Disclaimer: Freelancer.com does not promote gambling in any way.
The concept of artificial intelligence has been around for decades now, yet the development of technology today is making it possible for more advancements to be made within shorter periods of time. A great way to train and test AI is to have it play poker, due to the fact that dealing with poker, means having to deal with tons of imperfect information, thus making the game complicated even for advanced AI systems, similar to what real life has to offer.
In a recent infographic published by the folks behind PokerSites, we can learn more about the development of AI machines capable of playing poker, and find out who is more capable in this game: champion players, or AI machines.
Before anything else, it is important to talk a bit about the demographics surrounding artificial intelligence. Studies have shown that at this time, 35% of people over 55 trust AI, 62% of people aged 17-24 trust AI, and 71% of people over 50 believe that at some point in time, artificial intelligence assistants will be capable of simplifying their lives.
The problems that AI usually face both in real life and in poker
Some of the general real life problems that AI interfaces often face include dealing with imperfect knowledge, risk management, agent modelling, deception and unreliable information, all of which are problems that poker offers as well, in the forms of: not seeing the opponents’ hands, having to create betting strategies and considering their consequences, identifying patterns of players and finding ways to exploit them, learning to bluff efficiently, and handling the deceptive plays that opponents might make.
A quick history of Poker AI
The first basic, AI software for Poker, named Orac, was created back in 1984, and tasked to compete in the WSOP. Later on, in 1997, UoA released a more advanced system titles Loki, which was focused in beating Limit Hold’em variations. 2006 was the year when the Annual Computer Poker Competition first started, followed by the development of multiple great artificial intelligence systems focused on Poker, such as Polaris, Sartres, Cepheus, Slumbot, Act1, DeepStack and more.
Case Studies where AI beat pro poker players.
· When the Polaris system was first tested in a real-life poker scenario, against a Poker champion, before playing 200 hands in 2007, Poker professional Phil Laak was already down $1,500.
· In the final day of the Humans vs. Libratus Poker tournament, pro player Daniel McAuley was down $4,000 after a couple of seconds, and shortly later, down over $8,000, thus showing the clear win of the AI machine.
· Studies have shown that after 30,000 hands, the probability of beating the Cepheus poker bot will go down to around 5%.
· Not only this, but after 30,000 hands against Cepheus, 1 in every 20 people would finally believe that they are better players when compared to the AI machine.
Case Studies where pro poker players beat AI systems
· There are of course cases, when the human intelligence alongside with the ability to bluff and change patterns can be good enough to beat an AI system. Such is the case of poker professional Ali Eslami, who was over $800 by the mid-point of his session against the Polaris AI, and ended up finishing around $395. While it may not be much, it is still a clear win.
· By using the blind-squirrel model, Michael Bowling managed to beat the Cepheus poker AI, and win both of the 100-hand matches that were played.
· The complexity of an AI system can also have an important influence on the odds of the game, considering the fact that the possibility of being ahead of the Cepheus AI for instance, after a total of 100 hands is of 46%.
Regardless of these aspects, there are still many benefits that Poker AI has over humans and vice-versa. Some of the main benefits of Poker AI over humans include: an AI can easily identify weaknesses that player have, AIs do not feel the value of money, AIs, do not get tired and make poor decisions because of increasing fatigue, AIs are not afraid of risks, and AIs have no emotions, therefore aren’t even affected by tilt, thus AIs see the poker game in numbers, algorithms and learning patterns, rather than feel its emotion.
On the other side of the spectrum, due to human nature, and due to the status of Poker as a game where bluffing can get you a clear win, pro poker players have had enough time to create a variety of strategies, many of which are based on deception, bluffing and reading emotion.
When it comes down to comparing online poker bots against AI machines, people will quickly find out that online bots are actually far behind AI machines. In fact, no poker bot available on the market is good enough to beat the game of $1,000, and their turnover is of around 5bb for each 100 hands. Regardless of this aspect, the use of poker bots is an offence on all poker sites. If that wasn’t enough, due to their lack of artificial intelligence, most can be easily identified due to repeated use of the identical bet sizing, identical timing, repeating common lines, not responding in the chat box, and more.
Based on everything that has been outlined so far, clearly determining whether AI poker machines are better when compared to humans when playing poker, is not possible yet, due to the lack of extensive research, and human nature, which in this case, is beneficial to the player, and not the AI machine.