^{page 133 note 1} Addendum to note 2, page 140. The criteria of soundness in an argument (a) take in considerations which are special to the relevant subject matter, (b) do not in general preserve the principle of monoticity. If P justifies Q, it does not follow that (P & R) does. Soundness is ‘defeasible’ by adding extra considerations, e.g. ‘NN is not going to lecture today’.

But see Anscombe, G. E. M.'s critique in Socratic Digest, iv (Oxford, 1946) of ch. 3Google Scholar of Lewis, C. S.'s Miracles.Google Scholar

^{page 134 note 1} Something is said about this on pp. 140–1, n. 3.

^{page 134 note 2} But I do feel constrained to make about these questions a single (long) remark. If we could get the general problem sufficiently well under control to undertake the marxist question then the tu quoque argument might certainly come into its own again. But not in a way calculated to cheer either side in the controversy. It might, for instance, take the form of suggesting that capitalism and marxism are precisely alike in one important, special and causal respect – that both ideologies are formed and vitiated throughout by a special obsession, an obsession with a concept of work, or more specifically production, which capitalism communicated to marxism even as marxism reacted to it. It is this common aetiology which is anthropologically special, and philosophically and imaginatively so impoverished, in both positions. In a world where some Gresham's law propagates the cultures of the more powerful and industrialised nations it may come to take more imagination than it takes now to see that it is not logically constitutive of human rationality to delegate decisions of ends and their components to economists or cost accountants, or to find it a matter of surprise if the result disappoints. But even now the result of the preoccupation with economically quantifiable output has been to degrade, or sentimentalise (which is worse), the evaluation of absolutely everything which cannot be described or accounted for properly in such terms. (Witness our now etiolated and apologetic notions of ‘leisure’, ‘amenity’, and even ‘welfare’.) That is the remark. In this use of the causal argument, however, tu quoque discredits both positions and both the capitalistic and communistic forms of consciousness. But adjudication must really wait on the answer to the question which directly concerns me – why does not the causal determination of a belief as such corrupt it?

^{page 135 note 1} Proc. Arist. Soc., suppl. vol, XLI (1967)Google Scholar. I might also have cited an article in The Times newspaper (25 01 1969, p. 9)Google Scholar ‘Why the brain is more than a mere computer’ by Professor W. H. Thorpe, F.R.S.: ‘Even assuming the pre-Heisenberg conception of a physically determined brain, we cannot from the very nature of the case thereby “explain” our acquisition of knowledge of the external world or explain away our ability to make a personal choice. Whatever the direction in which future advancement of knowledge concerning the action of the brain may go, it cannot have this consequence; for if it were true none of our knowledge on any subject could be valid and so it is self-destructive.’

^{page 137 note 1} See Davidson, Donald's ‘Actions, Reasons, and Causes’, in Journal of Philosophy (1963).Google Scholar

^{page 137 note 2} Cf. H. P. Grice, ‘The Causal Theory of Perception’, in Proc. Arist. Soc., suppl. vol. xxxv (1961).

^{page 137 note 3} ‘Remembering’, in Philosophical Review, LXXV (1966).Google Scholar

^{page 138 note 1} ‘Is Justified True Belief Knowledge?’, in Analysis, XXIII 6 (06 1963).Google Scholar

^{page 138 note 2} Lack of space precludes a fuller expansion of the argument here, but I hope to do so elsewhere. It proceeds by letting the principle of sufficient reason push every kind of alternative proposal to its unsatisfactory conclusion.

^{page 139 note 1} A causal or nomological necessitas consequentiae, not consequents. It is true that nothing like these requirements applies to such concepts as believes fortuitously correctly, which qualifies with ‘know’ and ‘see’ by my definition of an epistemic concept. But the definition could be delimited further and in any case there are obvious limits to the usefulness or importance of such fortuitous concepts. What we plainly could not dispense with are the ones which are framed to exclude accidental correctness. And of the latter such as ‘know’ and ‘see’ I hope to make it seem credible that they all are and must be partly causal.

^{page 140 note 1} A thorough examination would bring into prominence what could have been said apropos of knowledge itself, that it is really too strong to specify belief in the necessary condition. Limitations of space prevent me from importing the qualifications and amendments I should wish here.

^{page 140 note 2} It is important to emphasise that the kind of definition of knowledge I have been defending works out in quite different ways for different sorts of case. A belief constitutive of knowledge of some future fact P, for instance, could hardly be the effect of P. What the definition suggests that we should look for is a linkage between some Q, which prompts the belief that P and P. Nor could the mathematical knowledge that there is no greatest prime be the causal effect of the structure of the number system. Here my definition of knowledge suggests that we look and see why the knower believes this. If the relevant Q, is the fact that he has been told so, then we must see whether his source was good enough for us to accept the proposition on the strength of this. (It is neither here nor there that this gives a very derivative sort of knowledge.) If Q. be the fact that he has reasoned about the matter, then is the fact that this particular man has reasoned good grounds to accept his conclusion? (Again this is all beside the point so far as mathematical truth is concerned. But I am talking about knowledge of mathematical truth, not mathematical truth.) To defend it against trivialisation and to get a fully satisfactory definition of knowledge of P it is necessary, for reasons into which I shall not enter, to add conditions (iv) and (v) as follows: Where x believes that P, x knows that P if and only if (i) x believes P, (ii) there is a Qsuch that x believes P because of Q, (iii) Q, and x believes P together yield a sound (not necessarily deductive) argument to the truth of P, (iv) this argument needs the premiss x believes P, (v) no subset of this argument's premises can be made to yield the conclusion that P. (A sound argument requires true premises.) But believes is too strong, v. note land p. 132.

