Feynman had it right: http://www.youtube.com/watch?v=lL4wg6ZAFIM
But it isn't just "studying man-made creations". It's about studying computation in general. Not just man-made machines that compute.
Mind you, information theory, which is often considered computer science, certainly applies to nature itself. Of course, it would be easy to claim that information theory isn't actually computer science. Personally, I consider most of computer science more math than science. In fact, I sometimes tell people I do math! (This is a great way to avoid further questions at borders or awkward parties.)
Saying that complexity theory studies "man-made things" is like saying that chemistry studies man-made molecules. It's technically true, but it confuses the methodology with the object of study: studying particular chemicals vs the underlying laws governing them or particular models vs the underlying laws of computation.
Paraphrasing from Sipser[1], the class of decidable and undecideable languages is the same for any "reasonable" model of computation where "reasonable" involves some basic things like not being able to do an infinite amount of work in a single step. So while any particular model (e.g. a Turing machine) is arbitrary, the class of languages is actually natural.
[1]: http://www-math.mit.edu/~sipser/book.html
This makes sense--any model of computation containing a countable number of instances will not be able to decide or even recognize every language. It also makes sense that these models would be able to recognize the same set of languages.
I think this also holds for complexity classes. That is, the complexity of a language in a class like P or NP is the same across different models as long as those models are all deterministic or non-deterministic.
So really, complexity theory is not about a man-made thing: it's about a deeper, fundamental truth. It's really awesome.