This is starting to become a fallacy rather than a cute tongue in cheek saying by AI researchers.
Once a problem that is used to push AI research is solved better than humans then it is usually no longer cutting edge AI research. Chess for example. Minimax search with tweaks for board eval. Done. Handwriting recognition - better than humans by a committee of NNs, but still considered an AI research problem. It is a benchmark for algorithms. That doesn't mean just because an algorithm can do it that we have general artificial intelligence.
A general AI will have to identify digits and play a decent game of chess and have a conversation.
That doesn't mean the problem domains that drive AI research are not AI after they are solved individually.
A lot of researchers think "this is a great problem, surely it will require human level general intelligence". When an algorithm is identified that solves the problem without needing general intelligence, they look for the next problem domain... or keep on improving in that domain (image recognition, but not necessarily of dog breeds, that's solved).
This intermediate focus gives direction to AI research while general intelligence gives purpose.
None of those researchers are saying solving X means solving AI. None are saying each of these domains is sufficient for AI, rather they might be necessary and are a good place to start to advance the field.
About the only behavior or skill, X, that researchers think may be sufficient for AI is an entity that can communicate well enough to pass an unbiased Turing test.
Artificial general intelligence is a long way off, no matter how many cats can be identified from a pile of photos.
Maybe it is a definitions issue. AI is whatever AI researchers are working on at the cutting edge to advance the field. AGI (general intelligence) is much broader and no algorithm has come close to demonstrating AGI.
Is anybody anywhere working in "human level intelligence"?
People like to talk about it, but I doubt any serious researcher claims to be solving it. It's a non-specified problem, I can't even imagine what a paper on it would look like. How would you know you achieved it?
In practice AI is those concrete problems. And the fact that they often stop being AI after solved is a real PR problem for the researchers.
