Multi-agent learning / Social games
The environment
has intentions.
Mule is building reinforcement learning environments for social games: hidden roles, private information, and agents whose incentives do not always align.
Learning in the presence of others.
In a social game, language is part of the action space. A message can reveal information, coordinate a team, or change what another player believes. Its consequences may only become clear many turns later.
We’re interested in environments that make these dynamics measurable—belief updates, strategic communication, cooperation, and adaptation to unfamiliar opponents.
For research and collaboration.
jerry@mulelabs.org ↗