Jump to Main Menu Skip to Main Content
Table of Contents
Back to Issue Page
Issue 1

The Right to a Human Decision in International Arbitration: Exploring the Limits of Party Autonomy and the Prospects for the Use of AI in Decision-Making

Introduction

Machine learning and natural language processing systems can significantly contribute to streamlining various tasks that take counsel and arbitrators considerable time.1Mahnoor Waqar, The Use of AI in Arbitral Proceedings, 37 Ohio St. J. on Disp. Resol. 343, 351 (2022) (“[The] introduction of AI systems which uses Natural Language Processing (‘NLP’) . . . would significantly reduce the tasks of a lawyer engaging in arbitration.”); Layan Al Fatayri, AI in International Arbitration:  What Is the Big Deal?, Blog of the Am. Rev. Int’l Arb. (Oct. 22, 2024) https://aria.law.columbia.edu/ai-in-international-arbitration-what-is-the-big-deal/ (“Through natural language processing (NLP) and machine learning algorithms, AI can efficiently sift through large volumes of documents to identify relevant information, reducing the time and costs associated with manual review.”).1  In fact, artificial intelligence (AI) is already being deployed in important activities, such as document review, legal research and evidence gathering.2Maroof Rafique, Why Artificial Intelligence Is a Compatible Match for Arbitration, 88 Arb. 310, 315 (2022) (“[A]utomation is increasing its share in the market for legal services.  According to some reports, 80% of leading law firms are already using applications of AI.  Arbitrators are using AI for e-discovery, procedural automation, and drafting of awards.”); Cristina Florescu, The Interaction Between AI (Artificial Intelligence) and IA (International Arbitration):  Technology as the New Partner of Arbitration, 18 Revista Română de Arbitraj 42 (2024) (“Nowadays, nearly all phases of an international arbitration are facilitated by technology, with tools broadly accessible for:  selection of arbitrators, case management, discovery, legal research, document review and hearings.”).2  Nonetheless, decision-making is an area where the use of AI requires careful consideration.

Under the existing legal framework of international arbitration, the enforcement of AI-generated awards and agreements to refer disputes to AI-driven arbitration is uncertain.  Although not expressly stated, several curial laws and arbitration rules presuppose that arbitrators are human beings, as evidenced by the provisions on the appointment, replacement and challenge of arbitrators.3Gizem Halis Kasap, Can Artificial Intelligence (“AI”) Replace Human Arbitrators?  Technological Concerns and Legal Implications, 2021 J. Disp. Resol.209, 238–39 (2021)(“[Some] international arbitration laws . . . do not explicitly state that arbitrators must be human beings but envisage them as such by requiring standards that only a human can meet or attribute certain characteristics that are exclusive to human beings.”).3  Some curial laws even stipulate that arbitrators must be natural persons.4Id. at 238 (“In [some] national laws, the arbitrator is explicitly required to be a natural person with full capacity, as opposed to being a legal person.  The most notable standards are set by the French Code of Civil Procedure, the Dutch Code of Civil Procedure, and the Portuguese Voluntary Arbitration Law.  All of which stipulate that only ‘a natural person’ or ‘individuals’ can act as an arbitrator.”).4  Moreover, the provisions on the making of awards establish formal and content requirements whose compliance is debatable in the case of AI-generated awards.

The deployment of AI in decision-making also causes disquiet about the negative impact it could have on the perception and legitimacy of arbitration.  One of the most important features of arbitration is the parties’ ability to select the members of the tribunal, as it enables the parties to appoint the most suitable arbitrators for each case.  In the conformation of the tribunal, the parties consider the professional experience and academic formation of candidates, as well as their public recognition and interpersonal skills.5Felipe Rusinque, The System of Party-Arbitration:  The Legitimacy of Party-Appointed Arbitrators and the Right to Make Unilateral Nominations, 27 Int’l Arb. L. Rev. 102, 125 (2024) (“[T]he ability to select arbitrators . . . gives the parties control over the composition of the tribunal, so they not only ensure that arbitrators possess the knowledge and skills required to solve the dispute, but also bring to the tribunal persons held in high esteem for their professional career and integrity.”).5  These individual attributes can hardly be replicated by machine-based systems, let alone the abilities to empathise and discern, which are key to grasp the views of the parties and conduct fact-finding enquiries.

