prg.ai meetup: The future of game theory according to Michael Bowling

The Prague AI scene is bustling with new activity, and the vitality of our local ecosystem is no secret abroad, either. To further support and broaden the horizons of our local AI ecosystem, prg.ai organised a meetup which gathered over 180 local academics, entrepreneurs, students, and industrial researchers under one roof. The event posed a great opportunity to forge new connections and reconnect with the local innovation scene, as well as to gain new insights — those were predominantly provided by Michael Bowling, who visited Prague from University of Alberta in Canada.

Michael Bowling’s lecture about a new approach to rationality in game theory

Michael Bowling is a full professor at the University of Alberta, Canada, where he teaches computer science and supervises doctoral students, yet he has also gathered valuable experience at DeepMind’s prestigious research department. In academia, Professor Bowling looks primarily at game theory and its underlying assumptions that enable computer models to play games by the likes of poker. In his research articles, he points out the shortcomings of some of the foundational principles of game theory, which are often considered unshakable, and offers promising alternatives that could mark the advent of a new approach to computational game theory.

During his Prague lecture, Professor Bowling talked about so-called hindsight rationality, a new approach to multi-agent machine learning that focuses on retrospective evaluation within a single life-cycle and comparison with targeted behavioural deviations instead of assessing models against a future optimum.

The lecture marked the culmination of a year-long academic project that aimed to find an alternative to Nash equilibria in game theory, and whose papers were published during prestigious AAAI and ICML conferences. Watch a recording of the full lecture on prg.ai’s YouTube channel below to learn more about:

• the flawed premises of traditional rationality;
• the innovative ideas behind hindsight rationality approach;
• how the new approach applies to diverse deviation sets;
• and what it implies for the field of game theory research.

Our thanks go out to the partners who made this event possible, including CzechInvest, DataSentics, Recombee and Impulse Ventures.  The event was also supported by Czech Technical University’s Faculty of Information Technology, which lent us its impressive lecture halls. In case you couldn’t join us, you can check out a short video from the event, produced by DataSentics.