumptionsaboutthosetheories’metaphysicalpreeminence.(Seenote10above.)


Occasionally14,theChurchlandsseemtosubscribetoaversionofco-evolutionMthatresemblesthisposition.Forexample,ChurchlandandSejnowskiemphasize“theimportanceofthesingleneuronmodels[amongthevarioussub-levelsofanalysiswithinneuroscience]asthebedrockandfundamentintowhichnetworkmodelsmusteventuallyfit”(1992,p.13,emphasisadded).15AlthoughtheChurchlandshaveavoidedthetraditionalmicroreductionists’fervoraboutthereplaceabilityofthereducedtheoryattheupperlevel(e.g.,ChurchlandandChurchland1990,p.256),theirrepeatedemphasisonlowerleveltheories’correctionsofupperleveltheoriesalsosuggeststhatselectionpressuresarelargelyunidirectional,especiallywhentheytreattheselowerlevelelaborationsasofapiecewithcorrectionsinintralevelcontextswheresubstantialontologicalmodificationissometimespartofthepackage.Co-evolutionMwillproverelevanttobutasmallpercentageofcases,atbest.Ontheonehand,ifco-evolutionMissupposedtoissueintheclassicalmicroreductionistprogram(presumably,itisnot),thenallofthefamiliarobjectionsandcaveatsapply–plusatleastoneimportantadditionalone.Thesortoftightintegrationwithadominantlowerleveltheorytowhichclassicalmicroreductionaspiresmustinevitablyrestrictresearchatthehigherlevel.Ifthereeverwasamicroreductionthatconformedtoallofthelogicalandontologicalconstraintsimposedbytheclassicalmodel,forexampleCausey’s(1977)version,itwouldendowthelowerlevelwithanexplanatoryandmetaphysicalprioritythatwoulddiscourageallmotivesfortheoreticalnoveltyatthehigherlevel.Itwouldencourageonlythosepathsofresearchatthehigherlevelthatpromisedtopreserveitstightfitwiththetheoryatthelowerlevel.Itseffect,inshort,wouldbetocheckimaginativescientificproposals.Ontheotherhand,ifco-evolutionMissupposedtoresultonlyintheweakeranalogicalrelationtowhichtheChurchlands’modelofapproximatereductionlooks,thenthepointsofreductivecontactmayprovelessextensivethantheknittingmetaphorsuggests,andthemicroreductionistcasefortheexplanatory,epistemic,andmetaphysicalpriorityoflowerlevelsendsupseemingsomewhatlesscompelling,especiallyoncewehaveteasedapartthedifferencesinthe“corrections”thatoccurininterlevelandintralevelcontexts.Thecaseforco-evolutionP,however,doesnotturnexclusivelyontheproblemsthetwocompetingconceptionsface.Scrutinyofactualcases,includingthoseincognitiveneurosciencetowhichtheChurchlandshavedevotedparticularattention,stronglysuggeststhattheoutcomeoftheco-evolutionoftheoriesisusuallyasco-evolutionPdescribes.Insteadofdrivinginexorablytowardcomprehensivetheoreticalandpracticalintegrationwherethelowerleveltheorygoverns,scientificopportunismisusuallyclosertothetruthinmostinterlevelforays.Atleastinitially,scientistsperiodicallymonitordevelopmentsatnearbylevelssearchingforeitherinterlevelsupport,tantalizingfindings,orboth.ChurchlandandSejnowski’ssurveyofproposalsconcerningtheneuralbasisofworkingmemoryisafittingillustration(1992,pp.297-305).Notonlydidtheconceptof“workingmemory”emergeoutoftheoreticaldevelopmentsinexperimentalpsychology,butsodidmanyofthefindingsthatguideneuralmodeling.Forexample,ChurchlandandSejnowskipointexplicitlytothediscoveryofashorttermmemorydeficitforverbalmaterialsinsomesubjects.Theyalsohighlighttheabilityofvariousinterferenceeffectsbothtodissociateworkingmemoryfromlongtermmemoryinnormalsubjectsandtodissociatesubsystemsofworkingmemory(linkedwithauditory,visuospatial,andverbalmaterials)fromoneanother.Thesediscoveriesinexperimentalpsychologyprovidedbothinspirationanddirectionforneuralmodeling.Theyalsoconstituteasetoffindingsthatanyrelevantneuroscientificproposalshouldmakesenseof.


