The Interpretability of Codebooks in Model-Based Reinforcement Learning is Limited

TL;DR

This paper critically examines whether vector quantization in model-based reinforcement learning offers true interpretability, finding that it is inconsistent, non-unique, and limited in aiding understanding of the model's concepts.

Contribution

The study provides empirical evidence that vector quantization does not reliably enhance interpretability in model-based reinforcement learning.

Findings

Codes are inconsistent across models

No guarantee of code uniqueness

Limited impact on concept disentanglement

Abstract

Interpretability of deep reinforcement learning systems could assist operators with understanding how they interact with their environment. Vector quantization methods -- also called codebook methods -- discretize a neural network's latent space that is often suggested to yield emergent interpretability. We investigate whether vector quantization in fact provides interpretability in model-based reinforcement learning. Our experiments, conducted in the reinforcement learning environment Crafter, show that the codes of vector quantization models are inconsistent, have no guarantee of uniqueness, and have a limited impact on concept disentanglement, all of which are necessary traits for interpretability. We share insights on why vector quantization may be fundamentally insufficient for model interpretability.