Artificial Intelligence systems, which are being developed in labs all around the world, are becoming increasingly good at beating humans at their own games.
But earlier this year, researchers at Carnegie Mellon University and Facebook Inc. have developed an AI machine called Pluribus, which beat five top poker players with ease. Will such technological advancements lead to the extinction of online poker?
Pluribus And The Limits Of A Human Brain
Well, in recent history there are several examples of AI machines beating people at their own games. For example; an AI called Libratus, also developed at Carnegie Mellon University, managed to beat a poker star in a one-on-one game back in 2017. Along with this, AI machines have beat humans at games such as Dota-2, checkers, chess, Jeopardy and Go.
But Pluribus has been considered as a huge leap forward in AI machines and their capabilities, and this is mostly due to the fact that it was able to beat not one, but five professional poker players at Texas Hold’em. Apart from it being able to beat multiple players at once, what sets Pluribus apart from previous AI victories, is how it made to claim victory over the pro players.
Pluribus is so advanced that it was actually able to master an integral human skill known as deception. The incredibly powerful computer was able to perform trillions of calculations in order to perfect its bluffing skills. Researchers from the University say that its winning strategy comes as a result of playing trillions of hands against five other clones of itself.
After each one of these rounds it played, it analysed the decisions made and if it resulted in winning the hand, it was more likely to choose such strategies in the future. In addition to this, Pluribus was able to learn to bluff as it noted that making a bet when having a weak hand can result in a win.
Learning these factors, the AI was able to implement a variety of different strategies while playing and also making real-time decisions and adjustments, as it played against some of the best players. The players who confronted Pluribus said that it had felt hopeless, like there was no way to win.
Looking at some statistics from the results of the games played, we can see how much better this powerful AI was able to perform. During the game between the professional player and Pluribus, there was a deviation of 1 big blind per 100 hands and an average of 0 big blinds lost/won per 100 hands. The results while playing against bots ranged from -0.5 to -4 big blinds per 100 hands. The humans, on the other hand, lost by 2.3 big blinds per 100 hands. This result shows that the humans lost by a significant amount.
Being so good at dealing with the uncertainty and deception (which surely are two key elements of poker), is what makes this AI machine so powerful and nearly impossible to beat. But, what’s even more impressive, is that Pluribus uses less than 128 gigabytes of memory and runs on two chips only when playing.
This is a drastic improvement when compared to previous AI machines. For example; during its one-on-one poker match, Libratus used 100 chips to run and IBM’s Deep Blue was made up of 480 custom chips while playing chess against Garry Kasparov.