1 We don't mean to exclude the position that narrow generalizations attribute states (to the subjects under their scope) simultaneously characterized both referentially and non-referentially. It may be that attribution of such states simultaneously takes care of the states’ referential properties and their ‘modes of presentation’ (however the latter are understood). In other words, we want to take a narrow psychology as one whose generalizations are to be read opaquely in the relevant contexts, without assuming those contexts to be referentially idle.

2 As Fodor himself might put it, exceptionhood on the cheap threatens to collapse laws of the form

1. (i)

   (i) Fs cause Gs ceteris paribus into laws of the form

2. (i)

   (i) Fs cause Gs unless they don't.(Fodor, J.A. ‘Making Mind Matter More,’ Philosophical Topics 67 [1990] 59–79Google Scholar)

3 We will use the term ‘Frege patient’ to refer to agents who

• (a) suffer from ignorance of the identity of the referents of some pair of co-denoting concepts in their possession, and

• (b) are apt to act on this incomplete information in a way which jeopardizes the success of their behavior.

All agents in Frege cases satisfy (a), but only some of them satisfy (b). So Frege patients constitute a proper subset of agents involved in Frege cases.

4 Oedipus is a Frege patient. He is not in epistemic equilibrium: he would have acted otherwise if he had known all the relevant facts, here the identity of Jocasta and Mom.

5 There is an alternative empirical reading of PIE: ‘Most agents are always in epistemic equilibrium in respect of the facts on which they act. Having all the relevant information- having all the information that God has -would never cause most agents to act otherwise than as they do.’ PIE on this second reading seems to be false, especially if R1 is true. But even if it were true, it is unclear how it could support (III). Again, some connecting premises are needed, and their truth would be moot. But most importantly, if the goal is to exclude Frege patients, then most Frege patients will turn out to be outside the scope of intentional psychology, which seems preposterous. Oedipus, like most Frege patients, lives an otherwise perfectly rational and epistemically responsible life. Even if (G) didn't cover him on two occasions, there were many other occasions it did; likewise for other intentional generalizations. To exclude someone from the domain of intentional psychology simply because he has been a Frege patient on a few occasions is simply unacceptable. (In personal communication, Fodor has confirmed that this was not his intention.)

6 Fodor sometimes gives the impression that he thinks agents who are not in epistemic equilibrium are irrational and that is why they are not covered by intentional generalizations. We'll come to this below.

7 Compare Fodor's, own remarks, in the first chapter of Psychosemantics: The Problem of Meaning in the Philosophy of Mind (Cambridge, MA: The MIT Press 1987CrossRefGoogle Scholar), about the psychology of Hermia, Lysander, and Demetrius in Shakespeare's A Midsummer Night's Dream.

8 T1 and T2 appear verbatim in Fodor labeled as such (The Elm and the Expert, 42). T3 also appears there, but we added the label for expository convenience. Fodor calls the first two premises ‘truisms,’ and he appears to regard T3 as more or less self-evident.

9 Fodor also offers a substantially weakened version of this thesis in a footnote (The Elm and the Experts, 122-3n.3). We will come to it below, in section IV.l.

10 That this reading is the intended one is suggested by Fodor's parenthetical remark on T1, where he notes: ‘if an agent has no views about what he would prefer if all the facts were in, then if he is forced to choose, the rational thing for him to do is flip a coin’ (The Elm and the Expert, 42).

11 For discussion of rationality of a different sort, see section IV.2, below.

12 For convenience, we call the belief involved a ‘higher-order belief’ even though it is a belief about one's preferences, not about one's beliefs.

13 But see J.J. Prinz, ‘Is Narrow Content Superfluous?’ for an interesting elaboration of doubts about T2. Available on-line at http: IIcsmaclab-www.uchicago.edu/philosophyProject/LOT/jjp1.html

14 Arjo, D. ‘Sticking Up for Oedipus: Fodor on Intentional Generalizations and Broad Content,’ Mind and Language 11 (1996) 231–45CrossRefGoogle Scholar, makes roughly the same point.

15 We don't mean to suggest that a historical/evolutionary explanation would be the only proper kind of explanation. Certainly, ahistorical and mechanistic explanations might also apply. These two kinds of explanations are not incompatible.

16 Note, however, that Fodor's officially stated aim is not to show that broad psychology is true, but to show that it might be true. He writes: ‘I therefore propose to argue, in this lecture, that it is plausible - not unreasonable to believe - that … for all we know, the laws of intentional psychology may well be broad … I pause for emphasis: I'm not going to argue that psychological laws should be broadly construed …. What I am going to argue is this: the considerations that have been supposed to show that an externalist construal of content won't meet the purposes of psychological explanation are, on balance, unconvincing. So maybe narrow content is superfluous’ (The Elm and the Expert, 28).

