The Similarity Approach to Counterfactuals: Some Problems Author(s): G. Lee Bowie Source: Nos, Vol. 13, No.

4, Special Issue on Counterfactuals and Laws (Nov., 1979), pp. 477498 Published by: Wiley Stable URL: http://www.jstor.org/stable/2215340 . Accessed: 01/04/2013 02:08
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.

Wiley is collaborating with JSTOR to digitize, preserve and extend access to Nos.

http://www.jstor.org

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

The Similarity Approach to Counterfactuals: SomeProblems

G. LEE BOWIE
MT. HOLYOKE COLLEGE

Although modal logicians have traditionally been interested in counterfactuals, only recently has the application of possible world semanticsto the problem of counterfactuals begun to yield formalsemanticswhich appear to capture the truth conditionsof ordinary The advance Englishcounterfactuals.1 has been broughtabout by definingsimilarity on structures sets of possible worlds.This use of similarity among possible worldsin the analysisof counterfactuals has been the subject of considerable theoreticaldevelopment. Through the work of Lewis,Stalnaker,van Fraassen,Thomason, et.al. , thebasic idea has been modified and generalized in the directionof higherorder quantification (over modalities),various indexings for selection functions,impossible worlds as limitsof sequences of possible worlds, counternecessaries,countercomparatives,counterfactualprobabilities, and so on, all acmodel-theoretic notions and axiomacompanied by elegant tics. Our natural tendencyto preserve the productsof great labor is, in thiscase, reinforcedby the power and elegance of the theoriesthemselves.But however elegant and powerful the similarity approach is in its full theoreticaldevelopment, thereare difficulties at itsfoundations-difficulties whichcast doubt on the use of the notion of comparative similarity of possible worlds in the analysisof counterfactuals. This paper, then,is concerned to pointout some of these whichsurroundthe basic assumptionsof the simidifficulties, larity approach in itsapplicationto theproblemofcounterfactuals. I shall not be arguing thatthe similarity approach does not capture the logic of counterfactualinference. In fact it
t 1979
NODS 13 (1979) byIndiana University 477

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

478

NOtS

conditional, A is the negation of A, and an A-world is a possible world in which A is true. Now the simplestsort of would be this: for possible worldsanalysisof counterfactuals each worldi there is a class of worlds C i such thata counterfactualALI->B is truein i iffA-RB is truein everyworldin C . sinceon itwe conditionaltheory, a strict Lewis callsthistheory can understandAL-*B as L(A-*B), where C i is thenjust the set of worldsaccessiblefromi withrespectto the modalityMl. In the case where Ci is just the set of worlds at which some formulaSi (whose choice may depend on i) is true, we may ALJ->Bis trueat restatethetheoryas follows:a counterfactual i iffA-RB is trueat everyworld (accessiblefromi) at whichS is true [iff(Si & A) -RB is necessary]. It is clear thatthere is no choice of Si (or of C i) which conditional.For choose A as any captures the counterfactual whatB is,it withS i. Then no matter incompatible proposition will (vacuously) be true in everyworld in which (Si & A) is true. Hence on this theory ALI->B will be vacuously true whenever A is incompatiblewith Si. More generally,if B is B is trueat everyworldat which(S i & A) is true,thenafortiori

case then B would be the case'. In addition,

approach, as developed by Lewis, appears thatthe similarity inference.Indeed, the doescapture the logicof counterfactual factthatit does is a verystrongargumenton its behalf. Nor of worlds is shall I be arguing that the notion of similarity vague and imprecise (although, to be sure, it is vague and imprecise).What I shall be arguingis thatno plausible undergeneratesa correctanalysis standingofthe notionof similarity of the truth conditions of ordinary counterfactuals.More as valid can verify whiletheoriesbased on similarity formally, all and only those ordinarycounterfactualswhich seem, in fact,to be valid, nonetheless no plausible understandingof will give us as true (in the intended the notion of similarity model) just those ordinary counterfactualswhich seem, in to be true.Thus whilethesimilarity approach generatesa fact, it not a successful analysis. successfullogic, does generate it will be useful to theories, Before proceeding similarity This will to reviewthe classicalapproach to counterfactuals. the successes of the both highlight similarity approach, and of the the connect failures similarity approach us to enable withthe failuresof the classical approach. I follow Lewis in 'if A were the using 'ALI->B' to representthe counterfactual
'->'

is the material

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

479

true at everyworld at which (Si & A & C) is true. Hence on a strict conditionaltheoryAL->B entails (A & C)L->B. But thisinferenceis not sound for counterfactuals. For example, if Mort were to go to the storetomorrowhe would buy bananas. It does not followthatif Mortwere to go to the storetomorrow and get shoton his wayto the checkoutstand, he would buy bananas. The argumentsabove show thatthereis no set of worlds such thatforeveryA and B, AL->B is trueiffA-RB is trueat everyworld in that set. The source of the difficulty is clear. When we ask whetherAL->B, we are asking,it would seem, about the truth value of B in relativelyfamiliarworlds in which A is true. Interpretingthe counterfactualas a strict conditional requires that we decide on a standard of familiaritybefore we know what A is. S encodes the standard of But havingchosen such a standard,we may then familiarity. wishto ask whatwould happen ifthatstandardwere not met. But we cannot do this.The standard has already been fixed. Thus the first objectionabove. Similarly, havingchosen some familiar setofworldsto look at in whichMortgoes to thestore, havingchosen to ignoreas too bizarreworldsin whichhe gets shotwhileshopping,we mayturnaround and ask whatwould happen ifMortwere to go to thestoreand getshot.Now we go wrong if we ignore worlds in which Mort gets shot,whereas beforewe would go wrongifwe were to consider such worlds. Thus the second objection above. Givena possibleworldsapproach to counterfactuals, it is clear thatin evaluatingthe truthof ALJ->Bwe wantto look at the truthof A->B in some set of worlds.From the discussion above it is clear that the choice of the set of worlds must be variable,dependent on A, and perhaps on B. takes thisinto The classical approach to counterfactuals account. According to the classical account, a counterfactual ALI->B is true iffthere is some true S such that B is true in everyworld in which (S & A) is true. In contrastto the strict thechoice of S maydepend on A (or on B). conditionaltheory, S in general willbe a huge conjunctionwhichwill,intuitively, tellus enough about whatour worldis likeotherthan the fact (if it is a fact)thatA, so thatwhen we conjoin it withA, B is entailed. For example, if we are wondering whether Mort would buy bananas if he were to go to the store tomorrow,S to determine would containnaturallaws,enough information

