It's really very simple. Carlssen can also play chess at very high level if you change the rules, for example of how one piece moves, and can do so at his first game with the new rules.
He can also answer a problem about anything else unrelated to chess, e. g. about how to start a fire while camping or whatever. He can use analogy in a field new to him. He can generalise. He doesn't need millions of samples to infer something. He can recognize an object in a picture having seen that object just once, and in a different position.
AI can do none of these things. AI scientists understand this very clearly. There is no "question" of the intelligence of machines being debated, because the proposition is ludicrous and in AI everyone knows that AI is most definitely not intelligent by ANY non-ridiculous definition of intelligence.
Only the public entertains this fantasies.