Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Oxford University Press and Mind Association are collaborating with JSTOR to digitize, preserve and extend
access to Mind.
http://www.jstor.org
1Philosophers on this side of the debate include Peirce (1931-58), Whewell (1860), Duhem
(1954), Geire (1983), Maher(1988), and Worrall(1989).
2 They include Mill (1843), Keynes (1921), Horwich (1982), Schlesinger (1987), Howson and
the interesting
problemof givinga generalaccountof the conditionsin
whichentailmentof a truthcountsas evidencefor a theory.Forour
purposeswe canunderstandentailmentof dataas relativeto a certain
experiment, andcorresponding classof mutuallyexclusivepossibleout-
comes.Relativeto experimentE, 'thedatum',refersto thatproposition
whichspecifiesthe uniqueactualoutcomeof E.
Second,we shouldbe clearon justwhatthe prediction/accommoda-
tion distinctionis. In a typicalcaseof successfulprediction,a theoryis
firstconstructed,then testedby derivingsome of its consequences,
whicharelaterdiscoveredto be true.In a caseof accommodation the
datumis alreadyknownbeforethe theoryis constructed.Thismight
suggestthatthe crucialdistinctionconcernsthe temporalorderof the-
ory constructionanddatadiscovery.Butwhilesomediscussionshave
focusedon thisdistinction,it seemsthatwhatreallymattersis not tem-
poralorder,but a causalrelation.Intuitively,a theoryis lesswellcon-
firmedif it wasdesignedto entailthe datum,thatis, the conditionof
entailingthatdatumactedas a constrainton the constructionof the
theory.4Of coursethe reasonwhya theorywasnot designedto entaila
datumis usuallythatthis datumwasnot knownat the time. Butif it
was known,yet the theorywas not constructedwith this datumin
mind,it seemsthatit shouldsupportthetheoryin thesamewayandto
the sameextentas it wouldhavehadit not beendiscovereduntilafter
the theorywasconstructed.Thefollowingdefinitionscapturethe dis-
tinctionwhichmattershere.
A theoryT accommodated D iff T entailsD, D is true,and Twasde-
signedto entailD, (thatis, the conditionthat the theoryentailD
actedas a constrainton the selectionof T as theacceptedtheory).5
T correctly
predictedD, iff T entailsD, D is true,and T was not de-
signedto entailD.
We can now statethe questionwhichconcernsus: In whatcircum-
stances,if any,shouldtheinformationthatT correctlypredicted,rather
thanaccommodatedD, giveus greaterconfidencein T?It is usefulto
distinguisha weakanda strongversionof predictionism:
heuristicnovelty.
5 Of course, a theory rarelyentails any specific experimentaldata on its own, but only in con-
2. Theanti-predictionistchallenge
Thereare a numberof reasonswhy the strongpredictionistthesis
seemshighlydubious,someinvolvinganalysesof individualcases,and
othersinvolvinggeneralarguments.First,manyof the caseswhich
mightbe takento illustratethe epistemicadvantageof predictionare
eitherhistoricallydubious,7or can be diagnosedas involvingsome
otherfactor,suchas simplicity,whichmakesthe epistemicdifference.
Whenwe arecarefulto constructa casewhicheliminatestheseother
differences,we often findthatthe epistemicadvantageof prediction
seemsto disappear.SupposeI watcha coin beingtossedfifty times,
landing heads every time. After five heads, I tentativelyform the
hypothesisthat the coin is double-headed,and correctlypredictthe
remainingoutcomes.You,on the otherhand,learnof the outcomes
afterthe sequenceis completed,andsimilarlyconcludethatthe coinis
double-headed. SurelyI haveno morereasonto believethishypothesis
thanyou,justbecauseI madean earlyprediction.Thiscaseeliminates
variousindependentfeatureswhichcanmakean epistemicdifference
(webothfullygraspthetheoryandits evidence).It is temptingto gen-
eralizeto the viewthatwheneverwe fullygraspthe theoryandall the
evidence,whetherthe datumwaspredictedor accommodatedmakes
no epistemicdifference.Second,a rathercompellingcasecanbe made
againstStrongPredictionism,as we will see presently.Andthird,the
standardandinitiallycompellingpredictionistargumentturnsout to
be flawedon closerinspection;I willexaminethisargumentin sections
3-4.
