Mealey’s International Arbitration Report recently asked industry experts and leaders for their thoughts on the American Arbitration Association’s International Centre for Dispute Resolution (AAA-ICDR) announcement of the launch of an AI arbitrator for use in construction arbitration. https://www.adr.org/ press-releases/aaa-icdr-to-launch-ai-native-arbitratortransforming-dispute-resolution/. 

We would like to thank the following individuals for sharing their thoughts on this important issue. 

• Omer Er, Partner, Michelman & Robinson, New York 
• Antoine K.F. Smiley, Partner, Reed Smith, Austin, Texas 
• Eugenie Rogers, Partner, Reed Smith, Dallas 
• Lisa Richman, Partner, McDermott Will & Emery, Washington, D.C. 
• Jessica Sabbath, Counsel, McDermott Will & Emery, Atlanta 
• Daniel R. Guadalupe, Partner, Pashman Stein Walder Hayden P.C., Hackensack, N.J. 
• Albert Bates Jr., Partner, Troutman Pepper Locke LLP, Pittsburgh 
• R. Zachary Torres-Fowler, Partner, Troutman Pepper Hamilton Locke LLP, Philadelphia and New York 
• Andrew Drennan, Managing Director, Alvarez & Marsal’s Disputes and Investigations, London 
• J. Laurens Wilkes, Partner and Chair of Infrastructure & Projects – Global Disputes, Winston & Strawn, Houston
• Gustavo J. Membiela, Partner, Winston & Strawn, Miami 
• Madison K. Haueisen, Associate, Winston & Strawn, Houston

Mealey’s: What are your views on the legal, ethical and practical implications of the American Arbitration Association’s International Centre for Dispute Resolution (AAA-ICDR) announcing the launch of an AI arbitrator for use in construction arbitration cases?

Er: The American Arbitration Association–International Centre for Dispute Resolution (AAA-ICDR)’s introduction of an AI arbitrator for documents-only construction disputes represents a significant paradigm shift, not merely in technology adoption, but in the institutional delegation of judicial function. Its implications necessitate rigorous scrutiny across legal, ethical, and practical domains.

Legal And Enforceability Challenges

The primary legal friction point revolves around due process and the enforceability of awards under the lex arbitri and conventions like the New York Convention. Although the AAA-ICDR maintains a “human-in-theloop” framework where a human arbitrator validates the AI-generated draft, the integration of algorithmic reasoning invites challenges to the tribunal’s proper constitution and impartiality. Parties may seek to vacate an award by alleging that the AI’s influence constitutes a form of arbitrator misconduct or improper delegation of the decisional mandate. Crucially, institutions must establish non-waivable disclosure protocols detailing the extent of the AI’s involvement and the dataset used, ensuring that party consent is both informed and explicit. Absent such transparency, a successful challenge based on a violation of public policy in the enforcement jurisdiction becomes plausible.

Ethical And Algorithmic Integrity 

Ethically, the initiative confronts the enduring problem of algorithmic opacity. While the AI system is trained on over 1,500 construction awards to promote consistency, this corpus inherently codifies historical human biases and prevailing adjudicatory trends. The risk is not merely error, but the systemic entrenchment of existing disparities or the suppression of novel legal arguments that were or are not present in the training data. For the human arbitrator, the challenge shifts from de novo decision-making to the critical assessment of machine-generated reasoning. To preserve the principle of independent judgment, the human element must function as an active editor and validator against bias, rather than a passive endorser of an ostensibly efficient but opaque output.

Practical And Institutional Implications 

From a practical perspective, the application to lowto-mid-value, documents-only construction claims is strategically sound. These disputes are characterized by predictable fact patterns and a high premium on speed, making them ideal candidates for leveraging AI to improve the cost-dispute ratio. Projected savings of 20–35% in time and 30–50% in cost could significantly enhance access to justice for smaller contractors and subcontractors. 

Ultimately, the AAA-ICDR is defining the early norms of AI-assisted adjudication. The success of this model and its expansion into other sectors will depend entirely on the robustness of its governance framework, ensuring that efficiency is never pursued at the expense of fairness, transparency, and the fundamental integrity of the arbitral process.

Smiley and Rogers: AI decision-making in arbitration was a matter of time. The first-of-its-kind AAAICDR model is narrow: two-party, documents-only construction disputes, where an AI system produces a draft award that a named human arbitrator can amend and sign. Even still, the AAA-ICDR is the first major institution to put its rules and reputation behind an AI-assisted decision-maker.

Consent remains the cornerstone, as users decide whether to utilize AI Arbitrator, and both parties must opt-in. Parties may review and confirm that the technology has correctly summarized their claims and submissions.

