(I) So far I have criticized Kuhn from a point of view which is almost identical with that of Lakatos. (There are some slight differences,such as my reluctance to separate theories and research programmes,but they will be disregarded. When speaking of `theories' I alwaysmean theories and/or research programmes.) I now want to defend Kuhnagainst Lakatos. More specifically, I want to argue that science bothis, and should be, more irrational than Lakatos and Feyerabend1 (thePopperian3 author of the preceding sections of this paper and of`Problems of Empiricism') are prepared to admit.1
This transition from criticism to defence does not mean that I havechanged my mind. Nor can it be completely explained by my cynicismvis-a-vis the business of philosophy of science. It is ratherconnected with the nature of science itself, with its conmplexity,with the fact that it has different aspects, that it cannot be readilyseparated from the remainder of history, that it has always utilizedand continues to utilize every talent and every folly of man. Contraryarguments bring out the different features it contains, they challengeus to make a decision, they challenge us to either accept thismany-faced monster and be devoured by it, or else to change it inaccordance with our wishes. Let us now see what can be said againstthe Lakatos model of scientific growth.
(2) Naive falsificationism judges (i.e. accepts, or condemns)a theory as soon as it is introduced into the discussion. Lakatosgives a theory time, he permits it to develop, he permits it to showits hidden strength, and he judges it only `in the long run'. The`critical standards' he employs provide for an interval of hesitation.They are applied `with hindsight'. They are applied after theoccurrence of either 'progressive' or of 'degenerating' problemshifts.
Now it is easy to see that standards of this kind have practicalforce only if they are combined with a time limit (what looks likea degenerating problem shift may be the beginning of a much longerperiod of advance). But introduce the time limit and the argumentagainst naive falsificationism reappears with only a minormodification (if you are permitted to wait, why not wait a littlelonger?) Thus the standards which Lakatos wants to defend are eithervacuous - one does not know when to apply them or they can becriticized on grounds very similar to those which led to them in thefirst place.
In these circumstances one can do one of the following two things.One can stop appealing to permanent standards which remain in forcethroughout history and govern every single period of scientificdevelopment and every transition from one period to another. Or onecan retain such standards as a verbal ornament, as a memorial tohappier times when it was still thought possible to run a complex andoften catastrophic business like science by following a few simple and`rational' rules. It seems that Lakatos wants to choose the secondalternative.
(3) Choosing the second alternative: means abandoning permanentstandards in fact though retaining them in words. In fact, Lakatos'sposition now seems to be identical with the position of Popper assummarized in a (because self-destructive) marvellous addendum of thefourth edition of the Open Society. According to Popper we do not`need any. . . definite frame of reference for our criticism', we mayrevise even the most fundamental rules and drop the most fundamentaldemands if the need for a different measure of excellence shouldarise. Is such a position irrational? Does it imply that science isirrational? Yes and no. Yes-because there no longer exists a singleset of rules that will guide us through all the twists and turns ofthe history of thought (science), either as participants, or ashistorians who want to reconstruct its course. One can of course forcehistory into such a pattern, but the results will always be poorer andmuch less interesting than were the actual events. No-because eachparticular episode is rational in the sense that some of its featurescan be explained in terms of reasons which were either accepted at thesame time as its occurrence, or invented in the course of itsdevelopment. Yes- because even these logical reasons which change fromage to age are never sufficient to explain all the important featuresof a particular episode. We must add accidents, prejudices, materialconditions (such as the existence of a particular type of glass in onecountry and not in another), the vicissitudes of married life,oversight, superficiality, pride, and many other things in order toget a complete picture. No- because transported into the climate ofthe period under consideration and endowed with a lively and curiousintelligence we might have had still more to say, we might have triedto overcome accidents, and to `rationalize' even the most whimsicalsequence of events. But-and now we come to a decisive point- how isthe transition from certain standards to other standards to beachieved? More especially, what happens to our standards (as opposedto our theories) during a period of revolution? Are they changed inthe Popperian manner, by a critical discussion of alternatives, or arethere processes which defy a rational analysis? This is one of thequestions raised by Kuhn. Let us see what answer we can give to it!
