THE AUTHOR:
Sarah Sedky, Senior Associate at K&A
For years, the conversation around artificial intelligence (“AI”) in international arbitration has largely focused on the arbitration itself: AI-assisted legal research, document review, drafting, predictive analysis, and the extent to which arbitrators may use AI in performing their mandate. Construction disputes suggest that we may be asking the question too late.
Long before a request for arbitration is filed, a major construction project has already generated an enormous digital record, with programmes and programme updates, progress reports, correspondence, notices, payment applications, variation logs, drawings, building information modeling (“BIM”) data, meeting minutes, photographs, and data generated by digital project-management systems. When a dispute eventually arises, counsel and experts are asked to reconstruct several years of project history from the record.
AI has obvious potential to accelerate that exercise. However, its more significant impact may be upstream. If AI becomes embedded in the way projects are administered, monitored and documented, it will not simply help lawyers analyze the evidentiary record, but may change how that record is created in the first place. Construction arbitration therefore offers a useful testing ground for the next phase of AI in dispute resolution.
Construction Disputes Have a Data Problem Before They Have an AI Problem
The complexity of construction arbitration is hardly new. The International Chamber of Commerce (“ICC“) Commission’s Report on Construction Industry Arbitrations has long recognized that construction disputes are factually and technically complex, involve large quantities of documents, and require tribunals to grapple with programmes, critical path analysis, computation of claims, and competing expert evidence. What has changed is the scale and nature of the underlying information.
Modern projects increasingly operate through interconnected digital systems, where the evidentiary record may no longer consist simply of documents deliberately created by individuals but include continuously updated schedules, digital models, structured databases, and system-generated information.
AI is therefore particularly attractive in construction disputes, as it addresses precisely the problem that makes those disputes difficult: volume coupled with chronology.
A tool capable of interrogating thousands of project records, correlating correspondence with programme updates, identifying recurring references to particular delay events, or tracing a variation through successive project documents could materially reduce the time spent assembling the factual architecture of a claim.
The broader arbitration community clearly expects this type of use to grow. The 2025 Queen Mary University of London and White & Case International Survey found that respondents expected to use AI for research, data analytics, and document review, while saving time was identified as the principal driver of increased AI adoption.
Construction, however, presents a more interesting possibility. What happens when the same analytical capability moves from the arbitration file to the live project?
From Dispute Resolution to Dispute Anticipation
Consider a familiar construction dispute. An event occurs. Months later, the parties disagree about whether it caused critical delay. Years later, counsel and experts reconstruct the event using contemporaneous documents, competing programmes, and witness recollection. Now consider a project environment in which technology continuously correlates programme movements, instructions, notices, site records, and cost information as the project progresses.
The question is no longer merely whether AI can help prove the claim; it is whether the technology can identify the emerging dispute while the project team still has an opportunity to do something about it. In fact, this possibility is already entering mainstream industry discussions. The International Federation of Consulting Engineers (“FIDIC”) has considered AI across procurement, contract administration, performance tracking, claims management, and dispute resolution, while repeatedly emphasizing the importance of reliable underlying data and human oversight.
The implications could be significant: (i) better real-time identification of contractual events could improve notice compliance; (ii) emerging delays could be analyzed closer to the event rather than retrospectively; (iii) gaps in project records could be identified while the relevant personnel are still available; and (iv) parties could test competing positions before those positions harden into formal claims.
In that sense, the most successful use of AI in construction arbitration might ultimately be an arbitration that never takes place. There is, however, an important counterpoint, as more information does not necessarily mean better evidence.
The Next Evidentiary Fight May Be about Provenance, Not Volume
Construction lawyers are accustomed to asking who authored a document, when it was created, and whether it reflects contemporaneous knowledge. AI-assisted project administration complicates those questions.
If a project report incorporates machine-generated analysis, what were the underlying inputs? If an automated system identifies a delay event, what methodology produced that conclusion? If an AI tool categorizes correspondence or generates a chronology during the project, can that output later be reproduced and tested? What happens when the underlying model has changed by the time the arbitration begins?
