What Do World Models Learn in RL? Probing Latent Representations in Learned Environment Simulators
This work addresses the interpretability of world models for RL researchers, providing incremental insights into how these models encode environment states.
The study investigated the internal representations of world models in reinforcement learning by applying interpretability techniques to two distinct architectures, IRIS and DIAMOND, trained on Atari games, finding that these models develop linearly decodable representations of game state variables like object positions and scores, with causal interventions confirming functional use.
World models learn to simulate environment dynamics from experience, enabling sample-efficient reinforcement learning. But what do these models actually represent internally? We apply interpretability techniques--including linear and nonlinear probing, causal interventions, and attention analysis--to two architecturally distinct world models: IRIS (discrete token transformer) and DIAMOND (continuous diffusion UNet), trained on Atari Breakout and Pong. Using linear probes, we find that both models develop linearly decodable representations of game state variables (object positions, scores), with MLP probes yielding only marginally higher R^2, confirming that these representations are approximately linear. Causal interventions--shifting hidden states along probe-derived directions--produce correlated changes in model predictions, providing evidence that representations are functionally used rather than merely correlated. Analysis of IRIS attention heads reveals spatial specialization: specific heads attend preferentially to tokens overlapping with game objects. Multi-baseline token ablation experiments consistently identify object-containing tokens as disproportionately important. Our findings provide interpretability evidence that learned world models develop structured, approximately linear internal representations of environment state across two games and two architectures.