Solver-resistant poker variants, is that something worth pursuing?
Having a poker variant you that is very computationally expensive to solve would make RTA cheating a lot harder online but so far no such poker variant has taken off. People tend to prefer games they are familiar with and generally do not want to learn any new game.
Example of a relatively simple variant that is much harder to solve is when you get up to 7 private cards for each player (such as 3 hidden cards and 4 cards that can be seen by others, 4 betting rounds, starting with 3 cards, no limit).
Another example is "best of 3 omaha" but that gets a bit messy for 3+ players (bigger problem IRL than online i think).
11 Replies
Probably just kicking the can down the road. By the time you get a game like that 'going', the computational power costs will be low enough to solve that as well.
Look on the bright side, you get to go to a casino, and actually see people to play the game we love. Much of the world is slipping behind a keyboard, avoiding all human interaction. Poker is forcing us away from that.
Probably just kicking the can down the road. By the time you get a game like that 'going', the computational power costs will be low enough to solve that as well.
I don't actually think computers becoming more powerful is the biggest problem.
The bigger issue is AI advancing to the point where it can beat all humans on the planet after a week of training. That doesn't require actually solving the game via brute-force methods.
Look on the bright side, you get to go to a casino, and actually see people to play the game we love. Much of the world is slipping behind a keyboard, avoiding all human interaction. Poker is forcing us away from that.
That works if you have a casino nearby to go to in the first place.
I am going to be stuck with only playing online poker (or IRL with play-money) for years to come. Well i should still be able to beat NL4 for a while.
I recently calculated that no limit best of 3 omaha would be 5 931 374 852 619 220 000 times harder to solve than NLHE (in large due to having 3 boards instead of just one).
preflop: 1 card per board, 4 holecards.
flop: 3 cards per board, 4 holecards.
turn: 4 cards per board, 4 holecards.
flop: 5 cards per board, 4 holecards.
There are 1755 distinct starting positions (same as the number of different flops in NLHE).
Not sure how fun the game actually would be to play though, never got around to testing it.
For a new game being unsolvable doesn't guarantee anything. The fact that there are GTO solvers for NLHE only shows how deep the game strategy is, so eventually it's good for the game of Texas Hold'em.
As long as you learn new things and conquer boredom when developing new variants, it is worth pursuing.
Not sure what you mean by "solvable".
GTO solvers don't actually solve games fully. Typically they start from the flop and only a limited number of bet-sizes are considered.
So what we are talking about for NLHE and PLO is "very weakly solvable".
The fact that there are GTO solvers for NLHE only shows how deep the game strategy is, so eventually it's good for the game of Texas Hold'em.
You don't need a GTO solver to understand how interesting of a game NLHE is.
It is definitely a problem for online poker that real time assistance provide such a massive advantage. Here one of the issue is the speed in which NLHE be solved. If you need to run it for a month on your computer and have 1TiB of ram it's less of an issue obviously.
Having solvers easily accessible does even the playing-field a bit for better or worse. I find myself avoiding PLO in part because i don't have access to any solver for it and i don't want to play at a big disadvantage against regs.
As long as you learn new things and conquer boredom when developing new variants, it is worth pursuing.
It can definitely be fun to invent new card-games and then try them out. I do have people to play them with which is nice.
I very rarely play NLHE with IRL friends since that wouldn't be a fair competition.
We usually play no limit poker8 (variant A) which i estimate is around 23 million times harder to solve heads up than NLHE with the same stack sizes (the game does play better with more players and deeper stacks though).
The notion that computer hardware is going to magically improve to solve all poker variant is very much false.
Let's take no limit poker8 (variant A) with 8 players.
Each player starts with 5 holecards
flop: 3 community cards added
turn: 2 community cards added
river: 1 community card added.
Number of card combinations (including equivalent hands) = FACT(52)/FACT(3)/FACT(2)/FACT(5)^8/FACT(6)
217 112 426 596 910 000 000 000 000 000 000 000 000 000 000 000
Then you have to multiply that with the size of the game-tree from the betting which is enormous even if you use a very simple tree with only one allowed bet-size (way more than a billion if everyone has like 256 times the initial pot and only 50% pot-size is used.
1000 billion computers checking 1000 billion combinations per second would need 6 880 024 652 437 833 years to check all the combinations.
It's not always practical to use a brute-force approach to solve a game. In the case of go very strong artificial neural networks eventually solved it properly.
I can see you're deep into numbers.
Can you tell me if 6 card stud with classic 5 card hand rankings is mathematically correct ?
We know 5 and 7 card stud are 100% correct. What about 6 ?
I couldn't find this anywhere.
I can see you're deep into numbers.
Can you tell me if 6 card stud with classic 5 card hand rankings is mathematically correct ?
That doesn't fit with the topic since a 6 card variant would be easier to solve than a 7 card variant.
You should probably learn basic combinatorics yourself if you are interested in trying to invest some poker variant.
I used libreoffice calc when figuring out a scoring system for 8 card poker ( to get a nice point distribution). I also used a random card generator to generate 621 hands and compared that against the estimated frequencies.
I might look into a poker variant where you select 6 cards out of 9 for a hand eventually but it's not really a priority.
I think the problem is that you have to come up with a game that is not just harder for computers but scales up harder for computers than it does for humans. Otherwise, I think the computer will gain an advantage as the complexity and combinations of the game increase. The game would be less solvable by the computer but the computer will be relatively stronger vs the human because combinations is a strength of the machine.
Personally, I don't see how this is done with a deck of cards and exact rule set.
I think you need incomplete rules and non-deterministic game structure but I have no idea what this would be or look like.
I think the problem is that you have to come up with a game that is not just harder for computers but scales up harder for computers than it does for humans. Otherwise, I think the computer will gain an advantage as the complexity and combinations of the game increase. The game would be less solvable by the computer but the computer will be relatively stronger vs the human beca
Computers were never able to beat the top humans at GO by relying on brute-force computations as far as i know., neural networks were very much needed in that case.
I don't think having a game with more cards would make it significantly harder for humans can be very good at learning to adapt to new scenarios.
The bigger issue is ai advancements resulting in very few if any humans being able to play better than an energy efficient ai (was probably the case even with pluribus).
I think you need incomplete rules and non-deterministic game structure but I have no idea what this would be or look like.
All games need to have a complete rule-set or it would no longer be a proper game. You don't want a situation where 3 judges decide which hand they like better.
Poker is already non-deterministic in the sense that which card that comes up is random, the more possibilities for different cards coming up the harder the game will be for computers to solve via brute-force methods.
Current mixed games are not completely solver proof but will give you 10 years or more easily.