Oneventhemostexactingphilosophicalstandards,thislastconsiderationisepistemicallysignificant.Theoreticalproposalsandtheresearchtheyspawnatthehigherleveldonotmerelycontributetotheprocessofdiscoveryatthelowerlevel.Theupperlevelscienceprovidesabodyofevidenceagainstwhichthescienceatthelowerlevelcanevaluatecompetingmodels.Thisevidenceisparticularlyuseful,preciselybecauseitfrequentlyarisesindependentlyoftheformulationofthespecificlowerlevelmodelstowhoseassessmentitcontributes.Ithelpstoassuretheindependenttestabilityofthemodelsinquestion.Ithasbeenwidelyconcededthatupperleveltheoriescanplayacatalyticroleintheprocessofdiscoveryatthelowerlevel.Indeed,sometimestheconceptualresourcesandresearchtechniquesofalowerlevelsciencearebasicallyinsufficienttoenablepractitionerseventorecognizesomeofthatlevel’sfundamentalphenomenawithoutaidanddirectionfromanupperlevelscience.(Lykkenetal.(1992)constitutesaparticularlyintriguing,recentillustration.)Intheprevioussectionwealsosawhowmicroreductionisticproposalstosubordinateupperlevelexplanationstolowerlevelexplanationsriskneedlesslydownplayingvaluableresourcesfordealingwiththeoftenhugecomputationalburdenslowerleveltheoriesentail.Upperleveltheorizing(e.g.,intransmissiongenetics)contributesusefullytoeverydayscientificproblemsolving,evenafterlowerlevelresearch(e.g.inmoleculargenetics)indicatesthemicrolevelstoryisfarmorecomplicated.ScientificendeavorsatdifferentlevelsregularlydisplaywhatRobertBurton(1993)hascalleda“strategicinterdependence.”Nowwecanseethatupperleveltheorizingalsoinitiatesresearchthatcancontributetolowerleveldevelopmentspertainingdirectlytojustification.Microreductionisticproposalstosubordinateupperlevelsciencestolowerlevelscienceseitherepistemicallyormetaphysicallyriskneedlessevidentiaryimpoverishment.Thevalueofthisevidenceturnspreciselyonthefactthattheresearcharosewithinacontextofscientifictheorizingandinvestigationsufficientlyremovedandsufficientlyautonomousofthelowerlevelresearchtoinsureanhonestcheck.Thesepsychologicalfindingsdonotoccurinisolation.Theyariseinthecourseofon-goingtheorizingandresearchatthepsychologicallevel.Theirvaluetoneurosciencerestsinpartonthefactthattheyemanatefromatraditionofpsychologicaltheorizingandexperimentationthatneurosciencehasnotdominated.Thisiswhyitisworthwhileforeachlevelofanalysistomaintainameasureofindependence.AsChurchlandandSejnowskinote,experimentalpsychologyhasacenturyoffindings(andtheorizing)fromwhichneuroscientistsandneurocomputationalmodelersmaydraw(1992,p.27;seetoop.240).NothingmoreclearlyillustratesthesortofscientificopportunismexplanatorypluralismenvisionsthanoneofSejnowskiandRosenberg’spapers(1988)indefenseoftheclaimthatNETtalkplausiblymodelsoperativeprocessesinhumanlearningandcognition.(Itisafairquestionatwhatlevelofanalysisconnectionistmodelingshouldbelocated.OnthecriteriaIidentifiedinsection2,itseemstooccuratalevelbelowthatofsocialandcognitivepsychology.ChurchlandandSejnowskiclearlyregarditasaformofneurocomputationalmodeling.ItisworthnotingthatMichaelGazzanigaplacesSejnowskiandRosenberg’s(1988)inthefirsthalfofhisbook,whichconcerns“NeurobiologicConsiderationsinMemoryFunction”ratherthaninthesecondhalf,whichconcerns“PsychologicalDimensionsofMemoryFunctioninHumans.”)NETtalkisaconnectionistsystemthatconvertsEnglishtextintostringsofphonemes.(SejnowskiandRosenberg1987)Itisathreelayer,feed-forwardnetworkthatemploysthestandardback-propagationlearningalgorithm.OnanygiventrialNETtalkreceivesseveninputscorrespondingtoawindowofsevenletters(includingpunctuationorspacesbetweenwords,iftheyhappentoarise).Thedesiredoutputisthecorrectphonemeassociatedwiththefourthiteminthewindow.The


