Artificial Intelligence in games has taken people by surprise. Playing chess and solitaire on our computers as a child and beating the computer was the hardest thing for us. But now, machine learning has allowed these computers to study their opponents and make informed decisions based on the data they receive.
The latest poker AI research has made intelligent systems like Libratus, Pluribus, and DeepStack, beating players left and right. Creating an AI poker system that could beat human opponents was quite challenging.
This is because poker requires strategy, intuition, and reasoning. So, developers designed AI-powered poker systems that are capable of winning. Let’s talk about these intelligent poker systems:
DeepStack: Winner of Two-Player Texas Hold’em Poker
The DeepStack team is from the University of Alberta in Edmonton, Canada. The team used a combination of algorithms and machine learning to create a poker game powered by Artificial Intelligence.
The AI poker game can win the No Limit Texas Hold’em variant of poker played between two players. DeepStack started with this variant because it is more challenging for the AI to learn and master the game’s randomness, player’s bluffs, and hidden cards.
The DeepStack team trained their neural networks with an enormous data set of 10 million poker game outcomes. So, the game’s AI depends on these outcomes to make the best moves.
DeepStack used its AI poker system against two famous professional players of the International Federation of Poker. The AI system played almost 45,000 games, and the results were ten times better than what professional poker players consider reasonable.
Libratus Competes in a 20-Day Poker Tournament
Tuomas Sandholm and Noam Brown developed the AI system Libratus. They both were from Carnegie Mellon University when they made this powerful AI in 2017. This AI poker system remained unbeatable by human players in two-player poker games.
Libratus used 100 CPUs (central processing units) to run and participated in a 20-day poker tournament. The system competed against four of the best Texas Hold’em poker players and played 120,000 hands. The results showed that it won the tournament with winnings as high as $1.8 million in chips.
Pluribus: A Collaboration with Facebook AI
Pluribus is a robot made by the collaboration of scientists at Carnegie Mellon University and Facebook AI. This bot isn’t an ordinary bot. Instead, this bot beat some of the best professional poker players in the world in a Texas Hold’em variant of six people.
This collaboration was the first of its kind, where the AI system played against more than one player at a time and simply could not rely on strategy to beat them. Pluribus, the bot, played 10,000 poker hands with impressive results, winning a million dollars against five players.
Stats show that Pluribus won $480 on each 100 hands-on par against its human competitors. These results are what poker players dream of achieving. The development team made Pluribus on their findings from Libratus.
Instead of processing the decision trees until the end of the game to make a move, Pluribus only determined the results of a few moves ahead to decide which move to make. Through reinforcement learning, the AI taught itself to look back at plays and evaluate its success based on the situations.
Before the developers used Pluribus to compete against human players, they played the AI system against itself for a trillion hands. Slowly, it competed against one professional poker player.
The player would alert the developers if the system made a mistake during the game. With this new information, the bot improved its results drastically. Ultimately, it competed against 5 players while developing mixed poker strategies based on the circumstances.
But are these Bots Better than Human Players?
The biggest upside to bots playing poker remains that these systems will never lose money. While losing a game can have serious money problems for human players, this doesn’t affect the bots.
Another vital point to notice is the ‘tells’ a human poker player can reveal during games through facial expressions and gestures. In contrast, bots can never reveal physical (tells) information that would give away their cards. Remember, these physical signs can make or break a poker match for humans.
Don’t expect these bots will talk with you. So, verbal tells to reveal gaming style is impossible for a bot. Humans competing against a bot is frustrating as they have to guess how the gameplay will progress each time.
Conclusion
While AI has changed poker games forever, enjoying this game is impossible with a bot. Playing with human players is far more thrilling than playing with machines. Poker is a game of strategies, intelligence, and wits. While AI systems utilize intelligence and strategies to the fullest, wit isn’t their strong suit. But who’s to say? Maybe these systems will become witty over time.