17 Assuming, of course, that cetera are paria in this case -which Fodor and others have given us independent reasons to doubt (see note 34, below). But such considerations will be largely bracketed for the purposes of this paper.

18 It is even doubtful that T3 entails that most rational actions are non-accidentally successful. But since the latter claim seems fairly intuitive, we won't make heavy weather of this point.

19 For a lovely Gettier-style example, see Prinz.

20 Though one might well suppose that pathological ('breakdown’) cases, which ordinarily are atypical in this sense, should count as exceptions. The problem is that atypicality need not imply pathologicality; Oedipus is a case in point. And there are various other serious problems with a statistical reading of PIE. It's not clear, for instance, how a reading could be given for R1 that applies only to occasions of acting, as in the case of R3. (In personal correspondence, Fodor has denied that the statistical reading was intended.)

21 In personal communication, Fodor has conceded that he needs (12), or something like it, to be the ‘unmarked case,’ but insisted that this is no problem for his view, since PIE is to be read as a ceteris paribus claim. But it's unclear how the principle could be read in this way. For PIE is supposed to specify an umbrella constraint on other things’ being equal in the intentional realm - a constraint governing the acceptability of candidate intentional generalizations in general. This makes it difficult to see how the usual sort of nomic hedging could be appropriate to it.

22 The cases we will describe are cases involving decisions under risk, where choices are made on the basis of the expected utility of each option. The cases we are interested in are those where the expected utility of the choice actually made comes out to be greater than that of the alternatives because the risk associated with the latter is very high, even though the subjective probability of its occurrence is low. See below.

23 Fodor might perhaps object that if your degree of conviction that you wouldn't change your mind after updating is low enough, then you are not making a rational choice after all. But remember the circumstances we are imagining are such that it is not optional for the agent to gather more information: she just can't. Nevertheless, she uses, in an epistemically responsible way, all the evidence she can responsibly gather. The demands of rationality, we take it, extend no further than this. See Bach, K. ‘Default Reasoning: Jumping to Conclusions and Knowing When to Think Twice,’ Philosophical Quarterly 65 (1984) 37–58Google Scholar for discussion of the inevitable tradeoffs between reliability and efficiency that real-world (i.e., resource-bounded) rational choice involves.

24 Also, note that in this example it is intuitively implausible to suppose that when I chose the orange candy over the green one, I had a relevant HB which was among the beliefs out of which I acted, in the sense that it was causally implicated in the production of my choice behavior. In personal communication, however, Fodor has indicated that he meant the relevant HB to be (at least) implicit, in the sense that were the agent to believe its negation, she would choose differently. Thus an agent would have the relevant HB implicitly just in case she would choose differently were she to believe that she might not stick with her original choice after full updating. But there are difficulties with this suggestion. Most obviously, it remains to be explained how, or in what sense, such implicit HBs could contribute to the production of an agent's choice behavior. To see why, suppose I lack the relevant HB: that is, suppose that, even were I to believe that I might not stick to my original choice after full updating, I might not switch. What would be the likely impact on my choice behavior? As far as we can tell, not much. Of course, Fodor might counter that the implicit HB is one out of which I act insofar as my choice of action would not qualify as rational were I to Jack it. But this again seems implausible.

25 Thanks to David Malament for suggesting the material for this paragraph.

26 Notice that the last two counterexamples are just less dramatic versions of the situation involved in Pascal's Wager. I'm trying to choose, on practical grounds, between two beliefs: the belief that God exists and the belief that He doesn't. As a rational epistemologist sensitive to using evidence in responsible ways, I think it unlikely that God exists; my confidence is not complete of course, but it is closer to 1 than to 0.5. But despite this confidence, I decide to adopt the stance of a religious believer on the basis of (a) a very powerful aversion to infinite torture in Hell, and (b) the assumption that such torture is exactly what unbelievers can expect if it turns out that God does exist. It is plausible that my choice is rational, but I certainly don't believe that if all the facts were in I wouldn't switch (and embrace atheism). What's more, I regard it as likely that if all the facts were in I would switch! I just can't take the attached risk. (Here we're assuming that under certain conditions ‘belief to degree n that P’ can be systematically translated to ‘belief that the probability of P is k.’)

