To the casual observer, arbitration might appear as a simple means of resolving a dispute.  Two parties that are unable to resolve a disagreement ask an independent third-party to step in to find a solution for them.  Viewed in this way, arbitration is not dissimilar to two friends asking a fellow patron at their local pub to settle a bet.  Any international arbitration practitioner would undoubtedly dismiss such a reductive analogy as ignorant of the complexities that exist at every stage of the arbitral process.  Commercial arbitration carries with it the challenges and intricacies of litigation before a court, compounded by the need for the parties to construct the metaphorical courthouse. 

The ability of parties to tailor the arbitral process to the precise needs of their case is an attractive feature when placed against the one size fits all structure of domestic courts.  Despite this benefit, the task of structuring an arbitration can be a daunting list of numerous significant decisions, including whether to utilize an arbitral institution, what procedural and substantive laws to apply, and where physical hearings, if needed, are to take place. 

Arguably, the most important of these decisions to both the conduct and outcome of the arbitration is the selection of the arbitrator or arbitral panel.1Douglas Pilawa, Sifting through the Arbitrators for the Woman, the Minority, the Newcomer, 51 Case W. Res. J. Int’l L. 395, 405 (2019).X1  As with all other decisions made during an arbitration, arbitrator selection requires parties to analyze relevant information and predict potential outcomes.2Charlie Morgan, Data Analytics in International Commercial Arbitration: Balancing Technology with the Human Touch, 9 Inside Arb. 23, 23–24 (2020).X2  This process is often frustrated by a lack of available resources and data.  Even when candidates are provided by an arbitral institution, specific information about how an individual has adjudicated prior cases or their arbitral practices, temperaments, and philosophies can be difficult to uncover.  Regular practitioners have traditionally relied on word-of-mouth recommendations, instinct, and intuition.  Less experienced practitioners, however, are often left with only the confidence that can be derived from an internet search or résumé review. 

In recent years, many areas of law have benefited from technological innovations related to artificial intelligence (“AI”).3See Jordan Bakst et al., Artificial Intelligence and Arbitration: A US Perspective, 16 Disp. Resol. Int’l 7, 12–15 (2022); Harry Surden, Artificial Intelligence and Law: An Overview, 35 Ga. St. U. L. Rev. 1305, 1331–32 (2019); Kathleen Peisley & Edna Sussman, Artificial Intelligence Challenges and Opportunities for International Arbitration, 11 N.Y. Disp. Res. Law. 35, 37 (2018).X3  This article will consider whether similar tools may be of value in the arbitrator selection process.  It will first provide brief explanations of artificial intelligence (II) and arbitrator selection (III).  The article will then consider both the benefits (IV) and areas of concern (V) related to the implementation of AI tools in identifying and selecting arbitrators.  Finally, it will provide some final thoughts on how AI tools will be used by practitioners moving forward (VI).

Artificial intelligence is difficult to define, with academics and other experts offering various, often conflicting, definitions.1European Commission High-Level Expert Group on Artificial Intelligence, A Definition of AI:  Main Capabilities and Scientific Disciplines at 1 (Dec. 2018), https://ec.europa.eu/futurium/en/system/files/ged/ai_hleg_definition_of_ai_18_december_1.pdf.X1  The European Commission High-Level Expert Group on Artificial Intelligence attempted to define AI as:

systems that display intelligent behavior by analyzing their environment and taking actions—with some degree of autonomy—to achieve specific goals.

AI-based systems can be purely software-based, acting in the virtual world (e.g. voice assistants, image analysis software, search engines, speech and face recognition systems) or AI can be embedded in hardware devices (e.g. advanced robots, autonomous cars, drones or Internet of Things applications).2European Commission High-Level Expert Group on Artificial Intelligence, A Definition of AI:  Main Capabilities and Scientific Disciplines at 1 (Dec. 2018), https://ec.europa.eu/futurium/en/system/files/ged/ai_hleg_definition_of_ai_18_december_1.pdf.X2

Generally, AI encompasses technologies that automate tasks that are traditionally thought to involve cognitive ability when performed by humans.3Surden, supra n. 3, at 1307–08.X3  At a foundational level, it is important to understand that modern AI does not actually use intelligence in the same manner as a living person, but, rather, performs complex tasks through pattern recognition and adaptation based on encoded knowledge, rules, and data.4Id. at 1308; Ryan McCarl, The Limits of Law and AI, 90 U. Cin. L. Rev. 923, 926 (2022).X4

