[2] Brown confines the role of judgement to circumstances in which no algorithm is known to the agent (139); about other situations, he says that "when rules are available... [a rational agent] will apply those rules" (174). This last is a mistake: an algorithm may require computations beyond the capabilities of the agent; or, more mundanely, the cost of applying the algorithm may exceed the reward to be obtained by applying it. However, this is not the main thrust of my criticism of this approach to rationality.
[3] Healy goes further than Brown, and argues that judgement is involved in rational decision even when rules are available. He writes that "we are effectively in the same boat regardless of whether we are engaged in a rule governed process... or in a non-algorithmic process.... Even in the former situation... we must appeal to judgement to establish that we have the appropriate rules at our disposal and how they are to be applied" (paragraph 5).
[5] The judgement exercised by an agent can hardly fail to conform to some set of rules (possibly probabilistic rules, or rules expressed in the terms of a theory of fuzzy sets, or something of the kind); to say it does not is to say that it is infinitely complex{2}, and an infinitely complex behavior seems clearly beyond the reach of a finite agent.
[6] Brown considers a similar objection. He writes:
[P]hysical systems, it will be maintained... operate in accordance with rules, even if these are statistical rules.... Thus if judgement is a natural phenomenon, it will be a rule-governed phenomenon.... In order to deal with this objection... we must distinguish cases in which a rule is being followed from cases in which behaviour conforms to a rule, but the agent is not following that rule.... In order to follow a rule, we must be aware of it and consciously guide our behavior by it.... [C]onsider a falling stone that moves in accordance with the rule that its acceleration shall increase in inverse proportion with the square of its distance from the center of the earth. The stone does not follow this rule, even though the rule accurately describes the stone's behavior. In general, showing that an item's behaviour can be accurately described by a rule shows only that it behavior conforms to that rule; considerably more is required to show that the rule is being followed. (174-5)I have two replies to this argument of Brown's. The first is simply to doubt whether the distinction between following a rule and conforming to a rule can be maintained if one takes naturalistically all of the concepts mentioned in Brown's declaration that "in order to follow a rule we must be aware of it and consciously guide our behaviour by it."
[7] The second and stronger reply is to deny that Brown's distinction is to the point, even if it can be interpreted naturalistically. Not all agents exercise judgement in a given area equally well; thus, even if behaving rationally is exercising judgement without following rules, but merely conforming to them, we can try to determine which rules it is more (or most) rational to conform to.
[8] This brings me to my main criticism of Brown's, and a fortiori of Healy's, theory of rationality: by failing to examine the internal structure of the skills underlying judgement, they set an unnecessarily low standard for rationality.
[9] Is the tacit belief underlying the skills that constitute judgement supposed to be true, i.e. is it supposed to be knowledge? There is no reason to suppose that a reliable skill or expertise requires more than local and approximate truth (cf. Hattiangadi 1987): since a skill is learned by training or by apprenticeship, it will be at least as local as the range of examples in the training set. Moreover, if the belief underlying one's skill is tacit, one will not have more than an approximate idea of the domain within which the skill is reliable: agents will not normally be able to answer the question "Is it rational to rely on skill S in situations of type T?" To answer this question, the agent would have to have the knowledge underlying the skill explicitly.
[10] Thus skills are tacit in two ways: the knowledge underlying a skill is known only tacitly, and from this it follows that the proper domain of the skill is known only tacitly.
[12] This is why it will pay to study which sets of rules may rationally be conformed to by which sorts of agent, in which sorts of circumstances (see my forthcoming a, b, c): agents of different cognitive capabilities will find that different forms of judgement are most rewarding when employed for their purposes. Not all expertise is equal, and while skills may be learned from authorities, they can be criticized, and need not be accepted on authority.
[13] The approach suggested by Brown and by Healy cuts inquiry short by treating judgement as a black box. The result is that on their approach rationality becomes nothing more than dogmatic conformance to tradition and to the norms of one's community. As Agassi and Jarvie have shown in an illuminating series of papers (Agassi and Jarvie 1987b, c, d, e), behaving in this way can be rational to a degree, but one can hope to do better.
Agassi, J. and I. C. Jarvie (1987b) "The Problem of the Rationality of Magic." In (Agassi and Jarvie 1987a).
Agassi, J. and I. C. Jarvie (1987c) "Magic and Rationality Again." In (Agassi and Jarvie 1987a).
Agassi, J. and I. C. Jarvie (1987d) "The Rationality of Dogmatism." In (Agassi and Jarvie 1987a).
Agassi, J. and I. C. Jarvie (1987e) "The Rationality of Irrationalism." In (Agassi and Jarvie 1987a).
Brown, H. I. (1988) Rationality. London: Routledge.
Polanyi, M. (1964) Science, Faith, and Society. Chicago: University of Chicago Press.
Polanyi, M. (1974) Personal Knowledge (Corrected edition). Chicago: University of Chicago Press.
Putnam, H. (1978) Meaning and the Moral Sciences. London: Routledge.
Ryle, G. (1971) Collected Papers. London: Hutchinson.
Reiner, R. (Forthcoming a) "Non-arbitrary Imperfect Rationality: New Directions for the Theory of Rationality."
Reiner, R. (Forthcoming b) "Arguments Against the Possibility of Perfect Rationality."
Reiner, R. (Forthcoming c) "Modelling Bounded Agents."
{2} Such a behaviour would be infinitely complex not merely in the sense in which a fractal curve is infinitely complex, but in the stronger sense that necessary and sufficient conditions for it cannot be finitely specified. Return