In recent years there have been great advances in artificial intelligence (AI), with games often serving as challenge problems and milestones for progress. AI breakthroughs such as the one from DeepMind AI, AlphaGo beating a South Korean master at the board game Go, received a considerable amount of media attention.

Poker has served for decades as such a challenge problem. No other game captures the challenges of hidden information as effectively and elegantly as poker. However, past successes in poker have been limited to two-player games where AI could effectively compute a Nash equilibrium strategy, tough in a very slow timely manner.

Multiplayer games present additional challenges where finding a Nash equilibrium in a much larger game tree requires much more resources and is far from guaranteed.

Joining us on today's show is Nikolai Yakovenko, a deep learning researcher at Nvidia and an experienced mixed games poker player.

We address the techniques used by Pluribus, the poker-playing AI bot developed by Noam Brown and Prof. Sandholm at Carnegie Mellon University. We dive into the tradeoff made by the AI and how it achieved superior performance to beat top professionals. Finally, we try to understand what this AI breakthrough means for artificial intelligence and the future possibilities it brings to the poker industry.

Interested in sponsoring the show and getting in front of hundreds of European tech-affine listeners? Write to me at blockvalley.eu@gmail.com

If you haven't left a review yet, please take 30 seconds of your time and leave me a review on iTunes, that would be much-appreciated support to the show!

Episode links
Nikolai's Twitter: https://twitter.com/ivan_bezdomny
Research article: https://science.sciencemag.org/content/early/2019/07/10/science.aay2400