The field of AI is currently dominated by the subfield of machine-learning, which relates to the ability of a model to improve its performance automatically with time and the accumulation of greater amounts of data.5McCarl, supra n. 7, at 926, 928; Surden, supra n. 3, at 1311; Eliza Mik, Caveat Lector: Large Language Models in Legal Practice, 19 Rutgers Bus. L. Rev.70, 77–78 (2024); What is Machine Learning (ML)?, UC Berkeley School of Information (June 26, 2020), https://ischoolonline.berkeley.edu/blog/what-is-machine-learning/.X5  The ability of a machine-learning algorithm to automatically update each time it runs allows for the analytical accuracy of the model’s outputs to increase as more data is analyzed.6UC Berkeley School of Information, supra n. 8.X6  Machine-learning tools often take the form of predictive models that generate predictions by recognizing patterns in large quantities of data.7McCarl, supra n. 7, at 928; Surden, supra n. 3, at 1311–12, 1314–15; Mik, supra n. 8, at 78; Artificial Intelligence (AI) v. Machine Learning, Columbia Engineering, https://ai.engineering.columbia.edu/ai-vs-machine-learning/.X7  The ability of these models is highly dependent on the quality and quantity of the data on which they are trained.8Mik, supra n. 8, at 78.X8

A perceived benefit of arbitration over litigation before a court is the ability of parties to select the finder of fact and law.1Sarah R. Cole, Arbitrator Diversity:  Can it be Achieved?, 98 Wash. U. L. Rev. 965, 974 (2021).X1  As arbitration is a creature of contract, the method of selecting the arbitrator or arbitral panel that will ultimately decide a dispute is dictated by the agreement of the parties.  In an ad hoc proceeding, parties are left to their own devices to identify mutually agreeable candidates.  Alternatively, parties may agree to allow an arbitral institution to administer their arbitration or to act as an appointing authority. 

Arbitral institutions maintain rosters of arbitrator candidates and offer differing methods for arbitrator selection and appointment.  For example, under the rules of the International Court of Arbitration of the International Chamber of Commerce (ICC), if the parties are unable to mutually agree upon an arbitrator or panel, the selection is made by the ICC.2ICC Arbitration Rules (2021), arts. 12–13.X2  In a contrasting approach, the American Arbitration Association (AAA) and International Centre for Dispute Resolution (ICDR), the international division of the AAA, use a list method for choosing an arbitrator or panel.3AAA Commercial Arbitration Rules and Mediation Procedures (2022), art. R-13; ICDR International Dispute Resolution Procedures (2021), art. 13.X3  Based on party input, the institution prepares a list of potential candidates from members of its roster of arbitrators.4AAA Commercial Arbitration Rules and Mediation Procedures (2022), art. R-13(a); ICDR International Dispute Resolution Procedures (2021), art. 13(6).X4  The parties are encouraged to agree on an arbitrator or arbitrators from the presented candidates.5AAA Commercial Arbitration Rules and Mediation Procedures (2022), art. R-13(a); ICDR International Dispute Resolution Procedures (2021), art. 13(6).X5  If the parties cannot reach consensus, each side strikes the names of any candidates that it finds unacceptable and ranks the remaining candidates in order of preference.6AAA Commercial Arbitration Rules and Mediation Procedures (2022), art. R-13(b); ICDR International Dispute Resolution Procedures (2021), art. 13(6).X6  To assist in this task, the institution provides a résumé of general information on each presented candidate.  Upon receiving the parties’ responses, the institution appoints arbitrators based on the party rankings in the order of mutual preference.7ICDR International Dispute Resolution Procedures (2021), art. 13(6).X7

Regardless of the method employed, arbitrator selection is among the most important steps in the arbitration process.  Arbitrators bring individual and distinct philosophies, experiences, temperaments, and procedures that will have an impact on how the arbitration is managed and the resolution of the dispute.  A practitioner must consider all the objective and subjective characteristics they wish to be reflected in the person or persons responsible for deciding their case, while simultaneously accounting for how the opposing party is performing the same task.8Pilawa, supra n. 1, at 405–06.X8  In doing so, the practitioner attempts to gain as much information as possible about potential candidates through reviewing professional backgrounds, surveying colleagues for their experiences and opinions, researching prior publicly available awards, reading relevant published articles and public speeches, identifying potential grounds for disqualification, and conducting interviews.9Id.X9  This process can be time consuming, expensive, and hindered by a lack of available information.  Despite these obstacles, the significance of selecting the correct arbitrator for the case mandates that practitioners perform all due diligence and make a decision informed by all available information.