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

480

NOtS

that the store will have bananas, to determine Mort's proclivitiestoward banana-buying,as well as informationabout earthquakes, homocidal maniacs and other factors which would, iftheyobtained,bar Mort'sbanana-buyingproclivities fromtheirnatural fruition. We have required that S be true, but clearly more is required. For wheneverA-RB is true,it will,togetherwithA, entailB. (When A--B is false,so is AEI-*B.) So ifwe can always conditionalwillcollapse choose A-*B as S, the counterfactual into the material conditional. Goodman terms 'co-tenable' those truthsS which are appropriate for conjoining with A A as a counterfactual If we are when entertaining hypothesis. wonderingwhatwould happen if Mortwere to go to the store tomorrow,then truthsof the sort noted earlier would be co-tenable with the proposition that Mort goes to the store tomorrow. Now, whyis A--B (when true) not alwaysco-tenablewith A? At first glance,thismightseem to be anotherof thosecases in whichwe would liketo be able to draw a distinction between "real" facts (like natural laws, the fact that there will be no etc.) and "phoney" homocidal maniacs in the storetomorrow, facts(like the factthatA--B). Having drawn such a distinction,we would thensaythata truthis co-tenablewithAjust in case it representeda real factand itwas (in some appropriate sense) compatiblewithA. Now quite apart fromthe factthatattemptsto so distinguish the real facts from the phoney ones have not been notable successes, the attempt would be misguided in this context.Martha's salaryis $100,000, and her propertytaxes are $5,000. Out of an equal desire to donate 10% of her salary annually and to donate twiceher propertytax annually,she gives$10,000 everyyear to the Flat Earth Society.Now, what would happen ifMarthawere president(of the U.S.). The fact that Martha always gives 10% of her salaryand the factthat the president's salary is $200,000 entail (together with the thatMarthais president)thatshe willgive$20,000. hypothesis But the factthatshe alwaysgives twiceher propertytax and the factthatthe presidentpays no propertytax on the White House entail (together with the hypothesisthat Martha is Yet all of thefactsalluded president) thatshe willgivenothing. "real" as factswhich typito seem (to me) to be as intuitively There are, it would seem, too cally support counterfactuals.

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

481

many "real" facts around, so we must make some further choice. Unfortunately, among the factsrelied on above, the ones thatwe wantare just those factsthatwouldremain factsif Martha were president.So we have come fullcircle,since in order to decide what would be the case if A were the case we mustfindthose truths whichare co-tenablewithA. And to do that,we must find which facts are "real" facts. But among those,theonlyones in whichwe are interested are thosewhich would remain factsifA were the case, so we mustagain know what would be the case if A were the case. If the truths co-tenablewithA arejust those whichintuiAEI-*Bthenthereis good tively "support" the counterfactual reason to believe thatin general the truthsco-tenablewithA willbe just those truths whichwould remain truthsif A were the case. For the claim that S supports the counterfactual ALi->B can alwaysbe defeated by showingit to be falsethatif A werethecase thenS would be thecase (as reflection on a few if whichwould remain truths examples shows). So onlytruths A were the case support ALI-RB. And while some proper would suffice to support all true counsubsetsof those truths terfactuals withantecedentA (forexample, thesetof all truths A->B whichwould remain truthsif A were the case), examination of cases makes it appear likelythatany truthwhich would remain true if A were the case can be enlistedin supALJ->B(withthe proviso,perhaps, portof the counterfactual thatitdoes notitself entailB). A solid argumentis not possible here, due to the factthatthe notion of "support" appealed to notion. The considerationsraised remainsa rough, intuitive above, however, make it appear plausible that it will be no easier to characterizecarefullythe truthswhich support the ALI->B than itwillbe to characterizethe propcounterfactual ositionswhichare counterfactually implied byA and to select fromthemthose whichare true (and, perhaps, whichdo not implyB). Thus if the truthsco-tenablewithA are just those whichsupport counterfactuals withantecedentA, it appears the classical thatin order to give an account of co-tenability, theory will have to give an account of the counterfactual conditionalitself.3 David Lewis provides an elegant account of counterfactualswhich,to be understoodas an analysisof the counterfactual conditional',requires (in addition to the semantics of possible worlds) only the intuitive notion of comparative

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

482

NOOS

over-allsimilarity taken as a 3-place relationamong possible worlds.On Lewis' account,given a possible world i, thereis a set of spheres associated withi. The set is nested, and closed Each sphere is a set of possible under union and intersection. if there is a sphere associated withi such worlds. Intuitively, thatj is inside it and k is outside it,then is more similarto i than k is. We need only consider the usual case where i is a member of every,non-emptysphere associated with itself. Thus we willspeak of the spheres associated withi as spheres iffthereis some worldin itat i. A sphere is A-permitting about whichA is true. Then ALI-*B is true at i iffeitherthereis no A-permitting sphere about i, or there is some A-permitting sphere about i such that A-HB is true at everyworld in that for we getthe following sphere. Phrased in termsof similarity, the non-vacuouscase: ALI-*B is true at i ifthereis a world at whichA is truesuch thatat no world at least as similarto i is A true and B false. in termsof spheres and the charThe characterization relation (j is acterizationin terms of the 3-place similarity more similarto i than k is) are equivalent. Given a systemof spheres,we can say thatj is more similarto i than k is ifffor some sphere, S, about i,j is inside S and k is not. Similarly, relation, we can define a given the comparative similarity systemof spheres as follows.For each world i, S is a sphere about i iffeveryworldin S is more similarto i than any world not in S.4 There is then a natural isomorphism between which preservestruth sphere-systems and similarity systems value of all formulasat all worlds. I have described Lewis' in termsof spheres,since the truthdefinisemanticsinitially I tion is simplerand since it is Lewis' preferredformulation. in the sequel, sinceitis a willrelyon the similarity formulation which I more convenientvehicleforexposing the difficulties raise. Naturally,in virtue of the close correspondence beand similarity tweensphere structures structures, any doubts about similarity generates corresponding doubts about spheres; so no purpose otherthan ease of expositionis served characterization. by the decision to focus on the similarity relation enters into Lewis' Finally although the accessibility account,itis not relevantto mydiscussion,and I have ignored it. (Ignoringitis equivalentto supposing itto hold betweenall possible worlds.)

