Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper analyzes theoretical bounds for agent-agnostic safety guarantees in partially observable, time-constrained environments, demonstrating that even perfect agents cannot guarantee safety. The authors introduce two realistic scenarios (infinite state space and signal mixture) to prove this limitation, highlighting the necessity of co-designing agents and their environments for robust AI safety. The key result is that environment-specific considerations are indispensable for any rigorous proof of AI safety or alignment.
Even perfect AI agents fail to guarantee safety in realistic, time-sensitive, and partially observable environments, fundamentally challenging agent-agnostic alignment approaches.
In the wake of the increasingly public discussion about AI alignment, recent work has tried to propose specific AI architectures that behave safely. However, the proposed arguments that seemingly demonstrate proved alignment mostly neglect the environment the agent needs to act in. We discuss theoretical bounds for agent-agnostic safety guarantees in environments that can only be partially observed and within which an action is required within limited time. We introduce two realistic scenarios, one with an infinite state space and one with signal mixture. In these scenarios, we prove that even a perfect agent cannot guarantee safe behaviour. It will be argued that for any proof of AI safety or alignment, the environment and associated safe actions need to be specifically considered together with the agent.