volume18,no.12 june2018
Epistemology from an
Evaluativist Perspective
Hartry FieldNew York University
© 2018 HartryFieldThis work is licensed under a Creative Commons
Attribution-NonCommercial-NoDerivatives 3.0 License. <www.philosophersimprint.org/018012/>
1. Skepticism, and the plan of the paper.
Thereisaclassofgloballyskepticalpositionsthatmanyphilosopherstakesomewhatseriously.Examples includeskepticismabout theex-ternal world, skepticism about inductive reasoning, and skepticismaboutlogicalinference.WhenIsaythatmanyphilosopherstakethemsomewhat seriously, I don’tmean that theyhave any inclination tobeglobally skeptical inanyof theseways,but just that they regardtheseformsofskepticismasposingaseriousthreatthatdemandsanon-dismissiveanswer.Butinmyview,theseskepticalpositionsposeathreatprimarilytothosewhohavebeentakeninbyanerroneousmeta-epistemology,aspeciesofepistemologicalrealismthatI’veelse-where(Field2009)deridedasthe“justificatoryfluid”picture.1
Theerroneouspicturehasitthatepistemologyisconcernedwithascertainingthe facts about justification,conceivedasanalogsofthe facts about the electromagnetic field,oraboutflow of electromagnetic energy,orsomesuchthing.Onthispicture,skepticismisthethreatthatjustifica-tioninthismetaphysicalsensemightnotexist,ormighthaveproper-tiestotallydifferentthanthoseourjustificatorypracticesarethoughttopresuppose.Andonthatconceptionof justification, the threatofvariousformsofglobalskepticismdoesseemveryserious.
Let’sconsiderasanexamplethecartoonversionofHume’sworryabout induction, as given for instance inWesley Salmon 1967. The“problemofinduction”,asSalmonposesit,istodiscoverthesourceofjustificationofourinductivepractices.Thefactthatpeopleengageinthesepracticesisirrelevant:theymightdosowithoutthebenefitof
1. I’mslightlyoverstatingthis, forat least inthecaseofskepticismabouttheexternalworld,theremightbepositiveargumentsforthinkingwegobadlywrong:e.g.arguments,basedoncosmological theories forwhichwehaveevidence, thatwe aremore likely to beBoltzmannbrainswith a bleak fu-turethanthenormallyevolvedbrainsinthekindofworldwethinkwe’rein.(Somewhatsimilarly,alabassistantinalabwithathousandbrainsinvatswho’s told that thesebrainsareallnowbeing fedexactlyherexperiencesmighthavereasontoworrywhethersheisreallytheassistant—aworrythatwouldpresumablybecomemorepressingwereshetoldthatinthenearfu-ture thebrains in vatswouldbedecoupled fromher and coupled insteadto someonebeinghideously tortured.)Theskepticalworries Imean tobedismissingareonesbasednotonanysuchpositiveargumentsforskepticalhypotheses,butmerelyontheabsenceofargumentsagainstthem.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –2– vol.18,no.12(june2018)
buildenoughconnectingpipesandthefluidwillappeartofillthem.Butthatgivesrisetothestandardobjection,whichseemsprettydeci-sive:lotsofcoherentmethodologiesforreasoningfromtheobservedtotheunobservedareobviouslybad.
Arealistcouldinsteadholdthatweneedn’tcreatethefluidex ni-hilo,sinceitistherefromthestart:wehavean“a priorientitlement”tobelievecertainthings(suchasthatwearen’tbrainsinvats)withoutevidence.Thisindeed,ifdevelopedinacertainway,mayleadtosome-thingclosetothemethodologythatIwillrecommend,butthenatureofthese“entitlements”isprima faciequitemysterious.IsitthatwhileGod gives us blackmarks if we believewithout evidence thatHill-aryClintonranachildpornringfromapizzaparlor,hedoesn’tgiveusblackmarks forbelievingwithoutevidence thatwearen’tbrainsinvats? (Ormaybe it’s just like thatexceptwithout theGod?)Also,wouldn’tthereberoomfordoubtsaboutwhetherthe“entitlements”reallyfavorourpracticeoverpracticesthatweregardasbadones,andshouldn’tthatmeta-skepticismtendtodiminishtheentitlement?
Perhapstheseremarksarebuiltontooloadedaninterpretationof‘entitlement’;indeed,perhapstheanti-realistviewtobesketchedsup-plies an anodyne interpretation of the nature of “entitlements” thatcouldberegardedasvindicatingtheentitlementview.2ButIthinkthatwithoutsuchaquestioningofepistemologicalrealism,itwillbehardtosetskepticismtorest.3
I concede that the label ‘epistemological realism’ is less than to-tallyclear.Ialsoconcedethatthereareviewsthatnaturallyfitunderthislabelforwhichthe“fluid”metaphorisinappropriate,andwhichthepreviousparagraphsdon’tadequatelytarget.I’llconsideronesuch
2. Forasophisticatedversionoftheentitlementview,seeWright2014.
3. Idon’twanttoquibbleabouttheword‘skepticism’.Indeed,inonesenseoftheterm,justabouteveryonetodayisaskeptic:we’veallgottenbeyondtheCartesianhopeforana prioriandincontrovertibleproofthatourinductivemethods couldn’t possibly lead us badly astray. Themain target of theseopeningremarksisthosewhodon’tbelieveinthepossibilityofsuchaproof,butdobelievein(orhankerafter)somekindof“metaphysicaljustification”shortofthat.
justification.Thefactthatweapproveofthesepracticeslikewisecutsnoice:wemightbewronginsodoing.Tosaythatthepracticeswork wouldbequestion-begging.Sure,they’veworkedwellinthepast,buttoconcludefromthisthattheywillcontinuetodosointhefutureistomake an inductive argument; if there isn’t already a justificatorysourceforinductivereasoning,thiswillcutnoice,andifreal justifica-tioniscounterinductive,itwillmakethingsworse.
Thebesthope,Salmontellsus,isforapragmaticjustification(or“vindication”)whichtriestoarguethatourinductivemethodsarebet-terthananycompetingmethods.Buthethinksthattheargumentfortheirsuperioritycannotrelyonempiricalconclusionsestablishedbyinduction: thatwouldbequestion-begging.What’s required instead,he thinks, is amathematical proof that our inductivemethods, andtheyalone,haveacertaincombinationofdesirableproperties;with-outthat,it’shardtoseewhyweshouldprefersciencetovoodoo(1967,p.55).Butashecametorealize,it isverydifficulttofindclearlyde-sirablepropertiesthatourmethodscanbemathematicallyprovedtohave,beyondveryweakpropertiesthatclearlyundesirablemethodshaveaswell.(Moreover, if theskepticismunderconsiderationwereextendedto include logicalandmathematicalreasoning,we’dbe inevenmoretrouble.)
Here’smycartoonofthecartoon:InSalmon’sview,what’sneededfor inductiontobereasonable is thatpremisesofaninductiveargu-mentpassjustificatoryfluidtotheconclusion.Givensuchapicture,itseemscorrectthatwithoutapositiveargumentforsuchjustificationweshouldn’tbelieveinit:ourdefaultpositionshouldbethe“skeptical”onethatthereisnosuchfluid.Andit’shardtoseewhatthatpositiveargumentforthe“justificatoryfluid”couldbe.Similarly,it’shardtoseewhatpositiveargumenttherecouldbethat,ifitexists,itflowsinthe“inductivedirection”asopposedtothecounterinductive.
Arguments from the regress of justification have some force onanepistemological realistpicture: theymakevivid (what shouldbeplausibleevenwithoutthem)thatyoudon’tcreate“justificatoryfluid”ex nihilo.Ofcoursethiscouldbedenied,andcoherencetheoristsdo:
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –3– vol.18,no.12(june2018)
“boats”thatwethinktotallyunacceptable.Presumablysomeonewhoisonsuchanunacceptableboatshouldn’tcountasjustified,howeverwellheorshedoesatmeetinglocalobjections.Butif“factsaboutjus-tification”areconceivedofinarealistspirit,itseemslikemeredogmatoassumethattheboatwe’refloatingonisoneofthe“justified”ones.Thisislikelytoleadtoskepticaldoubtsatthemeta-level:doubtsaboutwhetherone is justified.Anddoubtsaboutwhetherone is justifiedhavesometendencytoinduceskepticaldoubtsatthegroundlevel.
Ultimately, then, there is a tension between any Neurath’s boatmethodologyandepistemologicalrealistpositions:theepistemologi-calrealismtendstounderminethemethodology.(Shiftingfromsome-what loaded epistemological notions like “justification” to blanderones like “reasonable” helps only in that an epistemologically real-istunderstandingofthelatterislesslikely.)Iwon’ttrytoarguethatthereisnowayaroundthis,butmypreferredsolutionistogiveupontherealismandofferadifferentunderstandingofnotionslikebeingjustifiedorbeing reasonable.Oneway tounderstand themwhich Ithinkwoulddisarmtheskepticismisablatantlysubjectivistone:forabelieformethodtobejustifiedorreasonableisjustforittoaccordwiththestandardsof thebelieveror thebeliever’scommunity.This,however,strikesmeasahighly implausible line. It’snobetter tore-place‘thebeliever’scommunity’with‘ourcommunity’(orwith‘anas-sessor’scommunity’)4:thiswouldpreservethefundamentalproblemwith subjectivism, which is its attempt to explain normative termsasdescriptive.Mypreference is forsomethingmorealong the linesofGibbardianexpressivism(orMacFarlane-styleassessor relativism,properly interpreted:seeprecedingnote).Iwillnotenterintothedetailsofhowsuchaviewistobeformalized,butwillmakesomegeneral
4. MacFarlane2005issometimesinterpretedasmakingthelatterproposal,andsomeofhisremarksdosuggestit,butitseemstomeanincoherentview.Ithinkhispapersuggestsafarmoreinterestingview,verymuchakintoGib-bardianexpressivism,andindeedhehasrecentlysaid(2014)thatthediffer-encebetweenhisviewandGibbard’sisratherslight.Thedifferencehenoteswon’tmatterforpurposesofthispaper.
formofrealisminSection2,butarguethatittoofailstohandleskepti-cismproperly.ButI’mlessinterestedincomingupwithageneralar-gumentagainst“epistemologicalrealism”(howeverexactlythatmightbedefined)thaninsketchinganalternativeviewofepistemologythatIthinkdoesnotgenerateproblems(ofwhichskepticismisonlyone)towhichstandardwaysofthinkingaboutepistemologygiverise.
I’llsketchthealternativestartinginSection3.Oneaspectofitin-volves (to put it very roughly) focusing on sensible epistemic prac-ticeswithoutfetishizingsuchnotionsasjustificationandknowledge.There’s a lot of philosophical baggage currently built into these no-tions, which can to some extent be avoided by framing epistemo-logical questions in such termsas “Whatwould it be reasonable tobelieveinthissituation?”,“HowconfidentshouldIbe?”,andsoforth.Ofcoursethere’snoreasontobanishordinaryepistemologicaltermslike‘justified’and‘knows’:forinstance,‘justified’canbeusedtomean‘reasonable’.Butthenit’sunobviousthatitmakessensetotalkabout“thenatureofjustification”,andtheideathatforabelieftobejustifiedtheremustbea“source”of“thejustification”losesmuchofitsforce.
Suchanalternative to the sortof epistemology that takes skepti-cismseriouslydoesnotputbasic featuresofour inductivepracticesbeyond debate: there can be serious questions aboutwhich of ourinductivepracticesneedimprovement,andaboutthewaysinwhichthey shouldbe improved.Butdebates about thesematters arebestconductedinthespiritofthesailorsfixingNeurath’sboatwhileit’satsea,sothatonlyspecific local objectionsshouldworryusaboutextantinductivepractices.Section3willsketchwhatanepistemologymorefocused on such local improvements in basic methods might looklike,broadlyinkeepingwithsomeremarksofReichenbach.Section4dealswithanobjectionthatmightseemtounderminethecoherenceoflookingforlocalimprovementsinthisReichenbachianspirit.