Human oversight and involvement in this early model is critical. Legally, it anchors the award in a human decision-maker, easing concerns about whether an award by a wholly non-human “tribunal” would be enforceable under the New York Convention and national laws. Practically, parties are unlikely to entrust highstakes disputes to an unreviewable algorithm until the technology earns real trust. Ethically, retaining a human arbitrator ensures that someone with professional duties and moral judgment stands behind the decision, can correct biased or context-blind model outputs, and provides a real person whom parties can hold to account. Humans remain integral to the process. 

A deeper question is whether one can train a model to be “fair and reasonable” or “just and equitable” in the sense those phrases are used in many construction contracts, though if parties want machine-applied formulae, they tend not to draft in those terms. In time, however, party autonomy—the bedrock of arbitration—should allow courts to give effect to clauses explicitly entrusting those evaluative standards to AI-assisted decision-making. 

The use of AI in arbitration will also reshape advocacy. Practitioners are accustomed to persuading human arbitrators; the next skillset will be crafting submissions so that AI models are persuaded by them. It is easy to imagine parties using “shadow” models that imitate the AI Arbitrator to test and iterate arguments in advance, with a risk that those with the most sophisticated tools gain a new procedural advantage. 

For construction disputes, the upside is clear: Many lower value claims are paper-heavy, time-sensitive and uneconomic to litigate fully. AI-assisted decisions could become a digital analogue of dispute boards or statutory adjudication, providing rough, interim justice that keeps projects moving and reserving full arbitration for final decisions or for complex, high-stakes issues. In international work, we expect to see AI-assisted tiers in multistep dispute resolution clauses handling, for example, routine change-order and delay claims up to an agreed monetary threshold, with traditional tribunals dealing with higher value claims. The humans involved will continue to play essential roles to verify the technology’s outputs and ensure that it does not supplant human arbitrators in delivering just results.

Richman and Sabbath: The AAA-ICDR estimates that using an AI arbitrator will save at least 30-50% in costs and 25-30% in time. There might be other use cases as well not mentioned by the AAA-ICDR. For example, an AI arbitrator might be used for mock arbitrations.

An important drawback, however, is that an AI arbitrator is only suited for straightforward cases that can be decided by documentary evidence without witness testimony. But an AI arbitrator may not be able to determine the relative weight to give to contradictory or unreliable documents. These limitations will foreclose its use in many cases. Moreover, an AI arbitrator may not ultimately deliver the intended cost and time savings because AI decisions (1) must be reviewed and confirmed by a human arbitrator, and (2) may face a greater risk of challenge and potential vacatur.

AAA-ICDR has represented that an AI arbitrator was developed, “trained,” and refined on over 1,500 arbitration awards. Notwithstanding its anticipated reliability, human arbitrators are required to review the AI arbitrator’s draft award and to provide oversight and validation of the award. Indeed, AAA-ICDR’s Guidance on the use of AI Tools “requires arbitrators to retain complete control over decision-making.”  

The scope of human review and amount of deference to be given to the AI determination is not yet known and may vary among overseeing arbitrators. Relying on an AI determination without conducting an independent review of the evidence could result in a greater likelihood of error. Alternatively, if human arbitrators review and analyze the evidence themselves, using an AI arbitrator to make an initial determination may not ultimately render the process more efficient and economical.

Another consideration is that decisions rendered by an AI arbitrator will be highly scrutinized and may carry a greater risk of being challenged and potentially set aside than those rendered by a human. This is particularly true if the decision is not well-supported and reasoned, or if the seat of the arbitration has laws that require a human arbitrator as, for example, in France and the UAE. See Article 1450 of the French Code of Civil Procedure; see also Article 10(1) of the UAE’s Federal Law No 6 of May 3, 2018. This could result in vacatur of AI decisions on the basis that the procedure did not comply with the governing law.

AI has the potential to revolutionize arbitration procedure. But whether it will result in the intended cost and time savings remains to be seen.

Guadalupe: As a cautious advocate of AI, this is good news as an initial experiment. The AI arbitrator is limited to documents only and two party cases, and human arbitrators are overseeing the process to ensure fairness and accuracy. As chair of the AI committee in my firm, last year I predicted that AI arbitrators would arise to decide small disputes, consumer retail claims and other “less risky” areas of decision-making. And here we are. AI offers thoroughness, efficiency and a deep, multidimensional thinking that is soon catching up, and perhaps rivaling at times, human judgment. AI learns from and by itself and, each year, is becoming more reliable due to the relentless “machine-learning” process.