threeplacesoneithersideofthefourthitemprovidethenetworkwithinformationabouthowcontextaffectspronunciation.NETtalk’sperformanceisnothingshortofremarkable.ItcapturesmostoftheregularitiesinEnglishpronunciationandmanyoftheirregularitiesaswell.After50,000trainingtrialswithwords,itsaccuracywithphonemesapproaches95percentanditisvirtuallyperfectwithstressesandsyllableboundaries.Thecriticalquestionfornow,though,iswhatevidenceSejnowskiandRosenbergmightcitetosupporttheclaimthatNETtalkmodelsprocessesthatresemblethoseinvolvedinhumanlearningandcognition.Amodelofco-evolutionasexplanatorypluralismsuggeststhatattentiontothefindingsofexperimentalpsychologymightprovejustashelpfulhereasattentiontoresearchonneuralstructure,and,infact,notonlydoSejnowskiandRosenberglooktopsychology,theylooktooneofthosecenturyoldfindingsaboutremembering,viz.,thespacingeffect.Thespacingeffectisthefindingthatdistributedpracticewithitemsenhancestheprobabilityoftheirlongtermretentionmorethanmassedpracticedoes.Ifoccasionsforrehearsalarespacedoutovertime,theprobabilityishighthatmemoryperformancewillexceedthatfromemployingsomesmallnumberofmassedpracticesessionsofcomparabledurationattheoutset.Massedrepetitionfacilitatesmemorywhenretentionintervalsareextremelyshort.Inpracticalterms,thespacingeffectiswhycrammingforanexamisnotnearlysohelpfulasregular,dailypreparation,whereasretentionoftwonewtelephonenumberssuppliedbyDirectoryAssistancerequiresimmediate,massedrehearsal,iftheycannotbewrittendown.Inthecourseofinvestigatingthevarioushypothesespsychologistshaveofferedforexplainingthespacingeffect,researchershavedemonstrateditsrobustnessacrossahugevarietyofexperimentalsettings,materials,andtasks.Thus,SejnowskiandRosenbergsuspectthatitreflects“somethingofcentralimportanceinmemory”(1988,p.163).Consequently,itisbynomeanstrivial,ifNETtalkcanbeinducedtoexhibitthespacingeffect.Itwouldbeevenmorestriking,ifitsexhibitionoftheeffectwassimilarinformtodocumentedhumanperformance.BecauseofNETtalk’sarchitecturetheobviouscomparisoniswithstudiesofcuedrecall.SejnowskiandRosenbergchoseadesignafterGlenberg(1976).ThedesigncalledfortrainingNETtalkupinthestandardfashion,andthenpresentingitwiththecuesfromalistoftwentypairedassociateswherethosecueswerestringsofsixrandomlettersandtheirassociatedresponseswererandomlygeneratedphonemeandstressstringssixcharacterslong.(ThisinsuredthatNETtalk’sperformanceatthetestcouldnotbeafunctionofanyinformationithadacquiredaboutEnglishpronunciation.)Duringboththespacingintervalbetweentrainingopportunitiesandtheretentionintervalbeforethetest,NETtalkwaspresentedwithEnglishdistractorwordsthatwerepartofitsoriginaltrainingcorpus.Bothtrainingonthepairedassociatesanddistractorepisodesincludedfeedbackviabackpropagation.TheorderofthepresentationstoNETtalkintheexperimentwasasfollows:(1)2,10,or20presentationsofeachofthetwentypairedassociatecues;(2)aspacingintervalof0,1,4,8,20,or40distractors;(3)2,10,or20re-presentationsofeachofthetwentypairedassociatecues;(4)aretentionintervalof2,8,32,or64distractors;(5)atestofNETtalk’saccuracyincuedrecallofthetwentypairedassociates.Inshort,NETtalkdisplayedthespacingeffect:“AsignificantspacingeffectwasobservedinNETtalk:Retentionofnonwordsaftera64-itemretentionintervalwassignificantlybetterwhenpresentedatthelongerspacings(distributedpresentation)thanattheshorterspacings.Inaddition,a


