Collaboration between specialist researchers from Facebook and Carnegie Mellon University has resulted in a breakthrough in the area of Artificial Intelligence. They managed to develop an AI bot for poker that has been tested to win against even some of the greatest masters of the game.
With bluffing, unpredictable bet-size maneuvers, and a convincing proven victory to top it off, the bot known as Pluribus, is practically history in the making.
News Highlights
- Carnegie Mellon University and Facebook AI research scientists have created an AI poker bot.
- Pluribus won against five professional poker players, with an average winnings rate of $5 per hand.
- Questions regarding the future of online poker have been raised in light of this new discovery.
A Successful Research
Noam Brown has been working on his Ph.D. at the Facebook Artificial Intelligence Research (FAIR) Group. He has been collaborating with his former colleague, Tuomas Sandholm. They have come up with a unique poker-playing AI bot that can take on multiplayer challenges and achieve winning statistical rates.
The bot Pluribus has been tested in a six-player, no-limit Texas Hold’em poker table game, based two different scenarios. On the one hand, there were five copies of the AI against two poker pros – the six-time WSOP winner Chris “Jesus” Ferguson, and Darren Elias, record-holder in most World Poker Tour titles. On the other, a single copy of the AI went up against a total of 13 professional poker players and beat them in 10,000 hands.
Reaching the decisive margin of victory in both scenarios, co-creator Brown claimed the following:
It’s safe to say we’re at a superhuman level and that’s not going to change.
What’s the Science Behind AI Pluribus?
The basics of this success lies in the fact that Pluribus is able to perceive the poker game in a whole new way. It is able to make estimates for each hand it is supposed to play out. It draws up four different short-term scenarios for the following players’ moves and making the best choice based on that.
In this regard, the AI has been noted as having practically taught itself, based on practical experience in the gameplay. It makes unpredictable and different sized bets, as well as bluffs in order to take on its opponents, raising some existential questions all the while. After all, with bluffing being considered an inherently human characteristic, the matter came down to answer if it’s “the behavior or the result” that defines the practice.
What makes the concept all the harder to grasp and accept is the fact that poker is considered an imperfect and incomplete game. This is due to the unpredictability of the outcome, as well as the lack of preview over some elements of the game like the opponents’ cards.
Building on Past Achievements
This isn’t the first time Artificial Intelligence has been able to perform something that was considered to be inherently human. Just in the 1950s, chess was overturned as one of the most intelligence-provoking games into a simple AI bot algorithm. Soon afterwards, the game Go underwent the same process, with bluffing being just the latest one in the line.
In order to eradicate any malicious implications, co-creator Noah Brown came forward with a statement declaring his initial motives.
This whole idea that there is this, you know, mathematical strategy to the game, this perfect strategy that, if you can play it, nobody will be able to beat you”, is what seems to be the prompt towards this AI’s creation. The bot doesn’t view it as deceptive or lying in any way, it just views it as ‘This is the action that’s going to make me the most money in this situation’
Bear in mind that Pluribus, as it is today, has only been developed thanks to two previous versions: the 2015 Claudico and the 2017 Libratus. The former didn’t manage to achieve the statistical victory, only to be remedied by the latter. This taught researchers a valuable lesson. They stopped trying to sabotage and exploit the opponents and started improving on their own gameplay strategy.
What’s in the Cards for Artificial Intelligence?
The future of artificial intelligence seems to be getting all the more impressive considering how fast it took them to gain such valuable insight and produce tangible results. As impressive as Pluribus’ capabilities are, many have been wondering about its potential practical applications.
The primary concern has been swiftly eradicated, as Brown himself claimed that:
I have come to love the game, because these AIs have really shown there’s a whole additional depth to the game that humans haven’t understood, even brilliant professional players who have played millions of hands.
With some reassurance coming from one of the inventors, tensions have loosened up, and other guesses are being made. Some of the top areas that could benefit from such perfected Artificial Intelligence include health, security, all the way to transport and entertainment like self-driving cars and motorsports games.