On the recent victory of the artificial intelligence Libratus in a poker tournament

Good day to all! A few days ago, an article on GeekTimes was published: “ Take and win: the AI ​​won the poker tournament with four pros ” and I want to challenge a bit the significance of this victory.

First of all I want to say that I closely followed the events. I even managed to get answers to your questions from Carnegie Mellon University and professional player Jason Les . In general, I take the topic of winning AI in poker very seriously. This is not chess or any backgammon, where the computer is capable of much faster calculating the visible moves and probabilities.
')

Most people learn from their mistakes. Rarely when strangers. Society makes the same mistakes again and again. Artificial intelligence develops differently. If the autopilot makes a mistake, the other self-driving machines will become smarter. All new cars will be created with the full set of skills of their ancestors. Thus, the collective learning of AI can occur faster than that of people — Eric Schmidt, Sebastian Trun .

Amazing and subtle observation, with which I fully agree. But an important part of the game of poker is bluffing, where feelings and intuition are inherent in a living person. With all my love of technology, in such competitions I support us, because Skynet and the utopian future will acquire a completely different level of argumentation. Let me explain - I am convinced that the computer will surpass us everywhere, everywhere where there is some experience expressed in certain data. In the case where it is necessary to make a completely new decision that does not have a precedent in the past, the high probability of the correct choice will be behind the person. But agnosticism does not give me hope that it will be possible to calculate everything absolutely once.

I apologize for such a long intro with a deepening of my subjective perceptions in futurology. Why is this not the very cult victory? Because all the conditions were created for the success of the computer, because the advantage of AI in the high-frequency operation of data over a long distance is obvious.

Firstly : this is a win in unlimited Texas Hold'em with heads-up (1 on 1 game). I am not a professional poker player, but I have some basic knowledge of odds, pot odds and outs. The game of only two opponents has its own specifics, which is already a certain convention.

Secondly : The player's effectiveness was measured by the ratio of the amount of winnings to the amount of bets (for more details on MIT Technology Review I consider this to be a completely wrong measure. Imagine the situation. There are two bets of $ 1 that did not increase. When opened on the river (5 card) a computer with a stronger combination wins, a certain computer efficiency ratio is 2. Change the conditions, the AI ​​bet $ 2, and I raise to $ 6. Neuron folds cards, I take the bank. My efficiency is ~ 0.33, although the amount won is 2 times more.

Third : the lack of an emotional component. Inanimate money is a different perception of risk. The absence of the stack concept as such. The computer was trained at night, while people need sleep. But the training did not take place on the live conditions of poker.

For the reasons described above, I personally perceive this competition as a first-class advertising technology, and not the evolution of the machine. From the very beginning, in case of victory, the creators of AI predicted the use of technology in live negotiations, military strategy, cyber security and medicine. Success in a tournament can be a good demonstration for potential investors. At this point hypotheses began again, but everything can be. For example, the former director of the same Carnegie Mellon University relatively recently justified the propaganda of lies around IBM Watson. And the facts about the conditions of the tournament remain facts.

Thank you very much for your attention.