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

483

If we assume for any world i and propositionA, that if there is any world at which A is true then there is a closest world at which A is true (that is a world at which A is true whichis more similarto i than any other world at whichA is true),we get as a special case Stalnaker'stheory:That AL->B is trueat i iffB is trueat the closest(mostsimilar)world to i at which A is true. Lewis' theoryseems superior to Stalnaker's. It is more general, since Stalnaker'stheorycan be obtained as a special on the system case of Lewis' by placingsuitablerestrictions of spheres (or on the similarity relation). Stalnaker'smodels do not seem to be generatedbyour intuitive notionof similarity. For example suppose, contrary to fact,thatMortwere standing more than 120 feet away from the Washington Monument.It would seem thatgivenany worldin whichMortwere standing 120 + 8 feetaway, there is a world more similarto ours in whichMortis standingonly 120 + 8/2feetaway. Thus there seems to be no closestworld in which Mort is standing more than 120 feet away fromthe Washington Monument. Furthermore, it would seem as though two different worlds could be equally similarto a givenworld.Consider a worldin whichthree pool balls are lined up on a pool table, the 8-ball betweenthe 7-balland the 9-ball.Is a worldin whichthe 8-ball is 3 inchescloserto the 7-ballmore or less like the givenworld than one in whichthe 8-ballis 3 inches closer to the 9-ball?It would seem as thoughthetwoworldsare equally similarto the givenworld. But thiscannotbe so in Stalnaker'smodels. Thus notionof similarity. Lewis' modelsbetter captureour intuitive The more telling considerationis this:Stalnaker'stheory validates the scheme (ALI->B) v (ALIJ>B). But this does not seem correct.In the example above, which of the following is true: (i) if the 8-ballwere 3 inches closer to counterfactuals eitherball, it would be 3 inches closer to the 7-ball,(ii) if the 8-ball were 3 inches closer to eitherball it would be 3 inches closer to the 9-ball?Surelyifthe 8-ballwere 3 inches closer to eitherball then it would eitherbe 3 inches closer to the 7-ball or it would be 3 inches closer to the 9-ball. But thisdoes not seemsto have it forceus to accept either(i) or (ii),and intuition thatneither(i) nor (ii) is true. But either(i) or (ii) mustbe true on Stalnaker'stheory. Both Stalnaker and Lewis give formal semantics and axiomatics for counterfactuallogic. Structuresare defined

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

484

NOOS

over arbitrary sets,understoodas setsof possible worlds,and arbitrary orderingson those sets (or their equivalents,as in Lewis' spheres), understood as similarity orderings.Both into generatean analysisof counterfactuals in tendtheirsystems the sense that one gets the truth conditions for ordinary counterfactualsby looking at the truthconditions of their correspondingformulasin the standard interpretation, that whichtakesitssetof possibleworldsto be is,theinterpretation the real set of possibleworlds,and whichtakesthe relationon relationon the set of possible worlds to be the real similarity possible worlds.What I shall tryto show in the sequel is that thereis no plausibleunderstandingof the real similarity relation whichgives us the correcttruthconditionsfor ordinary in eitherLewis' or Stalnaker'stheory.For ease counterfactuals of expositionit willbe convenientto speak, 'a la Stalnaker,of the A-worldwhichis closestto ours, or the A-worldmostlike ours, or the A-world most similar to ours (these are interchangeable). All of my remarkscan be reformulatedso as to avoid assuming that there is a closest A-world,though generally at the cost of more complex sentence structure. Now, whenwe compare twoworldsforover-allsimilarity, in what respectsdo we compare them?It followsfromLewis' account that we must compare them in all respects. For if therewere respectsin whicha worldmightdiffer fromi which did not count at all in judging its overall similarity to i, then therecould be a worldwhichdifferedfromi but onlyin those respectswhchcounted fornothinginjudging overall similarity.This would entail the existenceof a world which was not identicalwithi, but whichwasjust as similarto i as i is to itself. But Lewis adopts the "centeringassumption,"whichrequires that no world be as similarto i as i is to itself.Furthermore, Lewis evidentlyintends that worlds be compared in their temporaltotality. That is, the basis of comparisonis the entire world,in itsspatialand temporaltotality, and thattotality is to be compared withthe spatial and temporal totality of other worlds in every possible respect of comparison. Although some respectsof comparison may count for littlein judging overall similarity, no respects of comparison can count for nothing. But thismethod of comparison does not seem to give us correct truthconditions. Consider: I am standing with my finger on the ultimate doomsday button. The button and

