Is Poker Bot Capable of a Reasonable Bluff?

(1 votes, average: 5.00 out of 5)

Loading...

The main problem with artificial intelligence is not about solving Chess, the game of Go, or Rubik’s Cube. Indeed, all these games have a vast number of decisions but still are very transparent regarding options. When it comes to playing poker, everything changes dramatically and is most clearly seen when trying to program a ‘thinking’ bot.

The Tricky Part

The main problem is to turn meaningful data into a working and useful game strategy. According to the gambling facts, finding and picking up a particular set of rules for a poker bot may help it achieve the optimal result in the game.

This principle can be seen in the game of Garry Kasparov’s against the chess supercomputer Deep Blue. In 1997 he lost after a series of matches. After that, Harry disapproved of this approach to playing chess.

When it comes to a game like poker, which is a game with incomplete information, things are somewhat different. A bot will not be able to win merely following a limited set of rules because some data is missing, that is, a player does not know all the cards. Such a conclusion applies to many other areas, such as gambling on the stock exchange, buying at auctions, and participating in business talks.

For instance, it is still possible to find the optimal set of rules for a game like Go which is open but has many more possible steps than chess. The Google DeepMind’s AlphaGo confirmed this fact. On the other hand, how to win at poker every time? Only continually adjusting your game strategy to the data based on your opponents’ play. Playing your hand the same way, you will become as predictable as a bot and most likely lose.

Poker has long attracted such famous scientists as Alan Turing and John von Neumann who were among the first to become interested in thinking machines. Presently, this interest is increasing.

Often poker is seen more as an art, rather than a science because this game depends heavily on ingenuity and initiative, and not on monotonous figures and calculations. Nevertheless, the best poker bots today make us doubt this idea. At the same time, our understanding of how a bot and a human create the game strategy and make decisions also changes.

In 2015, there was a very significant event in the world of poker. Developers from the University of Alberta presented their poker bot Cepheus which was able to solve the limit version of Texas Hold’em for two players which is a one-on-one game with restrictions on the maximum bet. The bot played according to equilibrium strategy which in every single situation depends on possible options with a certain probability. So that in the long term the player adhering to such an approach will not lose money.

It is noteworthy that no one taught Cepheus any poker strategy. It came to it due to losing billions of simulated hands.

Can Success Be Achieved in Other Poker Types?

To date, developers have switched their attention to other types of poker. The Limit version of Hold’em, solved by Cepheus, has a somewhat rigid framework which makes it possible to study the strategy more efficiently. However, it significantly reduces the relevance of such studies in regarding using them in the real world.

Undoubtedly, No-Limit Hold’em will be the next step as this game is by far the most popular poker version. The fact that the bet amount in this game is not limited and the player can go all-in any time significantly complicates the task.

Some bots already play No-Limit Hold’em with ease and even show some tactical thinking that is far beyond human capacities. Gradually evolving, bots begin to discover new ways of juggling risks and also find innovative decision-making options for playing games with incomplete information.

All the above-mentioned makes us think about the following question: ‘Which aspects of our behavior can be considered inherently human, and the thinking machines can adopt which of them?

Although poker has a reputation for a psychological game, one cannot say that in this game only people can deceive each other, that is, to bluff. In fact, as of today, the poker bot can come to the same decision based on the optimal strategy.

The best poker bots taught each other to bluff, show aggression, and even manipulate their opponents. In the end, we can say that Kasparov’s expectations that computers will learn that it is sometimes profitable to play as a human, not a machine are gradually coming true.