Of some concern is the data on which this AI arbitrator will be trained. Attorneys would have to learn how the AI arbitrator is trained, what are the applicable algorithms and how they would be applied. Will AI awards reflect biases of previous awards, such as perhaps favoring owners in home improvement disputes versus contractors (given the ubiquitousness of deceptive practices/ consumer fraud statutes). If so, eliminating and avoiding these biases will require effort.

Arbitration is based on consent, and before giving it, there would be an ethical obligation on attorneys to understand what technology is being used to cause the AI arbitrator to review and interpret the documents in the dispute. Another issue is whether state and federal courts will uphold awards issued by the AI arbitrator. How will statutory standards for confirming, vacating or modifying awards be applied? And who would be responsible for a defective award, the AI or its human “supervisors” or both? If a state or federal court, by way of expert testimony, finds that an AI arbitrator was partial, what impact will this ultimately have on the perception of fairness of ICDR awards?

These challenges can be addressed but they need to be identified now and trustworthy solutions developed. Despite these queries (pun intended), we have to accept the advent of this technology and the many benefits it will bring to adjudication of disputes, such as speed, accuracy and depth of knowledge. However, just as the confirmation process serves as a check on human awards, thoughtful guardrails should be established, such as human review or oversight, disclosure and transparency regarding the foundation for the AI arbitrator and statutory reform on standards to confirm, vacate or modify AI-issued awards.

Bates and Torres-Fowler: AAA-ICDR’s announcement of an AIenhanced arbitrator for twoparty, documentsonly construction disputes marks a cautious but noteworthy development in the arbitration industry’s efforts to respond to user demands for greater cost efficiencies. The system has been trained on case files from approximately 1,400 AAA construction documentsonly matters during a recent two-year period and is limited to documentsonly cases where the parties opt-into AI enhanced arbitration. Critically, fewer than ten human arbitrators are available; one is assigned to each case to oversee every stage of the process—including reviewing, editing, and challenging the AIgenerated draft award—and to execute and take responsibility for the final award. The process intends to enhance efficiency by augmenting human arbitration with AI—it is not (as least yet) a standalone “AI arbitrator.” The AI enhanced arbitrator option went live in early November 2025 and remains in a beta test at this stage. The use of AI in the arbitral decision-making process is a significant innovation; however, even with human arbitrators, the process raises legal, ethical, and practical concerns that all participants must consider.

Legal questions surrounding the enforceability of AI arbitrator awards will be central. While the awards in documents-only cases are skeletal in comparison to reasoned awards, party consent, procedural fairness, and transparency will remain the cornerstones of judicial review, and some courts are likely to approach AI-augmented awards with a dose of skepticism. AAAICDR’s reliance on human oversight is both prudent and necessary at this early stage given due process concerns and the institutional protections contemplated by the AAA-ICDR. Further, the AAA-ICDR has taken proactive steps to provide transparency about the specific model implemented, the specific data set utilized, the training and data testing performed, including the extensive human review and modification of outputs, and the training data and methodology.

Ethically, the hardest questions revolve around the issue of bias, disclosure, and confidentiality. Ostensibly, training the AI arbitrator with prior construction awards can encode historical judgment patterns that improve predictability but risk perpetuating embedded biases. Ongoing validation and documented quality controls—particularly from human oversight—will be essential to keep the model’s recommendations fair across participants, claims, and regions. Separately, maintaining confidentiality protections over the AI arbitrator tool will be paramount to ensure that parties maintain confidence and understanding that the information disclosed to the model is carefully controlled.

The fit for lower dollar, documents-only construction disputes is sensible: AI-assisted synthesis and issue framing can accelerate timelines, enhance efficiency, and reduce legal spending. Still, the timeline of the uptake is uncertain. Ultimately, success with this specific caseload will turn on demonstrable accuracy, transparent governance, and predictable outcomes that are competitively priced. The next chapters of this story are yet to be written but will develop very rapidly, as AI enhancements that can assist arbitrators and counsel with the efficiency of the arbitration process will inevitably become more widely available.

Drennan: The AAA-ICDR has announced that its AI Arbitrator will be available from November 2025. The tool is designed to handle two-party documents in construction arbitration cases. 

According to the AAA-ICDR, “the AI Arbitrator aims to deliver fast, cost effective, and trusted dispute resolution while maintaining fairness and transparency."

The key features sound impressive, such as¹ :

• Human in the loop framework: Human oversight is integral to the functionality. 
• Training and development: The AI Arbitrator has been trained on 1,500 AAA awards. 
• Efficiency and cost savings: The AI Arbitrator reports estimates of 20%-25% in time savings and 35%-45% in cost savings. 
• Ethical standards and privacy: The system adheres to rigorous ethical standards, ensuring unbiased outcomes.