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

485

tested,and are associated mechanismshave been thoroughly reliable. If I were to push it, the entire universe would explode-unless a loose piece of metal drops down as I push the buttonin such a wayas tojam the button.But itwon't.It is loose, but not loose enough. Yes, I see that if I pushed the button,the universewould explode. But whathappens in the worldmostlikeours in whichI push the button?There can be onlyone answer-the piece of metalfallsanywayand jams the button.For surelya worldin whichthe buttongetsjammed is far more like ours than a world in which it doesn't and the had to entire universe explodes. Even if the piece of nH~eTal come fromall the wayacross the room, or ifthe buttonhad to worldwould be more likeours than the resulting disintegrate, a world which explodes. In factwe can say withsome confidence that in the world most like ours where I push the button, the nearest relativelyunattached small object flies under the button and jams it, for surely a temporarylocal breakdownin the lawsof mechanicswould preservesimilarity farmore thanworldcataclysm. So, theexample continues,ifI push the button,the world would not explode. But we know that if I push the button, the world will explode. We can inspect the mechanism,do simulated trial runs, etc. So the theory is wrong. It is wrong because it makes change too If there is some simple way to keep the world on difficult. track,it will happen. If thingscan go right,theywill. If the objection stands, the theories we are considering approach would be clearly unacceptable, and the similarity would have to go. Now it mightbe thoughtthatthe objection can be met by placing more weight on laws of nature in This approach fails for several readeterminingsimilarity. sons. First,no matterhow much weightwe place on laws,we muststillplace some weighton particularmattersof fact.To do otherwise would be to do injusticeto more mundane counterfactualsand to run afoul of the centeringprinciple. But of weightswe decide upon, whateverreasonable distribution specbymakingthe resultsof preservingthe laws sufficiently tacular, and by making the breakdown in laws sufficiently unexciting (as in the example), we will find cases where a more than small,local violationof thelaws preservessimilarity strict adherence to the laws. withfalse ana counterfactual Secondly,in entertaining tecedent,the only worlds in whichthe antecedentis true but

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

486

NOOS

where the laws of this world are preserved mightbe worlds which are unacceptablydifferent fromours. As Lewis5says,
as we Suppose thatthe laws prevailingat a worldi are deterministic, used to think the laws of our own world were. Suppose a certain roulettewheel in thisdeterministic worldi stops on black at a timet, and consider the counterfactualantecedent that it stopped on red. What sortof antecedentworlds are closet to i? On the one hand, we have antecedentworldswhere the deterministic laws ofi hold without exception, but where the wheel is determined to stop on red by from those of i. Since the laws are deterparticularfactsdifferent ministic, the particularfactsmustbe different at all timesbeforet,no matterhow far back. ... On the other hand, we have antecedent-worldsthat are exactlylike i untilt or shortly before; where the laws of i hold almost withoutexception; but where a small,localized, inconspicuous mirathewheel to stopon red in violationof the cle attorjust beforepermits laws. Laws are very important,but great masses of particular fact count for somethingtoo; and a localized violation is not the most serious sort of differenceof law.

Thirdly, in testingscientifichypotheses we must take some of the laws ratherlightly. Suppose forexample thatwe are testingpart of the special theoryof relativity for the first time,bycheckingon theapparent positionof Mercuryduring a solar eclipse. Classical mechanics has it that Mercurywill has itthat appear to be whereitis, at pointC. Special relativity Mercurywill appear to be at point S, distinctfromC. Interestingly, Mercuryappears to be at S, and we inferthatclassical mechanicswas wrong. But let us entertainthe counterfactual hypothesis thatMercury had appeared to be at pointC. There are two hypotheses(at least): HI-that although the photons were disturbedby the mass of the sun, random displacement of the photons compensated for this disturbance, making was right Mercuryappear at pointC. On H I, special relativity even thoughMercuryappeared to be at C, because therewere compensatingrandom influences.H2 has it thatthe mass of the sun did not disturbthe passage of the photons,and classical mechanicswas right.Now the point is that in the actual case, we clearly reject HI as an explanation of our and assert hypothesizedfailureto confirmspecial relativity, instead that if Mercuryhad appeared at point C, then remechanics would have been incorrect.Our acceptlativistic ance of H2 in this counterfactualsituationrequires that we mechanicsis incorcount as closera worldin whichrelativistic

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

487

rect,a world in which the laws of nature are violated,than a world which is fullyaccounted for withoutviolationof any laws of nature, but which is so improbable as to be disregarded. We conclude with Lewis that "the preeminence of laws of nature among cotenable factual premises is only a matterof degree."6 thereis an easier way to meet the objection. Fortunately, It can be metbymakingclear thattheworldwe are comparing with ours is not being compared in virtue of its temporal totality.We must require only that its historyup to (and perhaps including)thetimeat whichA is true(forcounterfactual AEI-*B) is most like the historyof this world. In the example, we are to imagine standingin the room, fingeron so faris as much as the button;the stage is set,and everything possiblelikethingsare here. At thispoint-I havejust pushed about how close the worldsare; the button-we stopworrying we just sitback to wait and see what happens. In thiscase we needn't wait long-the world explodes. It no longer matters, from though, that the world has just become verydifferent ours. We just stopped comparing. We made the worlds as close as possible up to and including the time of the buttonpush, and then let naturetake itscourse. Since thatworldwas likeours,themetalwas loose but not too loose, the mechanism worked,etc. And because all thatwas the case, the worldblew up, just as thisworld would have if I had pushed the button. Once the basis of comparison is made clear, the counthe solutionraises new terexampleis blocked. Unfortunately, difficulties. They Paradigm counterfactualsare forward-directed. in if the future thingswere conjecture what would happen would in the past or present,or what happen in the different or earlier in the time at same futureif thingswere different be can also past-directed. the future.But counterfactuals in the structureof the There is an alarming instability Because we findit universewhichmakesitliable to cataclysm. we have inventedan earlywarningdevice (EWD) comforting, whichsends out a signal in advance of the universaldestruction.The device,forproperoperation,is situatedat thecenter of mass of the universe.There are twosocieties-one, Society Near, so close to thecenterof the universethatifit receivesa signal,therewillbe 3 yearsbefore the cataclysmbegins. The other,SocietyFar, is so farfromthecenteroftheuniversethat

