There is no difference between AI and non-AI save for the model the observer is using to view a particular bit of computation.

Eh, I'd be fairly comfortable delineating between AI and other CS subfields based on the idea of higher-order algorithms. For most things, you have a problem with fixed set of fixed parameters, and you need a solution in the form of fixed solution. (e.g., 1+1=2) In software, we mostly deal with one step up from that: we solve general case problems, for a fixed set of variable parameters, and we produce algorithms that take the parameters as input and produce the desired solution (e.g., f(x,y) = x + y). The field of AI largely concerns itself with algorithms that produce models to solve entire classes of problem, that take the specific problem description itself as input (e.g., SAT solvers, artificial neural networks, etc where g("x+y") => f(x,y) = x + y ). This isn't a perfect definition of the field (it ends up catching some things like parser generators and compilers that aren't typically considered "AI"), but it does pretty fairly, IMO, represent a distinct field in CS.