significantadvantageformassedpresentationswasfoundforshort-termretentionoftheitems”(SejnowskiandRosenberg1988,p.167).Moreover,althoughdirectcomparisonwasimpossible,NETtalk’soverallresponseprofileresembledthatofGlenberg’shumansubjects.Theinterlevelinteractionherebenefitsbothcognitivepsychologyandneurocomputationalmodeling.SejnowskiandRosenbergbrieflyreviewthetwomajortheoreticalproposalsforexplainingthespacingeffectincognitivepsychology,pointingoutthatneithertheencodingvariabilityhypothesis(e.g.,Bower1972)northeprocessingefforthypothesis(e.g.,Jacoby1978)canaccountforalloftheavailabledata.Theythensuggestafurtherhypothesisfocusingontheforminwhichinformationisencodedinaconnectionistnetwork,i.e.,ontheformofthememoryrepresentation.Theyproposethattheshort-termadvantageofmassedpracticeand,particularly,thelongertermadvantageofdistributedpracticeareatleastpartiallyexplicableintermsofthedynamicsofconnectionistnets.Crucially,SejnowskiandRosenbergdonotconstruetheirhypothesisascompetingwith(letalonecorrectingoreliminating)thetwopsychologicalproposals.(Theyhave,afterall,exploredbutonesetoffindingsconcerningcuedrecall.)Instead,theyemphasizeitscompatibilitywitheach.Theyclaimcorrectlythatitoffers“adifferenttypeofexplanation”at“adifferentlevelofexplanation”(1988,p.170).Theyexplicitlydiscusswaysinwhichthenotionsof“encodingvariability”and“processingeffort”couldmapontothedynamicsofconnectionistnetworks.Thesefinergrainedaccountsoftheseprocessesintermsofanetwork’soperationssuggestbasesforelaboratingthetwohypotheses.Iftheco-evolutionofresearchininterlevelcontextsyieldstheexplanatorypluralismforwhichIhavebeenplumping,thenitisnotonlythelowerlevelthatofferstheaidandcomfort,norisitonlythehigherlevelthatreceivesit.Astheneuralmodelingofworkingmemoryillustrates,heretoopsychologicalfindingsprovidebothevidentiarysupportandstrategicguidancetolowerlevelmodelingofbrainfunctioning.SejnowskiandRosenbergremarkthat“thoseaspectsofthenetwork’sperformancethataresimilartohumanperformancearegoodcandidatesforgeneralpropertiesofnetworkmodels”(1988,p.171).Theirprojectreflectsageneralstrategyforthetestingandrefinementofneurocomputationalmodelsthatreliesontherelativeindependenceofworkinexperimentalpsychology.Featuresofparticularnetworksthatenablethemtomimicaspectsofthehumanperformancethatpsychologydocumentsthemselvesdeservemimicryinsubsequentmodelingofhumancognition.WhatisespeciallyclearaboutthecontributionofhigherlevelsinthisexampleisSejnowskiandRosenberg’sexplicitacknowledgementofjusthowfar“guidance”cango.“WhenNETtalkdeviatesfromhumanperformance,thereisgoodreasontobelievethatamoredetailedaccountofbraincircuitrymaybenecessary”(1988,pp.172).TheircommentaccordsnicelywiththeaccountofexplanatorypluralismIhavebeendeveloping.Unlikethepictureofco-evolutioninspiredbythetraditionofmicroreductionism,apragmaticallyinspiredexplanatorypluralismpermitsnoaprioripresumptionsaboutlowerlevelpriority.SejnowskiandRosenbergreadilyallowthatourpsychologicalknowledgeenjoyssufficientintegritytoforcefullyurgefurtherelaborationofanalysesofbrainsystemsformulatedatlowerlevels.16Thiswouldbenoless(nornomore)acorrectionofthelowerleveltheory(oritsontology)thanarethelowerlevel“corrections”ofupperleveltheories(andtheirontologies)theChurchlandshavesometimesbeenwonttostress.Suchdivergences,then,arenotgroundsfordismissal.Theyare,rather,opportunitiesforadvance.Theco-evolutionofsciences(notjusttheories)atcontiguouslevelsofanalysispreservesthe


