My understanding is that the original use of AI was to describe algorithms that solved complicated problems, but only sometimes, and usually through heuristics. As opposed to provably correct algorithms like Dijkstra's.

I think that's how we still use it.