I kind of agree with 1, but not really with 2 and 3. It's easy to come up with trivial examples where it is both unreasonable and not feasible to follow those two, both for AI and non-AI scenarios.