Concerns about the use of AI in decision-making are not limited to arbitration.  The deployment of machine-based systems in the allocation of economic interests is a broader phenomenon that, as Aziz Huq describes, is spurring “anxiety about a shift of power from human to machine decision makers.”6Aziz Z. Huq, A Right to a Human Decision, 106 Va. L. Rev. 611 (2020).6

Indeed, the widespread use of automated decision-making (“ADM”) by corporate and state actors to allocate loans, insurance compensation, employment benefits and other economic interests7Margot E. Kaminski & Jennifer M. Urban, The Right to Contest AI, 121 Colum. L. Rev. 1957, 1960 (2021) (“[T]he use of AI in decision-making has increased.  The government and private sector now use algorithms to decide how to distribute welfare benefits, whether to hire or fire a person, whether expressive material should be removed from online platforms, whether to keep people in prison, and more.”); Ifeoma Ajunwa, Automated Governance, 101 N.C. L. Rev. 355, 357 (2023) (“In many ways, as consumers, we have already acquiesced to our algorithmic overlords.  We have increasingly delegated consequential business decisions about insurance, home purchase, or employment opportunity to automated algorithmic processes.”).7 is viewed with serious misgivings, not only by those directly affected, but by society at large.8Huq, supra note 6, at 613–14 (“[M]any people feel a loss of control over key life decisions.  Machines, they fear, resolve questions of critical importance on grounds that are beyond individuals’ ken or control.  Many individuals experience a loss of elementary human agency and a corresponding vulnerability to an inhuman and inhumane machine logic.  For some, ‘the very idea of an algorithmic system making an important decision on the basis of past data seem[s] unfair.’”).8  As a result, new regulations have been put in place to limit the deployment of AI in decision-making.  These regulations provide the basis for a new right aimed at ensuring human involvement in decision-making, to wit:  the right to a human decision.

This article discusses the legal issues and implications of the deployment of AI in arbitral decision-making in the light of the emerging right to a human decision.  Section II delves into the question of what constitutes a deployment of AI in arbitral decision-making and frames the concepts of AI-driven arbitration and AI-generated awards.  In turn, section III addresses the conceptual and normative foundations of the right to a human decision.  Section IV goes over the normative and legitimacy concerns raised by the deployment of AI in arbitral decision-making.  Lastly, section V outlines the possible formulations of the right to a human decision in international arbitration.

What Constitutes a Deployment of AI in Arbitral Decision-Making? AI-Driven Arbitration and AI-Generated Awards

This article focuses on the deployment of AI in arbitral decision-making.  Therefore, uses of AI in arbitral proceedings for purposes other than decision-making are outside the scope of this research.  With this clarification in mind, the first question to be addressed is, what constitutes a deployment of AI in arbitral decision-making?

To answer this question, it is necessary to pinpoint the tasks and duties involved in arbitral decision-making.  Additionally, this article frames the concepts of AI-driven arbitration and AI-generated awards, which facilitate the identification of AI systems intended for use in decision-making.

A.     Tasks and Duties Involved in Arbitral Decision-Making

Arbitral proceedings are comprised of different structural elements and stages.1Nigel Blackaby, Constantine Partasides and Alan Redfern point out the following “key elements of an international arbitration:”  (i) the agreement to arbitrate, (ii) the place of arbitration, (iii) the commencement of an arbitration, (iv) the appointment of the arbitral tribunal, (v) arbitral proceedings, (vi) the award of the tribunal and (vii) the enforcement of the award.  Nigel Blackaby, Constantine Partasides & Alan Redfern, Redfern and Hunter on International Arbitration ¶ 1.19 (7th ed. 2023).1  From the request for arbitration and the constitution of the tribunal to the making of the award and its enforcement, several steps must be taken throughout the entire arbitration process.  The procedural stage with which this article is concerned is decision-making.

Decision-making is the process whereby the arbitral tribunal occupies itself with resolving the dispute between the parties.2Mary Mitsi, The Decision-Making Process of Investor-State Arbitration Tribunals 21 (2019) (“The decision-making process signifies the specific process in which arbitrators engage in order to resolve a specific dispute.”).2  More precisely, the decision-making process encompasses a set of tasks aimed at deciding on the merits of the case, namely:  (i) evaluating the parties’ submissions; (ii) evaluating the admissibility and relevance of evidence; (iii) ascertaining, interpreting and applying the substantive law to the facts of the case;3Lucy Reed, The 2013 Hong Kong International Arbitration Centre Kaplan Lecture – Arbitral Decision-Making: Art, Science or Sport?, 30 J. Int’l Arb. 85, 87 (2013) (“The science of arbitral decision-making lies in rigorously assessing the evidence, methodically finding the relevant material facts, identifying the governing law [and] applying that law to the facts.”).3 and (iv) deciding on the claims, defences and counterclaims submitted by the parties.4Gary Born, International Commercial Arbitration § 13.04[A][6] (3d ed. 2021) (“The arbitrator exercises authority conferred by law, to resolve the parties’ disputes in a legally-binding manner.  As noted above, ‘the services of an arbitrator are principally and habitually those of settling a dispute between two or more parties.’ . . . [T]his is an adjudicative function and authority, which consists in hearing the parties’ submissions and evidence in fair, objective proceedings and rendering an impartial, reasoned decision that finally decides their rights on the basis of their submissions and evidence.”).4

