[I]t would be misleading to call my act practically irrational, since my mistake is only in my failure to respond to my epistemic reasons not to have this belief. It would also be misleading to call this act epistemically irrational, since it is not in acting in this way that I am failing to respond to these epistemic reasons.
We should not, I suggest, make either of these misleading claims. When some belief is epistemically irrational, this irrationality can be plausibly and usefully claimed to be inherited by any other belief that depends on this belief. But it is not worth claiming that some beliefs' irrationality is also inherited by any desire or act that depends on this belief... Our desires and acts are best called irrational only when, in having some desire or acting in some way, we are failing to respond to clear and strongly decisive practical reasons or apparent reasons not to have this desire, or not to act in this way.
I find this rather puzzling. After all, the fact that smoking is bad for your health is a genuine practical reason not to smoke (and, we may suppose, a "strongly decisive" one), to which the agent is failing to respond. So whether the smoker qualifies as practically irrational would seem to depend on whether this reason -- that smoking is bad for your health -- is sufficiently "clear" to the agent. But we have stipulated that this health fact is indeed "clear" according to the agent's evidence, such that they are epistemically irrational in failing to believe it. So why doesn't Parfit consider this a case of practical irrationality, by his own definition?
(Perhaps by "clear" he really means "believed", just as earlier he defines apparent reasons as "false beliefs about the relevant facts whose truth would give us some reason." But then we're missing any argument against the competing view that assesses practical rationality against evidentially "apparent" reasons. It sounds compelling -- undeniable -- to say that practical rationality must have something to do with the practical reasons we "appear", in some sense, to have. It's far less compelling to just assert that the relevant 'appearance' is found only in our explicit beliefs, rather than also being implicit in our evidence, and in what we rationally ought to believe.)
On my view, the smoker is making a mistake that relates to his (evidentially apparent) practical reasons. Granted, the mistake wholly derives from a prior epistemic misstep. But that doesn't mean that his mistake is "only" epistemic, since the epistemic error leads him to make a practical error: he smokes when he has clear practical reasons not to. The epistemic error becomes a practical error. That is to say, the agent makes an epistemic misstep that renders not just the initial belief, but also the downstream action, rationally unjustified or unwarranted. (Compare: in the belief case, I go wrong not "only" in believing that p and that if p then q; I am also rationally unjustified in my inferred belief that q -- though of course this rational error is wholly derivative of the earlier ones. There's a sense in which I haven't committed any additional missteps. But the flaw in my prior beliefs has "become" a flaw shared by my later belief also. The same flaw or misstep can infect - and render unjustified - multiple states. Since Parfit grants this in the case of beliefs, it's unclear why we shouldn't say the same thing in case of the practical "states" of desire and action.)
The best way I can make sense of Parfit's view is by attributing to him a distinctive method of carving up epistemic vs. practical rationality. I've been assuming that practical rationality is a matter of responding (e.g. acting) appropriately in light of the evidentially apparent [facts that would constitute] practical reasons. But note that there are two broad ways that one might fail to act appropriately: one might fail to believe the (evidentially) "apparent facts" in the first place, or one might incorrectly (perversely) interpret the normative significance of these factual beliefs. The former failure is the epistemically-derived practical irrationality we've been discussing. The latter is the more 'pure' form of practical irrationality found in malicious evildoers, Future-Tuesday indifferent agents, etc.
Now, I take it that Parfit wants to understand the latter, 'pure' kind of practical failure as the only genuinely practical failure. On this view, practical rationality is not a matter of "responding" (in my broad sense) to our evidence-relative reasons for action -- first noticing, then non-perversely interpreting them. It instead concerns only the interpretive element. We might say that (on this view) practical rationality is a matter of not being perverse. (Note that there's certainly nothing perverse about the smoker who irrationally thinks he's helping his health. The guy is acting stupidly, but -- given his beliefs -- hardly perversely. "At least his heart is in the right place," we might say. "Too bad he's such an incompetent fool.")
In support of this interpretation, I notice that Parfit later writes (in discussing an exception to the general rule that epistemic and practical rationality are independent) that "if we have irrational beliefs about practical reasons, and about what we ought rationally to want or to do, our having such beliefs makes us in one way practically irrational." This makes sense if by 'practically irrational' Parfit just means perverse. It's certainly true that someone who falsely believes that they ought to do evil (or be Future-Tuesday indifferent) is thereby "in one way" perverse. But merely having this belief doesn't suffice to make them practically irrational in the ordinary sense. After all, they might instead be rational akratics, and act entirely appropriately despite their crazy notions.
So, is this just a terminological dispute? It would be if Parfit meant his use of the term to be stipulative. But that is not the case. Instead, he writes:
I am using 'irrational' in its ordinary sense, to mean, roughly, 'deserves strong criticism of the kind that we also express with words like "foolish", "stupid", and "crazy"'. ... If we believe that one of two preferences deserves much stronger rational criticism, we shouldn't say that only the other preference is irrational.
So it seems we're disputing the character of a certain normative role (roughly: what makes an act/desire foolish). Parfit's view, if I've understood him correctly, is that perversity is what makes an act or desire foolish. I agree that this is one way that an act can be foolish. But an act can also be foolish in a way that derives from a prior epistemic error, as when an agent smokes for the sake of benefiting their health. If we believe that the smoker's act deserves rational criticism, we should reject Parfit's understanding of the epistemic/practical distinction. Perversity has an important role to play in our normative theorizing, but it isn't this one. Acts and desires can be foolish -- "deserve rational criticism", in the ordinary sense -- without being perverse.
Postscript: I think that what Parfit is really getting at here is a distinct but related role, namely, what makes a mis-step practical rather than epistemic. It's important to note that this is a distinct question from what makes an act or desire deserving of rational criticism, since -- as I've argued -- an act or desire can be foolish or unwarranted in a way that derives from a fundamentally epistemic mis-step.
Perhaps it's best to say that there are really two significant 'epistemic/practical' distinctions to consider in our theorizing about rationality. There is the ordinary state-based distinction between the rational status of epistemic states (beliefs), on the one hand, and the rational status of practical 'states' (acts or desires), on the other. Then there is Parfit's step-based distinction, between fundamentally epistemic missteps (stupidity), and fundamentally practical missteps (perversity).
What I've effectively been insisting is that an act might be irrational (in the ordinary sense -- unjustified and deserving of certain sorts of criticism) even if the agent's only mis-step was epistemic in nature. Contra Parfit, it doesn't follow that only their epistemic states are liable to rational criticism, for a single misstep may set awry multiple states, and an epistemic misstep may set awry downstream practical states. So long as we're clear on the difference, each distinction may well be "worth" talking about (depending on our purposes), and neither can be dismissed as necessarily "misleading" -- though of course either can be misleading, if conflated or used for the wrong theoretical role.