Interview, Fireside Chat
Michael Kearns: Game Theory and Machine Learning
- Algorithmic game theory is expected to provide methods for predicting or influencing platform outcomes when actor numbers are extraordinarily large and incentives are complex.
- Natural iterative system behavior is not guaranteed to reach an equilibrium even if one is theoretically possible.
- The intersection of machine learning and game theory, particularly no regret learning, is anticipated to enable interacting players to achieve equilibrium in a short number of steps.
- Driving apps are projected to nudge users toward a competitive Nash equilibrium by calculating selfish best responses based on the actions of all other players.
- Collective driving time is forecast to be higher, and potentially significantly higher, than under alternative solutions if all users remain in a competitive equilibrium.
- Social media and Amazon algorithms are described as colloquially directing users toward forms of competitive equilibrium.
- Machine learning is identified as the primary driver for platform optimization, including predictions regarding traffic locations, product preferences, and user satisfaction in newsfeeds.