It goes without saying that settling the dispute between the parties by means of a final and binding decision—an arbitral award—is the main objective of any arbitration and constitutes the primary obligation of the arbitrators.5Id. § 13.04[A] (“The arbitrator’s central obligation is to resolve the parties’ dispute in an adjudicatory manner. . . . [T]his is the basic object of any arbitration agreement and it is the essential reason that the parties appoint an arbitrator and present their cases to him or her.”).5  This primary obligation, in turn, entails specific duties for arbitrators, namely:  (i) to be and remain independent and impartial, including the duty of disclosure of potential conflicts of interest;6Id. § 13.04[A][2] (“Essential elements of the arbitrator’s obligation to resolve the parties’ dispute in an adjudicative manner are the related duties to be and remain impartial and independent and to conduct the arbitration impartially and independently.”).6 (ii) to conduct the proceedings and decide the case with care and skill, including the duty to devote the necessary time and ability to the arbitration;7Blackaby et al., supra note 9, ¶ 5.56 (“It seems appropriate to expect the same standard of professional care from a lawyer . . . who is serving as an arbitrator as would be expected in the course of his or her other professional work.  The parties to an arbitration entrust an arbitral tribunal with an important task, for which they are prepared to pay—often generously.  They therefore expect the arbitrator to perform the task with due care.”).7 (iii) to complete their mandate,8Born, supra note 12, § 13.04[E] (“[A]n arbitrator’s acceptance of his or her appointment entails an implied undertaking to complete that mandate, by issuing a final award.”).8 that is, to exhaust all the procedural stages of the arbitration and decide on all the claims, defences and counterclaims submitted by parties;9Id. § 13.04[A][4] (“The arbitrator’s obligations include deciding all of the disputes which are presented to him or her. . . . [A]n arbitrator’s failure to decide questions that are submitted to the tribunal constitutes infra petita, and may be grounds for annulling or refusing to recognize the award.”).9 and (iv) not to delegate or relinquish their duties and decision-making tasks.10Id. § 13.04[A][8] (“An arbitrator’s obligations include the duty not to delegate his or her responsibilities or tasks to third parties.”).10 

In principle, the use of machine-based systems to carry out any of the above-mentioned tasks or duties constitutes a deployment of AI in arbitral decision-making.  At this juncture, it is important to highlight the arbitrators’ duty not to delegate their duties and decision-making tasks.  Both the Silicon Valley Arbitration and Mediation Center (SVAMC) Guidelines on the Use of Artificial Intelligence in Arbitration (“SVAMC Guidelines”) and the Chartered Institute of Arbitrators (CIArb) Guideline on the Use of AI in Arbitration (“CIArb Guideline 18”) make express references to such duty, according to which arbitrators must personally carry out their decision-making tasks and refrain themselves from using AI systems that may compromise the integrity of the arbitration or the validity and enforcement of the award.11Silicon Valley Arb. & Mediation Ctr., Guidelines on the Use of Artificial Intelligence in Arbitration Guideline 6 (2024) [hereinafter SVAMC Guidelines]; Chartered Inst. of Arbitrators, Guideline on the Use of AI in Arbitration art. 8.2 (2025).11  This duty will be discussed in detail in section V.

Similarly, it is necessary to add a caveat that contributes to distinguishing the deployment of AI in decision-making from other uses of AI in arbitral proceedings:  arbitrators commonly employ clerical assistance in the performance of their tasks.12Born, supra note 12, § 13.04[A][8] (“Arbitrators obviously use clerical assistance, which can readily extend beyond typing, organizing files and the like, to dealing with administrative matters; these tasks will often be conducted either by secretarial or similar staff, but sometimes by junior lawyers or interns. . . . [I]n some instances arbitrators will appoint a ‘secretary’ to assist the tribunal.”).12  This by no means entails a breach of the duty not to delegate decision-making tasks, insofar as the assistance that arbitrators receive is limited to secretarial and clerkish activities, such as record keeping of hearings, receiving and sending communications between the tribunal and the parties, summarising the parties’ submissions and drafting procedural orders and boilerplate parts of awards.13Id. (“[I]t is common practice for secretaries (or other junior lawyers) to conduct legal research for tribunals, to organize files and witness statements (sometimes going beyond merely administrative work) and sometimes to draft administrative communications, procedural orders, or portions of awards.”).13  Since this clerical assistance does not imply the delegation of decision-making tasks, it could be provided by “AI assistants” or “AI secretaries.”14Danna Zhang, Integrating New Technologies in International Arbitration:  Thoughts on the Future of Our Profession, in Stockholm Arbitration Yearbook 2024, 100, 110–11 (Patrik Schöldström & Christer Danielsson eds., 2024) (“In the very near future, arbitrators will also likely be able to acquire personal AI secretaries who are fully tailored and trained in accordance with their individual preferences . . . .  Many lawyers will recognise these tasks as typical junior-level work:  drafting documents or email replies and preparing summaries for seniors to review.  And in many cases, this is not too different from the tasks typically performed by tribunal secretaries in international arbitration proceedings.”).14