Idon’tsaythatit’simpossibletocombinethe“Neurath’sboat”meth-odologywithanepistemologicalrealism—some“entitlement”views,e.g.Wright2014,seemtodoso—butIdothinkthattheepistemologi-calrealismmakessuchapictureawkward.Afterall,therearepossible
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –4– vol.18,no.12(june2018)
unreliably,theytypicallystarttobehavemorereliably.Thisissoformanybadinductivemethodsaswellasforgoodones;itisn’teasytoseehowtocomeupwithadefinitionofreliabilityaccordingtowhich,evenamong“equallypowerful”rules,thegoodonesaretheonesthatare“mostreliable”.5Thereisawidevarietyofexternalistgood-makingfeatures in an inductivemethod; I havenodoubt that onemethodscoring higher than anotherwith respect tomost of these externalgood-makingfeaturescontributestoitbeingthebettermethod,butIthinkitextremelyunlikelythatthereisanywayofcapturingthiswithanysimplenotionlikereliability.(Andashinted,suchexternalistfac-torsdonotexhausttherelevantconsiderations.)
Inthenextsection,I’llpresentmyfavoredalternativetobothjusti-ficatoryfluidapproachesandreliabilistapproaches.Butfirstlet’sturnto another kind of skepticism: skepticism about logical knowledge.LewisCarroll(1895)givesaclassicpresentationofoneproblemhere,theproblemofhowwecanknowanythingvialogicalinference.Theproblem,ashepresentsit,isespeciallysalientonajustificatoryfluidperspective:howdothejustificationsforAandforA→BcombinetosquirtjustificationintoB?(Ifyouprefer:Howdothey“transmitwar-rant to” B?)He considers the answer that it’s becausewe have jus-tificationfor thepremiseA˄(A→B)→ B,buthethenpointsout thatthatdoesn’t seem tohelpunlessA,A→B, andA˄(A→B)→ B squirtjuiceintoB,whichitselfreliesontheassumptionthatModusPonensisjuice-squirting.6
OnewayaroundtheLewisCarrollproblemistogoreliabilist:wedon’tneedaninternaljustificationoflogic,forlogicis“justified”intheonlysensethatmattersaslongasit’sreliable.Andreliabilisminthiscaseismorepromisingthanintheinductivecase,becauseinthiscase
5. There’sabitofdiscussionofthisinField2000,Section4.
6. There’sanironyintryingtojustifyModusPonensintermsofthevalidityoftheschemaA˄(A→B)→ B,technicallyknownas“PseudoModusPonens”:amoraloftheCurryparadox,atleastforthosewhoacceptrestrictionsonclas-sicallogictoensurenaiveassumptionsabouttruth,isthatModusPonensisacceptablewithoutrestrictiononlyifPseudoModusPonensisnotacceptablewithoutrestriction.
remarksinSection5aimedatthoseunsympathetictoanysuchviewandatthosewhothinkitcouldn’thelpwiththeskepticism.
Thepaper,then,presentsapackagethatcombinesamethodologyforepistemology(thathasantecedentsinNeurathandReichenbach,amongmanyothers)withanormativeanti-realism.Thetwopartsofthe package are to some extent separable.As I’ve already said, theNeurath’sboatmethodologymaybecompatiblewithamore realistviewofepistemology,thoughthereisatensionbetweenthem.Con-versely, theanti-realistviewofepistemology that Iwill recommenddoesn’trequirethemethodology:forinstance,itisprobablycompat-iblewithamorefoundationalistresponsetoskepticism.Indeed,onecouldmakefurtherassumptionscompatiblewiththeanti-realismthatwouldreinstatesomeoftheforceofthekindofskepticalargumentsfromwhichwebegan.I’lldiscussthatinSection6.Nonetheless,Idothink themethodology I’mproposing isverynaturalgiven theanti-realism,muchmorenaturalthanwithoutit,andforthatreasonIthinkitreasonabletopresentthetwopartsofthepackagetogether.(I’lluse‘evaluativism’asanameofthepackage,butitwillgenerallybeclearinanycontextwhetheritisthemethodologicalormetaphysicalaspectIhaveinmind,andIwillsometimesbeexplicit.)
2. Reliabilism and concept constitution.
The“fluid”metaphorismostnaturallyassociatedwithnon-naturalistformsofepistemologicalrealism.Onealternativetothatisakindofreliabilism;theideaistoforgothemysteriousfluidintermsofaper-fectlynaturalisticpropertyofreliability.
Idon’tthinkthisisthewaytogo.Inpartthat’sforaratherstandardreason—thatreliabilismdoesn’tseemtodojusticetotheinternalistaspectsofepistemology—but Iwilldefer thatcriticism,since induc-tiveskepticismisn’tthebestplacetoraiseit.
There’sanotherreasonthatIthinkmoredecisiveintheinductivecase:there’ssimplynonotionofreliabilityadequatetothejob.Onefeatureofinductivemethodsistheir“self-correcting”character:ifthesemethodsareappliedincircumstancesinwhichtheyinitiallybehave
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –5– vol.18,no.12(june2018)
ornotB”(whichisnotintuitionisticallyvalid,butisvalidinquantumlogicaswellasclassically).Accordingtothereliabilist,whetheroneofourthreecharactersisjustifiedininferringinthiswaydependsnotawhitontheirlogicalviews:ifthisDeMorganlawisnotinfactcor-rect,thennoneofthethreewouldbejustifiedinmakingtheinference,whereasifoneoftheotherlogicsisinfactcorrect,allthreeare;and that is all that can be said about the matter.
I’mnotatallopposedtotheideathatourepistemicapprovalordis-approvalofsomeone’sreasoningmightbetosomeextentconditionalonthe“external”questionofwhichlogicisactuallycorrect,8butIfindithardtobelievethatactualcorrectnessisthewholestory:ifinsomecontextMichaelslipsandreasonsasaboveeventhoughthatreason-ingdoesn’taccordwiththelogiche’sadvocating,thenhisreasoninghasaproblemthatHilary’sandSaul’s reasoningdoesn’thavewhentheyreasoninthatway;andthisissowhicheverlogiciscorrect.
The case can be made more decisive by imagining that the en-tire logical community unanimously supports, by arguments thatweallnowfindcompelling,a logic that isnotactuallycorrect (andmaybewillsomedaybeshownincorrectbyrevolutionarygeniuses);shouldn’ttheoverwhelmingtheoreticalsupportforwhatwilllaterbeseenasanincorrectlogiccountforsomething?Again,I’mnotdenyingthatthecorrectnessofthelogicmightbeonefactorintheevaluationoftheirreasoning;I’mjustdenyingthatitistheonlyfactor.9
8. Let’sputasideanydoubtsonemighthaveaboutthepresuppositionthatonelogicisuniquelycorrect.(Evenontheassumptionthatcorrectnessforlogicis truth-preservation by logical necessity, this could be questioned on thegroundsthatthereneedn’tbeauniquenotionoflogicalnecessity.)
9. I think the samekindof argument canbe given in the inductive case too,thoughforittohavemuchintuitiveforceweneedtorestrictittocaseswherethealternativemethodsarewithinthespectrumofreasonability:otherwisethe“external”aspectsofreasonabilityoverwhelmthephenomena.Considertwopeople,oneofwhomisgenerallymorecautious than theotheraboutinferencestotheunobserved,thoughthefirstisn’tcrazilycautiousandtheseconddoesn’tcrazilyjumptoconclusions.Ifonagivenoccasionthefirstpersongoesagainsthisusualmethodologybymakingan inference to theunobservedthatiswarrantedonlyontheother’smethodology,isn’tthereanimportantsenseinwhichheislessjustifiedthantheotherpersoninmaking
wehaveaprettyclearaccountofreliability:aformoflogicalinferencecanberegardedasreliableif,oflogicalnecessity,itpreservestruth.
Themainproblem(oratanyratethemainnon-technicalproblem)7 forreliabilityhereistheoneIdeferredinthecaseofinduction:exclu-sivefocusonreliabilitydoesn’tdojusticetotheinternalistaspectsofepistemology.
PaulBoghossian (2003)hasemphasizedone importantway thatitdoesn’t:according to theobviousversionof reliabilism, inferenceviaahighlyunobviousdeductiverulethatcanbeshownreliableonlyby an extraordinarily complicatedmathematical proof far beyond aperson’sgraspwould“justify”theperson’sconclusions,whereasintui-tivelysuchapersonismakingtotallyunjustifiedlogicalleaps.Maybesomesortof“higher-order”reliabilityconsiderationscouldbeinvokedinthehopeofgettingaroundthis,butitisfarfromobvioushowthatwouldgo,andsomeoftheworriesabouttheclarityofthenotionofreliability in the inductive casewould thenbe likely to arise in thedeductivecaseaswell.
For another way to see how unintuitive the thoroughgoing ex-ternalismabouttheepistemologyoflogicis,considerdebatesaboutlogic. Let’s suppose that Hilary, Michael, and Saul disagree aboutlogic:Saulbelievesinclassicallogic,Michaelinintuitionistlogic,Hil-ary inquantumlogic.Andlet’spretendthateachhascomeupwithawell-worked-out view according towhichhis favored logic is thecorrectone,withstrongprima facieargumentsfavoringhislogicovertheotherlogics.(Eachhasaprima faciereasonablereplytotheothers’arguments,sononeoftheargumentsisclearlydecisive;it’sthekindofsituationwheretheevaluationoftheargumentmightturnonveryhigh-level theoretical considerations of, say, the role of logic.)Nowconsidera typical logical inference that isvalidaccording tooneoftheselogicsbutnotaccordingtoanother:sayonefromapremiseofform“notbothAandB”tothecorrespondingconclusion“eithernotA
7. A technicalproblem is thatwhenweextend to logicsof truth,we cannolongerequategoodlogicswithonesthatpreservetruthbylogicalnecessity.There’sadiscussionofthisinField2015.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –6– vol.18,no.12(june2018)
damage typically associated with accepting such contradic-tionsis limitedbecausetheydon’t implyeverything. It isal-mostimmediatethatdialetheistsmustrejectModusPonensfor⊃,if⊃isdefinedasaboveintermsof¬and∨:ifoneac-ceptsbothAand¬A,onewillsurelyacceptbothAand¬A˅BnomatterhowabsurdtheB,soModusPonensfor⊃wouldrequireonetoacceptabsurdities.
• Inthecaseoftheordinaryconditional»,onesuchprominentviewisMcGee’s (1985;seealsoKratzer2012Chapter4),ac-cordingtowhich» obeystheExportationPrinciple
(A˄B)»C___________ A» (B» C).
That,withModusPonensfor», leadstotherule
A
(A˄B)»C___________B»C,
BoghossianhimselfoffersanalternativetoreliabilismwithinwhatI’m calling the “juice” framework (Boghossian 2003), in which the“juice” issuppliedby themeaningsofconcepts.Hispaper is framedaroundthequestionofwhythepremisesofaModusPonens“transmitjustification”toitsconclusion.Hisanswer(or,atanyrate,thecentralpartof it: seenote 12below) is that it’sbecause reasoning inaccor-dancewithModusPonensisapreconditionforhavingtheconceptif … then,whichisaningredientintherule.
A complication in discussing this is that there ismore than oneconcept if … then.As iswell-known, the “materialconditional”A⊃B(definedas¬A˅B)isnotagoodaccountoftheordinaryEnglish‘if…then’—witness‘IfIrunforPresidentin2020,I’llwin’,whichistrueonthe⊃account(andnotbecauseI’llwin).Ontheotherhand,theconditional» employedinsuchexamplesiscompletelyinappropriateforanothertaskoftheconditional,restrictinguniversalquantification:itmaywellbetruethateveryonewhowillbenominatedbyamajorpartyforthe2020electionisfemale,butitcertainlyisn’ttruethat∀x(xwillbenominatedbyamajorpartyforthe2020election» xisfemale),sincethatimplies‘TedCruzwillbenominatedbyamajorpartyforthe2020election» TedCruzisfemale’,whichisfalseontheabovestipula-tionfor»eventhoughitsanalogfor⊃isprobablytrue.Thepointofthis is just that indiscussingBoghossian’sclaim,weneed todecidewhetherwe’retalkingabouttherolethatModusPonensfor⊃playsinthemeaningof⊃,ortherolethatModusPonensfor» playsinthemeaningof».Thecasesarestructurallysimilar,butdifferentindetail.
