Bad Habits: Policy Confounding and Out-of-Trajectory Generalization in RL
This addresses a fundamental issue in reinforcement learning for AI systems, where agents may fail to generalize in dynamic environments, though it is incremental as it builds on existing understanding of policy biases.
The paper tackles the problem of reinforcement learning agents developing habits that are effective only under specific policies, leading to spurious correlations and poor generalization when deviating from typical trajectories; it mathematically characterizes this as policy confounding and illustrates its occurrence through examples.
Reinforcement learning agents tend to develop habits that are effective only under specific policies. Following an initial exploration phase where agents try out different actions, they eventually converge onto a particular policy. As this occurs, the distribution over state-action trajectories becomes narrower, leading agents to repeatedly experience the same transitions. This repetitive exposure fosters spurious correlations between certain observations and rewards. Agents may then pick up on these correlations and develop simplistic habits tailored to the specific set of trajectories dictated by their policy. The problem is that these habits may yield incorrect outcomes when agents are forced to deviate from their typical trajectories, prompted by changes in the environment. This paper presents a mathematical characterization of this phenomenon, termed policy confounding, and illustrates, through a series of examples, the circumstances under which it occurs.