The followingargument,basedon Collins (1994),bringsout just
how implausiblethe strongpredictionistthesiscan seem.Supposewe
knowthatD is trueandthatT entailsD. Wearealsothoroughlyfamil-
6. A newargumentfor StrongPredictionism
Our attemptsto savethe standardNo-CoincidenceArgumenthave
Weshould
failed.I wishto presenta newargumentwhichis persuasive.
stilltake
(ES) Thetheoristselecteda datum-entailing
theory
Aswe havenoted,thismightbe explainedby the
as our explanandum.
designhypothesis
(DS) The theoristdesignedher theoryto entailthe datum,that is,
knowingthe experimental outcome,sheselectedhertheoryon
the conditionthatit entailthisdatum.
WhatotherhypothesismightexplainES?Wemighttryto explainit by
supposingthatsheselectedhertheoryon theconditionthatit wastrue,
forthiswouldguarantee thatsheselecteda datum-entailing theory.But
unfortunately,
theoriesdo not comewithclearlabelsattacheddeclaring
theirtruth-value,so theycannotbe straightforwardly selectedby this
criterion.Theoryselectionmay,however,be moreor lesswellaimedat
thetruth.Thisnotionrequiresfurtheranalysis,but it mightroughlybe
characterizedas the degreeto whichthe causalchainof mechanisms
whichled to herselectionof the theorywerereliablyconnectedto the
facts.Obviouslythisis a matterof degree,but forthe sakeof simplicity
we canfocuson thetruthor falsityof thehypothesis
(RA) Thetheorist'sselectionof hertheorywasreliablyaimedat the
truth
by whichI meanroughlythatthe mechanismswhichled to herselec-
tion of a theorygaveher a good chanceof arrivingat the truth.This
hypothesisat least raisesthe theorist'schancesof holdinga datum-
entailingtheory,byraisingherchancesof holdinga truetheory.
7. Thearcheranalogy
TherelationsamongES,DSandRAcanbe illustrated by a simpleanal-
ogy.Wemayrepresentourtheoriesanddataon a mapof logicalspace,
whereregionson the maprepresentsetsof possibleworldsin whicha
propositionis true,theareaof a regionbeingproportional to theprob-
abilityof the proposition.ThedottedregionD representsour datum,
the outcomeof a certainexperiment.Thesmallcircularregionsrepre-
sentvarioustheories.Onlythosewhicharesub-regionsof D, entailthe
actualexperimental outcome,andregionTR,whichcontainsthe actual
world,is theonlytruetheory.
Now supposethat this map is drawnon the side of a barnand an
archershootsan arrowat it. Wedo not knowif the archeris aimingat
TR or even how good his aim is. Without seeing wherethe arrow
landed,we learnthatit landedwithina circlein regionD. Thequestion
whichconcernsus is whetherthe arrowlandedin TR.Theinformation
that the arrowlandedin D, shouldincreaseour confidencethat the
arrowlandedin TR,sinceTRis containedwithinD, andD is a smaller
regionthanthe wall.(Thisis analogousto the waythatlearningthata
theoryentailsthe datum,canprovideevidencethatthe theoryis true,
" Maher (1988)seems to be onto a similaridea, but develops it along differentlines. For criti-
cisms of Maher'sargumentsee Howson and Franklin(1991).
quiteapartfromwhether
thedatumwaspredicted
oraccommodated).
