Maybe I'm reading way too much into it, but when I read things like "AI will achieve human-level intelligence" I assume they mean that the AI will be fairly general purpose. Most of the applications I've seen, while impressive, have been pretty narrow.

Redefining problems is good when flaws in the original definitions are found.

AI has perhaps, maybe passed human intelligence in 0.005% of applications.

It's 2020 and we still have steering wheels in cars.

If we replaced the word "AI" with "statistics" we'd be a lot closer to describing where we are.