Abstract

How can one detect friendly and adversarial behavior from raw data? Detectingwhether an environment is a friend, a foe, or anything in between, remains apoorly understood yet desirable ability for safe and robust agents. This paperproposes a definition of these environmental "attitudes" based on ancharacterization of the environment's ability to react to the agent's privatestrategy. We define an objective function for a one-shot game that allowsderiving the environment's probability distribution under friendly andadversarial assumptions alongside the agent's optimal strategy. Furthermore, wepresent an algorithm to compute these equilibrium strategies, and showexperimentally that both friendly and adversarial environments possessnon-trivial optimal strategies.

 

Introduction (beta)

None

 

Conclusion (beta)

None