(4) That standards are not always adopted on the basis of argumenthas been emphasized by Popper himself. Children, he says, `learn toimitate others. . . and so learn to look upon standards of behaviouras if they consisted of fixed, "given" rules. . . and such things assympathy and imagination may play an important role in thisdevelopment'. Similar considerations apply to those grownups who wantto continue learning and who are intent on expanding both theirknowledge and their sensibility. We certainly cannot assume that whatis possible in the case of children- to slide, on the smallestprovocation, into entirely new reaction patterns- should be beyond thereach of adults and inaccessible to one of the most outstanding adultactivities, science. Moreover, it is likely that catastrophic changes,frequent disappointment of expectations, crises in the development ofour knowledge will change and, perhaps, multiply reaction patterns(ineluding patterns of argumentation) just as an ecological crisismultiplies mutations. This may be an entirely natural process, likegrowing in size, and the only function of rational discourse mayconsist in increasing the mental tension that precedes and causes thebehavioural outburst. Now- is this not exactly the kind of change wemay expect at periods of scientific revolution? Does it not restrictthe effectiveness of arguments (except as a causative agent leading todevelopments very different from what is demanded by their content)?Does not the occurrence of such a change show that science which,after all, is part of the evolution of man is not entirely rationaland cannot be entirely rational? For if there are events, notnecessarily argunments which cause us to adopt new standards, will itthen not be up to the defenders of the status quo to provide, not justarguments, but also contrary causes? And if the old forms ofargumentation turn out to be too weak a contrary cause, must they thennot either give up, or resort to stronger and more `irrational' means?(It is very difficult, and perhaps entirely impossible, to combat theeffects of brainwashing by argument.) Even the most puritanicalrationalist will then be forced to leave argument and to use, say,propaganda not because some of his arguments have ceased to be valid,but because the psychological conditions which enable him toeffectively argue in this manner and thereby to influence others havedisappeared. And what is the use of an argument that leaves peopleunmoved?
(5) Considering questions such as these a Popperian will reply thatnevw standards may indeed be discovered, invented, accepted, impartedupon others in a very irrational manner, but that there always remainsthe possibility to criticize them after they have been adopted andthat it is this possibility which keeps our knowledge rational:`What, then, are we to trust?' asks Popper after a survey of possiblesources for standards. `What are we to accept? The answer is: whateverwe accept we should trust only tentatively, always remembering that weare in possession, at best, of partial truth (or rightness), and thatwe are bound to make at least some mistake or misjudgement somewhere-not only with respect to facts but also with respect to the adoptedstandards; secondly, vwre should trust (even tentatively) ourintuition only if it has been arrived at as the result of manyattempts to use our imagination ; of many mistakes, of many tests, ofmany doubts, and of searching criticism.'
Now this reference to tests and to criticism wrhich is supposed toguarantee the rationality of science and, perhaps, of our entire lifemay be either to well defined procedures without which a criticism ortest cannot be said to have taken place, or it may be purely abstractso that it is left to us to fill it now with this, and now with thatconcrete content. The first case has just been discussed. In thesecond case we hare but a verbal ornament, just as Lakatos's defenceof his own `objective standards' turned out to be a verbal ornament.The questions of section 4 remain unansvwrered in either case.
(6) In a way even this situation has been described by Popper whosays that `rationalism is necessarily far from comprehensive orself-contained'. But the question raised by Kuhn is not whether thereare limits to our reason ; the question is where these linmits aresituated. Are they outside the sciences so that science itself remainsentirely rational, or are irrational changes an essential part of eventhe most rational enterprise that has been invented by man? Does thehistorical phenomenon `science' contain ingredients which defya rational analysis? Can the abstract aim to come closer to the truthbe reached in an entirely rational nmanner, or is it perhapsinaccessible to those who decide to rely on argument only? These arethe problems to which we must now address ourselves.
(7) Considering these further problems Popper and Lakatos reject`mob psychology' and assert the rational character of all science.According to Popper it is possible to arrive at a judgement as towhich of two theories is closer to the truth, even if the theoriesshould be separated by a catastrophic upheaval such as a scientificrevolution. (A theory T is closer to the truth than another theory,T', if the class of the true consequences of T', the so-called truthcontent of T', exceeds the class of true consequences of T without anincrease in the falsity content.) According to Lakatos the apparentlyunreasonable features of science occur only in the material world andin the world of (psychological) thought ; they are absent from the`world of ideas, [from) Plato's and Popper's "third world" '. It is inthis third world that the growth of knowledge takes place and thata rational judgement of all aspects of science becomes possible. Itmust be pointed out, however, that the scientist is unfortunatelydealing with the world of matter and of (psychological) thought alsoand that the rules which create order in the third world may beentirely inappropriate for creating order in the brains of livinghuman beings (unless these brains and their structutral features areput into the third world, a point that does not become clear fromPopper's account).2 The numerous deviations from the straight path ofrationality which we observe in actual science may well be necessaryif we want to achieve progress with the brittle and unreliablematerial (instruments ; brains ; etc.) at our disposal.
