[toc]Carnegie Mellon University artificial intelligence bot, Libratus, brought home more than the purse this year for its developers – it brought home an award.
The AI won the HPCwire Reader’s Choice Award for Best Use of AI.
According to Carnegie Mellon University, its developers aren’t leaving empty-handed, though.
The Supercomputing Center in Pittsburgh received five awards:
- Reader’s Choice Award for PSC Interim Director Nick Nystrom for Outstanding Leadership in HPC
- Reader’s Choice Award for Best Use of HPC in Energy
- Editor’s Choice Award for Best Use of High‑Performance Data Analytics
- Editor’s Choice Award for Best Use of HPC in Life Sciences
Poker professionals have taken on artificial intelligence before. This time’s competition, though, came back with statistically significant findings.
Back in January, Libratus, the AI developed by Carnegie Mellon University’s Tuomas Sandholm and Noam Brown, played 120,000 hands of heads-up no-limit hold’em over the course of 20 days from Pittsburgh’s Rivers Casino.
The event, “Brains Vs. Artificial Intelligence: Upping the Ante,” hosted four poker pros – Jason Les, Dong Kim, Daniel McAulay, and Jimmy Chou – competing for a $200,000 prize.
AI bests humans significantly
Libratus was developed from the prior year’s model, Claudico. The new version utilized a method to find equilibrium faster, identify hands that aren’t promising, ignore future paths, and use the Bridges computer to perform live computations with an endgame-solving approach and algorithm.
By the event’s end, the poker pros had taken quite a beating; Libratus had a collective $1,766,250 in chips, a statistically significant number.
This exciting find can spread into other realms of information gathering, including:
- Business negotiation
- Military strategy
- Cybersecurity
- Medical treatment planning
“The best AI’s ability to do strategic reasoning with imperfect information has now surpassed that of the best humans,” Sandholm said, co-developer of Libratus and professor of computer science with Noam Brown, a Ph.D. student in computer science, in a release from Carnegie Mellon, at the time of the finding.
AI successes show progress
The original goal was to set a benchmark for artificial intelligence, showing progress in the field as others had done before. That started with chess in 1997 (Deep Blue), with Jeopardy! in 2009, and with the board game Go this past year.
Poker is considerably more difficult than these games, Sandholm said. The AI can make decisions based on incomplete information, adjusting to bluffs, slow play, and more.
DeepStack, another computer that recently competed against human beings, defeated 11 other professional poker plays at heads-up no-limit hold-em, a study in Science reported. The bot underwent deep learning training to develop poker intuition for any situation.
Lengpudashi, another version of the Carnegie Mellon AI, beat six top Chinese poker players by $793,327 in virtual chips in 36,000 hands. This feat, similar to the others reported by Carnegie Mellon, was distinct in that the margin of victory was greater.
Lengpudashi took 220 milli-big-blinds per game vs. Libratus’ 147 milli-big-blinds per game.
