Virtuously Safe Reinforcement Learning
This work addresses safety issues in reinforcement learning for scenarios involving adversarial interference, which is incremental as it extends prior work on interruptible agents.
The paper tackles the problem of making reinforcement learning agents both interruptible and resilient to adversarial perturbations in perception, while ensuring safe exploration in the limit. It shows that agents can achieve these properties, addressing a trade-off between optimal policy adherence and escape from dangerous situations created by an adversary.
We show that when a third party, the adversary, steps into the two-party setting (agent and operator) of safely interruptible reinforcement learning, a trade-off has to be made between the probability of following the optimal policy in the limit, and the probability of escaping a dangerous situation created by the adversary. So far, the work on safely interruptible agents has assumed a perfect perception of the agent about its environment (no adversary), and therefore implicitly set the second probability to zero, by explicitly seeking a value of one for the first probability. We show that (1) agents can be made both interruptible and adversary-resilient, and (2) the interruptibility can be made safe in the sense that the agent itself will not seek to avoid it. We also solve the problem that arises when the agent does not go completely greedy, i.e. issues with safe exploration in the limit. Resilience to perturbed perception, safe exploration in the limit, and safe interruptibility are the three pillars of what we call \emph{virtuously safe reinforcement learning}.