O ° D C
fi \ _° o O / ' o O
O ,'o OrA O ,
O ' -o _°. un OO
archer,knowing the location of D, restricts his aim so that he is guaranteed to hit somewhere
within D. He may or may not attemptto hit some more specific region such as TR. The denial of
DSs is consistentwith his reliablyaiming at TR. Of course in one sense, if the archeris aiming at
TR, he must also be aiming at D, since TR lies within D. But there is anothersense the one rele-
vant to our discussion accordingto which the archermay aim at TR without aiming at D, that
is, without intendingto hit D, if he does not even know where region D is, or at any rate, if his
knowledgeof the location of D has no influenceon how he shoots. Similarlywith DS and RA, the
theorist may design her theory to entail the datum with or without also reliablyaiming for a true
theory.And she may reliablyaim at the truth, without designingher theory to entail the datum, if
she does not know the datum, or her knowledge of the datum plays no role in her selection of a
theory.
13 It is important here that TR is not just any sub-region of D, but a salient target, one which
stands out by being painted black.We do not know if the archeris aiming at any small region or
how good his aim is, but if he is aiming, he is somewhat more likelyto aim at TR, since it stands
out from the surroundingregions.
7.1 TwoobjectionslS
Objection
1:Yourargumentmightseemto involvea kindof illicitdou-
blecountingof evidence,by usingtheevidenceto providedirectinduc-
14 Hereit is important
thatthetruthis a salienttarget.Wedo notknowif theprocessof theory
selectionwasdirectedtowardanyspecifickindof theory,butinsofaras it was,it is mostlikelyto
havebeendirectedtowardthetruth(it wouldbe oddforthetheoristto tryto constructa specific
kindof falsetheory).
15 Theseweresuggested
byananonymous
refereeforMind.
tivesupportforthetheory,andforthetheorist'saimatthetruth,which
in turnsupportsthetheory.
Reply: Here it is importantto note that there are two independent
sourcesof evidencefortheoryT.ThedatumD mayprovidedirectevi-
dence for T.But it is not D, but ES,the theorist'sholdinga datum-
entailingtheorywhichprovidesevidencethatshewasreliablyaimedat
the truth,andhenceforthe truthof hertheoryT.It is onlythe signifi-
canceof thislatteritemof evidencewhichis affectedbywhetherD was
predictedor accommodated.
Objection2: Onefactorwhichdetermines a theorist'saimat thetruthis
how muchevidenceshe has to go on. A theoristwill havemoreevi-
denceto go on if she accommodatesa datumthanif she choosesher
theorywithoutthatdatum,butpredictsit. Hence,it mightseem,in the
a caseof accommodation, the theoristhasa betteraimat the truthand
hencewe havemorereasonto believehertheory.
Reply:Notice that whetherthe theoristpredictsor accommodatesa
datumgivesus no indicationat all as to her abilityat assessingevi-
dence,the reliabilityof herequipmentandmethods,or anythingalong
thoselines.Thatthetheoristaccommodated datumD entailsthatsheis
aimed at the truthjust in this she
respect: will comeup with a theory
whichentailsD, andsinceD is true,sheis somewhatmorelikelyto hit
upon a true theory than if her theory did not entail D. But in this
respect,sheis no betteroffthanif shehadpredictedD, forin eithercase
hertheoryentailsD.
8. Thelotterypredictionexample
Theadvantage witha casein which
of predictioncanbestbe illustrated
the datumD provideslittle or no evidencefor the theoryT,when T
merelyaccommodates D. Comparethe followingtwo cases:
Accommodation:Wereadin the paperthatJanewon the national
lottery.Fredproposesthe followingtheoryto explainthis fact:the
lotterywasriggedin Jane'sfavour.
Prediction:Beforethe lotteryis evendrawn,Fredproposesthetheo-
ry thatit is riggedin Jane'sfavour.WelaterdiscoverthatJanewon.
Inthesecondcasewe arefarmoreinclinedto believeFred'stheorythan
in the first.In the secondcasewe suspectthathe musthavebeenonto
something,thathe musthavehad somekindof reliableaccessto the
factsconcerningthe lotterysetup,to havebeenableto predictthe lot-
tery'soutcome.