Future Expansion: This scheme could potentially extend to other industries and case types.

As a construction expert and an early adopter of AI, I regularly use it to check for errors, summarise data, and search documents. AI’s usefulness has been a revelation and is comparable in value to the addition of another team member.

However, I am skeptical about its use in certain circumstances. A key issue is that construction arbitration is complex, and terminology nuanced. I have not seen evidence of AI being able to decipher these layers yet.

Without rigorous testing, clients may not fully trust an AI arbitrator. As arbitration is often a last resort where either of the parties may have perilous financial outcomes should they lose, the question as to whether they would gamble on an AI decision versus a human remains to be seen.

Endnotes 

1. https://www.adr.org/ai-arbitrator

Wilkes, Membiela and Haueisen: The AAA-ICDR’s new AI Arbitrator is a voluntary “analytical tool”¹ for documents-only construction disputes. After summarizing claims and submissions—and allowing party verification—the system, trained on AAA construction awards, drafts a recommended award for a human arbitrator to adopt or revise.² With these guardrails, the AI Arbitrator is being positioned to “call balls and strikes,” as Chief Justice Roberts put it³, subject to a replay review. While results, and case studies on those results, will help evaluate this system, it very likely will promote consistency and efficiency in awards for documents-only matters.

That said, the harder question is whether AI can “pitch or bat”⁴ in disputes involving nuanced and novel legal issues or complex facts. The AAA-ICDR acknowledges that the tool is not suitable for cases with “live witnesses or complex factual issues,” though its use may expand as the system matures.⁵ Ronald Dworkin’s “Hercules”—an idealized judge of “superhuman intellectual power and patience” who seeks the single right answer by selecting the interpretation that best fits and morally justifies the body of legal authorities—offers a useful lens.⁶ While AI’s processing power suggests a Herculean capacity to integrate vast authorities, practical computing constraints and AI’s inability to reliably assess credibility and weigh principles and the potential for AI hallucination prevent it from currently replicating Hercules in complex cases.

And, notably, the AAA-ICDR presently does not permit parties to review the AI-generated draft award given to the human arbitrator.⁷ That may be acceptable in the current role where AI is calling balls and strikes, but in complex cases, disclosure of the draft— or at least the AI’s role, rationale, and the arbitrator’s reliance—may be warranted to preserve transparency and legitimacy and ensure the duty of competency is satisfied and the required oversight maintained.

Additionally, even seemingly more straightforward document-only matters oftentimes do involve a certain level of decision-maker discretion. While “judicial activism” and “rough-justice” arbitral results are often criticized, decisions grounded in practical considerations are often embraced. The failed nomination of Robert Bork underscores how strict originalism, in tension with evolving norms, raised concerns regarding the legitimacy of potential future decisions.⁸ An AI Arbitrator trained only on some types of issues for some things creates similar questions, which will need to be evaluated over time. 

In sum, the AAA-ICDR’s AI Arbitrator certainly seems well-equipped to increase efficiency and help human arbitrators “call balls and strikes” in documents-only construction cases. But as the task shifts to pitching and batting—credibility disputes and novel legal questions—and the AI Arbitrator’s role expands, informed consent, clear disclosures, robust attorney oversight, and other ethical guardrails will be essential to preserving fairness and arbitral legitimacy.

Endnotes 

1. Am. Arbitration Ass’n–Int’l Ctr. for Disp. Resol., AI Arbitrator: A New Path to Dispute Resolution (Press Release, Oct. 29, 2025), at 
2. Id. 
3.  Nina Totenberg, Roberts Hearings Begin with Talk of Agendas, NPR (Sept. 12, 2005), https://www.npr.org/2005/09/12/4843158/ roberts-hearings-begin-with-talk-of-agendas 
4. Id. 
5. Am. Arbitration Ass’n–Int’l Ctr. for Disp. Resol., supra note 1, at 2. 
6. Ronald Dworkin, Law’s Empire 225–26, 239 (1986) 
7. Am. Arbitration Ass’n–Int’l Ctr. for Disp. Resol., supra note 1, at 2. 
8. Clyde Haberman, Want to Know Where Supreme Court Nominees Stand? Don’t Bother Asking, N.Y. Times (Mar. 19, 2017), https://www.nytimes. com/2017/03/19/us/supreme-court-bork-hearings. html. 

* This report was first published in Mealey’s International Arbitration Report. 

[Editor’s Note: Copyright © 2025, LexisNexis. All rights reserved.]