Railton ('Rational Desire and Rationality in Desire') argues that "desire is a compound, dynamic state", containing implicit expectations about the desired object, which serve to regulate the desire over time -- e.g. weakening the desire if the actual experience of its object disappoints our expectations. For example: one may desire to try some exotic fruit, with the expectation that it will offer a novel and pleasant taste. But this sounds to me like a purely instrumental desire, where the so-called "favorable expectation" represents what the person really wants or hopes for. So can Railton's account apply more generally, and make sense even of non-instrumental desires?
Let's distinguish five desire 'structures', or ways the object of desire may be related to the regulating favorable expectation:
(1) Instrumental desire: here what you ultimately desire just is the implicit 'expectation' (a novel and pleasant taste, say), and you desire the immediate object (the fruit) merely as a means to this end. Note that mere instruments are replaceable: you could just as well substitute any equivalent means in place of this particular object. [I typically find it unhelpful to count these as real desires at all.]
(2) [Pure] Ultimate desire: if I ultimately desire X, in the pure case this shouldn't come with any further "favorable expectations" besides the desired object itself. (Otherwise, one might think, it starts to look like what I really desire is the expected outcome, rather than the object X itself. But compare the "impure" possibilities below.) Pure ultimate desires thus aren't 'regulated' by any expectation beyond themselves, and hence don't really fit into Railton's account. I'm not sure how serious an objection this is: he might consider such 'pure' ultimate desires to be a theorist's invention that aren't really found in complex human psychologies. (More on this, below.)
(3) Conjunctive desire: Here one desires the object to be as one expects it. E.g. one desires that [one eats the fruit and it tastes novel and pleasant]. This differs from purely instrumental desire because this time you really do want this particular fruit, and not just any source of novel pleasant taste sensations. But it has to be tasty, or your desire is thwarted. [It seems a little odd to model this by separating the desired duo into an 'object' and an 'expectation', though. Why isn't it just a simple desire with a conjunctive object?]
(4) Conditional desire: Here one desires the object (in itself) conditional on its fulfilling one's expectations. This is much like the conjunctive case, insofar as it blocks instrumental substitution. The difference here is that if the expectation (e.g. tastiness) isn't satisfied, the desire as a whole is cancelled rather than thwarted.
(5) Contingent desire: Here one genuinely desires the object for its own sake. (So the desire is straightforwardly satisfied if you get the fruit, regardless of whether it turns out to be tasty.) But the desire just happens to be causally dependent on the expectation. That is, you have some brute (non-rational) psychological mechanism that will cause you to lose this intrinsic desire if the experienced object fails to live up to your expectations. Alternatively, the process might be rationalized by means of a higher-order desire to only have first-order desires that meet their expectations. (I've discussed this possibility before in relation to Railton's sophisticated hedonist.)
[I guess this last proposal is actually compatible with previous variations. E.g. one may have a conditional desire for A given B that is causally contingent on some third condition C.]
What's the best way to interpret Railton, or to understand how most of our ordinary non-instrumental desires fit his model? My best guess is that our ultimate desires are supposed to be mutually regulating, as complex #5-type structures. That sounds psychologically realistic too, since it does seem that (even) our ultimate ends do change over time, perhaps by becoming associated with other ultimate desires/aversions. Any thoughts?