ChatGPT, for an example, and unless it has been enhanced since last time I've heard of it, will tell you it can play chess, not even knowing it's a lie. Because ChatGPT has no chess engine to use, and is, therefore, unable to play chess properly.
ChatGPT will go pick moves and comments to the moves, in pre-existing texts from digitalized chess books and chess magazines, which will produce soon tons of illegal moves and inadequate comments. It's "doing conversation", not playing chess.
It doesn't even know what chess is, it doesn't even know it's currently "playing a game". It doesn't know what means "playing", it doesn't know what a game is.
Would you combine ChatGPT with a chess engine, ChatGPT would have a say about when to start the chess engine, and if it starts the chess engine. But even if talking to you during the game, it would still not take part in the chess calculations, and would have not a clue about what is going on in the game.
Etc. Etc. Etc.
AI is a thing, chess engine is an other, "chess AI" doesn't exist.
And if you want to quote the case of Alphazero, it's yes a self learning/teaching chess engine. Still just an engine. Can't give news from family and friends, just play chess. Of course, one could combine it with an AI, but that would still be two softwares working together.