Member-only story
Artificial Intelligence
DeepMind’s New Game to Improve Cooperation in Multi-Agent Models
Hidden Agenda is a social behavior game optimized for enabling cooperative behavior in reinforcement learning models.
I recently started an AI-focused educational newsletter, that already has over 100,000 subscribers. TheSequence is a no-BS (meaning no hype, no news etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. Please give it a try by subscribing below:
Cooperation is one of the most difficult elements to get right in multi-agent models such as those powered by reinforcement learning techniques. Effective cooperation dynamics in multi-agent have two main dimensions:
1) What levels of cooperation is needed for an agent to be effective?
2) Which are the right agents to cooperate with?
While there are many quantitative techniques that can be used to address the first point, the second remains largely unexplored. Recently, researchers from DeepMind and Harvard University published a paper proposing Hidden Agenda, a 2D social deduction game targeted to improve the cooperation dynamics in multi-agent models.
The challenge of cooperation in multi-agent models is very complex as it depends on the mechanics of each agents in a cooperation mechanism. Different ML agents can share goals but often have conflicting objectives which are not visible to the environment and, therefore, hard to quantify. This challenge is even more pronounced when operating in imperfect information environments. Social deduction games have been a popular mechanism to model cooperation under uncertain conditions. The essence of a social deduction game is to help players to deduct each other’s hidden goals.
Hidden Agenda, is a social deduction game based on multiple players from two…
