On the Structural Non-Preservation of Epistemic Behaviour under Policy Transformation
This work addresses the problem of understanding how internal information affects policy behavior in RL for researchers, but it is incremental as it builds on existing formalizations with controlled experiments.
The paper formalizes behavioral dependency in reinforcement learning under partial observability and establishes that probe-conditioned behavioral separation is not preserved under convex aggregation or gradient-based optimization, with experiments showing behavioral distance decreases in bandit and gridworld settings.
Reinforcement learning (RL) agents under partial observability often condition actions on internally accumulated information such as memory or inferred latent context. We formalise such information-conditioned interaction patterns as behavioural dependency: variation in action selection with respect to internal information under fixed observations. This induces a probe-relative notion of $ε$-behavioural equivalence and a within-policy behavioural distance that quantifies probe sensitivity. We establish three structural results. First, the set of policies exhibiting non-trivial behavioural dependency is not closed under convex aggregation. Second, behavioural distance contracts under convex combination. Third, we prove a sufficient local condition under which gradient ascent on a skewed mixture objective decreases behavioural distance when a dominant-mode gradient aligns with the direction of steepest contraction. Minimal bandit and partially observable gridworld experiments provide controlled witnesses of these mechanisms. In the examined settings, behavioural distance decreases under convex aggregation and under continued optimisation with skewed latent priors, and in these experiments it precedes degradation under latent prior shift. These results identify structural conditions under which probe-conditioned behavioural separation is not preserved under common policy transformations.