Arbitration Association Pilots AI Judges

- The pilot program is initially limited to two-party, documents-only construction cases, a sector chosen for its technical and data-heavy nature which is well-suited for machine learning. The AI model was trained on over 1,500 past construction arbitration awards and refined with input from construction law experts. - The system, developed in collaboration with McKinsey's Quantum Black AI team, uses a multi-agent system of about 20 AI agents to analyze claims, synthesize evidence, and draft a recommended award. This approach aims to reduce case time from a typical 60-75 days to 30-45 days. - A "human-in-the-loop" framework is central to the process; a human arbitrator, trained for AI-led cases, reviews, revises if necessary, and ultimately issues the final binding order. Parties to the dispute also serve as a level of oversight by validating the AI's understanding of their submissions. - The initiative is led by AAA President and CEO Bridget McCormack, a former Chief Justice of the Michigan Supreme Court, who views AI as a crucial tool to address the "civil justice crisis" where 92% of Americans cannot afford legal assistance. - Participation in the AI arbitrator program is entirely voluntary for both parties. If either party opts out, the case proceeds through the traditional human-only arbitration process. - In March 2025, the American Arbitration Association issued guidance for arbitrators on the use of AI tools, emphasizing their existing ethical obligations regarding accuracy, fairness, confidentiality, and independent judgment. - While the initial rollout is narrow, the AAA plans to expand the AI arbitrator's availability to more complex areas such as consumer and employment disputes, as well as cases with higher values.