How AI Libratus Plays Poker?

There is much talk about the 20 days competitive event that is currently on between Libratus and DeepStack, two computer bots offered by Carnegie Mellon University teams to contest against professional poker players.


Last year a similar event was held that resulted in the human players winning by a narrow margin over AI. This time, however, it seems that the poker bots are emerging victorious.

There are several characteristics that emerge in the ways the poker bots are playing hands against their human opponents. For instance, Libratus has showcased interesting attributes in the way it has decided on the hands and cards, showcasing something close to intuitive decision making. This is definitely a major advancement in the computer program. There is the final outcome to be seen of this 20 day event, but the games that the poker bots have been able to win also showcase certain ways the computer programs are behaving that will surely be considered advancement in AI research.

The pros were prepared to face an upgrade to what they had faced last year, named Claudico that has showcased some unusual though unwise strategies last year. The pros felt that the upper hand belongs to them as they work on bluffing techniques as well as instinctive decision making that are difficult for AI to replicate. However, they felt, after two days of games, that the AIs were definitely playing better this time. The differences noticed were that Libratus was making variations in bets that made it difficult to predict the hands it possessed.

Also, Libratus was taking more time to make decisions, sometimes more than ten seconds. This has definitely led to several one on one game winning in its favor. It remains to be seen whether the 120,000 hands will be completed in a time span of 20 days.