Training Generalizable Collaborative Agents via Strategic Risk Aversion

2026-02-24

AI Summary

Existing cooperative agents exhibit fragility when interacting with unfamiliar partners, primarily due to free-riding behaviors during training and a lack of strategic robustness. This work proposes incorporating strategic risk aversion as an inductive bias within a multi-agent reinforcement learning framework, guiding agents to proactively avoid inefficient equilibria in cooperative games. The approach not only effectively suppresses free-riding but also achieves collaborative outcomes that surpass Nash equilibria in efficiency. Evaluated across multiple standard cooperative benchmarks and large language model–based collaboration scenarios, the proposed agents consistently demonstrate stable cooperation with heterogeneous and previously unseen partners, significantly enhancing both generalization and robustness.

Technology Category

• Game Theory and Economic Paradigms: Cooperative Game Theory
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Safety and Robustness

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Many emerging agentic paradigms require agents to collaborate with one another (or people) to achieve shared goals. Unfortunately, existing approaches to learning policies for such collaborative problems produce brittle solutions that fail when paired with new partners. We attribute these failures to a combination of free-riding during training and a lack of strategic robustness. To address these problems, we study the concept of strategic risk aversion and interpret it as a principled inductive bias for generalizable cooperation with unseen partners. While strategically risk-averse players are robust to deviations in their partner's behavior by design, we show that, in collaborative games, they also (1) can have better equilibrium outcomes than those at classical game-theoretic concepts like Nash, and (2) exhibit less or no free-riding. Inspired by these insights, we develop a multi-agent reinforcement learning (MARL) algorithm that integrates strategic risk aversion into standard policy optimization methods. Our empirical results across collaborative benchmarks (including an LLM collaboration task) validate our theory and demonstrate that our approach consistently achieves reliable collaboration with heterogeneous and previously unseen partners across collaborative tasks.

Problem

• collaborative agents
• generalization
• strategic robustness
• free-riding
• unseen partners

Innovation

• strategic risk aversion
• generalizable cooperation
• multi-agent reinforcement learning
• free-riding mitigation
• collaborative agents