Part of their similarity is that in both cases, there’s at least oneprominentview(whichiswell-motivatedevenifnotultimatelycom-pelling)thatdeniesModusPonens.
• Inthecaseof⊃,theprominentviewisdialetheism,theviewthatundercertaincircumstances it’sallowable tosimultane-ouslyacceptbothasentenceanditsnegation,butwherethe
thatinference?Andisn’tthatsoindependentofwhichinductivemethodis“correct”,iftalkofcorrectnesshereevenmakessense?
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –7– vol.18,no.12(june2018)
case(atleastoncewe’veseenthattheyleadusastray),wecan’tregardreasoninginaccordancewiththerulesfor⊃and» aslegitimate,de-spitetherulesbeingmeaning-constituting.Meaningdoesn’thavetheepistemologicalcloutthatBoghossianrequires.12
Incidentally,evenifthemeaninglineworkedforthelogiccase,itseemshardtoapplyinaremotelyattractivewaytotheinductivecase,sinceinductiverulesdon’t involveanyspecialconnectives.13Pollock(1987,sec.4)didtrytogeneralizeittothatcase,byproposingthatev-eryempiricalconceptapersonpossessesissoshapedbythatperson’ssystemofepistemologicalrulesthattherecanbenogenuineconflictbetweenthebeliefsofpeoplewithdifferentsuchsystems;asaresult,thesystemsthemselvescannotberegardedasinconflict.Butthisviewiswholly implausible. I grant that there’sa sense inwhich someonewithevenslightlydifferentinductiverulesinevitablyhasslightlydif-ferentconceptsofravenandblackthanIhave,butitisnotasensethatlicensesustosaythathisbelief‘Thenextravenwillbeblack’doesn’tconflictwithmybelief‘Thenextravenwillnotbeblack’.Itseemshardtodenythattherewouldbeaconflictbetweentheseravenbeliefs,andifso,thesystemsofrulesgivegenuinelyconflictinginstructions.
Inanycase,thepointfromthelogiccaseremains:declaringcertaininductiverules“concept-constituting”doesnothingtoshowthattheycan’tbelegitimatelycriticized;itjuststipulatesthatthecriticismwillberegardedasacriticismoftheconcepts(here,raven, blackness,andallotherempiricalconcepts).Theoldinductiverulesarede-legitimat-edbythecriticism(ifthecriticismisgood),whateverone’sviewaboutwhethertheconceptshavechanged.14
12. ActuallyBoghossiandoesallowthatthemeaning-constitutingrulesofsomeconceptsmakethoseconceptsdefective,andthatthatunderminesanyjusti-ficationthattheirmeaningmightprovide.Butherulesoutthishappeningforconditionals—and,Ithink,forotherlogicalconceptsexceptfortransparently defectiveonesliketonk.Hedoesn’tseemtoallowforcaseswherethereisaserioustheoreticalissueastowhetheragivenlogicalconceptisdefective.
13. Thishasbeennotedbyothers,e.g.EnochandSchechter2006.
14. Pollock’s view is that it is our object-level concepts like raven that are de-terminedbyoursystemof rules.Aslightlymoreplausibleview is thatour
towhichthereareclearcounterexamples,10andso McGeerejectsModusPonensfor».11
Williamson2007(focusingontheMcGeecase,butthepointgener-alizes)haspressedtheclaimthatreasoninginaccordancewithModusPonens foraconditionalcan’tbeaprecondition forhaving thecon-ceptofthatconditional,becauseprominentandwell-motivatedviewskeeptheconceptwhilerejectingtheallegedprecondition.
Oneobviouswayaround the letterofWilliamson’s critique, andawaythatBoghossiantakes,istorejecttheviewthatthedialetheisthasthesameconceptof⊃thatnon-dialetheistsdo,andthatMcGeehasthesameconceptoftheordinaryEnglish‘if…then’asthoseofuswhokeepModusPonensforitbutrejectExportation.ButIdon’tthinkthisultimatelyhelps.ThoughWilliamsonhimselftakesastrongstandontheissueofwhentherehasbeenachangeofconcept,heneedn’t.Thebasicpoint is:maybe itwill somedaybe shown that reasoningwith thestandardlyacceptedrules for⊃and/or» leadsusastray insomecircumstances.Ifso,wewillwanttoreasonusingdifferentrules,forconnectivesthatwecancall⊃*and»*(thatwemayormaynotregardas“thesameconceptsas”⊃and»).Ifwewanttosaythattheconcepts⊃*and»*differfromtheconcepts⊃and»,fine:inthatcase,theconcepts⊃and» arebadonesthatwillleadusastray.Butinthat10. Ifyoudon’tlikeMcGee’selectionexample,letAandCbothbe‘I’lleatdin-
nertonight’andBbe‘I’llbebeheadedamomentfromnow’.(ButIthinkthisexamplemakesitprettyclearthatitisExportationratherthanModusPonenswhichisproblematic.)
Kratzer2012argues,fairlyconvincingly,thatit’samistaketorepresentor-dinaryconditionalsintermsofaprimitiveoperator»:instead,thosewithoutanovertmodalityhavetheformMust(q|p)wherethisisineffectabinaryop-erator(“ontheassumptionthatp,itmustbethatq”).Shetakesitthatwhen»issodefined,thenMcGeeisrightthatModusPonensratherthanExportationistoblame.Butthislastisfarfromobvious,itdependsonherinterpretationof“stackedrelativeclauses”(p.105)accordingtowhichMust(Must(p|q)|p)isequivalenttoMust(p|p˄q)andthustrivial,whichseemssurprisinggiventhat□(p⊃□(q⊃p))isinvalidinvirtuallyeverymodallogic.
11. Another view, perhaps related and probably more defensible, is that ofKolodnyandMacFarlane(2010),onwhich»violatesModusPonensforsomesentenceswithdeonticsorepistemicmodalsintheconsequent.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –8– vol.18,no.12(june2018)
thoughIcanseehow,ifyou’reintothatsortofthing,he’sprettygoodatit.”Butthereislittlepointintryingtobackofftoapositionofcom-pleteneutrality.
This isof immediaterelevancetothesortofskepticalargumentswe’vebeenconsidering.Froman“evaluativist”perspectiveitishardtoseethepointofthefoundationalistdemandfornon-circular“justifica-tions”ofourinductivemethods:thereisnoreasontothinkthatthele-gitimacyofinductivearguments,orargumentsbyModusPonens,isinperilunlesstheycanbenon-circularly“grounded”—say,inthemean-ingsof component terms.15To thequestion “Whyuseour inductivemethodsratherthancounterinductivemethods?”or“WhyuseModusPonensratherthanaffirmingtheconsequent?”itseemsperfectlyfinetogive theobviousanswer, “Thoseothermethodswouldyieldradi-callywrongresults”,andthisanswerneedsnofurtherdefense.Partofthereasonthisisfineisthatwecanbackoffquiteabitfromthedetailsofourmethodswithoutcompromisingtheanswer.Thatis,any alterna-tive inductive or deductive method that we can take remotely seriouslywillagree with ours that counterinduction or affirming the consequentleads to absurd conclusions. Evaluatingourmethods asbetter thanthosedoesn’trequireacompletelyneutralstandpoint.That’s inpartbecauseevaluatingtheminthiswaydoesn’tinvolveclaimingthatbe-liefs arrivedatbyourmethodshave somethingcalled “justification”forwhichwecansensiblyask,“Fromwheredoesitflow?”
Thismightsuggesteither(I)thatnoissueofjustificationcanariseforourmostcentraldeductiveorinductivemethods,or(II)thatthereisnoroleforasystematicepistemology.ButIemphaticallyrejectbothviews.
Regarding(I), I think that justificationofcentralpresuppositionsis important inabroadlydialecticalcontext.Supposesomeonechal-lengescurrentstandardsofdeductionandinductionbyofferingalter-nativestandardsthatsheregardsassuperior,oratleastthinksmightbe superior. (The someonemight even be ourself: ‘dialectical’ isn’t
15. Tosomeextentitisdesirabletosystematizeourevaluations,butthere’snoobviousreasonwhysystematizationneedstotakeafoundationalistform.
3. Systematic epistemology.
It’stimetosketchoutadifferentperspectiveonskepticalproblemsofthesortwe’vebeenconsidering.Let’sstartfromtheideathatthepointofepistemologyistoevaluateourownandothers’methodsofform-ingand retainingbeliefs, typically inorder to influence them to im-provethosemethods(ortoresistchangesinthemethodsthatwouldmakethemworse).
One feature of typical evaluations is that they aremulti-faceted:“Themoviepresentsacompellingsituationandhas imaginativecin-ematography,andtheleadactressgivesaknockoutperformance,butanimportantsubsidiarycharacterispoorlydeveloped,andthereisaholeintheplot.”Epistemologicalevaluationsareoftenlikethattoo:“Hisconclusionwasbasedonagoodthoughunpopularmethodofsta-tisticalinference—ironic,sinceheactuallyadvocatesadifferentmeth-odwhichwouldhave led to adifferent conclusion—andhemakesgooduseofveryextensivedata, thoughthere isaslightbias in themethodbywhichhecollectedthatdata,andthereisotheravailabledatathat,ifnotaccountedfor,wouldseemtounderminehisconclu-sion.”Itseemsalmostasabsurdtoevaluatebeliefsonasinglescaleofdegreeofjustifiednessasitistoevaluatemoviesorpiecesofmusicorliteratureonasinglescaleofdegreesofgoodness.Idon’tmeantosug-gestthatanepistemologicalrealistwouldhavetodisagreewiththis;butIdothinkthatthereisastrongtendencyintherealistliteraturetotalkasiftherewereasinglescaleofjustification.
Anotherfeatureoftypicalevaluationsisthatwemakethemusingourownbeliefs andpreferences.To somedegreewearewilling tobackofffromthesebeliefsandpreferences:“Ican’tstandFrankSinatra,
epistemologicalconceptslikereasonablearesodetermined:‘reasonable’justmeans ‘reasonable according to our (the assessor’s) rules’. That modifiedview doesn’t seem attractive either, but in any case, itwouldn’t serve Pol-lock’spurposes.Fortheadvocatesofalternativesystemsofruleswouldstillbeingenuineconflictaboutravens,andeachcouldraiseskepticalworriesaboutwhetheritmightn’tbebettertoshiftfromthesystemthatisreasonableintheirownsense(viz.,theirownsystem)tothesystemthatisreasonableintheotherperson’ssense(viz.,theother’ssystem).Allthatthemodifiedviewwoulddoisstripawaythenormativeaspectoftheterm‘reasonable’.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –9– vol.18,no.12(june2018)
onthebasisofevidence.Inmoreseriousmodelsthereislikelytobefarmoreopportunityforsuchvariations.Evenforasingleperson,it’shardtobelievethatthere’sauniquelybestchoiceofallsuchparam-etersforanidealizedmodelofthatperson;and(moretothepresentpoint)it’shardtobelievethattherangeofbestchoicesfortheideal-izeddescriptionofonepersonwillbethesameastherangefortheidealizeddescriptionofeveryoneelse.
Inanycase,thereisnoreasontorestrictourepistemologicaltasktothosethatarequasi-descriptiveofactualbeings:it’salsopossibleandIthinkimportanttoinventandstudymethodswithoutregardtowhetheranyoneactuallyemploysthem.Maybesuchmethodswouldbebetter.Thedetailedformulationofsuchmethodscanbethoughtofasquasi-descriptive inanexpandedsense: itquasi-describespos-siblebeings(whomightalsobesubjecttotiredness,inattention,anddrunkennessfromwhichthequasi-descriptionabstracts).
Thisbringsustotheevaluationofmethods:the“critical”and“ad-visory”tasksofepistemology,inReichenbach’sphrase.Herewestudythedifferentmethodsthatwe’veisolatedinthe(expanded)quasi-de-scriptivephase,seehowtheyperforminvariouscircumstances,andmakeanevaluationofthembasedonthis.Presumablytheevaluationistobecomparative:weneedtocompareeachmethodtootheravail-ablemethods,sinceinsomesensewecan’tdobetterthanusingthebestavailablemethod.Of course,wecan try tomakenewmethodsavailable,andifwethinkthatthebestavailableisn’tgoodenough,wewillbemotivatedtotrytodoso;ourdegreeofoptimismaboutfindingabetteronemightaffectthedegreetowhichwecallthebestavailableonedeficient.Wemightalsoinsomecircumstancesmakeabetthata bettermethod thanwe currently havewill yield a certain verdict,andgowiththatverdictdespitenothavingaverygoodbackingforit.Thereisnoformulaforhowtodoallthis;butoncewegiveuptheideaofasinglescaleofevaluation,thereisnoneedforone.Tryingtodecideexactlywhat isrequiredtobe“justified”distortsgoodepiste-mologicalpractice.