pluralityofexplanatoryperspectivesthatthedistinctionsbetweenlevelsimply,becauseleavingtheseresearchtraditionstotheirowndevicesisaneffectivemeansofinsuringscientificprogress.ReferencesBechtel,W.(ed.)1986:TheNatureofScientificIntegration.IntegratingScientificDisciplines,TheHague:MartinusNijhoff.Bechtel,W.1996:Whatshouldaconnectionistphilosophyofsciencelooklike?InR.N.McCauley(ed.),TheChurchlandsandtheircritics,Oxford:BasilBlackwell.Bechtel,W.andAbrahamsen,A.A.1993:ConnectionismandtheFutureofFolkPsychology.InR.G.Burton(ed.),NaturalandArtificialMinds,Albany:SUNYPress.Bechtel,W.andRichardson,R.C.1993:DiscoveringComplexity.Princeton:PrincetonUniversityPress.Bickle,J.1992:MentalAnomalyandtheNewMind-BrainReductionism.PhilosophyofScience,59,217-230.Bower.G.H.1972:Stimulus-SamplingTheoryofEncodingVariability.InA.W.MeltonandE.Martin(eds),CodingProcessesinHumanMemory,Washington:V.H.Winston&Sons.Burton,R.G.1993:Reduction,Elimination,andStrategicInterdependence.InR.G.Burton(ed.),NaturalandArtificialMinds,Albany:SUNYPress.Causey,R.1972:UniformMicroreductions.Synthese,25,176-218.Causey,R.1977:UnityofScience.Dordrecht:Reidel.ChristensenS.M.andTurnerD.R.(eds.)1993:FolkPsychologyandthePhilosophyofMind.Hillsdale,NewJersey:LawrenceErlbaumAssociates.Churchland,P.M.1979:ScientificRealismandthePlasticityofMind.Cambridge:CambridgeUniversityPress.Churchland,P.M.1989:ANeurocomputationalPerspective:TheNatureofMindandtheStructureofScience.Cambridge:TheMITPress.Churchland,P.M.andChurchland,P.S.1990:IntertheoreticReduction:ANeuroscientist’sFieldGuide.SeminarsintheNeurosciences,2,249-256.Churchland,P.S.1986:Neurophilosophy.Cambridge:TheMITPress.Churchland,P.S.,Koch,C.,andSejnowski,T.J.1990:WhatisComputationalNeuroscience?InE.L.Schwartz(ed.),ComputationalNeuroscience,Cambridge:TheMITPress.Churchland,P.S.andSejnowski,T.J.1990:NeuralRepresentationandNeuralComputation.InW.Lycan(ed.),MindandCognition:AReader,Oxford:BasilBlackwell.


