Abstract: Social interactions are a key aspect of human intelligence, but the study of intelligence has traditionally focused on single-individual experiments and measurements. The study of sociality is emerging as key to both understand human-relevant problems using AI tools, and to design more intelligent artificial systems. In this talk, I will discuss a body of work that aims to understand the necessary conditions for promoting cooperation in multi-agent systems and how they relate to humanity's hardest challenges.
We rely on a wide range of fields, like game theory, machine learning, reinforcement learning, and evolutionary biology. Some familiarity with the modelling of natural systems is beneficial. However, we will keep the discussions at a relative high level so the talk should be accessible to a broad audience.
Bio: Edgar is a senior research engineer at DeepMind working at the intersection of machine learning, evolutionary biology, and game theory. During his time in academia, his research focused on the evolution of cooperation, task partitioning, punishment, and their interaction with population structure. He has contributed to a wide range of applied machine learning projects at Google, including the index for Google Image Search, the development of the new voice of the Google Assistant using WaveNet, and the energy prediction for Google's wind farms. More recently, he has brought together his applied ML expertise with his expertise in cooperation and sociality into his work in multi-agent reinforcement learning, where he aims to understand and tackle the biggest (social) challenges humanity faces.