Barret & Stanford Prediction The ...

[ Pobierz całość w formacie PDF ]
Draft: Revisedversionto appearin ThePhilosophyof Science:An Encyclopedia,ed.
JessicaPfeiferandSahotraSarkar.New York: Routledge,Inc.)
Prediction
Jeff BarrettandP. Kyle Stanford
Departmentof Logic andPhilosophyof Science
Universityof California,Irvine
Introduction
is presumablyresponsiblefor the ideathat
predictionsmustbe aboutthe future. But in scientificcontextswe often characterizea
theory'spredictionsasits implicationsor entailmentswithout regardfor temporal
constraints,aswhenwe saya successful
theoryof cosmologypredictsthe existenceof
cosmicbackgroundradiationat all times. We alsousethe languageof predictionto
describedeclarativeassertionsaboutpastandpresenteventsmadein light of a theory,as
whenwe useevolutionarytheoryto predictthattheremustbetransitionalfossils
connectingreptilesandmammals.A temporalelementmight bepreservedby insisting
thatthesearereally casesof 'postdiction' or 'retrodiction' or evenshorthandpredictions
aboutfutureevidentialfmdings. But perhaps
thesecomfortableextensionsof our
thatthe centralelementin predictionis not
temporalbut epistemic. To predictis to makea claim aboutmattersthat arenot already
known,not necessarilyabouteventsthat havenot yet transpired.
is to predictit correctlyon the basisof a well-conflrlnedtheory. Predictive
languageseemsmostappropriatein caseswhenwe makeclaimsaboutunknownmatters
usingtools(like inductivegeneralization,scientifictheorizing,or sheerguesswork)that
canbecontrastedwith moredirectmethodsof ascertaining
the sameinformation(like
simplyobservingin the right placeand/orat theright time and/orunderthe right
conditions,or looking for physicaltracesof somepaststateof affairs). Although specific
philosophicalandscientific conceptionsof whatis immediatelygiven in experienceor
knowndirectly haveshiftedovertime, our predictivelanguagehascontinuously
respected
the fundamentalideathata predictionis a claim aboutunknownmattersof fact
whosetruth or falsity hasnot alreadybeenindependentlyascertainedby somemore
directmethodthanthat usedto makethepredictionitself.
it goesbeyond
whatwe alreadyknow mostdirectly, but this samefeaturerenderspredictioninherently
risky. Themostinterestingandusefulpredictionstypically concernmattersto which we
haveintersubjectiveaccess,sowe candiscoverthat theyarewrong whenthey are:
1
Whetherwe predictrainfall, recessions,or racetrackwinners,predictinganevent
or stateof affairsoften,perhapseventypically, involvessayingthat it will happenbefore
it occurs,andthis commonassociation
predictivelanguagemorenaturallysuggest
Of coursepredictioncannotbeassimpleasthat,becauseoneway to know
something
As this accountsuggests,successfulpredictionis valuablebecause
for predictionsconcerningunexperiencedmattersof fact, andhetook this to
illustratethatreasonor rationaljustification doesnot play anythinglike therole we
usuallysuppose
in our cognitive lives. In his 'skepticalsolution' to theproblem,Hume
arguesthat whatgeneratesour expectationsaboutunknowncasesis a primitive or
instinctivepsychologicaldispositionhecallscustom,which is not itself mediatedby any
processof reasoningat all. Customleadsus,automaticallyandwithout reflection,to
expectaneventof typeB on the appearance
of aneventof typeA just in caseB' shave
followed A's reliably in thepast. Thus,Humeoffersusa naturalisticex~lanationof the
psychologicalmechanismby which we makeempiricalpredictionsbut not anyrational
_Justification
for this practice. But this is not to saythat we makea mistakein relying on
custom:Not only do we haveno choicein the matter,Humeargues,but "Custom... is
thegreatguideof humanlife. It is thatprinciplealonewhich rendersour experience
usefulto us Without theinfluenceof custom,we shouldbeentirely ignorantof every
matterof fact, beyondwhatis immediatelypresentto thememoryandsenses"(29). The
fact thatthereis no rationaljustification for suchanimportantandusefulcognitive
function,hesuggests,simply illustratesthatNaturehassecured"so necessaryanactof
themind, by someinstinct or mechanicaltendency"ratherthanleavingit "to the
fallaciousdeductionsof our reason"(37). Thecentralaspectsof our cognitivelives, he
suggests,arenot productsof, nor evensubjectto, our reasoningfaculties. Insteadthey
are"a speciesof naturalinstincts,which no reasoningor processof thethoughtand
understanding
is able,eitherto produce,or to prevent"(30).
