Let’s talk about poker. Men in the Wild West have died over it. James Bond in Casino Royale outwitted over an evening of cards the shadowy conspiracy that would evolve into SPECTRE. People flock to Vegas thinking that, this time, they can beat the house at this game.
It’s a fascinating game—Poker. Now, it is no longer a game belonging to humans.
At Carnegie Mellon University, Artificial Intelligence has evolved to where machines are now playing poker better than the professionals. Recently, the college hosted a poker tournament between an AI system which was built by and four of the world’s top Poker pros. As this was less of regulation tournament play and a science experiment, the stakes were unlimited. You could bid as much as you would want. Carnegie Mellon employed machine learning to teach their A.I. how to play poker. They let their A.I. observe thousands of other poker hands to see how people played poker, and from there, the A.I. would learn the rules and—presumably—strategy.
120,000 hands were played between professional gamblers and cutting-edge technology, and by the time the tournament concluded, the A.I. bot had won $1.5 million. In a stunning outcome, the A.I. just cleaned everybody out. These poker pros still walked away with $20,000 as they had to work. 120,000 hands over multiple days is not an easy feat. As for their opponent, the A.I.’s main advantage was its ability to remain totally unpredictable. Libratus, the name of this A.I., won the respect of the pros it played against, but they don’t believe there are many tricks they can pick up from the system. However, they did realize Libratus’ approach was very mixed and randomized. It’s hard for humans to do that because they get locked in a certain way of thinking. Maybe they don’t really want to take a chance with a specific hand? Humans end up falling into a pattern of thought that becomes predictable.
The computer was not predictable at all.
One of the things Libratus does well is bluff, and it developed a really aggressive bluffing strategy. The A.I. learned how to be really ruthless. So if it would have a very bad hand that you would just lose on, Libratus realized it is better to bet the whole pot as though it’s got something great. Libratus was able to bluff out the humans, because humans will reason you’re not going to bet all $50,000 on a bad hand as a bluff. Libratus plays differently. “I have got to be super aggressive with a bad hand,” it reasons, and that would be what Libratus would do, essentially outbluffing professionals.
True, there are things beyond style of play that professionals use to match wits with other card players. You have tells, facial ticks, eye contact, and the like. Still, Libratus’ performance have people about how A.I. systems will evolve when they are in self-learning scenarios like this. Will they become more aggressive? What if they become weaponized? Is this a reasonable path we are heading down?
All I can speak for certainty is that Poker is the kind of game that brings out the best and worst in people, especially when stakes are high. I wonder if this is a glimpse to an uncomfortable truth that man and machine are not so different.
A research physicist who has become an entrepreneur and educational leader, and an expert on competency-based education, critical thinking in the classroom, curriculum development, and education management, Dr. Richard Shurtz is the president and chief executive officer of Stratford University. He has published over 30 technical publications, holds 15 patents, and is host of the weekly radio show, Tech Talk. A noted expert on competency-based education, Dr. Shurtz has conducted numerous workshops and seminars for educators in Jamaica, Egypt, India, and China, and has established academic partnerships in China, India, Sri Lanka, Kurdistan, Malaysia, and Canada.