Brendan wrote:
Hi,
For chess, "strategy" is mostly basic game theory. Start by analysing the board and giving it a rating (e.g. maybe based on how defended/exposed your pieces are, how defended/exposed your opponents pieces are, how many of your opponents pieces can be attacked, how many of your pieces opponents can attacked, and possibly with an underlying "this type of piece is worth this many points" system). Then for each move you can make, see if it improves the board rating, recursively (e.g. for each move you can make, for each response opponent can make, for each move you'd be able to make after that, ..). Finally make the move that improves the board rating the most.
Of course when computers do it it's typically "brute force search with optimisations";
I think the new "generation" of AI chess algorithms are a little different. Rather than pre-defining the "weight" of the pieces and the squares, I think they look at the board more like a bitmap image, and break it down into patterns, kind of like a facial-recognition, or object-recognition algorithm.
Here's a high level overview:
https://www.sciencealert.com/it-took-4- ... -alphazeroBut essentially the bottom line is that the new AI designs do not assign any "values" at all. They let the AI come up with it's own solutions. And the game that it's playing doesn't really matter. It would learn chess and checkers just as well, with no code changes. (Other than maybe the rules of the game.)