I know what you mean. Often it seems to me like the worst problem in developing in real AI is that the knowledge required is too scattered over a dozen disciplines whcih aren't really talking to each other - often because in order to really "talk to the other", you'd need to spend a lot of time actually studying that field.

And when somebody does spend all that effort and manage to get an interdisciplinary paper published, it goes ignored because as a result of its interdisciplinary nature, others don't understand it. Gah.