Firstlet'slookbrieflyatwhythe datum
(D) Janewon
doesnot renderthe theory
(T) Thelotterywasriggedin Jane'sfavour
veryprobableif T merelyaccommodatesD. WhileT does entailD, it
does so onlyat the expenseof beinghighlyimprobable.TheoryT,we
mightsay,inheritsthe arbitrarinessof D, for even if the lotterywas
rigged,we haveno morereasonto supposethatit wouldbe riggedin
Jane'sfavour,than we haveto supposethat Janewould win just by
chance.Indeedthe factthatJanewon hardlycallsforan explanationin
thefirstplace;someonehadto win,andit couldjustaseasilybe Jane,as
anyone.In Bayesianterms,we cannotethatthereis a weakertheory1>,
whichstatessimplythatthelotterywasrigged,whichis not confirmed
at all by Jane'swinning,sinceJaneis no morelikelyto win giventhat
the lotterywasrigged.Butnow since T entails1>, T canbe no more
probablethan 18. Thatis, Jane'swinningrendersthe hypothesisthat
the lotterywasriggedin Jane'sfavourno moreprobablethanthatthe
lotterywasriggedat all.
Butthereis somethingelse whichwe mightwantto explain,apart
fromJane'swinning,namelyFred'sholdingof a theorywhichentails
herwinning.Or rather,Fred'sholdinga theorywhichentailstheactual
outcomeof thelottery,(hisholdinga theorythatentailsJane'swinningis
significantonlyif Janewasthe actualwinner).Thequestionthatstrikes
us is, out of all the possibletheoriesconcerningthe mechanicsof the
lottery,howdidFredmanageto getone intohisheadwhichhappensto
entailthe actuallotteryresult?Now of coursein the accommodation
case,the answeris straightforward. SinceFredknewthatJanewon,he
couldselecthis theoryundertheconstraintthatit mustentailthisout-
come. Apartfromthis constraint,his theoryconstructionneed not
havebeenaimedat thetruth,it mayhavebeenjusta wildspeculation.
In the predictioncase,Freddid not selecthis theoryunderthe con-
straintthatit entailthe data,so we needa differentexplanation.The
naturalhypothesisthatcomesto mindis thatFredwassomehowrelia-
blyhookedup to the facts.On thisassumption,he is farmorelikelyto
comeupwitha theorywhichentailsthe actualoutcome.Itwouldbe an
extraordinary fluke,if he justguesseda theorywhichentailedtheactual
outcome.So in the casewhereFred'stheorypredictsthe data,we have
reasonto supposehe wasreliablyhookedup to the facts,whichin turn
givesus reasonto supposethathe is right.
challenge
9. Meetingtheanti-predictionist
Wearenow in a positionto see howthe newaccountavoidsthe prob-
lems of the standardNo-CoincidenceArgumentandmeetsthe anti-
predictionistchallenge.Wehavetwo potentialexplanatory hypotheses
for the fact ESthat the theoristchose a datum-entailingtheory:the
designhypothesisDS, andthe reliableaimhypothesisRA.Thedesign
hypothesisrendersthe reliableaiminghypothesisunnecessary,with
respectto explainingthe theorist'sentailment-success.Forwhatwe
haveherearetwo causalhypothesesconcerningthe processby which
the theorywasselected,eachof whichpotentiallyexplainsthe resultof
the selection.Thisis a caseof causal pre-emption. Perhapsthetheorist's
processof theoryselectionhada goodchanceof producinga true,and
hencedatum-entailingtheory.Butin a caseof accommodation,this
causalexplanationis pre-emptedby the factthatnon-datum-entailing
theorieswerenot even open to selection.The factthat,knowingthe
datum,the theoristrestrictedhertheoryselectionto datum-entailing
theoriesguarantees thatshewouldselecta datum-entailing theory,and
no furtherhypothesisregarding heraimat truthis necessaryto explain
herdoingso.
Wecannow see the plausiblenon-mysterious waythatinformation
concerninga certainpsychologicalprocessin the theorist'shead,
namelydesigninghertheoryto entaila certaindatum,is epistemically
relevantto the truthof hertheory.Thisinformation,DS,is relevantin
thatit screens offtheconfirmation of thehypothesisthatthetheorywas
reliablyselected,by the factthatthe theoryentailsthe datum.In doing
so, it diminishesthe supportthatthe theorist'sentailment-successpro-
videsforhertheory.
