I know they do in chess and go.
Is there a perfect information game when methods like alphazero that worked in chess and go fail in it?
I do not mean also to games when humans are unbeatable because there is an easy solution for them.
Do engines beat humans in all stategic perfect information games?
Moderator: Ras
-
Uri Blass
- Posts: 11165
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
-
towforce
- Posts: 12880
- Joined: Thu Mar 09, 2006 12:57 am
- Location: Birmingham UK
- Full name: Graham Laight
Re: Do engines beat humans in all stategic perfect information games?
Well they cannot beat us at football!
I am sure you're talking about intellectual games, though. In chess and go, strong humans used to win through superior pattern recognition. Now it looks as though computers are able to beat us in the pattern recognition game. Right now, it takes a lot of cost and resource for a computer to be able to match good humans at pattern recognition. In some areas, they seem to struggle even when they have that (I am astonished how long it's taking computers to learn how to drive given the resource that has been thrown at that problem), but that will become less true over time: computers are gaining in intelligence a lot more quickly than the domesticated apes right now.
I am sure you're talking about intellectual games, though. In chess and go, strong humans used to win through superior pattern recognition. Now it looks as though computers are able to beat us in the pattern recognition game. Right now, it takes a lot of cost and resource for a computer to be able to match good humans at pattern recognition. In some areas, they seem to struggle even when they have that (I am astonished how long it's taking computers to learn how to drive given the resource that has been thrown at that problem), but that will become less true over time: computers are gaining in intelligence a lot more quickly than the domesticated apes right now.
Human chess is partly about tactics and strategy, but mostly about memory
-
chesskobra
- Posts: 360
- Joined: Thu Jul 21, 2022 12:30 am
- Full name: Chesskobra
Re: Do engines beat humans in all stategic perfect information games?
You could create a game that is so large and complex that however you train your NNs or create hand crafted evaluations, the engine would still be making pretty much random moves. In such a case, an engine would be not much better IN PRACTICE than a human making random moves. But IN THEORY, methods like MCTS and their variations are quite general, so they are applicable to all games of complete information, with finite game trees.
-
Uri Blass
- Posts: 11165
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: Do engines beat humans in all stategic perfect information games?
I talk practically because in theory you do not need NN for beating humans in games and you can solve all games by alphabeta search if you are fast enough.chesskobra wrote: ↑Fri Feb 16, 2024 11:51 pm You could create a game that is so large and complex that however you train your NNs or create hand crafted evaluations, the engine would still be making pretty much random moves. In such a case, an engine would be not much better IN PRACTICE than a human making random moves. But IN THEORY, methods like MCTS and their variations are quite general, so they are applicable to all games of complete information, with finite game trees.