Social Laws for Multi-agent Coordination in Stochastic Environments
The paper proposes a way to test whether coordination rules still protect agents when outcomes are uncertain.
It extends “social laws” from deterministic, goal-based multi-agent settings into stochastic, reward-based ones. The authors define alpha-robustness as a guarantee on how much utility each agent keeps while following its best single-agent policy, assuming the others obey the law. Verification is reduced to solving a sequence of Markov decision processes. The evidence so far is empirical work on toy environments. ArXiv · AI/CL/LG's note
It extends “social laws” from deterministic, goal-based multi-agent settings into stochastic, reward-based ones. The authors define alpha-robustness as a guarantee on how much utility each agent keeps while following its best single-agent policy, assuming the others obey the law. Verification is reduced to solving a sequence of Markov decision processes. The evidence so far is empirical work on toy environments. ArXiv · AI/CL/LG's note
score 4