Here's philosophers' take on Kurt Godel:
http://plato.stanford.edu/entries/goedel/
I am, however, curious as to how other problems in computational complexity figure in philosophy, e.g., P vs. NP completeness. Wish I had the time to read the essay.
But maybe that's not a useful definition of free will. The conventional, fuzzy definition of free will has an omniscient narrator in it; "I do not have free will if the omniscient narrator knows in advance everything I will do" is a reasonable expansion of the conventional idea. But in my hypothesized universe, there is no omniscient narrator. There are only various entities with various degrees of computational power, with a varying but rather quantifiable amount of information that any entity can obtain about any other.
Suppose it can be demonstrated that it is computationally infeasible for any entity due to being too complex to ever predict the actions of any human-sized (or greater) other entity, even if granted all the remaining resources in the universe to compute the actions. Or suppose it is demonstrated to be quite easy. Either way, that would be an interesting philosophical contribution, no?
(I'm just sketching the idea here. I'm not trying to defend it or attack it. Oh, and one of the foundations of philosophy is that there is never One True Definition; all loaded terms in this post are ultimately ill-defined, and I've avoided the distraction of even beginning to nail them down on purpose.)
While the idea of God looking down on us from on high is a compelling story, if you take God away but leave determinism it doesn't do anything to resolve the problem: our actions are still pre-determined and cannot be changed.
Of course, whether or not that makes us free is a further question. A compatibilist [1] would claim that we are.
Also, FWIW omniscience is not logically incompatible with free will.
The incompatibilist argument is at heart very straightforward: free will is incompatible with determinism; determinism is true; therefore free will does not exist. When pressed on the second premise, the incompatibilist can further assert that indeterminism is incompatible with free will, and that determinism and indeterminism exhaust the possibilities (this is just an instance of the law of the excluded middle). Therefore regardless of which one of them holds, free will does not exist.
However, even this argument requires a definition of free will, in order to demonstrate its incompatibility with (in)determinism. It thus suffers from the same 'problem' that you allege compatibilism does: the compatibilist can simply say that the incompatibilist is defining the problem into existence.
I'm curious to know what you think an answer which didn't suffer from the problem you outline would look like.
Epistemology becomes much richer when fed with mathematical proofs of statements like this, and some older snap answers at least have to be reconsidered.
However, it's not in any way obvious that there are metaphysical implications to the limits this sort of result would place on knowledge. Facts are (for present purposes) simply to be identified with physical states of affairs. If some physical state of affairs is unknowable (and "System S will be in state R at time T" certainly looks like a physical state of affairs), does that mean it does not (or will not) actually obtain? I have a hard time seeing how one could make that argument, although of course that shouldn't stop you, if you think you have a good angle!
To address the particular example under discussion: if you take free will to consist of the impossibility of predicting an agent's actions and conclude that because this prediction is impossible, we are free, you still need to provide a further argument that your definition of free will is the correct one.
I'm not prejudging whether or not you could come up with a convincing argument to that effect—it might well be possible. But to my mind, the point of the omniscient narrator example is to add determinism to the system: because our actions are known in advance, they can be known in advance; if they can be known in advance, they must be determined. I take this to be the implicit argument when people suggest that we're not free because God knows what we will do.
Obviously there are different ways in which we can understand the modalities of time, and I won't try to pretend that this isn't a contentious area.
I subscribe to the belief that there are four types of knowledge and beliefs: ontological subjectivity, ontological objectivity, epistemological subjectivity, and epistemological objectivity. Various thoughts can be categorized into each one of these depending on what is being expressed.
Aaronson covers this, pointing out that those results are both in computability theory, not complexity.
Will need to read this, of course. My fault for replying before at least attempt to scan over the full essay linked from the post.
That sentence is messed up in a bunch of ways. You probably meant: "Turing-undecidable problems are a subset of NP-hard problems."
(Actually, are uncomputable problems necessarily NP-hard? This sounds obvious at first, but then you realize, wait, why would the reduction be polynomial time? If we assume our undecidable problem is at least as hard as the halting problem there's an obvious constant-time reduction, but what if it's intermediate? Is this something that's known?)
You're right to raise that question. It's a classical result of computability theory that there exists an undecidable problem that is strictly easier than the halting problem: http://en.wikipedia.org/wiki/Turing_degree#Post.27s_problem_...
But while that's certainly true in a sense, that doesn't imply what you stated -- just because they care about computability theory doesn't mean they care about this aspect of it.
In regard to your second point, you say "other problems in computational complexity". Again, you mentioned computability theory, not CCT. They are different endeavors. Aaronson was careful to say that this is not another essay on philosophy and computability.