Researchers studied reinforcement learning under external influence, proposing an algorithm to tackle the problem and providing guarantees for policy improvement. The work has implications for AI agents operating in dynamic environments.