However there is no need to pursue this objection further. There isno need to argue that real science may differ from its third worldimage in precisely those respects which make progress possible. Forthe Popperian model of an approach to the truth breaks down even if weconfine ourselves to ideas entirely. It breaks down because there areincommensurable theories.
(8) With the discussion of incommensurability, I come to a point ofKuhn's philosophy which I wholeheartedly accept. I am referring to hisassertion that succeeding paradigms can be evaluated only withdifficulty and that they may be altogether incomparable, at least asfar as the more familiar standards of comparison are concerned (theymay be readily comparable in other respects). I do not know who of uswas the first to use the term `incommensurable' in the sense that isat issue here. It occurs in Kuhn's `Structure of ScientificRevolutioms' and in my essay `Explanatiom, Reduction, and Empiricism'both of which appeared in l962. I still remember marvelling at thepre-established harmony that made us not only defend similar ideas butuse exactly the same words for expressing them. The coincidence is ofcourse far from mysterious. I had read earlier drafts of Kuhn's bookand had discussed their content with Kuhn. In these discussions weboth agreed that new theories, while often better and more detailedthan their predecessors were not always rich enough to deal with allthe problems to which the predecessor had given a definite and preciseanswer. The growth of knowledge or, more specifically, the replacementof one comprehensive theory by another involves losses as well asgains. Kuhn was fond of comparing the scientific world view of theseventeenth century with the Aristotelian philosophy, while I usedmore recent examples such as the theory of relativity and the quantumtheory. We also saw that it might be extremely difficult to comparesuccessive theories in the usual manner, that is, by an examination ofconseqence classes. The accepted scheme is as follows (Fig. I): T issuperseded by T'. T' explains why T fails where it does (in F); italso explains why T has been at least partly successful (in S); andmakes additional predictions, (A). Now if this scheme is to work thenthere must be statements which follow (with or without the help ofdefinitions and/or correlation hypotheses) both from T and from T'.But there are cases which invite a comparative judgement withoutsatisfying the conditions just stated. The relation between suchtheories is as shown in Fig. 2. A judgement involving a comparison ofcontent classes is now clearly impossible. For example, T' cannot besaid to be either closer to, or farther from, the truth, than T.
(9) As an example of two incommensurable theories let us brieflydiscuss classical celestial mechanics (CM) and the special theory ofrelativity (SR). To start with one should emphasize that the question'are CM and SR incommensurable?' is not a complete question. Theoriescan be interpreted in different ways. They will be commensurable insome interpretations, incomparable in others. Instrumentalism, forexample, makes commensurable all those theories which are related tothe same observation language and are interpreted on its basis.A realist, on the other hand, wants to give a unified account, both ofobservable and of unobservable matters, and he will use the mostabstract terms of vwhatever theory he is contemplating for thatpurpose. This is an entirely natural procedure. SR, so one would beinclined to say, does not just invite us to rethink unobserved length,mass, duration; it would seem to entail the relational character ofall lengths, masses, durations, whether observed or unobserved,observable or unobservable. Now extending the concepts of a new theoryT to all its consequences, observational reports included, may changethe interpretation of these consequences to such an extent that theydisappear from the consequence classes of earlier theories. Theseearlier theories will then all become incommensurable with T. Therelation between SR and CM is a case in point. The concept of lengthas used in SR and the concept of length as presupposed in CM aredifferent concepts. Both are relational concepts, and very complexrelational concepts at that (just consider determination of length interms of the wave length of a specified spectral line). Butrelativistic length (or relativistic shape) involves an element thatis absent from the classical concept and is in priciple exeluded fromit. It involves the relative welocity of the object concerned in somereference system. It is of course true that the relativistic shmemevery often gives us numbers which are practically identical with thenumbers we get from CM- but this does not make the concepts moresimilar. Even the case c- oo (or v- oo) which gives strictly identicalpredictions cannot be used as an argument for showing that theconcepts must coincide at least in this case: different magnitudesbased on different concepts may give identical values on theirrespective scales without ceasing to be different magnitudes (the sameremark applies to the attempt to identify classical mass with relativerest mass). This conceptual disparity, if taken seriously, infectseven the most `ordinary' situations: the relativistic concept ofa certain shape, such as a table, or of a certain temporal sequence,such as my saying `yes', will differ frolm the corresponding classicalconcept also. It is therefore vain to expect that sufficiently longderivations may eventually return us to the older ideas. Theconsequence classes of SR and CM are related as in Fig.2.A conmparison of content and a judgement of verisimmilitude cannot bemade.