In Reichenbach’s own practice, the evaluation of methods was
intendedtoexcludedebateswithoneself.)Then,totheextentthatherconsiderationsmoveuseventhoughtheydon’tultimatelyconvinceus,weneedtoconsiderwhatcanbesaidforwhyourstandardsarebetterthanthealternativeshe’ssuggesting.Therearesomebigissueshere,buttheywillcomeupmoreclearlyifIfirstturntomyalternativeto(II):thenatureofasystematicepistemology.
Oneimportantroleforepistemologyisthedevelopmentofformalmodelsofidealepistemicbehavior.Themodelsthathavebeendevel-opedsofarareextraordinarilyoversimplified—forinstance,Bayesianmodelsdon’thandlefailureoflogicalomniscienceoreventheinven-tion of new theories, they involve superhuman computational com-plexity,theytreatthenotionof“basicobservationpropositions”asablackbox,andattemptstosaywhichpriorsaregoodarehopelesslylimited (e.g. the continuaof inductivemethods are confined to lan-guages with only monadic predicates). Despite such extraordinarylimitations, Bayesian methods are extremely illuminating for theirresolutionsofawiderangeofpuzzles,anditisimportanttotrytode-velopfarmorerealisticmodelsthatincorporatetheirinsights.
What are these models models of? One thing to model is howpeopleactuallydothings.I’mtalkinghereaboutanidealizedmodel,onewhichabstractsawayfrommistakesduetotiredness,inattention,drunkenness,andsoon.(Theremaybesomedisputeastowhichfea-turesofourperformancearemere“performanceerrors”thatshouldbeidealizedaway,butthereisnoreasontoinsistonahardandfastdecisioninallcases:onecanlookformodelsthatidealizethefeatureawayandmodelsthatbuilditin.)Let’scallthetaskofcomingupwithsuchmodelsthequasi-descriptive taskofepistemology.(Theterminol-ogysomewhatechoestheopeningsectionofReichenbach1938.)
Wecansee,evenfromthecrudemodelswenowhave,thataquasi-descriptivemodelofonepersonisunlikely tobethesameineverydetailasaquasi-descriptivemodelofsomeoneelse: for instance, inthecontinuaofinductivemethodsthereareoneormoreparametersthatdeterminevariousfeaturesofcautionabouthowtomodifypre-dictionsaboutfutureinstances,orbeliefinuniversalgeneralizations,
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –10– vol.18,no.12(june2018)
beenmadeforvariousreasons,somemuchbetterthanothers:thebet-terones includedealingwithvaguepredicates inaway that resistsarguments that suchpredicateshavesharpboundaries,anddealingwithtruthandrelatednotionsinawaythatallowsnaiveprinciplestoholdwithoutleadingtoparadox.Advocatesofsuchproposalspresentvariousreasons foradopting them, in thehopeofpersuading thosewhoadvocate theuseof classical logiceverywhere (even forvaguepredicatesand/orfortruthinparadoxicalsituations)tochangetheirmind.Iftheydotheirjobwell,thereasonstheyprovidewon’tdependontheirpreferredlogic:thatis,theargumentsthattheygivewon’tuseanylogicalprinciplesthattheydisagreewiththeclassicistabout.16Forwecantypicallyshow,inabackgroundlogicneutralbetweenthetwoinquestion, thatacceptingone logic leavessuchandsuchpossibili-tiesforvaguenessandtruthwhileacceptingtheotherlogicleavessoandsootherpossibilities;andthentheadvantagesanddisadvantagesofeachcanbeassessed,againinawaythat isargumentativelyneu-tral.(By“argumentativelyneutral”Idon’tofcoursemeanthatitsargu-ments areneutralbetweenall logics,whichwouldbe impossible; Imeanthattheyareneutralbetweenthelogicscurrentlybeingdebatedbetween.)Doubtless,advocatesofthedifferentlogicswillbeinitiallyinclinedtoweightheadvantagesanddisadvantagesdifferently;still,novelarguments for theoverall advantagesof logicL*over logicLmayeventuallypersuadetheadvocateofLtotrytomodifyhermodesofreasoning.
16. Ofcourse,ifthepreferredlogicisineveryrespectweakerthanclassical,theneutral logicwill justbe thatweaker logic.Butnormally,analternative toclassical logic isweaker in thenonmodalclaims itaccepts,butstronger initsrejectionsandperhapsitsmodalacceptances:e.g.ifitisweakerinnotac-ceptingallinstancesofexcludedmiddle,itwillbestrongerinrejectingsomesuchinstances(asopposedtoacceptingtheirnegations),andinsomecasesmayacceptthatcertainnegationsofexcludedmiddleareatleastpossible.
Moreover,evenwereonelogicstrictlyweakerthantheother,thiswouldn’tpreventrationaldebate: for instance, theadvocateof theweaker logiccanmakethecasethatcertaincommitmentsofthestrongerlogicareuncomfort-able,andtheadvocateofthestrongerlogiccanmakethecasethattheweakerlogiciscumbersome.
supposedtobea prioriandfromacompletelyneutralstandpoint:hewastheinitiatoroftheattemptatana prioriandnon-circular“pragmat-icvindication”ofinductionofthesortI’vementionedinconnectionwith Salmon. But this part of Reichenbach’s story can be separatedfromtherest,andIthinkweshoulddropit.Whatwewantisamethodthatwillworkwellin a world like ours,andouronlyholdonthefeaturesofaworldlikeours is throughour inductivemethods.Wecanbackoffabitfromourownbeliefsandstandardsinmakingtheevaluationofmethods,buttherestrictiontoacompletelyneutralstandpoint ishopeless.
4. Defusing an objection to the methodology.
Thereisaworryonemighthaveaboutthis“evaluativist”methodology(indeed,Iconfesstohavingtakentheworrytooseriouslyinthepast):thatthecritical/advisorytaskwon’t,intheend,cutanyice,becauseeachmethodthatemergesinthe(expanded)quasi-descriptivephasewillenduprecommendingitself.
That is the worry that emerges from David Lewis’s two paperson “immodest inductivemethods” (1971, 1974).Lewis consideredassamplemethodsthemethodsofCarnap’scontinuum.Inthefirstpa-per,hearguedthattheonlymethodthatrecommendeditselfwasanobviouslyinadequateone.Inthesecond,heobservedthattherewasatechnicalerrorinthefirstpaper,whichwhencorrectedshowedtheproblemtobe,inasense,evenworse:everymethodintheCarnapiancontinuumdeclareditselfsuperiortoeveryothersuchmethod,sothatself-evaluationsimplyhasnoforce.
Lewis’s conclusions are basedon controversial rules bywhich amethodscoresitselfandothermethods,butIwillnotobjectonthatcount,sinceitishardtofindalternativescoringrulesthatleadtosat-isfactoryresults.Myobjection,rather, is to thesignificancegiventotheseimmodestyarguments.
Considertwokindsofproposalthatseemratheranalogoustopro-posalstoreviseourinductivemethods.
First,proposals torevisedeductivemethods.Suchproposalshave
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –11– vol.18,no.12(june2018)
requiresdeployinganobservationalvocabulary(thelanguageofrela-tivemotion)thatisneutralbetweenthetheoriesatissue.
Returningtothecaseofinduction,thepointisthatimmodestyar-guments,wheree.g.eachmethoddeclaresitselfbest,areanalogoustotheham-handedargumentsagainstconceptualrevisiondiscussedabove in the deductive and Copernican cases. Yes (putting asidequalmsabout thescoringmeasures theyrelyon), theadvocateofaparticular inductivemethodcanargue,using thatmethod, that thatmethodisbest.Butasinthedeductiveandperceptualcases,suchar-gumentsdonotprecludealternativeargumentsfortheoppositecon-clusion.Andthereisnoreasonwhythefirstargumentshouldtrumpthesecond.
Iwouldn’tbehappytostatethiscriticismbysayingthatbecausethefirstargument is circular, ithasno force.Arguments thatare insomesensecircularcansometimeshaveacertainkindof force: foronething,theycanservetoilluminatewhatthepositionbeingarguedforhastosayaboutalternatives.Anargumentthatiscircularinthiswaycanbeespeciallyusefulwhen(asinresponsetomostbrain-in-vatscenarios)thepositionitbegsthequestionagainsthasnothingposi-tivegoingforit.Myclaimisonlythatwhenthealternativedoeshavealotgoingforit, it isgoodpracticetotakethealternativeseriously:totrytogiveacomparativeevaluationofthetwopositionsthatisasneutralaspossible.
In all three kinds of fundamental conceptual change (deductive,perceptual, and inductive), we have arguments for competing con-clusionsastowhichalternativeisbetter.Sodifferentaspectsofone’stheoreticalstatearepushingusinincompatibledirections,anduntilwe’veresolvedwhichdirectiontogoin,we’reinanincoherentbeliefstate.Ihavegivensomevagueadviceabouthowtogoaboutreason-inginsuchcases:“Developeachofthealternatives,eveniftheycon-flictwithpriorobservationalpractice,logic,orinductivepractices.Seewhichdoesbest.”Butifonewantstodevelopsomethingmorepreciseinthesecasesofconceptualchange,we’llneedamodelofhowtodealwithinconsistentorotherwiseincoherentbeliefstates.
Butwait,doesn’teachpartyinadebateaboutlogichaveaneasyanswerthatsettlesthedebate?E.g.can’taclassicallogicianrespondtoanypositionwhosecoherencedependsonarejectionofexcludedmiddle,justbyusingexcludedmiddletoshowthatpositiontobein-coherent?(“Thepositionislogicallyinconsistent,whichisasbadasapositiongets!”)Ontheotherside,can’tanadvocateofnaivetruththeory respond toany truth theory in classical logicbyusingnaivetruthtoshowthatclassicalprinciplesleadtoabsurdity?
Theansweristhatofcoursesuchquestion-beggingargumentsareavailable,but that in thecontext inquestion, theyhave littledialec-tical force.Theremightbevalue instatingsucharguments ifdoingsomakesclearerhoweachpartyviewstheother’sposition;butoneshouldn’t view such arguments as trumping other considerations,presentedinalogicneutralbetweenthoseunderconsideration,thatmightcutintheotherdirection.
Another example with the same moral concerns observationalpractice.Consider“Feyerabendcases”(Feyerabend1975),caseswheretheoldobservationalpracticeisladenwithatheorythatcanbeques-tioned. Feyerabend’s own example concerned observations of thepathsoffallingobjects.Feyerabendinsiststhatwhenpre-Copernicansreportedobjectsasfallinginstraightlines,theydidn’tmean“straightrelative to theobserver”; theymeant “absolutely straight”.Let’splayalong with that—we can imagine an alternate history in which itwouldbeplausible,andhispointdoesn’treallydependonhistoricalaccuracy.
Wecouldimagineadogmaticpre-Copernicanusingtheoldobser-vationalpracticetodismisstheCopernicantheory:“We’veobservedthousands of bodies falling in straight lines, whereas CopernicantheorysaystheyfallincurvedarcsbecauseofthespinningEarth;soCopernican theory is decisively empirically refuted!” But obviouslythatwouldbeanabsurdmethodology: the rightmethodology is todeveloptheoldtheoryandthenewtheoryasbestonecan,andtrytocomparetheminasneutralawayaspossible,which inthiscase
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –12– vol.18,no.12(june2018)
waysofdealingwithinconsistentpremises.Andwecanthenusethiswayofdealingwith inconsistency, in thecasewherewehavegoodargumentsforsubstantialchangeoflogic,inductivemethodology,orobservationalpractices, competingwith theobvious (“question-beg-ging”)argumentsagainstsuchchange.
