For me the key point is that there’s a world of difference between spotting correlations between items in large datasets (something AI is very good at) and coming up with new theories as to why certain phenomenon have been observed. You could have all the data in the world regarding electromagnetic phenomenon, for example, and still be unable to understand what the CMB tells us about the very beginnings of our universe. Likewise you could have plenty of data about the orbits of the planets around our sun but you’d be unable to explain why Mercury’s orbit doesn’t precisely correspond to Newton’s calculations.

Large datasets are lovely when you want AI to tell you that you should show people who just bought a new set of golf clubs an ad for the new Lexus on days when it’s raining in the morning (for example) but AI is never going to give us a unified quantum field theory or anything equivalent to the Dirac equation. People get all worked up over AI because, even with a PhD in data science, most overlook the many hard limitations and instead wander into sci-fi musings. Which can be fun, but wholly uninstructive.