(10) In what follows I shall discuss a few objections which have beenraised, not against this particular analysis of the relation betweenSR and CM, but against the very possibility, or desirability ofincommensurable theories (almost all objections againstincommensurability are of this general kind). They expressmethodological ideas which we must criticize if we want to increaseour freedom vis-a-vis the sciences.
One of the most popular objections proceeds from the version ofrealism that I just deseribed in (9).`A realist',we said,`wants togive a unified account, both of observable and of unobservablematters, and he will use the most abstract terms of whatever theory heis contemplating for that purpose'. He will use such terms in order toeither give meaning to observation sentences, or else to replace theircustomary interpretation (for exanmple, he will use the ideas of SR inorder to replace the customary CM-interpretation of everydaystatements about shapes, telmporal sequences, and so on). As againstthis it is pointed out that theoretical terms receive theirinterpretation by being connected either with a pre-existingobservation language, or with another theory that has already beenconnected with such an observation language and that they are devoidof content without such a connection. Thus Carnap asserts that 'thereis no independent interpretation for LT (the language in terms ofwhich a certain theory, or a certain world view, is formulated). Thesystem T (consisting of the axioms of the theory and the rules ofderivation) is itself an uninterpreted postulate system. (Its) termsobtain only an indirect and incomplete interpretation by the fact thatsome of them are connected by the (correspondence rules) C withobservational terms'. Now, if theoretical terms have no 'independentinterpretation' then they cannot be used for correcting theinterpretation of the observation statements which is the one and onlysource of their meaning. It follows that realism as described by us isan impossible doctrine.
The guiding idea behind this objection is that new and abstractlanguages cannot be introduced in a direct way but must be firstconnected with an already existing, and presumably stable,observational idiom.3
This guiding idea is refuted at once by pointing to the way in whichchildren learn to speak and in which anthropologists and linguistslearn the unknown language of a newly discovered tribe.
The first example is instructive for other reasons also, forincornmensurability plays an important role in the early months ofhuman development. As has been suggested by Piaget and his school, thechild's perception develops through various stages before it reachesits relatively stable adult form. In one stage objects seem to behavevery much like afterimages - and they are treated as such : the childfollows the object with his eyes until it disappears and he does notmake the slightest attempt to recover it even if this would requirea minimal physical (or intellectual) effort, an effort moreover, thatis already within the child's reach. There is not even a tendency tosearch-and this is quite appropriate, `conceptually' speaking. For itwould indeed be nonsensical to `look for' an afterimage. Its 'concept'does not provide for such an operation.
The arrival of the concept, and of the perceptual inmage, ofmaterial objects changes the situation quite dramatically. Thereoccurs a drastic reorientation of behavioural patterns and, so one mayconjecture, of thought. Afterimages or things somewhat like them stillexist, but they are now diffcult to find and must be discovered byspecial methods (the earlier visual world therefore literallydisappears). Such methods proceed from a new conceptual scheme(afterimages occur in humans, not in the outer physical wvorld, andare tied to them) and cannot lead back to the exact phenomana of theprevious stage (these phenomena should therefore be called bya different name, such as `pseudo-afterimages'). Neither afterimages,nor pseudo-afterimages are given a special position in the new world.For example, they are not treated as evidence on which the new notionof a material object is supposed to rest. Nor can they be used toexplain this notion: afterimages arise together with it and are absentfrom the mind of those who do not yet recognize material objects; andpseudoafterimages disappear as soon as such recognition takes place.It is to be admitted that every stage possesses a kind ofobservational `basis' to which one pays special attention and fromvwhich one receives a multitude of suggestions. However this basis (I)changes from stage to stage ; and (2) it is part of the conceptualapparatus of a given stage, not its one and only source ofinterpretation.
