You can’t get an edge over someone playing game-theoretic optimal bluffing strategy in poker.
These types of “solved games” make some pretty hefty assumptions, such as limiting possible actions. But that can only really happen with online poker, when you do it live, you introduce a ton of variance that a good player can exploit.
Algorithms may have “solved” a game, but that doesn’t mean a human has. It’s the same idea with a game like chess, where we’ve developed essentially “perfect” computers that can compute every possible board state from a given point onward and give you an optimal move, which will give you the best possible outcome. Does that make chess uninteresting? No. At the highest levels, it’s less a strategy or tactical game and more psychological. The idea is to surprise your opponent and play something they aren’t prepared for which gets them into time trouble figuring out your plans, and the clock becomes a piece you can use against them. So the prep for a game is studying their past games and guessing what they might be preparing against you, and preparing something they won’t expect to use against them.
When dealing with humans, there will always be weaknesses to exploit, and that’s interesting. So the game of live poker remains interesting.
It’s the same idea with a game like chess, where we’ve developed essentially “perfect” computers that can compute every possible board state from a given point onward and give you an optimal move
Chess isn’t solved: chess computers have outplayed the best current human players, but they can’t always provide an optimal move, can’t look down branches far enough. Although they do use Minimax!
But it is similar to the extent that you can get not-perfectly-optimal play that will probably do better than a human.
When dealing with humans, there will always be weaknesses to exploit
After the first 10 moves or so, they can. There’s something like 9 billion possible chess positions after that point, and opening theory is well established, so it’s largely solved. Computers can calculate something like 100 moves deep (and nearly all branches), though they do use heuristic to eliminate unlikely branches.
There are some interesting games between top bots because of that heuristic, but any of the top bots will consistently beat a human because they can compute orders of magnitude more possible game states.
So it’s essentially solved, meaning that, in practice, a top AI will pretty much always beat or draw a top player. The difference in rating between a top bot and the top human player is something like the difference between a GM and someone aiming for IM, and we expect a similar performance difference.
These types of “solved games” make some pretty hefty assumptions, such as limiting possible actions. But that can only really happen with online poker, when you do it live, you introduce a ton of variance that a good player can exploit.
Algorithms may have “solved” a game, but that doesn’t mean a human has. It’s the same idea with a game like chess, where we’ve developed essentially “perfect” computers that can compute every possible board state from a given point onward and give you an optimal move, which will give you the best possible outcome. Does that make chess uninteresting? No. At the highest levels, it’s less a strategy or tactical game and more psychological. The idea is to surprise your opponent and play something they aren’t prepared for which gets them into time trouble figuring out your plans, and the clock becomes a piece you can use against them. So the prep for a game is studying their past games and guessing what they might be preparing against you, and preparing something they won’t expect to use against them.
When dealing with humans, there will always be weaknesses to exploit, and that’s interesting. So the game of live poker remains interesting.
Chess isn’t solved: chess computers have outplayed the best current human players, but they can’t always provide an optimal move, can’t look down branches far enough. Although they do use Minimax!
But it is similar to the extent that you can get not-perfectly-optimal play that will probably do better than a human.
That’s probably true.
After the first 10 moves or so, they can. There’s something like 9 billion possible chess positions after that point, and opening theory is well established, so it’s largely solved. Computers can calculate something like 100 moves deep (and nearly all branches), though they do use heuristic to eliminate unlikely branches.
There are some interesting games between top bots because of that heuristic, but any of the top bots will consistently beat a human because they can compute orders of magnitude more possible game states.
So it’s essentially solved, meaning that, in practice, a top AI will pretty much always beat or draw a top player. The difference in rating between a top bot and the top human player is something like the difference between a GM and someone aiming for IM, and we expect a similar performance difference.