Calibrating LLM Explanations to Human Decisions with Reinforcement Learning

Key points

• Explanations of LLM outputs are considered an important way to make model behavior legible, helping human users make good decisions on downstream tasks.1
• However, default training approaches for improving explanations target model task performance or human preference rather than human decision quality.1
• Our approach calibrates a human decision simulator on observed human decisions, applying a proxy reward model to optimize the LLM to generate explanations that help people to make good decisions.1
• By applying the HD-RL framework to the open-source Qwen 3.6 model, we generate optimized explanations that improve human performance, achieving a GLMM-estimated human accuracy of 63.7% in a randomized between-subjects experiment with repeated measures and outperforming RLVR, RLAIF, and other baselines by over 8%.1

Sources (1)

• [1]Calibrating LLM Explanations to Human Decisions with Reinforcement Learning
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 06:59 PM

  To bridge this gap, we introduce HD-RL, a reinforcement learning framework for optimizing LLM explanations for human decision making.

  Explanations of LLM outputs are considered an important way to make model behavior legible, helping human users make good decisions on downstream tasks.

Extractive summary: sentences quoted from the sources.