Me: "ah ChatGPT misunderstood when I said 'United' -- I meant Sheffield United not Manchester United".

Some pedant: "you shouldn't say 'understand': an AI is not even a robot. It is just a large language model, a set of matrices, it just predicts the next word, it can't really understand anything."

What would Wittgenstein say about this? We can use his concepts of language games and also family resemblances.

The pedant got what I meant. Would have done so whether I put 'misunderstood' in "scare quotes" or not. Do they fail to get what I mean because this instance of ChatGPT cannot fall in love, does not have kidneys and is not scared of death? No.

We can consider the multiple and different but overlapping language games in which we say that a human (or perhaps even a dog) understands or fails to understand something, and then compare them with the new language games in which we (unavoidably, but often in scare quotes) say that an AI understands or fails to understand something. We'd find that these games have family resemblances to one another, enough to explain why we use the word "understand" in all of them. And we can be relaxed about the fact things which are important for its use in one of these games (having a brain! having free will! having kidneys!) are absent from its use in another.

Nothing to see here, move along please.