Churchland,P.S.andSejnowski,T.J.1992:TheComputationalBrain.Cambridge:TheMITPress.Dennett,D.C.1987:ThreeKindsofIntentionalPsychology.TheIntentionalStance,Cambridge:TheMITPress.Enc,B.1983:InDefenseoftheIdentityTheory.JournalofPhilosophy,80,279-298.

Feyerabend,P.K.1962:Explanation,Reduction,andEmpiricism.InH.FeiglandG.Maxwell(eds),MinnesotaStudiesinthePhilosophyofScience,VolumeIII,Minneapolis:UniversityofMinnesotaPress.Fodor,J.A.1975:TheLanguageofThought.NewYork:ThomasY.CrowellCompany.Glenberg,A.M.1976:MonotonicandNonmonotonicLagEffectsinPaired-AssociateandRecognitionMemoryParadigms.JournalofVerbalLearningandVerbalBehavior,15,1-16.Greenwood,J.D.(ed.)1991:TheFutureofFolkPsychology.NewYork:CambridgeUniversityPress.Hirst,W.andGazzaniga,M.1988:PresentandFutureofMemoryResearchandItsApplications.InM.Gazzaniga(ed.),PerspectivesinMemoryResearch,Cambridge:TheMITPress.Hooker,C.1981:"TowardsaGeneralTheoryofReduction,"Dialogue20:38-59,201-36,496-529.Jacoby,L.L.1978:OnInterpretingtheEffectsofRepetition:SolvingaProblemVersusRememberingaSolution.JournalofVerbalLearningandVerbalBehavior,17,649-667.Kuhn,T.1970:TheStructureofScientificRevolutions(2ndedition).Chicago:UniversityofChicagoPress.Laudan,L.1977:ProgressandItsProblems.Berkeley:UniversityofCaliforniaPress.Lehky,S.R.andSejnowski,T.J.1988:NetworkModelofShape-from-Shading:NeuralFunctionArisesfromBothReceptiveandProjectiveFields.Nature,333,452-454.Lykken,D.T.,McGue,M.,Tellegen,A.,andBouchard,T.J.1992:Emergenesis:GeneticTraitsThatMayNotRuninFamilies.AmericanPsychologist,47,1565-1577.McCauley,R.N.1981:HypotheticalIdentitiesandOntologicalEconomizing:CommentsonCausey’sProgramfortheUnityofScience.PhilosophyofScience,48,218-227.McCauley,R.N.1986:IntertheoreticRelationsandtheFutureofPsychology.PhilosophyofScience,53,179-199.McCauley,R.N.1987:TheRoleofCognitiveExplanationsinPsychology.Behaviorism(subsequentlyBehaviorandPhilosophy),15,27-40.McCauley,R.N.1989:PsychologyinMid-Stream.Behaviorism(subsequentlyBehaviorandPhilosophy),17,75-77.


