With the ability and processing speed of the new age artificial intelligence (AI) technology, many complex human intelligence processes are being simulated using computers and machines. However, AI continues to be incapable of devising a strategic move in the game of poker. Since it involves a bluff or semi-bluff, it isn’t straightforward for AI to approximate to a single step from the various combinations.
Artificial intelligence and poker
The simulations from AI rely on an analytical approach to problems. The primary criterion for AI has been working out solutions, especially when there is imperfect information. Poker is the perfect game for this benchmark as it involves a bluff or semi-bluff component that cannot be deciphered by AI. For the last twelve years, a research line has been dedicated to developing game theories that can assist AI in giving out solutions during a non-definitive game approach.
Let’s compare the AI approach to chess and poker:
Chess is a more ambiguous game compared to poker. Besides unpredictable initial moves, there is always one best predictable move for the game. Therefore, an efficient Chess AI can easily defeat even the best chess champions.
However, in poker, the margin of registering a win can be turned around through bluffing. It is a component which poses a difficulty for AI to comprehend. Contrary to chess, where the next best move can be determined from ongoing movements, Texas Hold ’em (a poker variant) often requires steps based on player’s intuition even when the logic doesn’t synchronise with it. The unrevealed cards of the opponent folders further add up to the uncertainty that clouds judgments.
Let’s delve into this further by understanding the game plan and considering some of the scenarios that arise in a game of poker (Texas Hold ’em variant).
A player wins the game when:
- they are the last standing person, and the remaining players fold.
- two or more players remain at the end of the game and the winning player is the one who possesses the best combination of Community and Hole cards.
Scenarios that are likely to arise during the game:
Scenario 1: Player has a bad hand
Game plan: Players can choose to fold and avoid any chance of winning or bluff by making a bet. It thus tricks other opponents and makes them believe in a firm hand possession by the player. However, the game can turn tables if the opponents have a strong hand.
Scenario 2: Player has a decent hand
Game plan: Players can bet and bluff to protect their hand but should also be able to find out whether their opponents have a weak hand that the player can beat. If so, while betting, players should raise the amount that can make other players call and not fold. This strategy helps players with a decent hand make the best out of the game.
Scenario 3: Player has a firm hand
Game plan: If the player has a dominant, steady hand, they should make a reasonable bet which can make the players call and not fold.
Thus, Poker is a cognitive game where the player’s understanding of the opponent’s next move moulds their strategy. It can not be calibrated through a series of mathematical solutions for the best definitive step. Therefore these scenarios makes it difficult for AI to come up wth the next move.
So, is it impossible for AI to crack the code of poker?
An artificial machine can adapt to different players’ strategies by dividing the game into small parts and acclimate the moves as the momentum of the game picks up. Thus it involves understanding the weakness in the opponent’s strategy and manipulating it.
The present-day scenario of AI and poker
Present-day artificial intelligence is capable of challenging even the most talented poker players. A new milestone was achieved through Libratus. It defeated four of the best poker players in the world in a 20-day marathon. This tipped the scales in favour of AI because poker is a non-definitive game that contains incomplete information which is usually hard to be deciphered by AI. Libratus used the new equilibrium approximation technique and several other methods to analyse the outcomes as cards get revealed only at the end of the game.
This achievement of Libratus increases the possibilities of success of AI to solve real-life situations such as security interactions and negotiations.
However, Libratus cannot still comprehend the multiplayer no-limit Texas Hold ’em.
Conclusion
Although poker plays on the lines of incomplete information with bluff elements, technological developments in AI are advancing to overcome this hurdle. The days when AI handles real-life scenarios or completely takes over a game of poker are not far.