Objection: a mental model of how we deal with inconsistencywouldtellushowwedochangelogic,nothowweshould.Itwouldleaveopenthequestionofwhetheractinginthiswayisrational.
Reply:Oncewehaveamodelofhowwedochangelogic(orevenof how we might), the question of whether the model makes thechangerationalissimplyaquestionofevaluation:
Isactinginaccordancewiththemodelagoodthing(orwoulditbe,ifwedon’tactuallyactthatway)?
To answer this, wemust compare themodel to alternativemodels.Andit’shardtobelievethatthedogmaticmodelscouldwin.
Insummary,thekeyfeaturesofrationalrevisionofinductivemeth-odologyare:
• comingupwithanalternativemethodology (describedinenoughdetail),
• arguingformeritsofnewmethodologyovertheold(us-ingtheold,orwhat’scommonbetweenthetwo),and
• retraining ourselves to operate in accordance with themethodologyweconsiderbetter.
Stage2iscomplicated:theoldmethodologywillalwayshave(atleastcheap)argumentsthatitissuperiortothenew.Butthisdoesn’tpre-vent arguments in the other direction.We need toweigh the argu-mentsoneachside.Wehaveintuitivewaysofdoingthis,butaformaltheoryofhowwedealwithinconsistentinformationwouldbenice.
SuchamodelwassuggestedbyBrysonBrownandGrahamPriest(2004). They were dealing with classically inconsistent theories indomainswhere the appropriateness of classical reasoning is not indoubt17(suchasthetheoryofinfinitesimalsthatBerkeleycritiqued,orBohr’searlymodeloftheatom).Theirgeneralidea:
• Ourcognitiveprocessesaredividedinto“chunks”,withinwhichwereasonusingourlogic.
• Insteadofallowingfreepassageofinformationbetweenchunks,weimposerestrictions.
(E.g.inthecaseofinfinitesimals,thefirstchunkassumesthatinfini-tesimalsarenon-zero,andusesthistoderivesuchconclusionsasthatthederivativeof the functionx2 is the function2x+dx.Thischunkpasses that conclusion [though not the derivation] to the secondchunk,whichcontainsthepremisethatinfinitessimalsarezeroandsoconcludesthatthederivativeis2x.)Thismodelneedstobegeneral-izedabitifitistobeappliedtorevisionoffundamentalpractices(e.g.deductive, observational, or inductive). For this, we probably wantanindeterministicmodelofmentality,wheresomethinglikechanceplaysaroleinboth
• whattheories(e.g.logicaltheories)onethinksup,and
• whatevaluationonecomestooftherespectivemeritsofthetheories.
Presumably focusingononechunkandbecoming influencedby itsconclusionswilldiminishtheinfluenceofincompatiblechunks;sowedon’twant thefixed restrictionson informationtransferassumedbyBrownandPriest.
Butwhateverthedetails,thereislittledoubtthatwehaverational
17. Sothe‘paraconsistent’intheirtitleispotentiallymisleading.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –13– vol.18,no.12(june2018)
onwhat makes our beliefs justified.Themainproblemwiththeviewthatreflective equilibrium is “whatmakes a person’s beliefs justified” isthatsuchanapproachleavesnoroomfortheevaluator’sperspective.
Oneattempttogetaroundthisisstraightforwardlysubjectivist:ittreats an evaluator’s claims about justifiability or reasonableness asclaimsabouttheevaluator’sepistemicnorms.Alternatively,asclaimsaboutthenormsoftheevaluator’scommunity—akindof“groupsub-jectivism”.Butwhetherinindividualorgroupform,thisstrikesmeasnot theway togo: it leads to the idea thatevaluatorswithdifferentnorms,orfromcommunitieswithdifferentnorms,arejusttalkingatcrosspurposeswhentheyapparentlydisagree(orevenwhentheyap-parentlyagree).Abetterwaytogoistofollowtheexampleofexpres-sivistsaboutotherevaluativediscourse,e.g.morality.
The term ‘expressivism’ has been used for a bewildering varietyofviews,fromthenon-cognitivismoffigureslikeAyer,wholikenedevaluationstocheeringandbooing,torecent“quasi-realism”whoseadvocatestrytosoundsomuchlikenormativerealiststhat it’shardtoseewhatthedistinctivelyexpressivistfeatureoftheirviewcomesto.Perhapsatthispointthelabeldoesmoreharmthangood;perhaps‘evaluativist’isalessmisleadingterm,eveninthemoralcase.
And ‘expressivism’mightbeevenmoremisleading in theepiste-mologicalcase.Foramoralexpressivistislikelytothinkthatthereisacloseconnectionbetweennormsofmoralgoodnessanddesiresorpreferences,andinconversationI’vehearditassumedthatsuchacon-nectiontodesiresorpreferencesisbuiltintoexpressivism.Thisstrikesmeasinadequateeventothemoralcase,onceonegetsbeyond‘good’:Itaketheexpressivistideaformoralobligationtobethatthenormsofobligation(orlackofobligation)functionmorelikecommands(orpermissions)thanlikedesiresorpreferences.Butintheepistemologi-calcasethereseemsevenlessconnectiontodesiresorpreferences,18
18. There might be some connection at the “second-order” level: in evaluat-ingfirst-ordernorms,wemaybring inpreferencesabout theweightingoftruth-orientedproperties.(E.g.:Howmuchriskoffalsitybalancesthechancefortruthaboutagivensortofquestion?Whendoesahigherchanceofap-proximatetruthoutweighalowerchanceofexacttruth?Whendoesahigher
5. The metaphysics of normativity.
Let’sgetbacktotheevaluativistpicturesketchedinSection3.Ithassomeconnectiontoareflective-equilibriumpicture,ononeconstrualofthat.
Not ona construal that says that being justified consists in beingin reflective equilibrium. (That construalmakes the reflective equi-libriumviewprettymuchthesameas thecoherencetheoryof justi-fication.)Onthatconstrual,thereflective-equilibriumviewistotallyimplausible,inthattherearereflectiveequilibriathatwerightlyjudgeasidiotic.
Theconnectionofevaluativismtoreflectiveequilibriumisjustthatpropermethodologyconsistsofstrivingforreflectiveequilibrium,notforprovidingfoundationsforourbeliefs.(Thisleavesopenwhether,wereequilibriumachieved[!],therewouldbevalueincontinuingtolookforandevaluateothermethods.I’minclinedtothinkso:thateveninanequilibriumpositionwherethereisnolocalpressuretochange,itwouldstillbeofvaluetolookforothernearbyequilibriathatmightbesomewhatbetter.Butthis issueissomewhatacademic: therearealwaysconflictingpressureswhoseresolutionsweshouldlookfor.)
Similarly,theevaluativistpicturemayhavesomethingincommonwiththosewhoadvocateakindof“methodologicalconservatism”,ac-cordingtowhichthereisvalueincontinuingtobelievewhatoneal-readybelieves. (Afterall,advocatesofmethodologicalconservatismusuallyliketheNeurath’sboatmetaphor.)Butagain,ifthismeansthatwearetoexplainwhatitistobejustifiedinbelievingthatpintermsof factors thatprominently includeactuallybelieving thatp, itdoesnotseemapromisingapproach:thereislittlemerittoapersonwhoconservativelysticks to thestoryonwhich themoon landingwasaconspiracy, and employsmethods designed to immunize this viewfromcriticism.
Onecould try tofixup the reflectiveequilibriumand relatedac-countsbyaddingotherfactorsthatarerequiredtomakeourbeliefsjustified,butIthinkthatabetterapproachistoreorientfromafocus
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –14– vol.18,no.12(june2018)
onpar:anorm(hyperplan)issomethingthatassignstoeachpairofabelief(oraction)andworldanevaluationofthebelief(oraction)attheworld(whetherjustaspositive,negativeorneutral,orsomethingmorefine-grained).Whereasoneworld ismetaphysicallyprivileged(itrepresentsreality),thereisnoobviousreasontothinkofonenorm(orhyper-plan)asmetaphysicallyprivileged.Indeed,it ismetaphysi-callyprivilegedonlyiftheworldscontain“normativefacts”thatmakethenorms“correct”,andpresumablytheGibbardideawasthatthereisnoneedforthat.
I’vestressedthattheframeworkofnorm-worldpairsisanattempttomodelnormativelanguageinnon-normativelanguage.Morespe-cifically, thepointof theGibbard framework is to capture the logicofnormativity,asanevaluativist/expressivistseesit:inparticular,toshow thatdespiteFrege/Geach, theexpressivisthasnothing to fearabout the logicofembeddednormativeclaims.Obviously therearelimits on themodel: there is no hope of anymodel accurately cap-turing themeaning of normative language in non-normative terms,because thenormativity itself isacrucialpartof theirmeaning.Butattemptstomodelvaguelanguageusingnon-vaguelanguage(e.g.su-pervaluationalsemanticsorŁukasiewiczsemantics)canbeilluminat-ingwithoutprovidinganything likea translation; similarly formod-elingtensedormodallanguageinnon-tensedandnon-modalterms.Andso,Ithink,inthecaseofnorm-sensitivelanguage.
WhiletheGibbard/MacFarlanemodelisvaluable,therearespecif-icwaysinwhichonecouldbemisledbyit—thatis,bythecorrespon-dencebetweennorm-sensitiveclaims(e.g. “Youshouldbelieve thatp”or “‘Youshouldbelieve thatp’ is true”)andclaims thatexplicitlyrefertonormsinanorm-insensitiveway(e.g.“Accordingtonormn,youshouldbelievep”or“‘Youshouldbelievethatp’istrueinnorm-worldpair<n,@>”where@istheactualworld).Itisacrucialpartoftheevaluativistviewthatanorm-sensitiveclaimisverydifferentfromthecorrespondingnorm-insensitiveone.OnlytoaverylimitedextentisthatdifferenceclarifiedbytheGibbardframeworkbyitself(asGib-bardhimselfofcourserecognizes:hesupplementsitwithilluminating
soanyonewhoreadsanystrongsuchconnectioninto‘expressivism’isboundtofindtheideaofan“expressivist”epistemologyabhorrent.ForthatreasonI’llstickwiththeterm‘evaluativism’.
Here’sathumbnailsketchofthemetaphysicalaspectofevaluativ-ismasI’llunderstandit:Thekeyideaisthatjudgmentsaboutwhatisjustified,reasonable,andthelikecanbedividedintotwocomponents.Onecomponentisanormofevaluation;theotherisabeliefinanar-rowsense(“purebelief”)aboutwhatisjustifiedaccording to that norm.Becauseoftheevaluativecomponent,itisnaturaltodeclareanorma-tiveclaimsuchas“Itisreasonabletobelieveinquarks”not straightfor-wardly factual: incontrast to thestraightforwardly factual “Therearequarks”,theclaimaboutwhatisreasonabletobelieveinvolvesepiste-mologicalvalues.
Normativeclaimshaveaspecialkindofperspectival featurethatnon-normativeclaimsdon’t share—theperspectivebeing theevalu-ator’snorms.Somewhatsimilarly, tensedclaimshaveaperspectivalfeature that untensed ones about 4-dimensional reality don’t share,andmodalclaimshaveaperspectivalfeaturethatnon-modalclaimsaboutthehyper-universeofpossibleworldsdon’tshare.(Obviouslythenormativecaseisalsoimportantlydifferentfromthese,inawaytobediscussedbelow.)But justasonecan illuminate tensed claimsbygivinganuntensedaccountin4-dimensionalterms,andilluminatemodalclaimsbygivinganon-modalaccountintermsofpossibleworlds,soonecanilluminatenorm-sensitiveclaimsbyanaccountinlanguagethatisnot norm-sensitive.Gibbardhasdoneso,intermsofhisframe-workof“norm-worldpairs”.(SeeGibbard1986and1990;hislaterre-labelingof “norms” as “hyperplans”, inGibbard2003,doesn’t affecttheframework.19MacFarlane2005and2014hasofferedwhatcanbeinterpreted as a very similar framework [see note 4], though he fo-cuses lessonnormativity inapplying it.)Normsandworldsarenot
chanceofgetting to the truthquicklyoutweighanoverall lowerchanceofgettingtoiteventually?)
