Consider a computer program that had access to memory of large numbers of chess board positions from games that people had previously played, along with the next moves that had been made from those positions. (Those data could be compiled from written records of completed chess games.)

Given a board position, if the program found that board position in its data base, it could then select a next move that other players had made. Otherwise it could make a random move.

In what sense would the program "know" how to play chess? (And would it play well?) Certainly not in the way people know how to play chess. That's like the way LLMs "understand" English.