A verysimpleBayesiananalysisbringsthis out, by comparingthe
relationbetweenESandRA,firston the assumptionof mDS,andthen
assumingDS:
P(ES|RAA nDS) > P(ES|DS)
A nDS) > P(RA|'DS)
andso, P(RA|ES (1)
thatis, relativeto nDS,ESconfirmsRA.However:
P(ESIRAA DS)=P(ES#DS)
andso, P(RA|ESA DS)= P(RA|DS) (2)
So DSscreensoffthe
thatis, relativeto DS,ESandRAareindependent.
supportthatESprovidesto RA.Furthermore, addingthe assumption
that:
informationconcerninghertheoryselectionprocess,suchaswhethera
certaindatumwaspredictedor accommodated.
It seemsclearthatknowledgeof thetheorist'stotalevidencedoesnot
entirelyscreenoff the relevanceof this furtherinformation.Forwhile
informationconcerningthe evidencethatthe theoristhadto go on is
veryrelevantto how reliablyhertheorizingwasaimedat the truth,it
does not settlethe matter.It is usefulhereto considertwo important
factorslinkingevidenceandtheory.First,therearecertaina prioriepis-
temicconstraintson howevidenceshouldbe assessed in formingtheo-
ries.Wecanthinkof this in termsof a rangeof degreesof confidence
that an idealepistemicagentmighthavein a theory,givena bodyof
evidence(how wide this rangeis, that is, how tightthe a prioriepis-
temicconstraintsare,is an open question).Second,therearevarious
causalrelationswhicharenot knowablea priori:theseincludetherelia-
bilityof ourperceptual faculties,thetrustworthiness of varioussources,
the accuracyof our measuringinstruments,and so on. The crucial
point hereis thatthe degreeof reliableaim of theorizingdependson
bothfactors,neitherof whichis entirelytransparent to us.16
Concerningthe first,sincewe arenot idealepistemicagents,we are
falliblein our assessmentof evidence.Forinstance,constructionof a
theorymightinvolvecomplexmathematical derivationswherethereis
plentyof opportunityfor errors,evenif we doublecheckourwork.In
some cases,the inferencefromevidenceto theoryinvolvesintuitive
judgements, theprinciplesof whicharenot easyto spellout.A particu-
larlystrikingcaseof thisis our abilityto 'read'a person'sfacialexpres-
sions, even thoughwe cannoteasilysay how we interpretthe visual
cueson whichourjudgementsarebased.Wehaveno troubleforming
suchjudgements,butthe degreeto whichthe visualevidencesupports
ourjudgementmaybe in doubt,and no amountof doublechecking
ourinferencecanhelpus.
Supposenow the theoristcomesup withtheoryT via complexderi-
vationsandintuitiveinferencesfroma multifarious collectionof back-
ground evidence E. T entails D, a possible outcome of a crucial
experiment.Upon laterdiscoveringthatD is true,we havereasonto
increaseour confidencein herassessmentof the evidenceE. Forif her
assessmentof the evidencewas well attunedto the actualdegreeof
epistemicsupportbetweenthe evidenceandthevariouscandidatethe-
ories,shehada betterchanceof hittingupona true,andhencedatum-
entailingtheory.Supposeon the otherhandthatknowingthe datain
advance,she narroweddownthe pool of candidatetheories,by elimi-
16 Lipton (1991)makesa similarclaim that the theory'ssimplicity need not be transparent.
nating those which do not entail the datum, and then appliedthe
mathematicalcalculationsand intuitivejudgementsto selectamong
thisnarrower pool,in thelightof backgroundevidenceE.In thiscaseit
is no surprisethatthe resultingtheoryentailsthe datum,andhencewe
haveno furthergroundsfor confidencein her assessmentof the evi-
dence.
Similarpointsapplyto thevariouscausalrelationswhichtheprocess
of theoryselectioninvolves.Forinstancein choosingtheorieswe often
relyon the use of measuringinstruments,opinionof experts,andour
own perceptualfaculties,the reliabilityof whichis open to question.
