By 
The ongoing debate about how good artificial intelligence will get may just have been won for now by Facebook very own AI that was developed in partnership with researchers from Pittsburgh-based Carnegie Mellon University that beat five of the world’s best players in a series of six-player no-limit Texas Hold’em Poker games.
This being one of the most popular yet intense poker games across the globe, you might think that the researchers would have tested out the intelligence at a much easier and low-graded series but their confidence is clearly shown in their choice and it worked out for them after all. This was also a huge surprise to everyone considering poker has been a vital yet entirely difficult grand challenge for the field of AI.
This is all because poker involves hidden information- you do not know your opponents’ cards and there is a lot of bluffing involved alongside other strategies that even the likes of chess do not require. The new AI bot, named Pluribus, was put to the test in two formats of the game; “five AIs + one human player” and “one AI + five human players and ended up winning every game.
Start off your week with tech news recap and trending conversations in your inbox
https://www.facebook.com/FacebookAI/videos/2087259478050682/
Pluribus won an average of $5 per hand with hourly winnings of around $1,000, a result that was described as a decisive margin of victory by the poker professionals. “Pluribus is a very hard opponent to play against. It’s really hard to pin him down on any kind of hand,” Chris Ferguson, a six-time World Series of Poker champion and one of the 12 pros drafted against the AI, said in a press statement.
The victory has been considered a huge milestone in the world of AI by the minds behind Pluribus’ development and even though AI has reached superhuman levels before in games like chess and Go among others, this may just have surpassed a whole new difficulty level.
These results have been revealed to have been achieved through several innovations from Libratus, an AI bot that was used to beat human pros in two-player no-limit Hold ’em poker in 2017 alongside other algorithms and code developed by researchers from the university. But in real detail, Pluribus was integrated with a new online search algorithm that can efficiently examine and calculate options by searching just a few moves ahead rather than only to the end of the game.
To make it even better, Pluribus incorporates faster self-play algorithms for games with hidden information. The two advances were then combined to make the bot efficiently use a 64-core server with less than 512GB of RAM- an equivalent of less than $150 worth of cloud computing resources. The efficiency also comes next to the fact it was created in just eight days, all this standing in significant contrast to other AI milestone projects, which demanded equivalents of millions of dollars worth of computing resources to train.
This efficient machine learning went on ahead to prove that unlike popular belief, short-term incisiveness is really all one needs in the game. Pluribus was very good at making bluffs at the opponents, an ability that got the human pros marvelled by its “relentless consistency” and how it squeezed profits out of relatively thin hands.
“The AI doesn’t see bluffing as deceptive. It just sees the decision that will make it the most money in that particular situation. What we show is that an AI can bluff, and it can bluff better than any human.”
-Research scientist at Facebook AI Research, Noam Brown.
Does all this then mean that an AI has beaten humans to the punch in quick decision making in the world’s best poker game? With the actions portrayed, it goes a long way to show that as much as humans are the creators, we can also learn a few things from computers including taking risks with strategies that we’ve always seen as not worth it.
What may then happen best with this is if the same intelligence was incorporated into other industries that are looking to have AI as a core foundation with the examples of autonomous vehicles.
Comments