A furtherproblemof inductivejustification, arguablyanticipatedin Hume's
treatment,is clearly articulatedby NelsonGoodman(Goodman1954). Herethe problem
is not how to justify thebelief that unexperiencedcaseswill resembleexperiencedones,
but how to understand,categorizeor describeexperiencedcasessoasto knowjust what
it would belike for unexperiencedcasesto resemblethem. Our inductiveevidencefully
supportstheclaim that all emeraldsaregreen,for example,but it equallywell supports
theclaim thatthey areall grue,where'grue' means'greenif first observedbefore2010
andblue if first observedafterwards'. Thosewho believethatemeraldsaregruerather
thangreen,however,will haveexpectationsconcerningtheappearanceof emeraldsthat
divergesignificantly from our own startingin 2010. Nor canwe saythat thepredicate
grueis somehowartificially conjunctiveor really disguisesa change,Goodmanargues,
for it is only relativeto a setof predicates
thatregardsgreenandblue asnatural
thatit doesso. If we takegrueand,say,bleen(understoodas 'blue if first
observedbefore2010,andgreenif [lIst observedafterwards')asnaturalor primitive
predicates
it will be 'green' thatmustbedefinedin anartificially
conjunctiveway (i.e. 'grue if first observedbefore2010andbleenif first observed
afterwards).But of course,it wasthechoiceof greenandnot grueasnatural,primitive
or singularlyappropriatefor law-like generalization
thatwe soughtto defendin the first
place. Goodmanthusarguesthatanyattemptto useinductiveevidenceto projectfuture
or unknowncasesrelieson a setof entrenchedpredicates,andit is controversialwhether
theentrenchmentof onesetof predicatesratherthananothercanberationally defended.
Like Hume's custom,Goodman'sentrenchmentmay offer a kind of naturalistic
explanationof how we cometo makethepredictionsthatwe do, but not onethat seeksor
providesanyrationaljustification for our practices.
3
whatsoever
categories
for a language,
lawsinvokedarestatisticalprobability
statements.Herea specificeventis not logically implied by the boundaryconditionsand
laws,but only supportedto a certaindegree(175-177). For Hempel,theconclusionof
anyargumentof this form qualifiesasa predictionif E refersto anoccurrenceat a time
laterthanthatat which the argumentis offered. A fascinatingandcontroversialfeature
of this accountis the symmetryit assertsbetweenpredictionandexplanation:To explain
aneventby appealto a setof lawsandconditionsis simply to showthat we could have
predictedit usingthem.
is true (1980,57). As in theempiricisttradition moregenerally,
then,thedistinctionbetweenobservableandunobservablephenomenamustdo
significantwork. In orderto distinguishobservableandunobservableaspectsof the
naturalworld, vanFraassenadoptswhathecallsa semanticapproach.To presentan
empiricaltheoryis to specifya modelof thetheory:a domainof objectstogetherwith a
descriptionof the propertiestheycanhaveandtherelationsthey canbearto oneanother.
In presentingthe theory,onealsospecifiesthosesubstructuresof its modelthatare
candidates
for representing
the observablephenomena.And a theoryis empirically
givenin our phenomenalexperienceareisomorphicto the
observablesubstructuresof the model(64).
this distinctionis supposed
to begroundedin the actualobservationalcapacitiesof humanobservers,andin general
it is naturalscienceitself that tells uswhattheseobservationalcapacitiesare. A kitchen
tableis supposed
to beobservable,while a bariumatomis not. But eventhis simple
contrastillustrateswhy the distinctionis difficult to drawin a principled way, especially
if we appealto our bestscientific theories.Our besttheoriestell usthat whena single
bariumatomis illuminatedby a bright laserit canbemadeto scatteranamountof light
thatis easilydetectableby the nakedhumaneye. And thephysicalprocess
involvedhere
to how our besttheoriesdescribethe observationof a kitchentable
whenlight is scatteredby its constituentatoms. Van Fraassen
repliesthat a distinctionin
degreeof observabilityis still a distinction. But evenso,it remainsunclearnot onlyjust
how a distinctionof degreeshouldbedrawnhere,but alsothat it canbe drawnin sucha
way asto do theepistemicwork expectedof it. The upshotis that it remainspuzzling
just whatanygiventheory'sempiricalpredictionsareon this account.