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

488

NOOS

of the universe by the timeit receivesthe warningsignal, 2/3 remaining the and destroyed, been 13 willgo willalreadyhave 'Sohypotheses: counterfactual two the soon. Now, entertain Far 'Society EWD' and from the a signal receives cietyNear receivesa signal fromthe EWD'. In the firstcase, since the device has been tested and is highlyreliable, all works well. of the universesup Since we are onlycomparingthe histories to thetimewhen SocietyNear receivesthe signal,a universein whichthe signalis receivedand is genuine willhave a history more likethatof our universethana universein whicha signal of this highlyreliable deis received througha malfunction vice. We may say,then,thatif SocietyNear were to receivea signal,the cataclysmwould begin 3 years thence. On the otherhand, suppose SocietyFar receivesa signal (even the same signal).Now, any universein whicha signal is receivedbySocietyFar, and is genuine,is a universewhichhas already been 2/3destroyed.Since we are comparing the hisof our universe,such a toriesof such universesto the history from different would be grossly destroyed, universe,being 2/3 our intactone. We are in the same positionas beforewiththe doomsday button.For any universein which SocietyFar receives a signal which is genuine (hence a universe which is is more already2/3destroyed)thereis a universewhose history of our intactuniverse,in whichthe signal to thehistory similar resulted from a malfunction.Hence the counterfactual'if has begun' is false, SocietyFar receivesa signal,the cataclysm theory,even if we suppose Society Far to on the similarity receivethesamesignalwhichwe suppose SocietyNear to have received, fromthe same device. as our basisof comparison,we are So in choosinghistories stillcomparing too much. Perhaps what we are to do is to compare some worldwithours onlywithrespectto itstempoforpurposes of ral sliceat thetimeat whichA is true,ignoring, comparion, not only its future from that time (as we did above), but also itspast before thattime.Thus, forthe counAEJ->B,we pick the A-worldwhichis such thatits terfactual temporalsliceis mostlikeours,and see whatwill(did) happen thatunlesswe assume a strongformof in thatworld.Note first thissuggestionwillrequire thatwe abandon the determinism, assumptionrequires centeringassumption.For the centering But on the current thatno worldis as similarto i as i is to itself. suggestionwe are comparingonly temporalslices of worlds;

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

489

and unless some strongformof determinism holds, two distinctworldswillhave identicalslicesat a given time,differing only at some earlier or later time. Since we are comparing onlytemporalslices,twosuch worldswillbe as similarto each other as each is to itself.But this is not permittedby the centering principle. However, since we could adopt what Lewis calls "the weak centering principle,"8which requires no only thatno world be more similarto i than i is to itself, intractableproblem is created for the similarity approach. When we compare an A-sliceof another world withour When we were comparworld,whatare we comparingitwith? ing temporally completeworlds,thiswas no problem.But now thatwe are looking only at an A-slice of a possible world,we must presumablycompare it for similarity witha sliceof our world; and the question is, which slice? Evidentlywe do not compare theA-slicewitha currentsliceof thisworld.Suppose thatitis December, and myfootis in a castwhichwillcome off nextweek. If we are entertaining the counterfactual 'if I were I would need a new pair of skiboots', to go skiingin February, we would go wrongifwe compared slicesin whichI go skiing in Februarywithcurrentslices,since in the slice mostlike the currentsliceof our world,I have a caston myfoot.Hence we would be asking what will happen in a world in which I go skiingin Februarywitha caston myfoot,and thisis clearlynot going to give us the rightanswer. Do we compare A-sliceswitha sliceof our worldwhichis, as it were, cotemporaneous withthe A-slice?9That is, in the example above, do we compare A-sliceswith a slice of this worldtakenin February?Suppose, as a matterof fact,thatin but to cut offthe lefthalfof JanuaryI decide not to go skiing, myfootinstead. Thus in FebruaryI have onlyhalfa foot.We would go wrongifwe were to compare A-sliceswitha slice of thisworldtakenin February,forin thatslice I have onlyhalfa foot(and would need new skiboots). But in the real counterifI had gone skiingin February,I would not factualsituation, have decided to mutilatemy footin January,and would not have needed new boots. It is not as easy as it should have been to see where we went wrong. For in deciding that we had to compare slices, and not worlds,we tacitly assumed that we would compare A-sliceswithslices of our world. But in the skiingexamples, the intentis clearlythatamong all of the A-worlds,we chose

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

490

NOOS

the one whose currentsliceis mostlikethecurrentsliceof our world, and then see what happens there. Thus in the first skiingexample above, look among the worlds in which I go skiingin Februaryforthatworldwhose December is mostlike December in this world, and then see whether I need new boots in thatworld. In thatworld I do not have a cast on my foot in February,and I do not need new boots. The same procedure applied to the second example gives us a world in do not mutilatemyfoot,and do whichI go skiingin February, not need new boots. Now that the basis for comparison has been made clear, thingsseem to work right. we have stillnot got itright.Returnto the Unfortunately, example of the earlywarningdevice. On the presentsugges'ifSocietyFar receivesa tion,in evaluatingthe counterfactual signal,the cataclysmhas begun', we are to look among all of thoseworldsin whichSocietyFar receivesa signal forthe one whose currentslice resemblesour world's currentslice. But the current slice most of any world in which the signal is genuine willbe 2/3destroyed,and willbe less like our current slice than one in whichthe signal resultedfrommalfunction. stillturnsout false on the similarity Thus the counterfactual theory,when it should turn out to be true. Two more examples will help illuminate the problem. Consider the false counterfactual'if the Norman Conquest much the wayitis now'. had failed,the worldwould be pretty The presentsuggestionhas us look among all of the worldsin whichthe Norman Conquest failsfortheone(s) whose current slice is mostlike our currentslice,thatis, forthe one whichis of compensating mostlike ours now. But giventhe possibility whichcounteractthe failureof the Norman Condifferences quest, there is at least one world in which the Conquest fails which today is prettymuch like ours. Hence the world in which the Conquest failswhich today is most like our world today, will be a world which is prettymuch like our world 'if today.Thus on the presentsuggestion,the counterfactual the Norman Conquest had failed,the world would be pretty answer muchthewayitis now'comes out true.To gettheright in thisexample we have to compare A-slices (slices in which the Norman Conquest has just failed) withco-temporaneous compare historiesup to slices of our world, or alternatively, the time of the failure of the Norman Conquest with coof our world.While these procedures temporaneoushistories answerin thiscase, theyfailedin earliercases. giveus theright

