While artificial intelligence (AI) programs have long dominated games such as chess and scrabble, they have made quite awful poker players. At the beginning of 2017, however, AIs finally proved their mettle at the game of poker. This is after an AI system known as Libratus beat accomplished human gamers in heads-up, no-limit Texas hold’em poker. This is a complex form of poker, which relies largely on long-term betting strategies and game theory. As such, this feat represents a positive development in the types of challenges AI can resolve.
Let’s have a deeper look at Libratus AI system.
Libratus AI System
Libratus is an AI system that runs on a Pittsburgh supercomputer designed by Mellon, Sandholm and Brown who target Libratus to have superior balance within its learning strategies.
These researchers first tried to play against some professional human players with a similar AI about two years ago; they lost, unfortunately. This time, however, they beat the human players, hands down. Namely, across 20 days of play, Libratus outdid its four professional human competitors by at least $1.7 million. All of the four human players finished with a negative number of chips.
AI Using Multiple Technologies
Libratus depended on three different systems, which worked together. Modern AI is driven by a number of technologies; not just one. The defeated human players pointed out that Libratus had its act together. It did a little bit of everything perfectly. Namely, it knew not only when to bluff, but also when to bet low with good cards.
With the help of an algorithm known as counterfactual regret minimization, Libratus began by playing at random. In the end, after some months of training and over a trillion hands of poker, it reached a level where it could play in ways humans couldn’t. It could play a vast range of bets and randomize them so that competitors don’t have an easy time guessing what cards it holds.
AI Going beyond Human Comprehension
According to one of the players defeated by the AI system, Libratus split its bets into a number of different sizes, something no human can be able to do. In fact, the human players believe that Libratus’ game changed every day. Whenever they felt they had a clue in the machine’s strategy, everything would change.
Do you think that the machine’s designers changed the AI system between rounds? If so, was it fair?
If that really happened and it sounds unfair to you, you should realize that the human players also applied every trick at their disposal. While every player played his own game against Libratus, they shared tactics. You know why?
One of the players pointed out that while they were playing against one another, they were also rooting for humans’ victory.
The Significance of Libratus AI System
Poker may be just game, but the game theory displayed by Libratus may be applied in other fields such as financial trading, political negotiations, auctions, among other potential sectors. A machine that understands this kind of complex poker simulates the kind of human intuition these tactics require.
Libratus has the potential to be so much more based on its application use. Based on this data, AI could predict people’s habits as well as steer the world out of poverty given enough time.
Artificial Intelligence in the past has been a poor poker player, but it eventually proved its ability at the game after beating four accomplished human players. This added to the number of challenges AI can solve, and the future looks bright for what AI can accomplish.