McCauley,R.N.1993:Brainwork:AReviewofPaulChurchland’sANeurocomputationalPerspective.PhilosophicalPsychology,6,81-96.McCauley,R.N.(forthcoming):Cross-ScientificRelations:TowardanIntegratedApproachtotheStudyoftheEmotions.InB.ShoreandC.Worthman(eds),TheEmotions:Culture,Psychology,Biology.Nagel,E.1961:TheStructureofScience.NewYork:Harcourt,BraceandWorld.Neisser,U.1967:CognitivePsychology.NewYork:Appleton-Century-Crofts.Oppenheim,P.andPutnam,H.1958:UnityofScienceasaWorkingHypothesis.InH.Feigl,M.Scriven,andG.Maxwell(eds),MinnesotaStudiesinthePhilosophyofScience--VolumeII,Minneapolis:UniversityofMinnesotaPress.Richardson,R.1979:FunctionalismandReductionism.PhilosophyofScience,46,533-558.Schaffner,K.1967:ApproachestoReduction.PhilosophyofScience,34,137-147.Sejnowski,T.J.andChurchland,P.S.1989:BrainandCognition.InM.Posner(ed.),FoundationsofCognitiveScience,Cambridge:TheMITPress.Sejnowski,T.J.andRosenberg,C.R.1987:ParallelNetworksthatLearntoPronounceEnglishText.ComplexSystems,1,145-168.Sejnowski,T.J.andRosenberg,C.1988:LearningandRepresentationinConnectionistModels.InM.Gazzaniga(ed.),PerspectivesinMemoryResearch,Cambridge:TheMITPress.Thagard,P.1992:ConceptualRevolutions.Princeton:PrincetonUniversityPress.Wimsatt,W.C.1976:Reductionism,LevelsofOrganization,andtheMind-BodyProblem.InG.Globus,G.Maxwell,andI.Savodnik(eds),ConsciousnessandtheBrain,NewYork:PlenumPress.

1. The story is even more complex, since each level of analysis has both a synchronic and a diachronic moment for which separate theories have been developed. See McCauley (forthcoming). At the biological level, for example, cell biology is one of the synchronic sub-disciplines focusing on the structures within the cell whereas evolutionary biology is devoted to the study of change in forms of life over time. The Churchlands' have confined their discussions almost exclusively to synchronic examples.

2. One of the first, if not the first, is Wimsatt's (1976) classic discussion.

3. Although they concur with Churchland's judgment that the folk psychological notion of a unitary faculty of memory is probably wrong, Hirst and Gazzaniga (1988, pp. 276, 294, and 304-05) seem to adopt a far more sanguine view about the contributions of psychology (both folk and experimental) to our understanding of memory. They recognize that the fragmentation of 'memory' need not lead to its elimination. (See section 5 below.)


4. . . . and the position from which they have generally (though not unequivocally) retreated over the past few years.

5. See Churchland and Sejnowski 1990, p. 229, Churchland, Koch, and Sejnowski 1990, pp. 51 and 54, and Churchland and Sejnowski 1992, pp. 10-13.

6. Consider the discussion in Neisser (1967).

7. Interestingly, Ernest Nagel's The Structure of Science (1961), the locus classicus of traditional research on reduction, implicitly recognizes the importance of distinguishing between intralevel and interlevel contexts. Nagel consistently describes intralevel cases as involving the reduction of theories and interlevel cases as involving the reduction of sciences.

8. This is, in part, the result of the same considerations that motivate the Churchlands and Richardson's (1979) arguments that alleged reductions that conform to traditional microreductionistic standards can only be domain specific.

9. –or in neuropsychology, as Churchland and Sejnowski's (1992, p. 282) discussion of the role of the hippocampus in short term memory illustrates. See P.S. Churchland 1986, p. 361.

10. An interesting illustration arises in Churchland and Sejnowski's discussion of the major levels of organization in the nervous system (1992, pp. 10-11). Their diagram of the relevant levels tops out at the central nervous system with no mention of psychology. The obvious defense is to note that the diagram addresses anatomical structures of the nervous system only. Fair enough. What is telling, though, is a footnote (1992, p. 11, footnote 5) to this discussion. Churchland and Sejnowski concede that a more comprehensive account would include a social level above the central nervous system. At least for the purposes of this discussion, they seem not even to countenance the possibility that cognitive research may capture organizational structure of explanatory significance not immediately reducible to the neurophysiological. (See too Sejnowski and Churchland 1989, p. 343.)