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

491

Now consider one last example. Mort is sittingat the of him. Mort dinnertable witha dish of magic yogurtin front one is 6 feettall.The yogurt worksas follows:ifhe eats exactly teaspoon or less, he will remain 6 feet tall. As he eats more than one teaspoon, he will grow as a linear functionof the amount, until,if he eats it all, he will grow to 7 feettall. As a matterof fact,Mort abominates yogurt,magic or not, and is not going to eat any of it,although he may dip his spoon in it and play with it. Now, consider the counterfactual'if Mort were to eat some of thisyogurt,he would remain 6 feettall'. My intuitionis that neither this counterfactual,nor the associated counterfactual,'if Mort were to eat some of this yogurt,he would grow', is true.10The antecedent can be satisfiedin too manyways,and there is no way of projecting which way it wouldbe satisfiedif it were satisfied. But the similarity approach says thatif Mort were to eat he would remain 6 feettall. For a worldin some of theyogurt, bitofyogurt is more he is eatingonlya tiny whichceteris paribus similarto our world, in which he eats no yogurt,than any world in which he eats more. Hence in the closest world(s), Mort eats only a tinybit of yogurt and remains the same height. It might be thought that there will be features of the whichwould giveus a betteranswer.If, forexample, situation Mort were an inveterateglutton, given to extremes, then perhapswe could saythathe would eithereat none or eat itall. If, on the other hand, he were a man of delicate sensibilities, given to caution in new ventures,thenwe could say thatif he were to trysome, he would tryonly a tinybit, and would remain the same height. This would amount to saying that of the situation(includingMort'scharacter) thereare features in determining than the whichare more important similarity, mere quantityof yogurtwhich Mort eats. While such considerations might settle the issue more in thiscase, the general problemremains.For howfavorably ifin a counterfactual the ever we construesimilarity, ALJ-RB, antecedentis false but can be satisfiedin a number of ways, and if there is no way of projectingwhich way it would be if it were satisfied, then the similarity approach resatisfied, quires thatit be satisfiedin whateverway is closest to its not being satisfiedat all. So for any antecedent A, we have the followingcounterfactualtrue: if A were the case, then A

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

492

NOOS

would barelybe the case. Thus, ifMortweighed between 300 and 400 pounds, he would weigh between 300 and 310, between300 and 301, etc. Here thereis nothingin Mort'scharacter or constitutionwhich suggests any particular range within whichhisweight would fallifhe were to weighbetween 300 and 400 (except,of course, the range 300-400). We have no reasonable way of saying,forexample, whetherhe would weigh over 350 or under 350. But the similarity approach requires that he weigh between 300 and 301 if he were to weigh between 300 and 400. This is because worlds in which he weighs between 300 and 301 are, ceteris paribus,more like our world (in which Mort weighs, let us say, 180) than are worlds in which he weighs over 301.11 Let us reviewour resultsthusfar.In evaluatingthe truth AL->B at a possible world i (we value of the counterfactual suppose here thatA is falsein i), thesimilarity approach has us look at worldsj, in whichA is true,and whichare similarto i. We have asked what things one is comparing when one is evaluatingoverallsimilarity, and have investigated the following alternatives: (1) (2) (3) (4) (5) compare i andj in theirtemporal totalities, compare the history ofj up to the time that A becomes true withthe correspondinghistory of i, compare an A-slice ofj witha currentslice of i, compare an A-sliceofj witha co-temporaneousslice of i, and compare a currentsliceof i witha contemporaneous (current)slice ofj.

Althougheach of thesealternatives workedforsome counterfactuals,none of them works for all of the counterfactuals which we have considered. Furthermore,there are some counterfactuals forwhichnone of the alternatives works.The early warning device example and the example of Mort's weightare such cases. As is clear fromthe listabove, we have not exhausted the possible alternatives. For example, we mighttry comparingthe twoclosestslicesof i andj, whenever they happen to take place, or (if this is different)the two closest contemporaneousslices. We have selected for discus-

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

493

sion only the most natural candidates, but none of the unnaturalcandidateswhichI have investigated does anybetter. Could it be thatwhilethereis no singlenotionof similaritywhichgives us a uniformanalysisof counterfactuals, that there is a range of notions of similarity such that (a) each counterfactual is correctly analysedbyone notionof similarity in thisrange, and (b) the contextof assertionindicateswhich in thisrange is theintendednotionof similarity? In discussing Quine's examples,namely If Caesar had been in command [in Korea]he would have used the atom bomb versus If Caesar had been in command he would have used catapults, Lewis says,
In one context,we may attach great importanceto similarities and in respect of Caesar's characterand in respect of regudifferences larities concerning the knowledge of weapons common to commandersin Korea. In anothercontextwe mayattachless importance to these similarities and differences, and more importance to in respect of Caesar's own knowledge of similarities and differences weapons. The firstcontext resolves the vagueness of comparative in such a waythatsome worldswitha modernizedCaesar in similarity command come out closer to our world than any withan unmodernized Caesar. It therebymakes the firstcounterfactualtrue. The second contextresolvesthe vagueness in theopposite direction, making the second counterfactual true. Other contextsmightresolvethe vaguenessin otherways.A thirdcontext, forinstance,mightproduce a tie between the closest worlds with modernized Caesars and the closestworldswithunmodernizedCaesars. That contextmakes both counterfactuals false.12

Could itbe thatour problemis similar, and thatthe contextof assertion can be called on to indicatetheappropriatenotionof similarity? Note first thatifA is false,and if(A & B) and (A & B) are both possible,bychoosingappropriate similarity relationswe can alwaysmake ALI-*B trueat i, or make ACl-> B trueat i, or dividetheset make neithertrue.To make ALI-*B true,merely of worlds into two subsets,W1 and W2. Let W1 be the set of worldsat whichA->B is true,and letW2be the setof worldsat whichA->B is false. Order the worlds so thateveryworld in

