New Directions in Cooperative AI

Cultural evolution is the fastest known intelligence generating algorithm. In humans, it is responsible for the upward ratchet of our cooperative understanding, communication, commitments and institutions. In this talk, I will draw on the evolutionary psychology literature to define cultural evolution. I will explain how cultural transmission can arise in AI systems via reinforcement learning, leading to "cognitive gadgets" analogous to human thought. Finally, I will outline how cultural evolution offers an important, tractable and under-explored path towards safe, generally-capable Cooperative AI.
Speakers

Edward Hughes (DeepMind)

Discussants

Sarah Mathew (Arizona State University)

Christian Schroeder de Witt (University of Oxford)

Time

15:00-16:30 UTC 22 April 2022

Links
This seminar has now finished

Google calendar event

ICS file

Zoom meeting (passcode sent via mailing list)

Dr. Edward Hughes is a Staff Research Engineer at DeepMind. His research pioneers the field of Cooperative AI, algorithms that work in partnership with each other and humans. His work lies at a crossroads linking multi-agent reinforcement learning with cognitive psychology and economics. Edward received his PhD in Theoretical Physics from Queen Mary University of London on applications of string theory to particle scattering. Outside academia, Edward conducts the Godwine Choir, with whom he has recorded on the EM Records label and toured internationally.

Cultural Evolution as a Cooperative AI Generating Algorithm