Robert Kane notes a problem for compatibilism with Aristotle's conception of moral responsibility. Here's how the reasoning goes.
In order for S to be morally responsible for X, X must arise from S's character.
In order for S to be morally responsible for S's character, S's character must arise from S's character.
. . .
An infinite regress threatens. If it is necessary for someone to be morally responsible for X, that person's character must not only be the cause of the action but also the cause of the character that causes the action, and so on ad infinitum, then it is impossible for anyone to be morally responsible for his/her actions absent some sort of self-determining, originating act.
However, I do not see why the regress should take hold. We might admit that S must be responsible for his/her character and actions, but why should we think that S must also be responsible for the character that gave rise to that character and actions (and so on)? There is an implicit assumption that each step in the regress is necessary for moral responsibility for the next step. But why should this be so? We do not think that explanation for some event requires a complete causal history of that event in order to explain it.
Here are a couple of examples that suggest that the causal history of the acquisition of one's character is not in fact relevant to one's moral responsibility, and one might fail to be in control of the acquisition of one's character and still be morally responsible.
Imagine a Star Trek style duplicating device (or "transporter"). One steps in the device and a molecule-for-molecule duplicate emerges from the other end. We'll call this person S2 in contrast to the original S1. S2 is not in any way responsible for the acquisition of his/her character; S1 acquired the character and S2 is just a duplicate. But it would be odd to think that S2 is for this reason not morally responsible for his/her actions. S1 and S2 are physical and psychological duplicates, and any kind of moral responsibility S1 has S2 has as well (although not necessarily responsibility for having done Y or Z, just the general capacity for moral accountability).
The only response I can see is that perhaps S1 and S2 are in actuality the same person despite Parfit and Williams-style problems with such a view. So, an alternative thought experiment could derive from Donald Davidson's swampman.
Consider Swamp-person who is a perfect molecule-for-molecule duplicate of S who appears spontaneously out of swamp gas by pure coincidence. With no causal connection to S, it is not possible for Swamp-S to be identical to S. (We could suppose, anyway, that Swamp-S just has all the mental and physical capacities of a person without being a duplicate of any particular person.) Now Swamp-S is still not responsible for Swamp-S's character, but it appears that Swamp-S is morally responsible for his/her actions, just as responsible as any ordinary agent would be.
If these cases are genuinely ones in which the duplicate (or similar being) is not responsible for the causal history of his/her own character, and these beings are morally responsible for their actions, then the causal history of one's character is not relevant to one's general capacity for moral responsibility. A fortiori it is not necessary that one be morally responsibility for the acquisition of one's character. Moral responsibility appears not to depend on the means of acquisition of one's character. Therefore Kane's argument, depending as it does on the transitivity of moral responsibility from one act of construction of character to the next to, ultimately, one's action, is not sound. We can be morally responsible even if we do not go through this process.
And that's a good thing because we cannot initiate our own choices free from preceding causes. (But that's another post.)
These examples should show that Compatibilist notions of natural or normal production of character are largely unnecessary and irrelevant. The production of my desires by natural or normal means is not necessary for my moral responsibility since I could have it even in the Star Trek/Swamp-person cases, and those cases are as unnatural and abnormal as one can imagine.
And really it is implausible that the naturalness of the acquisition of one's character/desires is relevant. Suppose people on some world develop a craving for what they call "Food" and "Water", and that this craving develops along perfectly normal, natural developmental pathways for the beings in question. Why would a desire for food be any less exculpatory than a desire for drugs that were, let us suppose, given to the victim unnaturally? If each provides a similarly strong compulsion to act on that desire, and they have otherwise identical psychological capacities and states, then their naturalness or unnaturalness is irrelevant to the responsibility one has for the action.