How did Poker Robot Deepstack Defeat the Pros?

December 12, 2019 0 comments

Elon Musk famously said that toying around with artificial intelligence was “like summoning the demon”, but technology developers are continuing to progress with it regardless. A future in which robots are ubiquitous and do everyday jobs used to be confined to the imagination of sci-fi writers, but now it seems as though it is an inevitability. The scary thing is, a lot of the robots being created will be far superior to the human race in many ways. Deepstack, the poker-playing robot which defeated professional poker players, is a prime example of this.


Deepstack has Beaten a Number of Professional Players

Deepstack is a poker-playing robot which has been programmed to bridge the gap between games of perfect information and those of imperfect information. For instance, chess is a game of perfect information because the algorithm could study every single possible outcome of the millions of scenarios and work out the best way to proceed in any given situation through trial and error. Poker is a game that has imperfect information because it is impossible to anticipate how any number of players would act in a particular situation.

Michael Bowling describes the research program behind Deepstack

Bots can be programmed to be able to calculate odds and work out when their best chances of winning are, but they may not be prepared to deal with bluffs and bets which don’t conform to their predetermined understanding of the game. Deepstack, therefore, uses intuition which it has acquired through deep learning and reassessing its strategy with every decision.

In a 2016 study, Deepstack became the first ever AI to beat professional poker players in heads-up no-limit Texas Hold’em poker. In 44,000 hands, the machine managed to defeat eleven professional poker players. It won 49 big blinds out of 100, and only one of the wins was outside the margin of statistical significance. One of the players that Deepstack came up against was the Irish-born American poker expert Phil Laak, who has been a regular on Poker After Dark and High Stakes Poker in the past. Laak has a World Poker Tour title and a World Series of Poker bracelet, but he was outplayed by the AI machine.

Deepstack has even shown the ability to bluff in certain situations.


What are the Rules of Poker and why is it Preferred by Bots?

When broken down into simple terms, it is clear how AI robots can learn to build an effective strategy in poker. Despite it being a game of imperfect information, there are a vast number of common occurrences that happen in any given poker game. By learning the basics and then refining the techniques through the experience of playing against other players, AI and humans alike are able to develop their skills and reach a professional level.


No matter whether you are playing Omaha Hi-Lo or the more popular Texas Hold’em poker games, the basic strategy can be broken down into eight simple steps. The most important thing, to begin with, is knowing the hand rankings, with a royal flush being the highest value hand and a pair being the lowest. If nobody has a pair, it would be settled on a high card. The hand is made up out of the best five cards from the five community cards and the player’s hole cards.

Position on the table is highly important as well, and players need to learn how to bet or raise depending on where they find themselves. Being dealer is the strongest post-flop position on the table because it allows the player to see how the other players act before they have to make a decision. After gaining an understanding of these things, players then need to develop their pre and post-flop game plan, along with post turn and post river play. In each round of betting, the situation changes. If players choose to bluff on the flop, they need to decide whether it wise to fire another barrel on the turn. This requires them to assess the risk involved against the potential returns.

After learning these basics, players can improve their game through experience and by using mathematical skills. They also need to try to turn emotions off and stick to their optimum strategy. The reason that AI favors poker is that once the basics are learned, it is easy to follow a similar strategy in each game. Because they don’t have emotions, they can avoid going on tilt and making irrational decisions in the subsequent rounds. By playing a lot of games, they can analyze their choices from past games and work out how they could do things more efficiently in future games.
What are the other Poker Bots and how do they compare with Deepstack?

AI has been around in poker for some time, and there are numerous basic machines that are used to play against humans in offline games. Some people have even tried to create their own machines to beat poker sites and win money. But most of these creations are extremely basic and can easily be spotted by those who know what to look for. These rudimentary bots typically act after the same amount of seconds each time, make similar to identical bets depending on the situation, and use relentless aggression in certain spots.

There are a few created by statistical experts, however, which could be compared to Deepstack. One example is Pluribus, an AI machine designed by Facebook’s AI lab and Carnegie Mellon University. It is the first bot to beat humans in a complex multiplayer competition. For this reason, it differs from Deepstack which only took on players in a heads-up scenario.

Deepstack did what professional poker players do but on a much larger scale. Experts in the game learn the basics and then refine their strategy through experience to get to the top level. But AI machines are able to run through millions of different scenarios to develop an optimum strategy – something which would take humans a number of lifetimes to achieve. If AI can get this efficient at a game like poker, what else will it be able to beat humans at in the future?

No Comments so far

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.

Your data will be safe!Your e-mail address will not be published. Also other data will not be shared with third person.