Thatthe theoristcameup witha datum-entailing theorymayindicate
thatsuchcausalconnectionswereindeedreliable,butonlyif hertheory
selectionprocessdid not involvenarrowingdownthe candidatetheo-
ries in the light of the knowndatum.Foras before,if the measuring
devices,opinionsof colleaguesandso on, did not leadto theoryT on
theirown,but only whenvariousnon-datum-entailing theorieswere
alreadyeliminated,it is no surprisethatthe selectedtheoryentailsthe
datum,andhencethetheorist'sentailment-success wouldbe no indica-
tion of thereliability of thesedevices.
So StrongPredictionismis vindicated.Evenif we knowall the evi-
denceon whichthe theoristbasedher theory,the factthat a certain
datumwaspredictedratherthanaccommodated, mayprovidefurther
evidencefor the theory.Whenit comesto our actualtheoreticalprac-
tices however,the strongthesisis not particularlyrelevant,sincewe
typicallydo notknowalltheevidenceon whicha theorywasbased.The
multifariousconsiderationswhichleadto the acceptanceof a theory
areoftentoo subtleand complexto be easilycommunicated.Indeed,
evenin our owncase,we do not typicallykeepcarefultrackof all the
reasonswe everhadfor adoptinga certaintheory.Wemightcometo
questionjusthowgood all our reasonswere,in whichcasethe predic-
tivesuccessof ourtheorycansuggestthatour reasonswerenot badat
all.
The degreeto which predictionhas an epistemicadvantageover
accommodationcan be seen now to dependon how well we under-
standtheprocessbywhichthe theoryin questionwasselected,andthe
bearingof the new predictedor accommodateddata.In some casesa
datummightprovideoverwhelmingsupportfor the theoryby itself,
even if we know nothing of the independentevidence, and hence
whetherthedatumwaspredictedor accommodated canmakelittledif-
ference.Thecoin'slandingheadsfiftytimesforinstance,wasalltheevi-
dencewe everneededto concludethat it is double-headed;whether
11. Super-strongPredictionism
WhatI havecalled'StrongPredictionism' is strongenoughto be dis-
putedby many,whileweakerthanthatwhichsomepredictionists may
accept,andsomeanti-predictionists see as theirrealtarget.Thestruc-
tureof my argumentforStrongPredictionism wassimilarto the argu-
mentsforWeakPredictionism. In eachcasethe epistemicsignificance
of predictionto thetruthof a theorywasin a certainsenseindirect. The
factthat the theorypredictedratherthan accommodatedthe datum
increasesourestimateof someintermediate factor,whichin turncon-
firmsthe theory.In the caseof WeakPredictionism,the intermediate
factorswerethetheory'ssimplicityandbackground evidentialsupport.
Forthe strongthesisit wasthetheorist'saimat thetruth.
Wecan imaginean evenstrongerthesis,callit Super-strong Predi-
tionism,accordingto whichthe factthata theorypredictedratherthan
accommodateda datumis evidenceall by itselffor the theory,quite
apartfromour estimatesof these intermediaries.It is not clearthat
anyonehasexplicitlyendorsedthisthesis,butit maybe whatsomehave
in mind,andmaybewhatmanywhoconsiderthemselvesopponentsof
predictionismarereallyopposedto. So it is worthnotingthatmy dis-
cussionin no waysupportsSuper-strongPredictionism.Indeedmy
accountof when and how predictionmattersmight underminethe
temptationto acceptthe super-strong thesis.Forthis temptationmay
be due to a simplisticgeneralization fromcaseswherepredicteddata
seem to carrymore weight, the antidoteto which is a more subtle
accountwhenandhow predictionmatters.In this waymy defenceof
predictionismmaybe welcomedby thosein the broadlyanti-predic-
tionistcamp.