The suggestion that we regard observability as a question to be settled by natural
scienceis perhapspromising. But how could our best theories tell us what is observable?
If they tell us something about the nature of the world and our place in it, then our
theories might be expected to tell us how and the circumstances under which we are able
to make reliable inferences from our measurements. Perhapsthis captures the basic idea
behind the observable-unobservabledistinction after all: It is in just those circumstances
5
eclipse). Hempelalsoallows for whathecalls inductive-statisticalpredictionswherethe
argumenthasthe samebasicforIn,:.but-the
A morerecentcharacterizationof therelationshipbetweentheoriesandtheir
predictionsis given in BasvanFraassen'sconstructiveempiricism. On this account,
acceptinga theoryinvolvesnothingmorethanbelievingthatwhat it saysabout
observablephenomena
adequate
if the appearances
To identify a theory'sempiricalpredictions,then,onemustknow how to specify
theobservablesubstructuresof its model. For vanFraassen,
is preciselyanalogous
belaid uponsucha coincidenceby
personsof scientific attainments(S~stemof Lo~c. ill. xiv, 6; cited in Musgrave
1974,2).
Mill's amazementnotwithstanding,versionsof this Whewellianintuition havebeen
defendedby 'personsof scientificattainments'asotherwisediverseasClavius,
Descartes,Leibniz, Huygens,Peirce,andDuhem. While decidedlylesspopular,the
Millian view hasalsohadinfluential champions,suchasJohnMaynardKeynes(see
Giere 1983,Section3).
is of greaterconfirmationalsignificancethanany amountof
accommodationof existingevidence.But theclaim of a specialconfirmational
significancefor predictiondoesnot requiresuchextremes.For predictionassuchto
enjoya specialconfirmationalprivilegeit seemssufficientthatpredictinga given
phenomenonprovides(or would haveprovided)greaterconfIrmationfor a theorythat
doessothanthe mereaccommodationof thatsame~henomenondoes(or would have).
Let uscall anyview havingthis consequence,
just described,a
form of ~redictivism. Predictivistthemeshaverecentlyloomedlargein debatesoverthe
progressiveness
includingtheextremes
to
confIrmation(especiallyBayesianism),andthe so-calledmiracledefenseof scientific
realism.
Imre Lakatosis widely creditedwith havingreintroducedtheconcernwith the
confirmationalsignificanceof novelprediction,specificallyin connectionwith his
'methodologyof researchprograms'. Lakatos'sbold claim wasthatit is
Qnl.Y
the ability
of the successive
or acceptability.But evenLakatos'sown work includes
severalcompetinglines of thoughtaboutthenatureof novelty. At timeshe seemsto
construethenoveltyof a predictionfor a theorypurelytemporally,thoughhis most
famousaccountholdsnovel predictionto consistin predictingphenomena
that are
"improbableor evenimpossiblein the light of previousknowledge"(1970;cited in
Gardner1981,2), andhe later acceptedElie Zahar'srevisionistproposalthatthe novelty
of a fact for anhypothesisrequiresonly thatit "did not belongto theproblem-situation
which governedtheconstructionof the hypothesis"(1973,103). Eachof theselinesof
thoughthasbeenmorefully developedby laterthinkersevenasthey havelost any
immediateconnectionto concernsabouttheevaluationof researchprograms.