19. Thenormsinquestionare“complete”ormaximallydetailednorms,justastheworldsarecompleteormaximallydetailedpropositions.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –15– vol.18,no.12(june2018)
the logic right, including inparticularhow theembeddingofnorm-sensitiveclaimsinsidelogicaloperatorsworks.20
In sum, the evaluativist framework has it that normative claimshaveanorm-sensitivity,withaspecialpragmaticrole.Thisframeworkallowsthefollowing:
• Despitetheirnotbeingstraightforwardlyfactual,norma-tiveclaimsplayanimportantcognitiverole:theycertain-lyaren’tmerelycheersorboos.
• Andtheycanberationallyevaluated,inpartbybringingintoconsiderationothernormativejudgmentsandinpart
20.These remarks add something to thediscussion in Field 2009,which alsostressed thepragmaticsofdisagreement,andpresented theviewas in thespiritofthe“assessorrelativism”ofMacFarlane2005,thoughwithacoupleofsignificantdifferences.Thatpaperofferedamotivationforcallingevaluativ-ism“relativistic”,whichisthatthe Gibbard modelingisclearlyrelativistic:tosaythatSistrueatnorm-worldpair<n,w>isanotationalvariantofsayingthatSistrueatworldw,relativetonormn.Focusingontheactualworld,whichismetaphysicallyfixed,thisbecomesjusttruthrelativeton.
Quiteproperly, I refrained fromconcluding from this that “true relativeton”istheonlynotionoftruth,oreventheprimaryone,inthenormativedomain.Rather,Itooktheprimarynotionoftruthtobethedisquotationalone,whereTrue(‘p’)isequivalenttop,sothat“True(‘p’)”inheritswhatevernorm-sensitivitythereisin“p”.ThismeansthatinaGibbardmodel,“True(‘p’)”likepwillneedtobeevaluatedatnorm-worldpairswhen‘p’isnormative.
However,IdidsuggestthattheGibbardmodelsuggestsakindof“relativ-ity”,notprimarilyintruthbutinground-levelnormativenotions,butinher-itedintotruthclaimsfromthatground-levelrelativity.Becauseof theprag-matic featuresofnormativediscoursementionedabove, the relativityhadtobeofanunusualkind, closer to the “assessor relativism”ofMacFarlane2005thantoprototypicalrelativism.Butthetalkof“relativism”hereseemsoptional:onecouldarguethatany“relativism”hereissimplyanartifactofmodeling norm-sensitivity in norm-insensitive language, so that evaluativ-ismitselfshouldn’tberegardedasarelativistdoctrine.Isuspectthattheissueofrelativismisn’tclearenoughfortheretobeanypointto insistingeitherthatevaluativisminvolvesrelativityorthatitdoesn’t.(Eithersidecanaccom-modatethepointthatit issometimesusefultobackoffthenormativitybysaying,“Well,it’sjustifiedrelativethesenorms,andthesenormshavesuchandsuchadvantages”:thatisn’tdecisivethat“therewassomekindofrelativ-ityinthenormativenotionsallalong”.)“Norm-sensitivity”nowstrikesmeaslesscontentiousthan“norm-relativity”.
remarksonthepragmatics).Theframeworkdoesallowustoholdthat thereisadifference:itallowsustoholdthatclaimsexplicitlyrelativ-ized tonormscanhave their truthvaluedeterminedwhollyby theworldcomponentofanorm-worldpair,sothattheexplicitrelativiza-tionthrowsawaythenorm-sensitivity.Butitdoesn’ttelluswhatthenorm-sensitivityamountsto.
Indeed,theframeworkitselfdoesn’tsayanythingaboutthedeepdifferences there are between norm-sensitivity and the kind of per-spectivity one has in the temporal ormodal cases. The differencesarisefromthefactthatwecanevaluateatemporallysensitiveasser-tionas“objectivelycorrect” if it is truerelativetothetimeintendedbytheutterer(typically,thetimeofutterance),andaworld-sensitiveassertionas “objectivelycorrect” forapossibleutterer if it is trueattheintendedworld(typically,thepossibleutterer’sown).Normativediscoursedoesn’tworklikethat:indeed,foranevaluativisttherejustisnosuchthingasobjectivecorrectness;thereisonlycorrectnessinthesenseofdisquotationaltruth(where“‘p’istrue”inheritswhatevernorm-sensitivitythereisin‘p’).
Thedifferencebetweenthetemporal/modalcasesandthenorma-tive casebecomes especially vividwhenone thinks about disagree-ment.Ifitwerepossibleforpeopleindifferenterastocommunicatewitheachother,therewouldn’tbedisagreementbetweenthemwhenone asserted “Theworld’s human population is now over 3 billion”andtheotherasserteditsnegation;similarlyif“residentsofdifferentpossibleworlds”couldcommunicatewitheachother.That’sbecausetheir“objectivecorrectness”conditionsarecompatible.But itseemsto be of the essence of norm-sensitivity that disagreement doesn’twork thisway; fromanevaluativistperspective, thereareno“objec-tivecorrectness”conditions in thiscase.There isn’tanythingwithintheGibbardframeworkofnorm-worldpairsthatexplainsthespecialpragmaticsofdisagreement(as,again,Gibbardrecognizes),butthereisalsonothingthatrulesouttherebeingsuchaspecialpragmatics:theframeworkisjustsilentonthematterofwhatconstitutesagreementanddisagreement.TheroleoftheGibbardmodel,asIsaid,istoget
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –16– vol.18,no.12(june2018)
don’tpositacounterfactualdependenceonouracceptanceofthosenorms.
SharonStreet (2011)hascomplained,quiteplausibly, that recentquasi-realistviewshavegonesofartowardacceptingwhattherealistsaysthattheyloseanyepistemologicaladvantageoverrealism.(Inthepresentcontext,thekindofepistemologicaladvantageconcernswhatIearliercalledmeta-skepticism:skepticismaboutclaimsofjustificationorreasonableness,which,aswe’veseen,canindirectlyleadtoground-leveldoubts,e.g.abouttheexternalworld.)Butthecurrentproposalisnotquasi-realistinthatsense:unlikethequasi-realist,theevaluativistmakesnoclaimtobejustliketherealistasregardsobjectivity.Afterall,as faras themetaphysicsgoes, theevaluativist is inexactagree-mentwiththeindividualorgroupsubjectivist(inStreet’sterminology,theconstructivist).Theonlydifferenceisinthewaythatmetaphysicsisaccommodatedinthetreatmentoflanguage.Iregardtheevaluativ-istviewofhowtoaccommodateitasfarmorenaturalthanthesubjec-tivist/constructivist,butonmattersofnon-counterfactualobjectivityandonconsequentmattersofepistemologytheyseemtomepreciselythesame.22
Iconcedethatthereismoretobesaidaboutboththedistinctionbetweennormsforvaluationand“purebeliefs”,andthekindofnon-counterfactualobjectivitythatisconnectedtoit.23Quasi-realistsseem
22.ArefereehassuggestedthattheepistemologicalchallengethatStreetisad-dressing(the “reliabilitychallenge”; Iwon’t take thespace toexplain it) isgenerated solelyby the counterfactual aspect of objectivity. I disagree: foranevaluativist,thechallengeismetbyouractualacceptanceofnormsthatapplyevenincounterfactualcircumstanceswhereweacceptdifferentnorms.(Someoftheliteratureonthereliabilitychallengeengendersconfusionoverthis,by talkingabout “mind-independence”or “attitude-independence”am-biguously: a claimcanbe sensitive toour actual attitudes, in theGibbard-MacFarlanesense,withoutbeingcounterfactuallydependentonthem.)
23. Thedifficultyofachievingcompleteclarityonthenon-counterfactualnotionofobjectivityhassomeparallelinthedifficultyofachievingcompleteclarityonthenotionofdeterminacy.Justasthetruismthat‘Joeisbald’istrueifandonlyifJoeisbaldwouldseemtoleaveopenthatthereisaquestionofwheth-erthereisadeterminatefactofthematterwhetherJoeisbald,similarlythetruismthat‘thejokewasfunny’istrueifandonlyifthejokewasfunnywouldseemtoleaveopenthatthereisaquestionofwhetherthereisanobjective
bybringing into consideration straightforwardly factualclaims.
• We can perfectly sensibly apply the words ‘true’ and‘false’ tonormativeclaims:“Ifwhathesaid is true, thenIshouldn’tdoX”isperfectlysensible,evenifnormativeclaimsareamongthethingshesaidthatarecentraltomyconditional conclusion that I shouldn’t doX. (Notonlyis itsensibletosoapplythenotionsoftruthandfalsityto normative claims, but disallowing such applicationswoulddefeatthemainpurposesthatthenotionsoftruthandfalsityserve.)
• Connectedwith the last point, it’s natural to say that Ibelievesuchnormativeclaims:it’sjustthatthisisn’tpure belief;ithasinadditionanevaluativeelement.
• Moreover,manynormativeclaimsclearlyhavewhatwemight call counterfactual objectivity:we can properly saythatIwouldn’tbejustifiedinbelievingtheEarthflatevenif I hadverydifferent epistemic standards thatdictatedsuchbelief; indeed,wecanpointout that it is inconsis-tentwiththestandardsweaccepttopositivelyevaluatebeliefinaflatearthbypeoplewiththosestandards.
The bulleted claims go some of theway toward “quasi-realism”:muchofwhattherealistsayscanbesaidbytheevaluativist/expressiv-istaswell.Butnotall:inparticular,andincontrasttoBlackburn1993andthemorerecentGibbard(e.g.2003),Ithinkthatthereisanissueofobjectivitythatgoesbeyondtheissueofcounterfactualobjectivity,andthattherealistbelievesinthatfurthersortofobjectivitybuttheexpressivist,quiteproperly,doesn’t.21Ithastodowiththefactthat,ontheexpressivistview,ournormativeclaimsarise out ofournormsbut
21. Somewhatrelatedtothis:advocatesofquasi-realismoftensaythatacceptingtheirdoctrinedoesnot in anywayaffect ground-levelnormativepractice;whereasamainthemeofthepresentpaper is thatevaluativismdoeshavesucheffects.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –17– vol.18,no.12(june2018)
Afinalwordonobjectivity:it’sbesttoviewitascomingindegrees.Therearesomenormsofevaluationwhichhavesomuchgoing forthemthatformanypurposestheycanberegardedasobjective:thesemightinclude
(i)somelogicalnorms,
(ii)somenormsofcomparativeevidenceforstatisticalhypotheses(e.g.“IfanexperimentEwasperformedandledtoresultR,thenE˄Rfavors a statistical hypothesis that gives R higher probability in cir-cumstances E over another statistical hypothesis that gives it lowerprobability”),and/or
(iii)someformofthe“PrincipalPrinciple”.
Idon’tmeanthatsuchnormsareentirelyuncontroversial:theyaren’t,especiallywhen itcomes to thedetailed formulation. (Therearede-batesaboutthedetailsof“thecorrectlogic”andaboutthebestformofthePrincipalPrinciple;andviewsofstatisticalinferencethatseemin conflictwith the comparative likelihood rule (ii) arewidespread,thoughnotaswidespreadas theyoncewere.) I’mdoubtful that thecontroversiesoversuchrulesoughttocountasobjectiveinthewaythatcontroversiesovertheexistenceofgravitationalwavesare,24butifsomeonewantstoarguethattheseareperfectlyobjectivematters,I’mnotgoingtoputupafight.Incontrast,thenon-comparativeevalu-ationofhypotheses(evenstatisticalones),andtheevaluation(evencomparative)ofhypothesesthatarenotpurelystatistical,bothmakeafarmoreserioususeofeitherpriorcredences(inaBayesianframe-work)orsomethingthatplaysasimilarroletothatinanon-Bayesianframework.(Thenon-comparativeevaluationofhypothesesinvolvessomething like a prior credence function over the space of alterna-tivehypotheses;andanyevaluationofhypothesesthataren’tpurely
24. Evenwhenthechoicebetweentwonormsisn’tobjective,therecanbecon-siderableadvantagesofonenormover theother; rationaldebateover thenormconsistsinpointingoutsuchcomparativeadvantages.
to question that there is such a further aspect to objectivity, but asCrispinWrighthasoftenargued(e.g.Wright1992),thisisexception-ally implausible: for instance, claims aboutwhat is funny can havecounterfactualobjectivity,butit’shardtobelievethattheyareobjec-tiveinanyverydeepsense.Ofcoursetheevaluativistaboutnormsinmoralityand/orepistemologywillgrantthattherearekeydifferencesbetween the form that the non-objectivity takes in these cases andin the case of humor—for instance, our epistemological andmoralnormsarewaymorehighlystructuredthanisoursenseofhumor,andfarmoredeeplyentwinedwithourgoals.Butmypointwasn’ttoas-similatethenormativetothecomic,butsimplytosaythatthereisfarmoretoobjectivitythancounterfactualobjectivity.
