Karl Popper actually proposes a nice criterion to distinguish science from non-science. He argues (whether persuasively or no) that scientific claims are falsifiable, where non-scientific claims need not be. That allows us to distinguish between things like astrology, whose claims are generally not falsifiable, and mathematical astronomy, whose claims are falsifiable.
Which brings me to an interesting question. If a hypothetical alien civilization develops technology using different types of math than us, different analogies than us (particles/wave, etc.), and different beliefs than us, is it still science? I would argue yes. If they can create sophisticated technology that is predictable, the end effect is the same that we humans have.
Which brings me to my last question. Is string theory science? Again, I would say yes. Because although its predictions are not currently capable of being tested, they are in principle capable of being tested with more sophisticated technology. There is a clear path to developing a test of its predictions.
So I think "science" can best be described as the pursuit of prediction.
http://www.youtube.com/watch?v=EYPapE-3FRw
I have no idea what you mean by "an arbitrary probability of 95%+"
(This would be opposed to a lot of religious beliefs, where you either have faith, or don't. You can't really prove God doesn't exist.)
I'm not sure where philosophy lies. Somewhere between the two, it seems.
In order to approach science, we must assume that nature is lawful, deterministic and understandable. Of course, we all know that there are certain 'laws' of nature that exist and have been proven beyond doubt, but the majority of science consists of findings that result from studies that rely on an arbitrary probability of 95%+.