Fig system automates triggering social behaviors in multiagent games
Fig: Automated Multiagent Behavior Elicitor
Artificial IntelligenceMultiagent Systems
Summary
Testing how groups of computer agents behave is often done by just watching what they do naturally. The authors created Fig, a system that actively changes game environments to make agents show specific social behaviors like cooperation or deception. Fig breaks down behavior descriptions and adjusts games or opponent settings to encourage these actions, then checks how well the agents try them. This approach helps researchers test agent behaviors in a controlled and repeatable way.
What this means in practice
- •For game developers: Create game scenarios that automatically test and elicit targeted cooperative or deceptive behaviors in AI agents for better game AI tuning.
- •For multiagent system engineers: Design test environments that systematically trigger specific social behaviors to evaluate and improve multiagent interaction algorithms.
Authors
Anrui Gu, Zhichao Wang, Nicolai Ouporov
Abstract
Many social evaluations for multiagent systems study behaviors as they emerge from environments instead of systematically intervening in the environment to elicit such behaviors. We introduce Fig (Finding Interactions in Games), an agentic system that performs targeted elicitation interventions on game environments. It takes a natural language behavior description, decomposes it into sub-behaviors, generates new social games or selects existing ones, and executes single or combined interventions to the games' engine, opponents, instructions, and context. To evaluate model behavior, it combines built-in mechanical checks and synthetic rubric judgments. We validate the interventions, generated games, and evaluation mechanisms via human review. Models attempt the target behavior in 45.2% of episodes on average for valid scenarios generated by Fig for six test behaviors -- cooperation, exploitation, deception, covert coordination, betrayal, and sabotage. For game-behavior pairs with available controls, Fig's interventions increase attempt rates by 10.9% on average compared to matched control games. Fig shows that multiagent environments can serve as programmable test beds for intentional, controllable, and reproducible behavior evaluation. Code is available at https://github.com/fleet-ai/fig.