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

494

NOOS

W1 is more similarto i than any world in W2. Order worlds within W1 and W2in anywayyou like. Then among A-worlds, any in which B is true will be more similarto i than any in whichB is false,so ALI->B willbe true at i. To make AC-> B true at i, merelylet W1 be the set of worlds at whichA-> B is to trueand W2the setof worldsat whichA-> B is false.Finally, make neitherAL->B nor ACl-> B true,letall worldsbe equally similarto i. Thus, whileit is true thatforeach counterfactual there is some measure of similarity which gives the correct truthvalue, this is trivially true; there is also a measure of which gives us the incorrecttruthvalue. similarity Secondly,thechoice among the five(or more) alternative construalsof similarity above does notseem to be a choice as to how to resolvevagueness. It does not seem to be a question,as in Quine's example, of how to weigh various respectsof similarity.In that example, in the contextof a discussion about militarypsychologyand the relative tendencies of military leaders towardoffenseor defense,we mightsaythatifCaesar were in Korea, he would have used the atom bomb. However, in discussingthewaysthatgeneralsadapted theirattackto the vagaries of terrain,contrastingCaesar's methods in Alesia withhismethodsin Britain, we mightsaythatifCaesar werein Korea, he would have used catapults.Note thatin the former case we would put emphasis on the word 'Caesar', whereas in the lattercase we would place emphasis on the word 'Korea'. The problem we face here, however, is not a problem of bases of context,for we may need to choose two different comparison in the same context.Change the doomsday button example slightly, supposing the buttonto be an ancient, but stilloperational,artifact leftby a lostcultureof aliens. In showingthe buttonto a group ofjournalists,a general might say, "If I pushed this button, the universe would be destroyed."As we have seen, to get the correcttruthvalue for thiscounterfactual we see whathappens in the worldin which the general pushes the buttonwhose history up to the timeof the buttonpush is mostlike our cotemporaneous history. But suppose the general now adds, "and if Attillathe Hun had the universewould have been destroyed." pushed thisbutton, Now we go wrongifwe compare histories up to the present;to truth value we mustcompare histories getthecorrect up to the time of Attilla.The choice of comparison basis has changed. But thecontextof utterancehas remained the same. Hence it

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

495

cannotbe contextwhichdeterminesthechoice of comparison Thirdly,in some of our examples there is no plausible choice of comparison basis which seems to give the correct truthvalue. In the example of Mort's weight,any plausible choice of comparisonbasis forceseitherALI-*B or AD}-+ B to be true,whereaswe wantbothto be false.In theearlywarning device example, to make the counterfactual'if Society Far were to receive a signal,the universe would be 2/3destroyed' truein the givenworldi, we mustfindworldsin whicha signal is received and compare them withi in respect to theirhisnot up to the timethe toriesup to the timethe signalis emitted, (keep the signal is received. But change the example slightly contextof utterancethe same if you like) so that instead of the signal 3 years before the cataclysmbegins, the emitting device emits a signal only after ?2 of the universe has been and in comparinghistories up to thetimethe signal destroyed, is emittedwe will make the counterfactualfalse. To make it onlyup to some earliertime, truewe have to compare histories some timebefore the cataclysm is sufficiently underway.For in the changed example, any world in which the signal is emitted and is genuine will be a world which is already '/2 different from destroyedand which is thereforesufficiently the given world i that worlds in which the signal is emitted willmore similar.Note finally thatthere throughmalfunction 'ifSociis no obvious relationship betweenthe counterfactual ety Far were to receive a signal, the universe would be 2/3 destroyed'(includingitscontextof utterance)and thetimewe In theunmodifiedexample we muststopcomparinghistories. stopped comparing historiesat the time of emission of the signal; in the modified example we stopped comparing at unsome earlier time before the cataclysmwas sufficiently derway.The choice of timesdoes not seem to be determined or by the contextof utterance(the itself bythecounterfactual and contextwere the same in both cases), but counterfactual ofthedevice. Thus thereis ratherbythedetailsoftheworking no way in which the counterfactual along withits contextof utterancedeterminesan appropriate basis for comparison. hypothesis There is no questionthat,fora counterfactual A, thereare respectsforwhichthe propositionsthatwould be the case ifA were the case are just the propositionswhichare the case in those A-worldsmost similarto ours in the given
basis.13

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

496

NOOS

respects.ALI->B will always be truejust in case B is true in those A-worldsmostsimilarto ours withrespectto the truths which true wouldremain ifA werethecase. This much was clear fromour discussion of the classical approach to counterfactuals. Unfortunatelyit is not generally the case that the A-worldsmostlikeours withrespectto thetruths whichwould remain truthsif A were the case coincide withthe A-worlds most like ours. In many cases the A-worldswhich wouldbe actual if A were true will be less similarto ours than other A-worldswhichcouldbe actual ifA were true.When thisis the in termsof over-allsimicase, the analysisof counterfactuals laritywill give us the wrong truth-value. Thus in the EWD example, theworldwhichwould have been actual had a signal been receivedwas less like the givenworld than otherworlds in which the signal was received. Given the counterfactual hypothesis values of A, thereare manywaysto adjust thetruth in order to accommodateA. In manycases, otherpropositions the adjustmentwhichwouldbe made ifA were the case would bring about a world much less like ours than other worlds whichcould be broughtabout byless drasticadjustments.It is perhaps a brute fact that some changes would bring about more drastic over-all changes than they need bring about. There is certainly no reason to expect thatthe changes which wouldresult from a change in the truth-valueof A are the minimum changes compatible with logical necessity. But when the changes which would be brought about by A are more drastic overall then they need be, the similarity approach will give us the wrong answers. Since thereis some notionof similarity, namelysimilarity withrespectto those truthswhich would remain truthsif A werethecase, whichgivesus the right answerforthe counterfactualAL-4B, it is no surprisethat the similarity approach seems to provide an acceptable logic for the counterfactual thereseems to be no non-circular conditional.Unfortunately, therequired notionof similarity whichdoes wayofexplicating not itself involvecounterfactual discourse. The problemis no different fromthatfaced by the classical approach, wherein there seemed to be no non-circularway of explicating the required notion of co-tenability. The similarity approach, understoodas an analysisof the truthconditions or ordinary counterfactuals,requires that when we specify for the formal theory an interpretation