In recent years, legal practitioners have become increasingly familiar with AI tools in the context of case management and certain common activities, including document review, legal research, and case analysis.1Bakst, supra n. 3, at 12–15.X1  For example, attorneys, litigation funders, and other interested parties use AI to predict outcomes and the likelihood of success in pursuing litigation.2Surden, supra n. 3, at 1331–32; Peisley & Sussman, supra n. 3, at 37.X2  In the courtroom, judges are utilizing predictive machine-learning algorithms trained on past crime data to assess criminal defendants’ potential risk of recidivism, flight, and danger to the community to better inform bail and sentencing decisions.3Surden, supra n. 3, at 1332–33.X3  

One could assume that similar tools might be adopted for the job of arbitrator selection because the task is heavily reliant on analyzing data to predict which individual or individuals will provide the best chance of success on a case.  In fact, the AAA-ICDR recently announced the beta launch of AAAi Panelist Search, a generative AI-powered panelist selection tool, to assist the institution in preparing lists of arbitrators.4AAA-ICDR Launches New AAAi Panelist Search to Enhance Panelist Selection with AI Technology at 1, AAA-ICDR (Oct. 10, 2024), www.adr.org/sites/default/files/document_repository/Press-Release_AAA_Launches_AAAi%20Panelist_Search_AI_Technology.pdf.X4  The AAA-ICDR boasts that the new tool will improve “its ability to conduct broader and deeper searches across the AAA-ICDR Roster for potential candidates.”5Id.X5

For the individual practitioner, the most enticing benefit from utilizing AI in the selection process is the promotion of efficiency and a reduction of cost.  An AI model trained on arbitrator data, such as academic and professional backgrounds and prior awards and writings, would be capable of identifying potential candidates based on case specific information much faster than a human attorney.  The model might also be able to eliminate candidates for potential conflicts that would not be obvious to a human sifting through arbitrator résumés and word-of-mouth recommendations.  In doing so, practitioners could reach a well-informed, data-driven decision much more efficiently than through traditional means.

The use of AI tools in the selection of arbitrators might broaden the pool of considered candidates.  In adopting an independent system trained on data from multiple sources, a practitioner is no longer limited to only those candidates with which he or she is familiar, that were provided by an institution, or that were recommended by a colleague.  Reliance on the tool could eliminate, or reduce, the influence of bias, either conscious or unconscious, on the practitioner’s valuation of candidates.  The implementation of AI tools might also eliminate some barriers, such as differing languages, that might have otherwise impaired consideration of a potential arbitrator candidate.

As with any nascent technology, the benefits of AI tools must be considered against their shortcomings.  In the case of arbitrator selection, the use of AI tools presents two significant questions.  First, in a field in which confidentiality is viewed as a major benefit, does the lack of available public information impair the reliability and accuracy of a predictive model?  Second, whether the adoption of an AI model would undermine current efforts to address other problems with arbitrator selection, specifically efforts to broaden and diversify the pool of arbitrators.

Machine-learning models are reliant on the availability of training data.  As more data is analyzed, the model returns more accurate and reliable results.  The limited availability of data is a major hurdle to the use of AI in international commercial arbitration.1Peisley & Sussman, supra n. 3, at 37.X1  Unlike with international investor-state arbitration cases and disputes before international bodies employing arbitration-like dispute resolution methods, international commercial arbitration is traditionally a confidential process and awards and orders are rarely published or made available in an unredacted form, if at all.2Id.X2  Some efforts have been made to collect and curate data on arbitrator procedural practices and individual user experiences through initiatives such as Arbitrator Intelligence,3Arbitrator Intelligence, https://arbitratorintelligence.vercel.app/.X3 Dispute Resolution Data,4Dispute Resolution Data, https://www.disputeresolutiondata.com/.X4 and Global Arbitration Review’s Arbitrator Research Tool.5Arbitrator Research Tool, Global Arbitration Review, https://globalarbitrationreview.com/tools/arbitrator-research-tool.X5  The value of the data offered by these services is somewhat limited because it largely relies on surveys completed by individual arbitrators and practitioners and information voluntarily released by arbitration providers.  Data of this nature often lacks contextualization and is susceptible to reflecting the biases of those that provide the data. 

