This Superhuman Poker AI Was Trained in 20 Hours!

❤️ Check out Weights & Biases here and sign up for a free demo:

📝 The paper "Superhuman AI for multiplayer poker" is available here:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
313V, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Brian Gilman, Bruno Brito, Bryan Learn, Christian Ahlin, Christoph Jadanowski, Claudio Fernandes, Daniel Hasegan, Dennis Abts, Eric Haddad, Eric Martel, Evan Breznyik, Geronimo Moralez, James Watt, Javier Bustamante, John De Witt, Kaiesh Vohra, Kasia Hayden, Kjartan Olason, Levente Szabo, Lorin Atzberger, Lukas Biewald, Marcin Dukaczewski, Marten Rauschenberg, Maurits van Mastrigt, Michael Albrecht, Michael Jensen, Nader Shakerin, Owen Campbell-Moore, Owen Skarpness, Raul Araújo da Silva, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Zach Boldyga.

Károly Zsolnai-Fehér's links:

#Poker #PokerAI

[HAL--]

First get rich in online poker, then publish the paper

[collector]

The AI clearly has the best poker face

[ChristianIce]

Ok.
Is it illegal to use an AI on online poker casino?
I ask for a friend.

[MobyMotion]

Always great to see papers with accompanying blog posts that make them accessible, while still explaining the broad ideas. Highly recommend anyone reads it if they haven't already.

[andreaaristokrates]

How to get rich (funding for science): Step 1. Set up online poker. Step 2. Sell AI and let it battle.

[thatguywhocooks]

So there's a lot of misinformation floating around about this A.I.

#1 the sample size was only 10, 000 hands. Which to some of you may seem like a lot but to actual professional poker players they consider a sample size of at least 100, 000 hands to see if you're a true winning player and to negate most variance.

#2 the bot had an extreme difficulty making computations with varying stack sizes and its performance worsened the more the stack sizes varied.

So, while impressive I would not consider it better than humans until I've seen more data

[Virsconte]

Okay, I'm not a machine learning expert, but I don't understand where the "secret sauce" is in this paper. What, if anything, was stopping people from using this approach a long time ago? The paper mentions that the training algorithm isn't very different from ones used by previous poker AIs, and it didn't take a huge amount of money or specialized hardware (about $144 on commercial cloud services!) to train, or even to run (two high-end CPUs worth about $4, 000). So why did nobody reach nearly this level of performance before?


I feel like I must be missing something important here. Usually, even if it's not clear *why* it works, I can at least get a sense of *what* was different. Here I just have no idea, but clearly they did something significantly better than anyone before.

[C-H]

Thanks both of u ... always love these segments

[DJBremen]

Lol you just know they've been "testing" this in real rooms online for cash

[d_b_]

Could you talk more about the variance reduction technique? Someone imported all the hands and found that the bot lost.

[ohokcool]

In the advert, I like how you said WAN-D-B like it's a database, instead of W-AND-B for weights and biases lol

[osujziC]

In "Example Hand 2", I don't think that huge check-raise on the River (4.7k raise in a 2k pot) with that hand was a good play. P6's call in response was also questionable, his hand was only beating bluffs at that point.

[DaverendeDodo]

Lol that 'trap' with the QJ was a disaster, with that river raise you only get called by better hands. On top of that, humans trap their good hands all the time lol

[RubenKelevra]

Online poker: exists
AI: I'll end this man whole career

[HansLemurson]

"This program was brought to you by Weights, and Biases" That reminds me of Sesame Street where they'd list numbers and letters of the alphabet as sponsors for the episode.

[claxviith]

"What a time to be alive!" in the sense your uncle will have to quit online poker and will stop taking loans from Mom.

[xrbqeos]

Okay pluribus is pretty good although he played in a 6max format against players who weren't experts in 6max cash but in mtt's (tournament format) Which is pretty important.
Also I heard that after the end of every hand the stack sizes were reset to 100 big blinds. That means the bot doesn't know how to play short stacked or deep stacked poker only the standard 100bb. Also some of the hands played by pluribus were analysed in a gto (game theory optimal) solver and it basically showed that even the AI wasn't playing close to a perfect strategy.
I'd like to see a rematch but this time against the top 6 max players and without resetting the stacks. Only then you could say that AI can beat humans.

[dmitriys]

The players have fixed stack size in this paper. If you consider real poker environment, player's stack can varies from 1bb to 300bb. Sure the most popular stack size is 100bb, but in real life there are can be different stack sizes, which can change strategy because you play with different potential pod odds

[TheBroadwood]

It was a genius play if you think about it. The back-check on the turn of P6 was actually really bad because the AI had to know that KQ or AQ is going to bet 3 streets. So either the AI gets a massive amount of value for having the higher kicker or it wins the pot anyway, because no weaker hand than QT is going to call the reraise on the river. I think its quite profitable in the long run.

If P6 would have c-betted the turn. The river would have been a check-check and P6 wouldnt have lost so much chips.

[PeterOtt]

Really liked your animation work on the poker table decision tree, well done Károly!