Ihopeit isalsoclearthatonmyviewtherearedeepdifferencesbetweenepistemologicalnormativityandmoralnormativity.Whereasthereispresumablyconsiderableindirectconnectionbetweenmoralnormsandpreferencesaboutwhatkindofworldonewants to livein,it’shardtoseemuchanalogousconnectionintheepistemologicalcase,atleastatthefirst-orderlevel(seenote18).Morefundamentally,inevaluatingeitherepistemologicalormoralnormsintermsofhowwelltheysatisfygivenfactualdesiderata,oneneedstouseepistemo-logicalnormsbutdoesn’tneed tousemoralnorms.This lastdiffer-enceiswhatmakesreliabilist-likeviewstempting:itmakestemptingthatwhatmatterstoepistemologicalgoodnessisapurelyfactualmat-ter,somethinglikethe“truth-conduciveness”oftherulesbywhichabeliefisformedandretained.Butsuccumbingtothetemptationisn’tinevitable,andI’vementionedsomeproblemswithdoingso.Aboveall,Idon’tthinkthere’smuchhopeinmakingtheideaof“truth-con-duciveness”atallclear;andeveninsofarasitisclear,exclusivefocusonactualtruth-conducivenessdemotestoomuchthe“internalist”fea-turesofepistemologicalevaluation.
factofthematterwhetherthejokewasfunny.Theremaybemorethanjustaparallelhere:maybeindeterminacyisjustaspecialkindofnon-objectivity.Inanycase,thepositivetaskofexplainingeitherdeterminacyorobjectivitywithcompleteclarityisdifficult.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –18– vol.18,no.12(june2018)
“entitlements” idea discussed in Section 1, seem to let it in throughthebackdoorviaskepticismabouttheentitlements.Thenormativeanti-realismdoesforeclosethat meta-justificational routetoskepticism.Nonetheless,itwouldbepossibletoadheretotheanti-realismwhileadoptingnormsthatwouldleadtoskepticismbyotherroutes.
Onesuchnorm(“CompleteOpen-Mindedness”)isthatweshouldtakeseriouslyanyhypothesis that iseversuggested,nomatterhowsilly:keepitasalivepossibilityunlessonehasnon-question-beggingreasons toeliminate it.Adopting thismethodologywouldallow forinitial justification inclaimsabout theexternalworld;butonce thebrain-in-vat (or even theCartesiandemon)hypothesis is suggested,themethodologywouldnolongerallowfortheacceptanceofclaimsaboutthephysicalworld.(Atleast,notunlessnon-question-beggingreasons could be provided to eliminate it, and that presumably isimpossible.)
Theobviousanti-realistresponsetothisissimplythatwedon’tac-ceptCompleteOpen-Mindedness,andshouldn’t.Descartes’powerfulwritingpersuadedmanyphilosophers to takeanapplicationof thatnormatleastsomewhatseriously:seriouslyenoughtohaveabadcon-scienceaboutgoingonasbeforewithoutprovidingreasonsagainstthedemon.But thenormalsohasconsequences thatnotevenDes-cartescouldhavetakenseriously.For instance,consider thehypoth-esisthattheworldisasscientistsbelieve,andwillremainsoaslongasnoonehopsaroundtheSouthPole91timesononelegwhilesing-ing“TwistandShout”infalsetto;whereasifsomeonedoesthat,globalwarmingwillbereversedandplentyoffoodwillbecomeavailabletoeveryoneinperpetuity.25Iventuretosaythatevenhavingraisedthishypothesis,noonewillevertestit,despitethebenefitstomankinditpromises.That’sbecausewedon’tacceptthenormofCompleteOpen-Mindedness.Andweshouldn’t:ournormsaboutwhatnormstoac-ceptaresuchastodictatethatacceptingitwouldbeathoroughlybadidea.
25. IrecallHilaryPutnamgivingasimilarexampleinaclassmanyyearsago.
statisticalinvolvessomethinglikepriorcredencesofauxiliaryhypoth-eses.)ItisprimarilyherethatI’dwanttoinsistonasignificantlevelofnon-objectivity.Thoughevenhereitisamatterofdegree(wherethedegreesarevagueandnotlinear-ordered):ifaclaimisgiventhesameevaluationbyallmethodswecantakeseriously,thenitshouldcountashighlyobjective.Myviewisthatdichotomizingbetweenthe“objec-tive”andthe“non-objective”,inacontext-independentway,isn’tveryuseful. (Understoodascontextual, avaguedichotomymakes sense:there’snothingwrongwithcountingclaimswithalowdegreeofwhat-everkindofnon-objectivityissalientinthecontextas“objective”,aslongaswe’reclearthatthat’sallwe’redoing.)
6. Skepticism again.
Let’s looka littlemoreat the impactof theevaluativist/expressivistpictureonepistemology.I’vementionedskepticismalready,butwillsay abitmore about it in this section. In thefinal sections I’ll givetwomoreexamples,thefirstofwhichhasafairlycloseconnectiontoissuesdiscussedinSections3and4,andthesecondofwhichissome-whatfurtherremoved.
Inotedearlyon(footnote1)thatthediscussioninthispaperdoesnot target all formsof skepticism.And in addition to the examplesmentionedinthatfootnote,itdoesnotdirectlychallengeskepticismbased on Benacerraf-style arguments against certain forms of Pla-tonism:theretheskepticismisbasednotonlackofinitialjustificationbutonanapparentundercuttingofthatjustification.(ThediscussioninthispaperdoeschallengetheanalogofBenacerraf-styleargumentsfor normativity, but only by challenging the normative realism onwhichtheyarebased;itisaseparatequestionwhattosayaboutspe-cificformsofmathematicalPlatonism.)
Moreimportant,thenormativeanti-realismrecommendedinthispaper does not totally foreclose even the kinds of skepticism thatwere targeted: itmerelymakes themfar lesswell-motivated.Partofthereasonwhythoseformsofskepticismcanseemcompellingisthateventheformsofnormativerealismdesignedtobluntit,suchasthe
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –19– vol.18,no.12(june2018)
leadtoparadox,Idon’tthinkhegetsatwhatseemstometobewrongintheconception).Hesays:
…wehavetotrytofigureoutwhatistruefromtheevi-denceavailabletous.Todothis,werelyonasetofepis-temicrulesthattellusinsomegeneralwaywhatitwouldbemost rational to believe under various epistemic cir-cumstances.Wereasonaboutwhattobelieve;andwedosobyrelyingonasetofrules.[Boghossian2008,p.472]
So the ruleshere are rules governingwhat is rational to believe; buttheyalsoserveasomewhatmoredescriptivefunction,inthattheyareinvolved in the reasoning processes ofsomeonewhorationallybelieves.Idon’tthinkthattheseneedbethesame;indeed,Ithinkiteasiertomotivate talkof rules involved in the reasoningprocessesofbeliev-ers(rationalandotherwise)thantomotivatetalkofrulesofwhatit’srationaltobelieve.
I’m sympathetic toBoghossian’s claim thatwe reasonby relyingonasetofrules.Talkofrulesisprobablyimportantinthequasi-de-scriptivephaseofepistemology:thatis,inthetaskofgivingidealizeddescriptionsofhow,atagiventime,anactualorpossibleagent(ormaybeanactualorpossiblecommunityofagents)wouldfunctioninabsenceof“performanceerrors”.(By“rules”Idon’tmeanjustexplicitlyformulatedrulesthat theagentconsults; likeBoghossian, Imeantoincluderulesthattheagentfollowsblindly.)AsI’vesaid,thereneedn’tbeaclearmodel-independentdistinctionbetweenwhatcountsasaperformance error andwhat doesn’t.Wemodel epistemological be-haviorby idealizeddescriptions,wherebehavior thatdoesn’tfit thedescriptioniscountedasperformanceerror;whatcountsasaperfor-manceerrorononeidealizationmaybebuiltintotheidealizationonanother,anddifferentsuch“levelsofidealization”maybeusefulfordifferentpurposes.Therulesarelevel-dependent.
Itmaywellbethatinanyidealizeddescriptivemodel,someruleswill be fundamental in the sense that themodel doesn’t allow any
Again, anormative realist could raiseameta-justificational issuehere:what is the“objective justification”(notbasedonmeta-normsthatcouldbequestioned)of theclaimthatweshouldn’taccept theCompleteOpen-Mindednessnorm?Anadvantageoftheevaluationistmetaphysicsisthatitobviatesthatquestion.
Theremaystillbeanissueforthenormativeanti-realist:giventhatwedon’taccepttheCompleteOpen-Mindednessnorm,whatnormofcomparablegeneralitydoweaccept?Thequestionhasapresupposi-tion:thatwedoacceptnormsofcomparablegenerality.Isuspectthatthatpresuppositionisfalse,thoughthisisnotanissueonwhichIwanttotakeastand.26Isuspectthatwecan’tdomuchbetterthansay,“Weoughttobeopen-minded,withinreason,buteachpersonmustdecideforhim-orherselfjustwhatalternativehypothesesareworthtakingseriously.”
7. Epistemic rules.
Inadditiontoskepticism,thereareotherwaysinwhichtheevaluativ-ist/expressivistpictureaffectsground-levelepistemology.Forinstance,Ithinkthattheevaluativistpictureundercutsacertainconceptionof“rulesof rationalbelief”.That conception tends to lead, inparticular,toaconceptionoffundamentalrulesofrationalbeliefthat are immune from rational revision.ButwhatIseeasthebasicerror intheconcep-tioncomesbeforethat.ThepictureIrejectisstatedveryclearlyintheopeningparagraphofPaulBoghossian2008(andthoughtheburdenofthatpaperisthattheconceptionofrulesdescribedthereseemsto
26.Theconnectionbetweenevaluativismandtheissueofwhethertolookforcompletelygeneralnormsisaweakone.Idothinkthereismorepressureto aspire to complete generality in norms if one takes there to be ameta-physicallybasedcorrectnessofnormsthanifonesimplytakesournormstobeproductsofourbiologicalandculturaladaptation toourcircumstances,forwhichtalkofcorrectnessdoesn’tarise.ButIacknowledgethatthereareparticularistrealistswhoverymuchdownplaythepressurestowardsystema-ticityfromarealistperspective,andalsothatachievingcompletegeneralitymighthavesomekindofappealfromtheanti-realistperspectiveaswellasfromtherealist.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –20– vol.18,no.12(june2018)
For another thing, speaking of “rules of rational belief” stronglysuggests that theonly factor in an evaluationof the rationality of agivenbeliefiswhetheritwasproducedviaagoodruleandwithoutperformanceerror.Thatisunobvious.
Another problemwith the “rules of rational belief” terminologyis that it strongly suggests that there isa setofoptimal rules:goodagentsaretheoneswhofollowthoserules,orsomeapproximationtothem.Ifthereissuchasetofoptimalrules,whichIdoubt,thatneedsan argument, and nothing in the quasi-descriptive picture suggestsanysuchargument.
Evenmore important than theoptimality issue is the suggestionthattherearerulesofrationalbeliefthat are fundamental in a model-in-dependent sense.Nothingabouttheroleofrulesinthequasi-descriptivepictureofanepistemologicallygoodagentprovidesthebasisforsuchaclaim.Whentherearemultiplelevelsofdescriptionintermsofrules,howarewetodecidewhichlevelcorrespondsto“therulesofrationalbelief”?Indeed,inthetoymodelofthe2-tapeTuringmachine,itisn’tatallclearwhatshouldcountasa“ruleofrationality”.