12.TheNo-MiraclesArgumentfor scientificrealism
I will concludewith someverygeneralsuggestionsas to howmy
accountof theepistemic advantage of prediction mightbeappliedto a
defenceof scientific realism.l7 Thosewhobelievethatourcurrentsci-
entifictheoriesaretrue,or atleastapproximately true,mustfacethe
factthatno matterwhatdatawehavesupporting a theory,thereare
numerous alternative theorieswhichentailthesamedata(orindeed
coincideinalltheirobservable entailments). Thechallenge forthereal-
istis to explainwhysometheoriesaremorelikelyto betruethanthe
manyothertheorieswhichentailthe sameobservabledata.One
response to thechallenge is to appealto further principlesof confirma-
tion;forinstance, it mightbeargued thatsometheories provide a better
explanationof the datathanothers,wherethe criterionfora good
explanation goeswellbeyondmereentailmentof data.Thisis notthe
strategy I wishto focuson.ThereiswhatI believeis supposed to bean
independent argument, whichclaims,in Putnam's (1975)words,that
scientificrealism 'istheonlyphilosophy whichdoesn'tmakethesuccess
of sciencea miracle.' (p.73)Itwouldbea miracle,it is sometimes sug-
gested,thatanaeroplane shouldflymesafelyhometo Sydneyif the
aerodynamical theoriesonwhichitsdesignis basedwerenottrue.Aer-
oplaneflightis a trickybusiness. Ofcourseit ispossibleforawildlyfalse
theoryto entailthecorrect results butwhywouldanysanepersonget
in a planeif hedidnotthinkthetheoriesonwhichitsdesignwasbased
werecloseenoughto correct?
TheNo-Miracles Argument involvesaninference to thebestexpla-
nationof thesuccessof science,wherethissuccessjustconsistsin the
factthatourtheoriesentailcertaindataconcerningsay,aeroplane
behaviour. Theclaimisthatthetruthofthesetheories canexplaintheir
success, whichseemsfairenough,sincenecessarily, truthsentailtruths.
Butlet'slookmorecloselyatwhatthissuccessconsistsin.LetD specify
theaeroplane behaviour thatourtheoryTentails.SinceTnecessarily
entailsD, T is successful if andonlyif D is true.Soto explainthefact
thatTis successful, isjustto explain D.Butnowthereis something odd
abouttheideathattruthcouldplaysomeexplanatory role,overand
abovethe factswhichobtain,if ourtheoriesaretrue.If anything
explains thefactthataeroplanes stayup,it is (roughly) thatthepressure
on theunderside of a movingairfoilis greater thanthepressure on its
OriginalArgument Argument
No-miracles
Explanandum: D T is successful
Explanans: T T is true
18 point.
Theexampleis adaptedfromLevin(1984),whousesit to makea somewhatdifferent
19 Thisdoesnot depend on a controversial accountof truth,butmerelythatprop-
deflationary
ositionsof theformit is truethatP andP,areknownto be necessarily andhenceone
equivalent,
cannotexplainanymorethantheotherdoes.
20 whois explicitaboutit.
Leplin(1997)is onerecentproponentof theargument
23 Something like this has been noted by proponents of the No-Miracles Argument, such as
Musgrave(1988), Lipton (1991),Leplin (1997)and Psillos (1999). This might partly explain why
they insist on the standardversion of the argument,the versionwhich I have arguedfails to meet
the antirealistchallenge.
selectionexplainswhythereareonlysuccessfulones.Perhapsif Fredhad
been in the habitof proposinglotteryriggingtheorieseverydayfor
years,we shouldfindit less remarkable thathe managedto come up
with a successfulone eventually(similarly,perhapsJane'ssuccessin
pickingreliableusedcarshasnothingto do withautomotiveexpertise,
but ratherherpolicyof buyingdozensof cars,andditchingthosethat
stop working).The plenitudeof Fred'sattemptedtheorieswouldnot
explainwhy,in the caseof Jane'slotteryhe managedto hit upona suc-
cessfultheory.Heretheremaybe no explanationat all,thatis, he was
justlucky.Buttheywouldmakehis successlesssurprising, andhence
lessin needof an explanationin termsof his reliableconnectionto the
truth.
