I've noticed that AI can't really answer hypothetical questions. Since it trains on existing data it can't imagine things it has not encountered during training. But there are all kinds of jobs that involve solving problems that have never been solved. Imagine being a physicist and wondering what would happen if, at the moment of the big bang, physical constants like the speed of light were not constant. What would be the specific implications of that wrt to particle physics? Since there is no data on such a thing, and perhaps no one has ever entertained that thought, AI will be of no use. It's relatively easy to think of similar situations in any field where solutions require imagining things that have never been imagined.