The familiar principle of “garbage in, garbage out” acquires evidentiary significance. A highly sophisticated analysis of incomplete or incorrectly classified project data may simply produce a more persuasive version of the wrong answer. Similarly, an AI-generated chronology may appear objective while embedding assumptions that neither counsel, the expert, nor ultimately the tribunal can readily identify.
This may shift the discussion in construction arbitration from what the project records show to how the information was collected, structured, processed, and interpreted. Data governance may therefore become part of dispute preparedness.
Parties implementing AI-enabled project systems should be thinking not only about operational efficiency, but also about preservation, auditability, and provenance. As a result, the digital infrastructure of the project may eventually become part of the evidentiary infrastructure of the arbitration.
The Expert Moves Center Stage
The same tension arises in expert evidence, with delay and quantum analysis being obvious candidates for AI assistance, particularly as large datasets, repetitive calculations, and the comparison of multiple sources of project information are areas in which technology can offer substantial efficiencies.
Yet an expert’s function is not merely computational. The difficult questions in construction disputes often involve judgment: (i) which programme provides the appropriate analytical baseline; (ii) whether an event actually drove critical delay; (iii) whether two causes operated concurrently; (iv) whether particular costs were caused by the event claimed; and (v) whether the assumptions underlying an analysis are reasonable.
AI may enable experts to analyze more information faster, but it does not eliminate the need to explain why a conclusion follows from that information. Indeed, as analytical tools become more powerful, tribunals may need experts to be more transparent about methodology. The relevant question will be not simply “Did you use AI?” but “What did the tool do, what did you ask it to do, what information did it rely upon, and which parts of the ultimate opinion remain yours?”
That distinction mirrors the direction emerging in arbitration more broadly. The 2025 Chartered Institute of Arbitrators (“Ciarb”) Guideline on the Use of AI in Arbitration stresses that users remain responsible for AI outputs and that arbitrators cannot delegate their decision-making function. Similar principles of responsibility, competence, and non-delegation appear in the Silicon Valley Arbitration & Mediation Center (“SVAMC”) Guidelines and the Vienna International Arbitral Centre (“VIAC”) on AI.
Construction Is Already Becoming the Laboratory
Perhaps the clearest indication of where this is heading is that one of the most ambitious institutional experiments with AI has begun with construction disputes.
In November 2025, the American Arbitration Association-International Centre for Dispute Resolution (“AAA-ICDR”) launched its AI Arbitrator for eligible two-party, document-only construction cases. The system analyzes submissions, identifies issues, and prepares a proposed award, while the parties are given an opportunity to validate its understanding of their submissions and a human arbitrator reviews the analysis and ultimately issues the award.
The choice of construction is revealing. A document-only construction case offers precisely the environment in which AI’s strengths can be tested against arbitration’s fundamental safeguards, with a dense documentary record, recurring categories of claims, and significant pressure to reduce time and cost, but a continuing need for human judgment and accountability.
The experiment also points towards a broader question. If AI can assist with the record at the beginning of the project, the claims when the dispute crystallizes, the expert analysis during proceedings, and the tribunal’s review at the end, where exactly should the human decision-maker enter, and where must the machine stop?
The Real Question Is Not Whether AI Belongs in Construction Arbitration
The debate is therefore moving beyond whether AI should be used in arbitration. It already is, and the available evidence suggests that its use will increase. For construction practitioners, the more useful questions are becoming narrower. Which parts of a project or dispute should be automated? Which AI-generated outputs should be preserved? When should the use of AI be disclosed? How should experts explain AI-assisted analysis? What degree of technological literacy should be expected from counsel and tribunals? At what point does assistance become delegation?
Construction arbitration is therefore well placed to provide answers to those questions, as it sits at the intersection of two industries undergoing rapid technological change. The opportunity is considerable, as AI allows project teams to identify disputes earlier, counsel to understand complex records faster, experts to interrogate larger datasets, and tribunals to navigate cases more efficiently. However, efficiency is not the same as reliability, and more data is not necessarily more truth.
The real promise of AI in construction disputes may therefore be less about replacing human work and more about supporting better-informed judgment throughout the process, from project data to arbitral award.
Conclusion
XXXXXXXXXXXXXXXXXXXXXXXXXX




