Retributivists who proclaim our moral obligation to punish criminals have displayed, on their own behalf, a type of argumant which I shall call Moral Balance. There are three versions of Moral Balance. According to Moral Balance I, retaliatory punishment restores the equality disturbed by the criminal. Moral Balance II philosophers admire the proportion between morality and welfare which punishment can yield. Those who hold with Moral Balance Ill are fascinated by the equilibrium of social benefits and burdens set by punishment. All three versions are flawed as follows: they imply that crime is morally justified.

Moral Balance 1: Equality

One smiles and tells the expert chef how good the sauce béarnaise is, not so much to inform him about the sauce (he knows better than we do how good it is) as to assure him that we are enjoying it and that we appreciate his efforts. But when a man kneels in his pew and repeats a litany of thanksgiving it is not, it seems, that he means to be informing God of anything—not even of his thankfulness. For God, unlike the chef, has no need of information.

Philosophers in the tradition of Berkeley say that the first step in gaining knowledge from perception is to report or describe one's perceptual data, or that which one sees (feels, hears, smells, tastes) ‘immediately'. Further, perceptual data (1) are existing things of some sort, and (2) always are exactly as they appear to be since, as H. H. Price says, “in the sphere of the given … what seems, is”. However, these two claims about perceptual data are sometimes incompatible, as the following case shows. Suppose a man looking at a speckled hen reports that it has many speckles but is not able to report the exact number of speckles it has. If (2) above is correct, the man's perceptual datum, as opposed to the physical hen itself which he sees—according to Berkeleians—in a merely indirect sense, would have to be indeterminate in character. Yet we assume that everything which exists (cf. requirement (1)) is and has to be completely determinate. A person's knowledge is indeterminate if he knows that the planet Mars has either two or three satellites and doesn't know which. But Mars itself at any moment has to have some definite number of satellites.

Along with more customary claims, Richard Taylor advances a new argument against a straightforward indeterministic analysis of free will. The analysis at issue is straightforward in the sense that it considers an occurrence to be a freely performed action, for which someone is fully responsible, if and only if it is causally contingent (i.e. causally possible but not necessary) in a particular way. Since Taylor's assumptions lead him to regard this position as a non-starter, the new part of his argument is not expressly designed as anti-indeterministic. He assumes that if an action were merely causally contingent relative to its antecedent conditions, then it would be random in a sense which involves that no-one could be responsible for it. Unlike the determinists who assume this too, he believes, however, that causal contingency is a necessary, though not sufficient, feature of any freely performed action, for which someone is fully responsible. The new part of his argument is expressly designed to support this intermediate position by establishing the existence of some further factor in agency which in one of its roles supplements the contingency of any action that is freely performed.

The dispositional analysis of sentences ascribing powers and abilities to things and people runs into many difficulties. The situation is especially complex and confusing when such ascriptions are brought to bear on particular occasions, and carry some implication about the opportunities available on those occasions. Some of these difficulties are produced by superficial confusions, but I shall concentrate on those that beset any version of the dispositional analysis. Three difficulties will be examined, one presented by Lehrer, and two by Austin.

Lehrer presents his difficulty as a refutation of Moore's dispositional analysis of “X could have acted otherwise”, which was “If X had chosen, he would have acted otherwise”. I shall introduce his objection by relating it to the distinctions drawn in Part I.· Expressed in my terminology, his point is that Moore's analysis presupposes that choosing is not an S-factor, and yet it might conceivably be an S-factor.

The account of belief adumbrated by Ryle in The Concept of Mind is, I think, a very tempting one despite its relative vagueness. According to this account, a belief that such and such is the case is a disposition of a certain kind. More specifically, it is a tendency or a propensity to behave and to react in more or less definite ways under certain circumstances. Thus “to believe that the ice is dangerously thin,” Ryle writes, “is to be unhesitant in telling oneself and others that it is thin, in acquiescing in other people's assertions to that effect, in objecting to statements to the contrary, in drawing consequences from the original proposition,. and so forth.” Moreover, he continues, “it is also to be prone to skate warily, to shudder, to dwell in imagination on possible disasters and to warn other skaters.” In short, to believe that p is to have “a propensity not only to make certain theoretical moves but also to make certain executive and imaginative moves, as well as to have certain feelings.”

Austin's lecture on this topic contributes little to the problem of freedom of the will, and so in my discussion of his ideas I shall stop short of the difficult part of that problem. His most important positive suggestion is that hypotheticals should be divided into two classes, conditionals and pseudo-conditionals. He claims that neglect of this distinction has been the cause of mistakes in certain forms of the dispositional analysis of the statement that an agent could have acted otherwise, and he then goes on to criticise all forms of that analysis using arguments which do not depend on the difference between the two kinds of hypothetical.

This discussion will be in two parts. In the first I shall take up Austin's distinction between conditionals and pseudos and criticise it and develop it at length, because it seems to me to be the most important thing in his lecture. Since I shall not have space for comment on fortuitous errors made by defenders of the dispositional analysis of the claim that an agent could have acted otherwise, I shall concentrate on the essential features of that analysis.

The principle of legal paternalism justifies state coercion to protect individuals from self-inflicted harm, or in its extreme version, to guide them, whether they like it or not, toward their own good. Parents can be expected to justify their interference in the lives of their children (e.g. telling them what they must eat and when they must sleep) on the ground that “daddy knows best.” legal paternalism seems to imply that since the state often can know the interests of individual citizens better than the citizens know them themselves, it stands as a permanent guardian of those interests in loco parentis. Put in this blunt way, paternalism seems a preposterous doctrine. If adults are treated as children they will come in time to be like children. Deprived of the right to choose for themselves, they will soon lose the power of rational judgment and decision. Even children, after a certain point, had better not be “treated as children,” else they will never acquire the outlook and capability of responsible adults.