So we at leastget a glimpseat howwe mightrenderit non-miracu-
lousthatscientistsmanageto comeup withtheorieswhicharemarve-
louslysuccessfulin applicationssuchas aeronautics.Veryroughly,it
mightbe a matterof trialand error.Plentyof dud theoriesaresug-
gestedandrejectedwhentheyfail,renderingit unremarkable thatsci-
entistscurrentlyholdsuccessfulones.Therearetwocomponentsto the
explanation, as therearetwofactsto explain:
(S1) Scientistscurrentlyholdsomesuccessfultheories
(S2) Scientistscurrentlyholdonlysuccessfultheories
VanFraassen's pointaboutthe harshenvironmentinto whichtheories
arebornexplainsS2, whilethe multipleattemptsat successfultheories
explainS1.
Thismereglimpsehowever,seemsinsufficient to groundanyconclu-
sions. We need to look at such factorsas just how remarkablethe
successesof scienceareandtheratioof successfulto unsuccessful theo-
ries,amongthosetheoriesthathavebeenproposed,not merelythose
currentlyheld.At leaston the faceof it, it doesnot seemthatscientists
havebeenchurningout so manyunsuccessful theoriesthata fewstar-
tlingsuccessesareto be expected by chance. What I hopeto havedone
in the precedingdiscussionis clarifyhowthedebateshouldgo.Asmat-
ters stand,it seemsthat we need to explain,or at least rendernon-
miraculous,the factthatscientistsmanageto come up with so many
remarkably successfultheories.An explanationmightlie in the mecha-
nisms of theory selectionbeing directedtowardthe truth, and the
acceptanceof suchan explanationshouldincreaseour confidencein
thetruthof scientifictheoriesin general.Butthisexplanation threatens
to be undercutby the alternativeexplanation that the mechanisms of
theoryselectionaredirectedtowardsuccessfultheories,withoutbeing
morespecificallydirectedtowardthe truth.VanFraassenGs attemptto
showhow this mightworkis suggestive,yet not compellingwithouta
lot moredetail.Howmatterswillstandwhenthe detailsareconsidered
is an openquestion.24
References
Achinstein,Peter1994:'Explanationv. Prediction:Which CarriesMore
Weight?',in Hull, Forbes,and Burian(eds) 1994,pp. 156-64.
Boyd, Richard1984:'The CurrentStatusof ScientificRealism'in Leplin
(ed.) 1984,pp. 41-82.
Brush,Stephen1994:'Dynamicsof TheoryChange:The Role of Predic-
tions' in Hull, Forbes,and Burian(eds) 1994,pp. 133-45.
Collins, Robin 1994: 'Againstthe Epistemic Value of Prediction over
Accommodation'.Nous,28,pp. 210-24.
Duhem, Pierre1956:TlXe AimandStructure ofPhysicalTheory.Prince-
ton: PrincetonUniversityPress.
Earman,John)(ed.) 1983:Minnesota Studiesin thePhilosophyofScience)
Vol. 10, Minneapolis:Universityof MinnesotaPress.
Fine, A and Leplin,J. (eds) 1988:Proceedings of thePhilosophyof Science
AssociationMol. 1, East Lansing Mich.: Philosophy of Science
Association.
Giere,RonaldN. 1983:(TestingTheoreticalHypotheses'in Earman1983
pp. 269-98.
Gooding, D., Pinch T. and SchafferS. 1989:TheUsesof Experiment.
Cambridge:CambridgeUniversityPress.
Hull, D., Forbes)M., and BurianR. (eds) 1994:Proceedings ofthePhilos-
ophyof Science AssociationVol. 2) East Lansing,Mich.: Philosophy
of ScienceAssociation.
JudyThomson, participants
24 Thanksto Adam Elga,Ned Hall, PatrickHawley,Bob Stalnaker,
in my 2001 EpistemologySeminar,and an anonymous refereefor Mind, for helpful comments on
drafts.
VanFraassen,Bas C. 1980:TheScientificImage.Oxford:OxfordUniver-
sity Press.
Zahar,Elie 1973:'Why did Einstein'sProgrammeSupercedeLorentz's?
(I)'BritishJournalforthePhilosophy of Science,24, pp. 95-123.