Artificial Intelligence Takes Over Humans in Poker
Last year we saw some pretty fundamental changes in the world of poker and otherwise. Far out technological advancements were seen in 2017 and they tend to grow exponentially with each year passing. Now, Artificial Intelligence has always been a pinnacle, the epitome of robotics. However, the less known fact is that it has existed in poker since 1984, when a basic Poker AI software called Orac was created to compete in WSOP.
The pitting of gaming machines against human player’s and the AI’s eventual win over them, has been a long journey. Artificial Intelligence has succeeded in defeating some of the best human players in the world.
It was just last year that the world’s most advanced self-learning AI bot, Sofia was awarded a citizenship by the Kingdom of Saudi Arabia, giving the world its first robot citizen.
Robots beating humans at chess sounds so 2016, now we have AI’s beating poker pros, amassing millions of dollars doing so while bluffing! Believe it or not, these robots bluff to lure their opponents in.
On January 30, 2017, Libratus Poker AI, a supercomputer developed by Carnegie Mellon University, beat some of the world’s best professional human poker players. This was accomplished playing the most difficult and challenging Poker variant, Heads-Up No-Limit-Hold’em. The bot managed to win $1.76 million in chips over the course of the session!
Libratus’s game strategy is much more advanced than any of the human players’ minds can compute. It is based on 15 million core hours of computation, deriving a super self-learning game algorithm. Libratus analyses each game, finds mistakes and corrects them to play differently in the next hand. This even includes bluffing based on randomization, making it impossible for its opponents to guess when a bluff occurs in the human sense – no reading of the computer’s body language as it doesn’t have any.
There are many other technical aspects that make AI inherently better than humans.
Firstly, AI doesn’t get tired. It can stay 100% focused for as long as it’s powered.
Secondly, AI doesn’t feel the value of money in a human way. No excitement, temptation or fear of losing can throw AI off. Let’s imagine that Libratus has a 10% chance of winning 20,000 against 90% chance of winning nothing. What will it choose? The 10% risk, of course. Try to make the decision that easily, humans!
Next, AI can spot weaknesses in its opponents. Vulnerability has no place in a poker game with AI. But how do they do it? Well, data derived from previous games gives them that wisdom!
Ultimately AI has no emotions. Emotions are the enemy of decision making, especially in poker.
Playing poker is dealing with a lot of imperfect information which makes the game quite complicated like many real-life situation, this is one of the main reasons why big universities AI departments are researching on poker.
The research has gone so far as Poker bots having their own WSOP tournaments. In 2005, the World Series of Poker Robots was held! Today, there are Annual Computer Poker Competitions for AI’s.
The bots making headlines in Poker are mainly – DeepStack, an AI bot from the University of Alberta that can beat poker pros, an unnamed AI that crushes best poker players, and of course our hero, Libratus.
Inside Libratus – The Poker AI that Out Bluffed the Best Poker Pros
Libratus sets its own poker strategy based on 15 million core hours of computation at the Pittsburgh Supercomputing Centre (PSC). It’s 30,000 times faster than an average modern day desktop, consists of 274 Terabytes of RAM and costs over $9.65 million.
Libratus has 3 modules, each performing a different task.
- The first module tries to find strategies that work for every situation and learns poker rules.
- The second one is the decision maker for each action during a hand.
- The third one provides consistent updates, recording and folding the new information from each move into the system.
Instead of a fixed built in strategy, Libratus relies on an algorithm that computes the strategy. Libratus analyses its own play every night and corrects mistakes, this enables it to play differently each day.
It randomizes its actions to make opponents uncertain of when it is bluffing, thereby, doing so better than humans.
Libratus’ algorithm is not only specific to poker, but to any imperfect-information. It lacks a Deep Neural Network and learns poker rules from scratch.
In 2017, Libratus made the most decisive blow in the history of human’s vs poker AI. 4 best players challenged Libratus and played 30,000 hands each. The result – Libratus beat each player and won at an astonishing rate equivalent to over $1.7 million.
You don’t have to worry facing a bot online since they are banned. Don’t break a sweat, unless you play against one in real life.