Attempt 2 at a response:

For every theory, there exists another theory (an infinity of them, actually) that is indistinguishable based on the evidence. To accept any theory at all, one needs some sort of inductive bias (which means what it sounds like), the classic example being Occam's razor.

My point is that there may be ways of learning, based on evidence, good inductive biases. Obviously relying on evidence must end at some level, though, or there would be infinite regress.

Back to proving that I suck at poker.