Artificial intelligence is improving day by day, thanks to its scalable learning methodologies and its fast rate of adaptation. Facebook has now developed an AI that is capable of defeating professional players in Hold & # 39; em Poker without limit for six players.
The AI is called Pluribus and has broken records in the world of AI bots that defeat the experts in the field of games. This is not the first time we have witnessed an AI that surpasses human players in games, but beating humans in games involving more than two players is really an achievement.
Facebook claims that they have made Pluribus play against the best poker players in the world, which even include two winners of the Main Event of the World Series of Poker. The tests were conducted in two formats: the first format was to stack the AI against five human players, while the second was five AI bots against a human player. In both cases, Pluribus has won with a decisive margin.
" " If each chip was worth a dollar, Pluribus would have won an average of approximately $ 5 per hand and would have won approximately $ 1,000 / hour by playing against five human players.These results are considered a margin of victory decisive by poker professionals " read his blog post.
To achieve this level of precision, the bot uses a self-game technique to teach several strategies to win without previous samples. It is worth noting that all these processes are done using fewer resources than other bots that have won against humans. Take a look at the video to see Pluribus in action.
"The most stimulating thing about playing against Pluribus was responding to their complex preflop strategies, unlike humans, Pluribus used multiple preflop upload sizes, trying to respond to non-linear open ranks was a fun challenge that differs from games humans ". said Seth Davies, a professional poker player.
The hallucinating fact about Pluribus is that the company managed to train the plan strategy in just eight days on a 64-core server with less than 512 GB of RAM. Also, GPUs were not used to accelerate the process. The cost of these computational resources at the typical rates of cloud computing translates into approximately less than $ 150 to train the entire AI, which is crazy. This clearly shows the possibilities of implementing artificial intelligence to perform a multitude of tasks.
However, Facebook's intentions to create this AI are not just to be a champion in Poker. Instead, it is an experiment to expand the limits of AI and to see to what extent technology can evolve to solve more complex problems and meet difficult use cases.
So, what do you think about Pluribus? Let us know in the comments.