27 Note that in both of these cases, the counterexemplification of T1i depends upon the fact that the agent assigns steeply asymmetric utilities to the possible outcomes of action. One might suppose that such cases need not be taken all that seriously, in the sense that intentional psychology can safely idealize away from them (personal communication with Fodor). But there are problems with this move. First, it's not clear that T1i can be read as a garden-variety ceteris paribus claim, any more than PIE can(see note 21,above).Second, at this point appealing to ceteris paribus clauses has become just too cheap, and in fact, ad hoc. What's needed here is an independent motivation for abstracting away from the cases in question, that is, something other than a prior commitment to broad psychology. But what that motivation might be is anyone's guess. At least it's unclear why asymmetries in an agent's utilities should be thought to interfere with the realization of decision-theoretic laws in anything like the way that, say, friction and air resistance can be held to interfere with the realization of laws governing the motion of spheres on inclined planes. (For more on the general topic of ceteris paribus laws and exceptionhood, see Pietroski, P. and Rey, G. ‘When Other Things Aren't Equal: Saving Ceteris Paribus Laws From Vacuity,’ British Journal for the Philosophy of Science 46 [1995] 81–110.CrossRefGoogle Scholar)

28 Where the notion of justification is internalist, as noted. At least until recently, Fodor himself used to insist on an internalist notion of rationality: ‘according to the present view, questions of rationality are assessed with respect to the vehicle of a belief as well as its content; whereas questions of truth are assessed with respect to content alone …. It's because the vehicle of his belief that his mother was eligible was, say, “J is eligible” rather than, say, “Mother is eligible” that [Oedipus]'s seeking to marry his mother was not irrational in face of his abhorrence of incest’ ('Substitution Arguments and the individuation of Belief.’ Reprinted in A Theory of Content and Other Essays [Cambridge, MA: The MIT Press 1990], 17n.10). See also ‘Methodological Solipsism Considered as a Research Strategy in Cognitive Psychology,’ reprinted in Representations: Philosophical Essays on the Foundations of Cognitive Science (Cambridge, MA: The MIT Press 1981 ), 241-3, where the same claim is elaborated in terms of an internalistically understood notion of content, rather than in terms of vehicles of content.

29 We don't think this is likely, but since Fodor's writing on this issue isn't very clear and there are passages which seem to suggest that he regards Oedipus’ behavior as rationally defective (or at least ‘rash'; see The Elm and the Expert, 46), and since it would nevertheless be instructive to see whether an externalist notion of rationality can come to Fodor's rescue, we will proceed with the discussion. Such an externalist notion of rationality is explicitly in play in the work of Ruth Millikan; see, e.g., Millikan, ‘White Queen Psychology; or, the Last Myth of the Given,’ reprinted in White Queen Psychology and Other Essays for Alice (Cambridge, MA: The MIT Press 1993)Google Scholar, especially sections 1-5.

30 See The Elm and the Expert, 43-7.

31 We think that Frege cases are ubiquitous in a broader sense according to which most intentional agents have many co-denotational concepts or extensionally equivalent thoughts- some of which may not be known to be so to their hosts, and only a few of which happen to provide the occasion for action, in which case their hosts become Frege patients (see note 3, above).

32 If Fodor has e-rationality in mind, his claim doesn't make much sense (and may be question-begging). For supposing that e-rational actions are ipso facto caused by true beliefs, it is already built into the claim that successful actions result from true beliefs; so what is the point of claiming that if T is not granted there is nothing to connect the e-rationality of an action to its success? Irrespective of whether T is granted or not, if the question is what connects the e-rationality of an action to its success, then the question answers itself: the truth of the beliefs out of which the agent acts.

33 Prinz makes much the same point.

34 This assumes, of course, that the notion of content required by a narrow psychology is non-problematic and the project of constructing one is viable. We haven't touched on this issue. If Fodor's long-standing arguments against the viability of such a notion - especially if worked out in terms of functional-role semantics - are cogent, then we may have to learn to live with the fact that intentional psychology can't explain why Frege patients do what they do. (Note too that Fodor's own notion of narrow content as a mapping from contexts to broad contents is no help with the explanation of Frege cases involving concepts expressed by proper names; see Aydede, M. ‘Has Fodor Really Changed His Mind on Narrow Content?’ [Mind and Language 12 (1997) 422–58] for details.CrossRefGoogle Scholar) This might be the real insight behind Fodor's argument. On the other hand, the successful folk practice of explaining interpersonal Frege cases suggests that there must be a viable notion of narrow content- at least if we assume that this practice is, as Fodor, likes to say, ‘intentional through and through’ (Concepts: Where Cognitive Science Went Wrong [New York: Oxford University Press 1998], 7)CrossRefGoogle Scholar.

35 We would like to thank Sara Bernal, Jonathan Cohen, Jerry Fodor, Melinda Hogan, David Malament, Eric Margolis, Mark Moyer, and Jesse Prinz, as well as an anonymous referee for this journal, for helpful feedback. Portions of this paper were delivered at the Pacific AP A meeting and at the 91st meeting of the Southern Society for Philosophy and Psychology (SSPP) in April1999; we would like to thank the audiences for their comments and questions.