Of Errors and Echoes: Responsibility-Aware Credit Assignment for Tool-Use Agents

Key points

• We introduce Error-Chain-Aware Optimization (ECHO), a responsibility-aware credit assignment framework that separates responsibility identification from credit allocation.1
• A frozen judge combines retrospective trajectory analysis with decision-time assessment from the policy-visible prefix to distinguish newly introduced errors, inherited consequences, missed corrections, and successful corrections.1
• ECHO then counts shared downstream consequences once across responsible decisions and constrains rewards for repairing self-induced errors through source-funded recovery, producing bounded action-level supervision alongside terminal supervision.1
• These results support decision responsibility as a useful basis for fine-grained credit assignment in tool-use agents.1

Sources (1)

• [1]Of Errors and Echoes: Responsibility-Aware Credit Assignment for Tool-Use Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 12:52 AM

  Reinforcement learning improves tool-use agents, yet fine-grained credit assignment remains challenging even with step-level feedback.

  We introduce Error-Chain-Aware Optimization (ECHO), a responsibility-aware credit assignment framework that separates responsibility identification from credit allocation.

Extractive summary: sentences quoted from the sources.