It is possible to measure progress for solving chess.Dann Corbit wrote: ↑Tue Feb 12, 2019 7:59 pmIt is also possible that the savage pruning will hide solutions.
Indeed, we see this happen from time to time when SF goes blind (as do all aggressive pruning engines at times) and plays for the draw instead of the win.
On the other hand, the exponential march forward of both hardware and software will eventually make a sort of pragmatic solution for a given position become realistic.
Currently, top hardware can reach 70 plies.
When I started fooling around with chess engines, it took about 12 minutes to get to 15 plies and 18 plies was a 24 hour effort.
In particular, if graphics cards continue to march forward as they have been, then a server with a boatload of GPUs in it might be able to simply solve chess rigorously by and/or proof search.
Not today, obviously. Maybe 30 years from now?
Perhaps a cooperative effort with a billion GPUs scattered about the internet.
Take a big pgn and find what is the percentage of positions that engines can solve(solve meaning or saying that one side win or saying that it is a forced draw).
Note that we need a special engine to solve draws because today it is not clear if 0.00 is a forced draw or not a forced draw(I am also not sure if a mate score is a win but I do not know about positions when there is a wrong score and it may be interesting to take all the tablebases positions to search without tablebases to see if there is a single case of wrong mate score because of hash collision).
I would like some tool that get a position(that include also information about previous moves of the game) and tell me if it can find by search that it is a forced draw(for example if both sides can force a draw by repetition or by the 50 move rule).