Speaker: Alexander Sasha Vezhnevets

Institution: Google DeepMind

Date: 14:00 BST, 10th September, 2025

Talk Title: Concordia, a multi-actor generative agent based modelling system

Abstract: This talk introduces Concordia, a  multi-actor generative agent based modelling system. Concrodia’s design is drawing inspiration from tabletop role-playing games (TTRPGs) and modern game engines. The core of this framework is the Entity-Component architectural pattern, which enforces a crucial separation between engineering (reusable components) and design (hierarchical composition and configuration). This separation enables rapid iteration and high modularity, ensuring system robustness. The approach supports diverse user motivations, allowing for tailored scenario configuration to align with specific goals. We will talk about various possible applications of generative agent based modelling like creating synthetic users, in silico social psychology experiments, forecasting and so on. 

Bio: I’m an AI research scientist and a meditator, based in London, UK. I’m a staff research scientist at Google DeepMind. Throughout my career I worked on computer vision, hierarchical reinforcement learning, multi-agent reinforcement learning and generative agent based modelling. My current interests are multi-agent systems, social cognition, computational social construction and cultural evolution. I’m also interested in bridging ideas in AI, social sciences, and social psychology. I’m also a meditator and a contemporary Vajrayana practitioner. My practice is mostly based on Dzogchen and Tantra (Yidams) from Dzogchen perspective. I write about my practice, the places it takes me and how it relates to my life on substack. I’m part of the evolving Ground community.