Thatisespeciallysoifonemakestheassumption(whichisn’tex-plicit in theBoghossianquote,butoften takenaspart of the “rulesofrationalbelief”picture)thatthe“rulesofrationality”arerationally indefeasible.Ifwetaketherulesofrationalitytoincludetypicallogicalor inductivemethods,suchasmightbeincludedintheinstructionsinthetoptapeatagiventime,thenthetoymodelisoneinwhichtherulesofrationalitycanchange.Andtheymightchangeinanintuitive-lyrationalway:thechangemightcomeaboutbynormaloperationofthemachine,ratherthanbyamalfunction,andthisnormaloperationmightbeanintuitivelygoodoneinthat
(i)itchangesthemethodonthetoptapeonlyinsituationswhereadeficiencyinthatmethodhasbeenexposed,and
(ii)thekindofchangesitmakesseemwell-designedforcorrectingthosedeficiencies.
considerations toundermine them.(Such fundamental rules, if theyexist,needn’tbedeterministic:theycouldallowthatincertaincircum-stancesan“internalcoinflip”dictateshowtoproceed.)Allowingforfundamentalrulesinthissensemightseemtomakeaproblemfortheidea,implicitinSections3and4,thatevenbasicdeductiveandinduc-tiveandperceptualmethodologycanbechallengedundercertaincir-cumstances:don’twewanttohaveepistemologicalmodelsthatbuildinlogicalandinductiveandperceptualrules,butalsomodelsthatdealwiththeprocessofchallengingthem?Onewaytodealwiththisistosimultaneously employmultiplemodels, at different levels of ideal-ization.Consider,asacrudeillustration,atwo-tapeTuringmachinewherethetoptapecontainsinstructionsforinductivemethodM.Themethodonthetoptapeisappealedtoconstantlyinrewritingthebot-tomtapeusedforordinarypractice.Onlyinveryexceptionalcircum-stancesdotheoverallrulesofoperationofthemachine(themachinetable,whichI’mimaginingtobebuiltintothearchitectureandthusnot explicitly appealed to) dictate rewriting the top tape. From thepointofviewoftheoverallTuringmachinerules,Mismerelyadefaultprogramthathasdefeaters.Butanyonewhowantstogiveamanage-abledescriptionofthebehaviorof themachineatagiventimewillappealnottothemachinetablebuttothedefaultrulesinM.
Torepeat,I’mconcedingthataquasi-descriptiveepistemologywillpositrules,evenperhapsrulesthaton that level of descriptionarefunda-mental.And,ifthequasi-descriptionisofanepistemologicallygoodagent,perhapswecouldcalltherulesitposits(atthatlevelofdescrip-tion)“rulesofrationalbelief”(atthatlevelofdescription).Butspeak-ingthiswayispotentiallymisleading,onanumberoflevels.
Foronething,itmightsuggestthatthenotionofrationalitywillap-pearintherules;andindeed,Boghossianthinksitdoes.Forinstance,heformulatestheruleofModusPonensas“Ifyouarerationallyper-mittedtobelieveboththatpandthat‘ifpthenq’,thenyouareprima facierationallypermittedtobelievethatq”(Boghossian2008,p.472).Butonapicturethatseparatesthequasi-descriptivefromtheevalua-tive,theruleswillnotthemselvesemploynormativeconcepts.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –21– vol.18,no.12(june2018)
fromthegripoftheideathatepistemologyisengagedin“describingthenormativefacts”.
For example,manyBayesianswho reject the idea that for everyepistemological situation there is a uniquely rational epistemic cre-dencefunctiontohaveinthatsituationthinkthatit’simportanttoaskwhat the “constraints”areona credence function—whichcredencefunctionsaccordwith“therulesofrationality”andwhichdon’t.Fromanevaluativistperspective,it’shardtofindanythingsensiblethatthiscanmean.Itisofcoursetruethattherearecredencefunctionsthatitwould be idiotic to employ: e.g. the aforementioned credence func-tionthatprotectstheviewthatthemoonlandingwasahoaxfromallcounter-evidence.Butthatdoesn’tmeanthatthere’sanypointinlook-ingfora“ruleofrationality”that“constrains”usnottoemploysuchacredence function (in somemetaphysical senseof constraining thatisdifficulttomakesenseof,exceptperhapsinatheologicalsenseofdivinepunishment).AsDickJeffreyoncesaid,“Thefactthatitislegaltowearchainmailoncitybuseshasnotfilledthemwithclankingmul-titudes”(Jeffrey1983,p.145).
Onemight,Isuppose,trytointerprettalkof“rationalconstraints”moresubjectively, in termsof the features thatwewouldrequireofa credence function ifweare todeem it rational (or in termsof thefeaturesthatanevaluatorthat we would deem goodwouldrequireofacredencefunctioniftheevaluatoristodeemitrational).Threepointsabout this: First, our evaluationof credence functions isn’t the yes/noaffairthattalkofsatisfyingorfailingtosatisfyrationalconstraintswouldsuggest:forinstance,acredencefunctionmightbeprettygood,butslightlydeficientinnotallowingenoughcredencetocertainkindsof hypothesis. Second, even if constraint talkwereweakened to ac-commodate this (e.g.allowing “constraint” tocome indegrees), thesuggestionseemstorequire thatourevaluationsareorought tobesystematizableinaveryparticularway:thatweemploygeneralcon-straints-to-a-certain-degreeoncredencefunctions,andevaluateacre-dencefunctionasrationaltopreciselytheextentthatitmeetsthesegeneralconstraints.AsIsaidneartheendofSection6,thisassumption
So,inthetoymodel,therulesofMaren’tacandidateforindefeasiblerationalrules.Andtheonlyotherrulesthatplayedaroleinthatde-scriptionweretherules(notexplicitlyrepresentedones)thatarebuiltintothemachinetable.Butthemachinetableisfartoo“low-level”tobenaturallyviewedasdescribing“rulesofrationality”.
Perhapsitwillbesaidthatifthemachinetableoperateswellinitsrevisionofthemethodsonthetoptape,itwillaccordwithcertainheu-ristics(thatitneedn’texplicitlyrepresent).Butsuchheuristicsneedn’tincludegeneraldeductiveandinductivemethods;theycanbemerelyrulesforchangingthetoptape;theyaren’tasubstituteforwhat’sonthetoptape,soitwouldbeoddtoconfinetheterm‘ruleofrationality’tothem.(Moreover,evaluatingachangeinMasrational inagivencaseneedn’t require a judgment that themethodused to change itwouldingeneralleadtorationalchanges.)
Themoral then is thatweneed to cleanly separate thequasi-de-scriptivetaskfromtheevaluative(whilegrantingthattheevaluationofthemethodsproducedinquasi-descriptiveaccountshasconsider-ablebearingontheevaluationofbeliefs).Oncewemakethissepara-tion,certainargumentsfortherationalindefeasibilityofmethodswetaketobegoodonesevaporate.27
8. Rational constraints.
Manyphilosopherswhohavegonesomewaytowardthesortofeval-uativismI’vebeenrecommendingdon’tseemtometohavebroken
27. Thereare related salutaryeffects too. For instance, theargument thatBog-hossianhasgiven inseveralpapers forregardingstandarddeductiveruleslikeModusPonensas“basicrulesofrationality”seemstobebasedonthesupposedinevitabilityofthefundamentalemploymentofsuchrulesinade-scriptiveaccountofanyagentthatreasonsproperly. I’mskeptical that it isinevitable,atthebasiclevel:aTuringmachinecanbeprogrammedtoinferaccordingtoModusPonens,butitsbasicrulesdon’tincludeModusPonens(thoughofcourseweuseModusPonensinreasoningaboutwhatsuchama-chinewoulddo).Asnoted,wepresumablydon’twanttocallthebasicTuringmachinerules(oranalogously,therulesgoverningtheevolutionofthestateofaneuralnetworkinhumans)“rulesofrationality”,butthepointismoregeneral:maybethe“basicrulesofrationality”,iftherearesuch,arenotrulesthatinvolvelogicalnotionsatall.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –22– vol.18,no.12(june2018)
Gibbard, Allan 1986. “An Expressivistic Theory of Normative Dis-course”.Ethics96:472–85.
------------------1990.Wise Choices, Apt Feelings: A Theory of Normative Judg-ment.HarvardUniversityPress.
------------------2003.Thinking How to Live.HarvardUniversityPress.Jeffrey,Richard1983.“BayesianismwithaHumanFace”. InJohnEar-
man,ed.,Testing Scientific Theories(UniversityofMinnesotaPress),pp.133–56.
Kolodny,Niko,andJohnMacFarlane2010.“IfsandOughts”.The Jour-nal of Philosophy107:115–43.
Kratzer, Angelika 2012. Modals and Conditionals. Oxford UniversityPress.
Lewis,David1971.“ImmodestInductiveMethods”.Philosophy of Science 38:54–63.
----------------1974.“SpielmanandLewisonInductiveImmodesty”.Phi-losophy of Science41:84–5.
MacFarlane,John2005.“MakingSenseofRelativeTruth”.Proceedings of the Aristotelian Society105:321–39.
---------------------2014.Assessment Sensitivity: Relative Truth and Its Appli-cations.OxfordUniversityPress.
McGee,Vann1985.“ACounterexampletoModusPonens”.The Journal of Philosophy82:462–71.
Pollock,John1987.“EpistemicNorms”.Synthese71:61–95.Reichenbach,Hans 1938.Experience and Prediction: An Analysis of the
Foundations and the Structure of Knowledge. University of ChicagoPress.
Salmon,WesleyC.1967.Foundations of Scientific Inference.UniversityofPittsburghPress.
Street,Sharon2011.“Mind-IndependenceWithouttheMystery:WhyQuasi-RealistsCan’tHaveItBothWays”.Oxford Studies in Meta-Eth-ics 6:1–32.
Williamson, Timothy 2007. The Philosophy of Philosophy. BlackwellPublishing.
Wright,Crispin1992.Truth and Objectivity.HarvardUniversityPress.
seemsfar fromobvious.Thirdandmost important, Idoubt that thesubjectivistreadingcapturesthespiritbehindtypicaltalkofrationalconstraints:afterall,mostofusrecognizethatourownjudgmentsastowhatisrationalaren’tthelastword,sothatacredencefunctionthatdoesn’tmeetthe“constraint”ofwhatwedeemrationalmightturnouttohaverealadvantages.28
References
Blackburn, Simon 1993. Essays in Quasi-Realism. Oxford UniversityPress.
Boghossian,Paul2003.“BlindReasoning”.Proceedings of the Aristotelian Society Supplement77:225–48.
-------------------- 2008. “EpistemicRules”.The Journal of Philosophy 105:472–500.
Brown, Bryson, and Graham Priest 2004. “Chunk and Permeate, aParaconsistent Inference Strategy. Part I: The InfinitesimalCalcu-lus”.Journal of Philosophical Logic33:379–88.
Carroll, Lewis 1895. “What the Tortoise Said to Achilles”. Mind 4:278–80.
Enoch,David,andJoshuaSchechter2006.“MeaningandJustification:TheCaseofModusPonens”.Noûs40:687–715.
Feyerabend,Paul1975.Against Method.NewLeftBooks.Field,Hartry 2000. “Apriority as anEvaluativeNotion”. InPaul Bog-
hossianandChristopherPeacocke,eds.,New Essays on the A Priori (OxfordUniversityPress),pp.117–49.
--------------- 2009. “EpistemologyWithoutMetaphysics”. Philosophical Studies 143:249–90.
---------------2015.“WhatIsLogicalValidity?”.InColinR.CaretandOleT.Hjortland, eds.,Foundations of Logical Consequence (OxfordUni-versityPress),pp.33–70.
28. I’vereceivedagreatdealofhelpfulcommentaryonpreviousdrafts,whichhaveledtobigimprovements.ThankstoPaulBoghossian,DavidEnoch,JimPryor,StephenSchiffer,ElliottSober,LisaWarenski,CrispinWright,andtwoanonymousreferees.
hartryfield Epistemology from an Evaluativist Perspective
philosophers’imprint –23– vol.18,no.12(june2018)
-------------------- 2014. “On Epistemic Entitlement (II): Welfare StateEpistemology”.InDylanDoddandEliaZardini,eds.,Scepticism and Perceptual Justification(OxfordUniversityPress),pp.213–47.