If a random person asked me what AI was (and if he looked genuinely interested), I’d tell him this:
I don’t know what AI is.
There is no fixed definition for AI. The definition changes with every new discovery made and with every new AI programmed. The point of complexity beyond which a program can be called an AI changes ever so often (it increases more often than it decreases). When the idea of AI first came out, AI was idealised as a computer that could act like a human. A computer that could think for itself and make decisions. However, they didn’t succeed in doing that and lowered their expectations of what an AI should be.
Another thing that happens is that as we accomplish a certain task at a higher level, we increase the AI threshold. We say that this is not AI and AI is better than this.
AI can’t be classified. It can’t be given a fixed definition. AI is always the next level. Of course, a lot of AIs do exist now (video games) but they aren’t actually looked at as the AIs that we dream of and lose the “magic” that we expect in an AI.