I think if you are looking to write self-modifying (adaptive) code, I don't think doing anything on the Bitcoin transaction/mining level is interesting problem for this type of programming. Bitcoin transactions are bound by a known set of rules, and the core problem of generating blocks in mining revolves around using brute force to get a certain hash output.

I think the most interesting place for adaptive algorithms is in a trading application. Having your program attempt to 'learn' any patterns that may exist in Bitcoin trading is an interesting problem - and one that is also very applicable to the stock market as well not just Bitcoin.