Enhancing RL Safety with Counterfactual LLM Reasoning

TL;DR

This paper introduces a method that uses counterfactual reasoning with large language models to improve the safety and explainability of reinforcement learning policies after training.

Contribution

It presents a novel approach combining counterfactual LLM reasoning with RL safety enhancement, which is a new application of LLMs in RL safety.

Findings

Improves RL policy safety post-training.

Enhances explainability of RL policies.

Demonstrates effectiveness of counterfactual LLM reasoning.

Abstract

Reinforcement learning (RL) policies may exhibit unsafe behavior and are hard to explain. We use counterfactual large language model reasoning to enhance RL policy safety post-training. We show that our approach improves and helps to explain the RL policy safety.