Confidentiality of awards is the norm in international commercial arbitration and even when data is made publicly available, it is often incomplete.  For example, the ICC publishes awards in partnership with the legal database Jus Mundi.6Publication of ICC Arbitral Awards with Jus Mundi, ICC, https://iccwbo.org/dispute-resolution/resources/publication-of-icc-arbitral-awards-jus-mundi-not-icc-publication/.X6  However, only information pre-approved by the parties is made available for publication and other information is redacted.7Id.X7  

It is not uncommon for arbitrators to issue unreasoned final awards, with the parties choosing to forego the cost of a full draft with discussion of the arbitrator’s reasoning.  In such instances, the ultimate outcome of an arbitration is known, but how the arbitrator weighed evidence and reached the result is left a mystery. 

Arbitral panels also present a challenge because their awards are often the result of unseen compromise.  This coupled with the relative rarity of dissenting opinions in international commercial arbitration8See Dissenting Opinions and Why They Should be Tolerated, Arbitration Journal (Mar. 12, 2019), https://journal.arbitration.ru/analytics/dissenting-opinions-and-why-they-should-be-tolerated/.X8 can make it difficult, if not impossible, to attribute a particular result or philosophy to a single member of the panel.

Another concern with the adoption of AI tools in arbitration is the potential to further perpetuate existing biases and to weaken efforts to expand the pool of appointed arbitrators.  International arbitration has been criticized for its corps of commonly appointed arbitrators not adequately reflecting the ethnic, racial, and gender makeup of the community as a whole.9Cole, supra n. 12, at 969.X9  The multiple factors contributing to this problem are beyond the scope of this article, but relevant for the purposes of this discussion is the behavior of clients and counsel.10See Pilawa, supra n. 1, at 430.X10  Practitioners that regularly practice in the area of international arbitration are prone to relying on personal experience and personal relationships in selecting candidates that might be sympathetic to a particular argument or defense.11Peisley & Sussman, supra n. 3, at 37.X11  Practitioners and clients that are less accustomed to international arbitration practice often place considerable weight on arbitrator experience and show a preference for former judges, experienced litigators, and arbitrators with name recognition.12Cole, supra n. 12, at 984–85.X12  Because minorities are traditionally underrepresented among judges, senior lawyers at major law firms, and business executives, they are appointed to arbitrator roles comparatively less than older, white, western males.13Id.; Deborah Rothman, Gender Diversity in Arbitrator Selection, Disp. Res. Magazine 22, 25 (Spring 2022).X13  The limited number of diverse appointments in the past impacts AI tools in the present.  An AI model is limited by the quality of its data, and the majority of existing data on which AI models might be trained was generated by non-diverse arbitrators.  This introduces the potential for AI models to develop biases toward white and male candidates.14See Hunter Cyran, New Rules for New Era:  Regulating Artificial Intelligence in the Legal Field, 15 Case W. Res. J. L., Tech, & Internet 1, 31–35 (2024); Amy Cyphert et al., AI Cannibalism and the Law, 22 Colo. Tech. L. J. 301, 304–06 (2024).X14

Further limiting the ability of AI models to consider diversity is that the data that does exist often lacks necessary identification information.  Diverse representation can involve a wide range of characteristics, including, but not limited to, gender identity, race, ethnicity, age, religion, national origin, sexual orientation, language, disability, veteran status, and socioeconomic status.  The use of these criteria in the selection of an arbitrator is only possible if the data is available for the AI model to analyze.  Because professional résumés and biographies are often focused on education and professional accomplishments, an AI model would be unable to reliably account for other criteria that a party might consider relevant.

The lack of information is a major hinderance to the use of AI in arbitrator appointments.  There is simply no central repository of information for AI models to draw upon, and the information that is available is often incomplete or self-serving to the individual or institution by which it was prepared.  In the absence of large quantities of quality data, the advantage that AI tools can provide over traditional arbitrator selection methods is minimal.

As with other areas of law and life, AI tools will continue to become common place in the practice of international arbitration.  Practitioners and arbitrators will undoubtedly use AI in conducting research, managing discovery, reviewing documents, and drafting submissions and awards.  It remains to be seen whether AI will have a significant impact on the selection of arbitrators.  At this time, the lack of publicly available data presents a major weakness to any machine-learning model designed to identify qualify candidates or predict outcomes.  Modern models cannot replace the experience and insight of seasoned practitioners or correct for existing institutional deficiencies, but with technological improvements and the accumulation of greater quantities of available data on which to train predictive models, it is reasonable to expect that AI tools will be regularly utilized in the near future in the selection of arbitrators.