to confirmationcanrecognizea specialconfirmationalsignificancefor novel
prediction;andif not, whetherthis weighsagainstsuchapproaches
to confirmationor
againstthe legitimacyof our predictivistintuitionsinstead. Suchapproachesare
describedastaking into accountonly 'logical' andnot 'historical' relationsbetween
theoryandevidence,or alternatively,only thecontentof theoriesandevidenceandnot
.,
fulfillment are,indeed,well calculatedto impressthe uninformed...But it is
strangethat anyconsideraQ~.,stressshould
Enthusiastshavesometimesgonesofar asto claim that only predictionsof novel
phenomenaareof anyconfirmationalsignificanceat all or that anypredictionof a novel
phenomenon
of researchprograms,the adequacyof variousapproaches
theoriesin a researchprogramto makesuccessfulnovel predictionsthat
bearson its progressiveness
The secondissueof recentinterestconcernswhetherstandardphilosophical
approaches
evidence,this itself would seemto call for somekind of explanation,in light of thegrip
thatpredictivistintuitions seemto_po!dQnour ordinarythinking aboutconfirmation.
to invite a starkconflict of powerfulnonnativeintuitions. Thereis something
especiallyimpressiveaboutsuchfamouscasesof novel predictionasgravitationallight
bendingandthe Poissonbright spot,but it seemsperfectlyfair to askwhy the temporal
orderor otherhistoricalcircumstancesof discoveryshouldhaveanybearingon the
confinnationalsignificanceof theevidencefor a theory. After all, whetherwe already
knewabouta phenomenonor not doesnot haveanyimpactwhatsoeveron how
convincingthe theory'saccountof thatphenomenon
Indeed,any seriousassessment
of theepistemicsignificanceof novelprediction
is. Why shouldit makeany
differencewhetherthe datawerepredictedby a theoryor actedasa constrainton the
developmentor selectionof that sametheoryin thefirst place? The theory'sfit to the
data,the auxiliary assumptions
requiredto obtainthatfit, the theory'sintrinsic
plausibility, andthe like remainpreciselythesame,whateverthe order,manneror other
circumstancesof their discovery. It seemsperverseto treatsuchapparenthistorical
accidentsasrelevantto the degreeof confirmationconferredon thattheoryby the
evidenceat hand.
Predictivism'sdefendershaveturned,therefore,to specifyingcriteria for
genuinelynovel predictionin a way thatseeksto avoiddependenceon apparently
arbitraryor epistemicallyinsignificantfeatures.Meanwhile,their opponentshavesought
to showthatthe apparentsignificanceof novelpredictionis a productof its confusion,
conflation,or frequentassociationwith somethingelsethatis of genuineepistemic
importance.It is, however,sometimeshardto seemorethana rhetoricalor
terDlinologicaldifferencebetweenthepositionsof thosewho seekto creativelyrefine our
conceptionof novel predictionsoasto guarantee
its epistemicsignificanceandthose
in somesense,of a given resultfrom the
formulationof the theoryfor which it countsasa novelprediction. Variousformulations
counta givenphenomenonasnovel for a giventheoryif andonly if it wasnot part of the
'problem-situation'that led to the theory(Zahar1973),wasnot actuallyusedin the
formulationof the theory(Worrall 1978;1989),wasnot known to sometheoristwho
formulatedthetheory(Gardner1982),or fits thehypothesisdespitethe latter's not
havingbeendesignedfor thatpurpose(CampbellandVinci 1983). Theseaccountsdiffer
mostcentrallyin thepreciserole knowndatamustplay in the formulationof an
9
seems
who seekto explainthe apparent
importanceof novelpredictionasdependentuponthe
genuineepistemicsignificanceof somethingelsealtogether. Sometimesbothcamps
appealto the sameor similar relationshipsbetweentheoryandevidence,andit is not
alwaysclearwhethera given authorevenmeansto beexplainingthe epistemic
significanceof novel predictionor explainingit away. A similar ambiguityinfectsthe
discussionof the confmnationalrole of noveltyin actualhistoricalcases.
Complicatingthis dialecticalsituationaretwo competingstrandsof thinking
abouttheepistemicsignificanceof novelprediction,whetherreal or merelyapparent.
The first, sometimescalled 'heuristic', strandholdsthattheepistemicsignificanceof
genuinenoveltyis a matterof the independence,
[ Pobierz całość w formacie PDF ]