^{page 140 note 3} Only amplification along such lines can begin to complement the standard quick answer to the causal argument for perceptual scepticism (see page 134). It is not enough to say that the concepts by which we describe the perceived object (which the sceptic doubts is truly revealed in perception) are themselves formed around the human senses and may only thus gain their place in thought and speech. We can only block the sceptic's attempt to get behind the way in which meanings are fixed for descriptions of objects if we show why a human sense must have a causal relation to its object. But even this is not enough without a decent account of the relevant distinctions between primary and secondary qualities and their respective titles to be real properties of objects.

^{page 142 note 1} Even with ‘quod’, which can take the indicative in Latin, the subjunctive would have to be used to phrase Q in ‘Credit P quod Q’. For the form we are considering involves indirect speech, Q, being a reason the man himself could cite as his reason for believing. The question arises: what about the Latin for my canonical form ‘He believes P because he believes Q?’ On my view ‘Credit P quod credit Q’ is compatible with ‘Credit P quod (subjunctive) Q,’ and leaves it quite open whether or not Q is something x could give or could not give as a reason for believing P. It is undetermined whether or not the causality is of the relevant (Humean) species. ‘Credit P quod credat Q’, on the other hand, means that he could cite his believing Q as his reason for believing P. Unless he were psychoanalysing himself Q would have to be some sort of grounds of his for believing P. I am not here or anywhere advancing a causal account of the evidence-conclusion relation itself. I am only attempting to show that a causal theory of what it is to believe on grounds Q, is not necessarily involved in various simple-mindednesses of which it will otherwise be accused. The subjunctivity of Q, in ‘Credit P quod Q’ might suggest the hypothesis that what is believed is rather the proposition ‘P quod Q’ (the subjunctive being that of a dependent clause within existing oratio obliqua). But it seems to me that a man can believe (be willing to assent to, etc.) ‘P quod Q’ without Q being his reason for believing P. Thus if, as seems likely, ‘Credit P quod Q’ (with subjunctive Q) is really a form for his giving his reason, the suggested analysis is wrong. It certainly seems very unlikely.

^{page 144 note 1} Cf. Aristotle, , De Motu Animalium, 701Google Scholar, and, for a committedly causal interpretation of this determination, see Davidson, , in Journal of Philosophy (1963).Google Scholar

^{page 145 note 1} This is not a behaviourist account of belief. This sensitivity is a disposition. Not everything which is sensitive always reacts to its proper irritant, and we cannot in general define dispositions by any set of observation statements or ‘if … then …’ statements. Properly construed, dispositions are rather inscrutable – even straightforward ones such as brittleness or flexibility – and their inscrutability well matches the inscrutability of belief.

^{page 145 note 2} My debts here will escape no reader who knows Dummett, M. A. E.'s ‘Truth’, in Proc. Arist. Soc. (1959)Google Scholar or B. A. O. Williams's further development of a similar theme in ‘Consistency and Realism’, in Proc. Arist. Soc. suppl. vol. XL (1966).Google Scholar

^{page 145 note 3} And this in one coherent account of the difference between doing X, which it so happens you want to do, and doing X because you want to (see Davidson, , in Journal of Philosophy (1963))Google Scholar. The philosophy of action so far lacks a proof that this is the only conceivable account, it must be admitted, but it is at least one account of this difficult difference.

^{page 147 note 1} It is not enough for the argument to move from ‘necessarily if P, x does A’ and P to ‘x necessarily did A, and therefore had no real choice’. For this apparently apes the patently invalid argument □(P→Q), P, ⊢ □Q. I believe that another argument and some supplementary premises will do justice to what worries the libertarian. But this is not easy to construct.

^{page 151 note 1} A final point about (iii). Even if it showed that a man could not be in a position to predict all events or every component of a world state it could not show that there was any one event or component which he was unable to predict. ∼ ◇ (x) (A predicts x) ⊃ (x) ∼ ◇ (A predicts x) is not a logical truth.

^{page 151 note 2} It is sometimes said that he could not predict them himself though another predicter could predict them. But his movements are no less the values of variables than anyone else's are. Given the assumptions just mentioned, what conceivable difference could it make who did the computations? And even if it did make a difference it is not clear why two predicters could not use one another as predictive devices. If there is alleged to be an interference problem that is dealt with below under argument (ii). If it is alleged that this would not count as knowledge, does the objector have an account of knowledge which does not lead to a regress of reasons ? As will be evident from my discussion of knowledge in Part 1, I do not believe any theory of knowing can define ‘knowing’ in the traditional rationalistic manner and exempt the notion from all dependence on the external fact that the knower is simply (but non-accidentally) right.

^{page 154 note 1} Unless to make one remark. Some people say (1) that quantum mechanics are ultimate and reflect ultimate reality; electrons simply cannot have both determinate velocities and determinate mass – they are not that sort of thing; and (2) that quantum mechanics show the limits of knowledge; for the act of getting to know interferes with what is known. Can one really have it both of these ways? And see again note 2, page 151.