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

THE SIMILARITY

APPROACH SOME PROBLEMS

497

whose set of possible worldsis the set of real possible worlds, and whose similarity relationis the real (but possiblyvague) similarity relationon possible worlds,the resultinginterprewhichare, in fact, tationmakestruejust thosecounterfactuals true.14 Although foreach counterfactual, there is trivially, a relationwhichworksforthatcounterfactual, there similarity appears to be no similarityrelation specifiable in nonwhichsatisfies the constraint above. We counterfactual terms, conclude thatthe similarity approach failsto give an analysis of the truthconditionsof ordinarycounterfactuals.
REFERENCES of Three Logics of Counterfactual [1] David Lewis,"Completenessand Decidability Conditionals,"Theoria37(1971): 74-85. [2] , Counterfactuals, (Harvard University Press, 1973). [3] Donald Nute, "Counterfactuals,"Notre Dame Journal of Formal Logic XVI, 4(October 1975): 476-482. of Words [sic],"Journal , "Counterfactuals and the Similarity [4] ofPhilosophy LXXII, 21(1975): 773-8. in Logical [5] RobertStalnaker,"A Theory of Conditionals,"in N. Rescher,Studies Theory, (Oxford: Blackwell, 1968). and R. Thomason, "A SemanticAnalysisof Conditional Logic," Theoria [6] 36(1970): 23-42. [7] Richmond Thomason, "A Fitch-StyleFormulation of Conditional Logic," LogiqueetAnalyse52(1970): 397-412. [8] Bas van Fraassen,"The Logic of ConditionalObligation,"Journal ofPhilosophical Logic I(1972): 417-38. NOTES 'Earlier versionsof thispaper have been read to colloquia at the University of Michigan and the University of Massachusetts.I am gratefulto the participants of those groups,and especiallytoJohnG. Bennett,fortheirmanyhelpfulsuggestions. 2See David Lewis, [1, 2]; RobertStalnaker,[5]; RobertStalnakerand Richmond Thomason, [6]; Richmond Thomason, [7]; Bas van Fraassen, [8]. All subsequent referencesto Lewis will be to Counterfactuals, and all subsequent referencesto Stalnaker will be to "A Theory of Conditionals." 3Lewis observes the equivalence of the Classical theory with his under the assumptionthatthe truths co-tenablewithA arejust thosewhich(accordingto Lewis' theory)would remain truthsif A were the case. Cf. Lewis, p. 69-70. 4Cf. Lewis, p. 48. 5Lewis,p. 75. 6Lewis,p. 75 70f course the situation willvarydepending on our individuationprinciplefor temporal slices. Consider, forexample, the factthatthe paper which you are currently readingwas written byme. Is thata factabout thecurrentsliceof theworld?If so, then no world could have an identicalcurrentslice unless in thatworld,at some timein the past,I wrotethispaper. If not,thena worldin whichsomebodyelse wrote

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions

498

NOS

this paper could have an identical currentslice. I shall ignore such problems,altheyare seriousproblems. thoughto theextentthatwe taketemporalslicesseriously, on Stalnaker'saccount. 8This is not a possibility identification 9I shallignore the considerableproblemsinvolvedin cross-world of times. and AEIR-Bare not contradictories. ?Rememberthaton Lewis' account ACEJ-RB "See Donald Nute, [3, 4]. This particularproblem can be avoided withinthe A-HB is trueat is trueati iff ACEJ-RB proposed byNute. In hisformulation, framework similarto i to warrantconsideration.In thisexample, itcan be all worldssufficiently claimed thatworldsin whichMort'sweightis towardthetop of the 300-400 range are similarto the given world to warrantconsideration. also sufficiently 12Lewis, p. 67. conditional 13Lewis uses an analogous argumentto show thatthe counterfactual depends on contextof utterance.Cf. conditionalwhose strength cannot be a strict Lewis, p. 13. "4Naturally we must also specifyvalues for the non-logical constants of the language.

Volume30 Number3 September1979 Contents,

AbsoluteSpace and Conventionalism DAVID ZARET Accountof TheoryChange Towardsan Objectivist ALAN CHALMERS and Choice FREDERICK SCHICK Uncertainty, Self-Knowledge, Articles

forthe Journal The British of Science Philosophy

Miller'sSo-Called Paradox of Information COLIN HOWSON and GRAHAM ODDIE Small Cauchy'sVariablesand Ordersof theInfinitely GORDON FISHER Analysis Century The Conceptof 'Variable'in Nineteenth JOHN P. CLEAVE
ISAAC LEVI

Discussions

Review Article of (review Supportand Surprise:L. J. Cohen's viewof InductiveProbability L. J. Cohen: TheProbableand theProvable)

Reviews HARRY M. BRACKEN on A. Simon: Models of Discovery; on Herbert Dice of theGods; D. M. MACKAY on NEIL COOPER on W. Ehrenberg: G. Pitcher: Berkeley; A. J. DALE on D. Holcroft(ed.): Revolution in Philosophy; A. Sloman: The Computer PapersinLogic andLanguage;JOHN BELL on JonBarwise(ed.): HandbookofMathematical ina World ofChance Logic; VERNON PRATT on W. H. Thorpe:Purpose
L. JONATHAN COHEN

Announcements

of Science by Society forthe Philosophy Publishedforthe British Hall, Aberdeen AB9 2XT, Scotland Press, Farmers Aberdeen University

and John Worrall John Watkins Editors:

Price ?2.50 ($8.00) Subscription Price ?7.00 ($22.00)

This content downloaded from 68.51.72.48 on Mon, 1 Apr 2013 02:08:08 AM All use subject to JSTOR Terms and Conditions