In our discussion on Jung, I brought up the issue of free will with respect to the existence of the unconscious, and I wanted to explore this a bit further:
Compatibilism is the doctrine that free will and determinism are in some way compatible, but since these terms were designed to contradict each other, any claim to be a compatibilist requires an account of how this is possible. There needs to be some level of analysis of the situation by which we're (either collectively, or you can look at individual cases) free and another by which we're not.
Kant's compatibilism was initially explained to me in school as such: It's like we're robots programmed to think we're free. Of course, since we're programmed, we're really not free, but feeling free is all that's required for moral responsibility.
Actually reading Kant myself, I saw that his position was closer to the reverse of what I've just stated: In the world of experience, science shows us (or more precisely, it bases itself on) the fact that everything is caused, so thus we are, according to this analysis of phenomenal reality, not really free. However, an underlying assumption behind the way we treat each other and ourselves is that we really are free, so even though we technically can't say anything about the world behind phenomena, we're forced by practical reason to assume that in reality (as noumena) we're free.
I'm not too concerned here with Kant scholarship. The point is that there's in both analyses a surface level of analysis whereby we're judged free in the first case but not in the second, and a deeper level whereby we're judged the opposite way. The second case is funny in that the surface level, the level that science acts at, is actually deeper than the level of experience itself, whereby we think we're free. So you might classify my second characterization of Kant here as having three levels:
1. The level of immediate experience (whereby we feel free normally, as opposed to when we're in chains and don't feel free, though in that case at least our thoughts and micro-movements still feel free; you can also imagine feeling possessed due to some neurological difficulty, with your body or even thoughts not obeying your commands)
2. The level of scientific analysis (science, by doing a systematic analysis of experience, is getting at something non-obvious to ordinary experience)
3. The level of the world-in-itself (which Kant thinks we can't know about directly but thinks that we need to make certain assumptions about due to practical reason; Schopenhauer then runs with this view to posit that the whole world has a Will)
One can add to this a possible level 1.5: The level of considered experience. Marx and Freud both pointed out instances where we feel free, but on (possibly lengthy) consideration of our motives, we realize that we were being influenced by factors that we do not consider ourselves allied with: we were not really free because we were being tricked in some way, either by ideology or by our own psyche.
Classical compatibilism is concerned largely with levels 1 & 2 (I can feel free, and that's enough, even though science says I'm really not), though if you take the problem of compatibilism to be fundamentally theological (God knows what I'm going to do, so how can I be free?), then you're really concerned with 1 & 3 (I feel free, but since God caused and knows everything, I'm really not).
You can agree about what the relevant levels are to consider but still disagree about which level matters for which purposes. William James (and Sartre, and Ayn Rand, among others) argues that level 1 is all that matters, that the claim of #2 is either factually incorrect or at least totally irrelevant to matters of ethics, guilt, punishment, self-motivation, and the like. One can just define #1 as determinate of whether or not an action is free.
Modern neuropsychologists who argue that free will is an illusion because, e.g. criminal brains have a certain structure that others don't, are arguing that #2 really does matter, and should affect how we treat others through praise/blame and public policy.
Likewise, Nietzsche took #2 to be the Enlightened view given what science has taught us, but his phenomenology of the will focused also on #1 and its relation to #1.5. You might see #2 as the ultimate unknowability of the self: our decisions are caused ultimately by factors in our psyche unknown to us, but we can at least become partially examined, as opposed to being an unreflective, #1-focused individual who gets a thrill out of, e.g. smoking as a way of rebelling against his parents without being aware that this is his motive.
The view of freedom shared, I think, by Frithjof Bergmann and Robert Pirsig (among others) is that freedom is at once an immediate experience, a feeling of being free, but also the result of lengthy introspection. Bergmann describes freedom as identification with your current action: you're free if you're doing something that you really want to do, where people's psyches are opaque enough that figuring this out might take some work, might take not just thinking but trying out things, seeing what "feels right," actually developing desires. This self exploration I described in the Heraclitus episode as a "stable ambivalence:" we are not in an epistemic position to determine whether we are discovering or building a self in such a circumstance.
I think this concept is equally applicable to the free will problem itself: we can self-explore and self-explore and so enable ourselves to make choices that are increasingly free, i.e. that wouldn't be overturned upon further reconsideration. However, there's never a final insight; we can never really be sure of our motives, and if Freud and co. are right, our ignorance is not just a matter of us not having sufficient knowledge of the facts of our desires. If that were the case, we could say "you made a free choice; it just wasn't a considered choice." No, according to Freud, it's not a matter of a conscious decision made on the basis of incomplete evidence but a decision made by a mechanism that we--in ways that could be crucial to a given decision--don't understand. You could characterize this (as Jung does) by saying that the agent, the self that makes the decisions, is a larger entity than the individual conscious ego, or you could just deny that there really is any agent: that decisions are a result of a mechanical process that is affected both by calculations based on data (i.e. our knowledge, conscious or not, of the circumstances of the case including our own desires, probable outcomes of our action, the environment in which we're performing the action, prior relevant occurrences, etc.) and other motive forces that either lack rationality entirely or at least involve a "decision-making" component that is foreign to our conscious decision-making.
So you might say that we are free to the extent that we are rational (where I'm taking rationality not only in the sense of explicit reasoning but in whatever else is going on when we self-discover), but can never know really how rational we're being, how partial our examination has been. For Sartre or Rand or anyone else to insist that you know that you had a choice and so were really free (and so deserving of guilt/culpability) is simply a gross oversimplification that denies the fundamental insights of Nietzsche, Freud, and Jung.