We gamers love a great challenge and appreciate a good artificial intelligence(AI) adversary. However, few of us have dealt with an AI teaching itself how to be better during gameplay.

In most computer games, when the difficulty level is raised, the computer responds by making itself harder to kill, the player easier to kill, and maybe use a pre-programmed strategy reserved for the level of difficulty.

However, computer scientists from Carnegie Mellon University created an artificial intelligence (AI) that defeated four of the top no limit Texas No Hold’em Poker Players.

And this AI is worlds apart from traditional gaming AIs.

Creating an AI that can defeat human poker players has long been the Holy Grail for AI scientists. Poker is a game where players work with limited information about the status of the game, including how they are doing compared to their opponents. During one game of poker, there are approximately 10^161 decision points, and this is with imperfect information

The challenge occurred at the Rivers Casino in Pittsburgh in January of 2017 and the results stunned the worlds of artificial intelligence, gambling, and gamers.

If the thought of playing a game against a good AI is appealing to you, check out this video that shows some of the best video game AI available today.

This is how AI will beat you at thinking

Recently the scientists who created the powerful poker-playing AI, named Libratus, revealed how the AI machine operates, including how it makes decisions during imperfect gaming environments.

Dr. Tuomas Sandholm a professor of computer science at Carnegie-Mellon and one of the creators of Libratus recently published a paper in the journal Science.

He explained how the AI works.

The Libratus AI has a three-step process for gameplay.

  • First, the AI breaks the game down into small bites for faster computation. It creates an abstraction for the game and also generates its own strategy for the beginning of the game. It also makes a rough strategy for further stages in the game. One way it does this is it groups similar hands and bets together.
  • As the game nears the end, a second module in the AI creates another abstraction for the game, but this one is more detailed and refined. This allows the computer to make nuanced decisions based on the state of the game. It then creates sub-games and compares them to make the best choice. During this time it will also create a second strategy, based on the rough strategy made in the first steps and the outcomes of opponent’s decisions.
  • Finally, the AI examines the size of the bets of the other players to discover any problems with its strategy. It then computes strategies to improve its playing position. This is done by a self-improver module. Yes, the AI has a module that makes itself better.

By the way, it is doing this quickly and undisturbed by sights, sounds, smells, or anything else that can break a human player’s concentration.

So, you think you can beat it?

The AI is impressive in design and function. How impressive? Well, during the competition against the human poker players, Libratus won $1.8 million.

Scientists are hoping to adopt the AI for use in gaming, cybersecurity, financial applications, and for rehabilitation help.

Dr. Sandholm’s paper is titled “Superhuman AI for heads-up no-limit poker: Libratus beats top professionals.”