Considering developments such as these we may suspect that thefalmily of concepts centering upon `material object' and the family ofconcepts centering upon `pseudo-afterimages' are incommensurable inprecisely the sense that is at issue here. Is it reasonable to expectthat conceptual changes of this kind occur only in childhood? Shouldwe welcome the fact-if it is a fact- that an adult is stuck witha stable perceptualvworld and an accompanying stable conceptual systemwhich he can modify in many ways but whose general outlines haveforever become immobilized? Or is it not more realistic to assume thatfundamental changes, entailing incommensurability, are still possible,and that they should be encouraged lest we remain forever excludedfrom what might be a higher stage of knowledge and of consciousness?Besides, the question of the mobility of the adult stage is at anyrate an empirical question which must be attacked by research andcannot be settled by methodological fiat. An attempt to break throughthe boundaries of a given conceptual system and to escape the range of`Popperian spectacles' is an essential part of such research.
(II) Looking now at the second element of the refutation-anthropological field work- we see that what is anathema here (and forvery good reasons) is still a fundamental principle for thecontemporary representatives of the philosophy of the Vienna Circle.According to Carnap, Feigl, Nagel, and others the terms of a theoryreceive their interpretation, in an indirect fashion, by being relatedto a different conceptual system which is either an older theory, oran observation language. Older theories, or observation languages areadopted not because of their theoretical excellence (they cannotpossibly be: the older theories are usually refuted). They are adoptedbecause they are `used by a certain language community as a means ofcommunication'. According to this method, the phrase `having muchlarger relativistic mass than. . . ' is partially interpreted by firstconnecting it with some prerelativistic terms (classical terms;commonsense terms) which are `comnmonly understood' (presumably as theresult of previous teaching in connection with crude weighingmethods). This is even worse than the once quite popular demand toclarify doubtful points by translating them into Latin. For whileLatin was chosen because of its precision and clarity and also becauseit was conceptually richer than the slowly evolving vulgar idioms, thechoice of an observation language or of an older theory as a basis forinterpretation is due to the fact that they are `antecedentlyunderstood', it is due to their popularity. Besides, ifprerelativistic terms which are pretty far removed from realityespecially in view of the fact that they come from an incorrecttheory- can be taught ostensively, for example, wwith the help ofcrude weighing methods (and we must assume that they can be so taught,or the whole scheme collapses) then why should wve not introduce therelativistic terms directly, and without assistance from the terms ofsome other idiom? Finally, it is but plain commonsense that theteaching, or the learning, of new and unknown languages must not becontaminated by external material. Linguists remind us that a perfecttranslation is never possible, even if we use complex contextualdefinitions. This is one of the reasons for the importance of fieldwork where new languages are learned from scratch and for therejection, as inadequate, of any account that relies on (complete, orpartial) translation. Yet just what is anathema in linguistics is nowtaken for granted by logical empiricists, a mythical `observationlanguage' replacing the English of the translators. Let us commencefield work in this domain also and let us study the language of newtheories not in the definition factories of the double language model,but in the company of those metaphysicians, experimenters,theoreticians, playwrights, courtesans, who have constructed new worldviews! This finishes our discussion of the guiding principle of thefirst objection against realism and the possibility of incommensurabletheories.
(12) Next I shall deal with a mixed bag of asides which havenever been presented in a systematic fashion and which can be disposedof in a few words.
To start with, there is the suspicion that observations which areinterpreted in terms of a new theory can no longer be used to refutethat theory. The suspicion is allayed by pointing out that thepredictions of a theory depend on its postulates, the associatedgrammatical rules as well as on initial conditions, while the meaningof the primitive notions depends on the postulates (and the associatedgrammatical rules) only: it is possible to refute a theory by anexperience that is entirely interpreted in its terms.