A meta-level comment: the physicalist holds that metaphysical manifestness (which, remember, is physical manifestness for the physicalist) constrains what will count as satisfactory explanation, whereas the pragmatist proposes that explanatory success should constrain metaphysical commitment. If that diagnosis is correct, the on-going negotiation in the Churchlands' work I described in section 3 is, at its root, one about competing norms.

11. I should emphasize that I am speaking of the elimination of folk psychology as an explanatory construct within scientific psychology. The elimination of the principles of folk physics centuries ago in physics has had little effect on its persistence among the folk.

12. The illustrations the Churchlands (1990) offer in support of their “overview of the general nature of intertheoretic reduction” (p. 249) proceed in the following order:

(1) the reduction of Kepler's laws to Newton's (intralevel);


(2) the reduction of the ideal gas law to the kinetic theory--emphasizing (p. 250, some emphasis added) that “this reduction involved identifying a familiar phenomenal property of common objects with a highly unfamiliar micro-physical property” (interlevel);

(3) the reduction of classical (valence) chemistry by atomic and sub-atomic (quantum) physics (interlevel);

(4) the reduction of Newtonian mechanics to the mechanics of Special Relativity (intralevel);

(5) the elimination of phlogiston by Lavoisier's oxygen theory of combustion (intralevel).

13. But just as progress in tracing the relevant biological systems preserved the vitality of organisms without vitalism, so too is progress at tracing the relevant psychological systems slowly revealing how we can preserve the cleverness and wondrous experiences of intelligent creatures without dualism. The interlevel influences of neuroscience will no more co-opt or eliminate psychological theorizing than the interlevel influences of chemistry co-opted or eliminated physiological theorizing.

14. As noted near the end of section III, the Churchlands more often seem to endorse an account of co-evolution resembling co-evolutionP. In Churchland and Sejnowski 1990 (p. 250) and 1992 (p. 240), they not only advocate a form of explanatory pluralism, but they explicitly include the psychological sciences.

15. Conceding that it will not involve a single model nor direct explanations of higher levels in terms of events at the molecular level, Churchland and Sejnowski, nonetheless, aspire to a “unified account” of the nervous system, where “the integration [will] consist of a chain of theories and models that links adjacent levels” (Sejnowski and Churchland 1989, p. 343).

16. If neurocomputational modeling of networks constitutes a higher level of analysis than does the study of particular neurons (and it certainly seems to on Churchland and Sejnowski's view – 1992, p. 11), then Churchland and Sejnowski's (1992, pp. 183-188) take on recordings of single cells' response profiles in the visual cortex is an illustration of just the sort of circumstances that the Sejnowski and Rosenberg citation allows for--one in which higher level research impels a reevaluation of lower level doctrines.

Churchland and Sejnowski (following Lehky and Sejnowski 1988) argue that neurocomputational research on the visual system's ability to extract shapes exclusively from information about shading reveals that the conventional interpretation of the function of receptive fields of neurons in the visual cortex may well be wrong. That interpretation, which arose from single cell studies, holds that these neurons function as edge and bar detectors. Churchland and Sejnowski maintain that this interpretation ignores the cells' projective fields. Hidden units in Lehky and Sejnowski's model developed receptive fields with similar response profiles, however these orientations were the result of training the network on the shape from shading task. “In a trained-up network, the hidden units represent an intermediate transformation for a computational task quite different from the one that has been customarily ascribed . . . they


are used to determine the shape from the shading, not to detect boundaries” (Churchland and Sejnowski 1992, pp. 185-186).

Explanatory Pluralism And The Co-Evolution Of Theories In Science …mechanism.ucsd.edu/~bill/teaching/w17/phil151/mccauley... · 2017-01-12 · Explanatory Pluralism And The Co-Evolution

Documents

Explanatory Pluralism And The Co-Evolution Of Theories In Science …mechanism.ucsd.edu/~bill/teaching/w17/phil151/mccauley... · 2017-01-12 · Explanatory Pluralism And The Co-Evolution