When artificial intelligence becomes unbeatable in card games

One of the last strongholds that resisted algorithms has just fallen. Artificial intelligence was stronger than humans at chess, go, but not yet at card games. This is now ancient history.

Card games resisted algorithms again and again. But that’s ancient history. A few days ago, for the first time, a French artificial intelligence beat 8 world bridge champions. 67 games out of 80 were won by the machine against humans. This is unheard of !

A few months ago it was poker champions who were beaten hands down by another algorithm. All this may seem anecdotal, but it is a real breakthrough in the field of AI because these card games require very human qualities, very difficult to reproduce by the machine.

An artificial intelligence that thwarts human psychology

In chess, it’s raw computing power that counts. But bridge – or poker – has a more psychological side. These are games with partial information, we don’t know which cards the opponent has, so we have to make decisions in an uncertain universe. It is necessary to react permanently according to what the adversary or partners does in the case of the bridge. There is a notion of intuition, psychology and that for a machine, it’s very complicated.

If we take the example of poker, a few months ago, a team of professional players faced Libratus, an AI developed by researchers at Carnegie University in the United States. They played for three weeks a total of 120,000 hands, to have a representative sample, because in poker, variance, in other words luck, can have a huge impact in the short term. In the end, they lost the equivalent of $1.76 million.

>>> Find all the chronicles of Anthony Morel in podcast

Unpredictable strategies for humans

Artificial intelligence has two big advantages: first, the machine never gets tired and never goes into “tilt”, that is to say does not get angry or alter its game when it starts to lose, which all humans do.

The machine learned “by itself”: it was given the basic rules of no limit hold em. She played against herself, until she played optimally. What makes his strategies have sometimes taken his human opponents completely by surprise, by playing a very aggressive game which can disconcert or even sometimes very strange bets such as “donk bets” which are considered beginner’s errors, is i.e. betting hugely aggressively to win a paltry bet.

Suddenly, the machine is much better than humans at bluffing. According to what his opponents have explained, the big problem is that his decisions are completely unpredictable and come out of the patterns of human logic.

AIs not just for gaming

A new generation of algorithms will arrive, not based on the exploitation of an enormous quantity of data (vision of Google, Facebook…) but capable of making optimal decisions with incomplete information. Bridge AI could have applications in fields such as health and research on rare diseases, where we do not have a lot of data, but also in aeronautics, education, cybersecurity.

AI for poker is of great interest to the US army, for example, for military simulation, since AI is able to surprise its opponents with confusing strategies. These AIs which are able to reason from incomplete information could also help to make forecasts, we could use them in climate forecasts or even language understanding, but also to cheat on poker sites. These automated bots could make it possible to earn large sums of money illegally.

Anthony Morel (edited by MM)

We would love to say thanks to the writer of this article for this remarkable material

When artificial intelligence becomes unbeatable in card games

Find here our social media accounts as well as other pages related to it.https://www.ai-magazine.com/related-pages/