Another point that is often made is that there exist crucialexperimets which refute one or two allegedly incommensurable theoriesand confirm the other, for example: the Michelson-Morley experiment,the variation of the mass of elementary particles, the transversalDoppler effect refute CM and confirm SR. The answer to this problem isnot difficult either: adopting the point of view of relativity we findthat the experiments which of course will now be described inrelativistic terms, using the relativistic notions of length,duration, speed, and so on, are relevant to the theory and we shallalso find that they support the theory. Adopting CM (with, or withoutan aether) vwe again find that the experiments (which are nowdescribed in the very different terms of classical physics, roughly inthe manner in which Lorentz described them) are relevant, but we alsofind that they undermine (the conjuction of classical electrodynamicsand of) CM. Why should it be necessary to possess terminology thatallows us to say that it is the same experiment which confirms onetheory and refutes the other? But did we not ourselwes use suchterminology? Well, for one thing it should be easy, though somewhatlaborious, to express what was just said without asserting identity.Secondly, the identification is of course not contrary to our thesis,for we are now not using the terms of either relativity, or ofclassical physics, as is done in a test, but are referring to them andtheir relation to the physical world. The language in which thisdiscourse is carried out can be elassical, or relativistic, orordinary. It is no good insisting that scientists act as if thesituation were much less complicated. If they act that way, then theyare either instrumentalists (see above, section 9) or mistaken : manyscientists are nowadays interested in formulae while we are discussinginterpretations. It is also possible that being well acquainted withboth CM and SR they change back and forth between these theories withsuch speed that they seem to remain wvithin a single domain ofdiscourse.
(13) It is also said that in admitting incommensurability intoscience we can no longer decide whether a new view explains what it issupposed to explain or whether it does not wander off into differentfields. For example, we would not know whether a newly inventedphysical theory is still dealing with problems of space and time orwhether its author has not by mistake made a biological assertion. Butthere is no need to possess such knowledge. For once the fact ofincommensurability has been admitted the question which underlies theobjection does not arise (conceptual progress often makes itimpossible to ask certain questions; thus we can no longer ask for theabsolute velocity of an object- at least as long as we take relativelyseriously). Yet is this not a serious loss for science? Not at all!Progress was made by the very same `wandering off into differentfields' whose undecidability now so greatly exercises the critic:Aristotle saw the world as a superorganism, that is, as a biologicalentity, while one essential elelment of the new science of Descartes,Galileo, and of their followers in medicine and in biology is itsexclusively mechanistic outlook. Are such developments to beforbidden? And if they are not, then what is left of the complaint?
A closely connected objection starts from the notion of explanation,or reduction, and emphasizes that this notion presupposes continuityof concepts (other notions could be used for starting exactly the samekind of argument). Now to take our above example, relativity issupposed to explain the valid parts of classical physics, hence itcannot be incommensurable with it! The reply is again obvious. Whyshould the relativist be concerned with the fate of classicalmechanics except as part of a historical exercise? There is only onetask we can legitimately demand of a theory and it is that it shouldgive us a correct account of the world. What have the principles ofexplanation got to do with this demand? Is it not reasonable to assumethat a point of view such as the point of view of classical mechanicsthat has been found wanting in various respects cannot have entirelyadequate concepts, and is it not equally reasonable to try replacingits concepts by those of a more successful cosmology? Besides, whyshould the notion of explanation be burdened by the demand forconceptual continuity? This notion has been found to be too narrowbefore (demand of derivability) and it had to be widened so as toinclude partial and statistical connections. Nothing prevents us fromwidening it still further to admit say, `explanation by equivocation'.
(14) Incommensurable theories, then, can be refuted by reference totheir own respective kinds of experience (in the absence ofcommensurable alternatives these refutations are quite weak, however).Their content cannot be compared. Nor is it possible to makea judgement of verisimilitude except within the confines ofa particular theory. None of the methods which Popper wants to use forrationalizing science can be applied and the one that can be applied,refutation, is greatly reduced in strength. What remains are aestheticjudgements, judgements of taste, and our own subjective wishes. Doesthis mean that we are ending up in subjectivism? Does this mean thatscience has become arbitrary, that it has become one element of thegeneral relativism which Popper wants to attack? Let us see.
To start with, it seems to me that an enterprise whose humancharacter can be seen by all is preferable to one that looks`objective', and impervious to human actions and wishes. The sciences,after all, are our own creation, including all the severe standardsthey seem to impose upon us. It is good to be constantly reminded ofthis fact. It is good to be constantly reminded of the fact thatscience as we know it today is not inescapable and that we mayconstruct a world in which it plays no role whatever (such a world, I
venture to suggest, would be more pleasant than the world we live intoday). What better reminder is there than the realization that thechoice between theories which are sufficiently general to provide uswith a comprehensive world view and which are empirically disconnectedmay become a matter of taste? That the choice of our basic cosmologymay become a matter of taste?
