AI as arbitrator for certain low-value construction disputes at AAA-ICDR

AI arbitrator – current and future uses

The American Arbitration Association’s international division, the International Centre for Dispute Resolution (AAA-ICDR), has announced that an AI arbitrator will be available from November 2025 in documents-only construction disputes (potentially limited to cases involving $25,000 or less under the AAA Construction Rules). 

Determination by an AI arbitrator will be available as an opt-in pathway within AAA-ICDR administered cases, requiring the consent of both parties. This builds on previous AAA-ICDR AI innovations, including (i) offering its panel of arbitrators access to Clearbrief, an AI tool designed to allow aid in the drafting and analysis of legal submissions; (ii) its October 2024 announcement of the AI-enabled AAAi Panelist Search to find arbitrators and mediators; (iii) the 2025 AAA-ICDR Guidance on Arbitrators’ Use of AI Tools, and (iv) launching a professional development course on the use of AI for law firm leaders. AAA-ICDR explained that the AI arbitrator has been trained on over 1,500 construction arbitration awards, and developed with input from construction disputes practitioners and arbitrators from AAA-ICDR’s Construction Panel, in collaboration with QuantumBlack, AI by McKinsey.

The AI arbitrator reportedly will deliver a reasoned decision based on the documentary record, with tools available to explain the basis of the outcome. The AI arbitrator will evaluate the merits of claims, generate explainable recommendations, and prepare draft awards, appearing to have greater functionality than an AI assistant previously announced by the Guangzhou Arbitration Commission.  It also appears to go beyond AAA-ICDR’s Guidance on the use of AI Tools which “requires arbitrators to retain complete control over decision-making”.  AAA-ICDR explains that humans will remain in-the-loop to evaluate reasoning as part of its focus on validating results, safeguarding trust, transparency and due process.  

AAA-ICDR has also confirmed that it plans to expand eligibility for the AI arbitrator to other industries (with the insurance sector mentioned as the next possible opportunity), dispute types and higher-value claims in 2026.

Opportunities for construction disputes 

AAA-ICDR expects the AI arbitrator to dramatically reduce the time and cost of arbitral proceedings, predicting that there will be at least 30 to 50% savings in costs and 25 to 30% savings in time: material savings in a sector that is especially reliant on cash flow, tight timelines and the ability to continue projects. These potential savings arguably go even further than previous initiatives targeting sectors that need swift dispute resolution to enable relationships to continue (including e.g., the UNCITRAL Model Clauses on Specialized Express Dispute Resolution; see our previous article on this topic).

Certain construction disputes appear well-suited as a use case to test AI arbitrators, especially those dependent on documentary records, schedules, quantification, and contractual mechanics, rather than on novel points of law. 

Risks of AI arbitrator

But use of an AI arbitrator, even with the parties’ consent, does not come without risks, including those inherent to the general use of generative AI, like data privacy and confidentiality, the risk of hallucinations, data biases and regulatory compliance. 

Relying on an AI arbitrator could lead to an increased risk of set aside and enforcement issues. Many arbitration laws and rules work on the assumption that human arbitrators exercise independent judgment, and some national arbitration laws go as far as requiring a human arbitrator (e.g. Article 1450 of the French Code of Civil Procedure or Article 10(1) of the UAE’s Federal Law No 6 of 3 May 2018). The EU AI Act classes decision making as a ‘high-risk’ activity, that requires human oversight. In circumstances where AI use by arbitrators has already led to an application to vacate an award, the risk exists of an arbitration award made by an AI arbitrator being challenged under the New York Convention on the basis that the arbitral procedure was not in accordance with the law of the country where the arbitration occurred (notwithstanding the parties’ consent to AI arbitrator use). Non-transparent or non-explainable reasoning in an award may also increase the risk of a set-aside application. 

AAA-ICDR is alive to these risks and it will be interesting to see the details of their mitigation methods as they unveil further information about the AI arbitrator in the coming weeks. 

Conclusions

Document-only, lower-value construction disputes appear to be an apposite initial use case for AI arbitrators (a concept hotly debated in recent years). It will be fascinating to see whether this innovation is adopted by parties and is capable of achieving significant time and cost benefits, or whether it introduces new time-consuming procedural skirmishes, especially at the enforcement stage, even by parties who expressly consented to this process. For parties, it will be important to consider and obtain advice on how to avoid these issues (including by carefully considering local law in jurisdictions of the seat and of potential enforcement) and mitigate risks generally associated with AI use in decision making. Many international arbitration institutions have already started considering and addressing these issues, with the SCC, VIAC, CIArb, SVAMC and AAA-ICDR issuing guidelines or notes iterating the importance of effective human oversight of the use of AI and human control over decision-making in arbitration proceedings.