Secondly, matters of taste are not completely beyond the reach ofargument. Poems, for example, can be compared in grammar, soundstructure, imagery, rhythm, and can be evaluated on such a basis (cf.Ezra Pound on progress in poetry).4 Even the most elusive mood can beanalysed, and must be analysed if the purpose is to present it ina manner that can either be enjoyed, or that increases the emotional(cognitive, perceptual) inventory of the reader. Evwery poet who isnot completely irrational compares, improves, argues until he findsthe correct formulation of what he wants to say.5 Would it not bemarvellous if this process played a role in the sciences also?
Finally, there are more pedestrian ways of explaining the samematter which may be somewhat less repulsive to the ears ofa professional philosopher of science. We may consider the length ofderivations leading from the principles of theory to its observationlanguage, ad we may also draw attention to the number ofapproximations made in the course of the derivation (all derivationsmust be standardized for this purpose so that an unambiguous judgementof length can be made; this standardization concerns the form of thederivation, it does not concern the content of the concepts used).Smaller length and smaller number of approximations would seem to bepreferable. It is not easy to see how this requirement can be madecompatible with the demand for simplicity and generality which, so itseems, would tend to increase both parameters. However that may be-there are many ways open to us once the fact of incommensurability is
understood, and taken seriously.
(15) I started by pointing out that scientific method, as softenedup by Lakatos, is but an ornament which makes us forget thata position of `anything goes' has in fact been adopted. I thenconsidered the argument that the method of problemshifts, whileperhaps useless in the first world might still give a correct accountof what goes on in the third world and that it might permit us to viewthe whole `third world' through `Popperian spectacles'. The reply wasthat there is trouble in the third world also and that the attempt tojudge cosmologies by their content may have to be given up. Sucha development, far from being undesirable, changes science froma stern and demanding mistress into an attractive and yieldingcourtesan who tries to anticipate every wish of her lover. Of course,it is up to us to choose either a dragon or a pussy cat for ourcompany. I do not think I need to explain my own preferences.
1/ The indices are intended as an ironical criticism of Lakatos[1968b] where the practice of splitting a guy into three was firstintroduced. (Also cf. this volume, p.18 h.) This practire has createda lot of confusion and has slowed down philosophers in their attemptto find the weak spots of critical rationalism.
2/ I am here referring to Popper [Epistemology without a KnowingSubject, 1968] and Popper [On the Theory of Objective mind,1968b]. Inthe first paper birdnests are assigned to the `Third World' (p. 341)and an interaction is assumed between them and the remaining worlds.They are assigned to the Third World because of their function. Butthen stones and rivers can be found in this third world, too, fora bird may sit on a stone, or take a bath in a river. As a matter offact, everything that is noticed by some organism (and therefore playsa role in his Umlvelt) will be found in the third world which willtherefore contain the whole material world and all the mistakesmankind has made. It will also contain `mob psychology'.
3/ An even more conservative principle is sometimes used whendiscussing the possibility of languages with a logic different fromour own. Thus Stroud, in his [Conventionalism and the Indeterminacy ofTranslation, 1968], discussing, and not just stating the principle,says that 'any allegedly new possibility must be capable of beingfitted into, or understood in terms of, our present conceptual orlinguistic apparatus' from which it follows (172) that 'any"alternative" is either something we already understand and can makesense of, or it is no alternative at all'. What is overlooked is thatan initially ununderstood alternative may be learned in the way inwhich one learns a new and unfamiliar language, not by translation,but by living with the members of the community where the language isspoken.
4/ Popper has repeatedly asserted, both in his lectures, and in hiswritings that while there is progress in the sciences there is noprogress in the arts. He bases his assertion on the belief that thecontent of succeeding theories can be compared and that a judgement ofverisimilitude can be made. The refutation of this belief eliminatesan important difference (and perhaps the only important difference)between science and the arts and makes it possible to speak of stylesand preferences in the first, and of progress in the second.
5/ Cf. Brecht [1964), p.119. In my lectures on the theory ofknowledge I usually present and discuss the thesis that finding a newtheory for given facts is Iike finding a new production fora well-known play. For painting, cf. also Gombrich [Art and Illusion,1960)