B.     AI-Driven Arbitration and AI-Generated Awards

Although arbitration is characterised by offering its users a flexible and versatile alternative for the resolution of their disputes, in practice arbitral proceedings tend to be lengthy and costly,15Queen Mary University of London & White & Case LLP, 2025 International Arbitration Survey:  The Path Forward:  Realities and Opportunities in Arbitration (2025) https://www.whitecase.com/sites/default/files/2025-06/white-case-qmul-2025-international-arbitration-survey-report-web.pdf (“Arbitration is commonly praised for its flexibility and ability to meet the needs of its users.  Yet concerns persist about delays and inefficiencies throughout the arbitral process.  Time and cost of the proceedings consistently arise among the most significant drawbacks for users.”).15 especially when dealing with complex and high-value cases.  For this reason, it is not surprising that arbitration users consider that AI has the potential to bring about a significant shift in the way arbitral proceedings are conducted, in terms of expeditiousness and cost-effectiveness.16According to the 2025 International Arbitration Survey of Queen Mary University and White & Case, arbitration users selected “saving time” and “reducing costs” as the two main reasons for the use of AI in arbitral proceedings.  Id. at 28–29.16

One potential development currently being explored is the use of AI to conduct the proceedings and render the award.17The AAA-ICDR announced the launch of an “AI arbitrator” intended for use in construction-related disputes and indicated that AI will also support other industries in 2026.  Am. Arbitration Ass’n-Int’l Ctr. for Dispute Resol., AAA-ICDR to Launch AI-Native Arbitrator, Transforming Dispute Resolution (Sept. 17, 2025), https://www.adr.org/press-releases/aaa-icdr-to-launch-ai-native-arbitrator-transforming-dispute-resolution/.17  In this manner, instead of arbitrators using AI tools to perform specific tasks in the course of the proceedings, the entire arbitration is led by a full-service automated system that handles both case management and decision-making.18Horst Eidenmüller & Faidon Varesis, What is an Arbitration?  Artificial Intelligence and the Vanishing Human Arbitrator, 17 N.Y.U. J.L. & Bus. 49, 62 (2020) (“A fully automated ‘robot’ AI arbitrator system, which would substitute human arbitrators, requires performance of a variety of tasks . . . A full-service AI arbitrator system would require a fully automated case management process of accepting cases, performing conflict checks, managing multiple appointments, cases, and hearings at the same time, without the involvement of human input.  In addition, it would need to perform factual analysis of a given case, including e-discovery of facts and document analysis . . . Based on this analysis, the system would have to render a final and binding decision as well as give the reasons for reaching this decision.”).18  This is what this article terms “AI-driven arbitration.”

As with other arbitrations, the intended outcome of an AI-driven arbitration is an award, which in this case is referred to as an “AI-generated award.”  From a technical standpoint, an AI-generated award is the output of the algorithmic processes executed by the AI to decide on the merits of the case—the AI processes the parties’ submissions and the evidence produced in the proceedings through its inner workings and renders the award.

It is worth noting that AI-driven arbitration would likely require a certain degree of human oversight to ensure that the system operates correctly.  If this oversight is limited to verifying the correct functioning of the system, the arbitration will remain entirely driven by AI and the award will continue to be considered the AI’s output.  The scope of the human oversight of AI-driven arbitration and the review of AI-generated awards will be discussed in section V. 

The Concept of the Right to a Human Decision and Its Different Formulations and Normative Manifestations

From the granting of government benefits and privileges1David Freeman Engstrom, Daniel E. Ho, Catherine M. Sharkey & Mariano-Florentino Cuéllar, Government by Algorithm:  Artificial Intelligence in Federal Administrative Agencies 6 (Feb. 2020) (report to the Admin. Conf. of the U.S.), available at https://law.stanford.edu/wp-content/uploads/2020/02/ACUS-AI-Report.pdf (“The government’s AI toolkit is diverse and spans the federal administrative state.  Nearly half of the federal agencies studied (45%) have experimented with AI and related machine learning (ML) tools.”  “Adjudicating government benefits and privileges, from disability benefits to intellectual property rights” was one of the governmental tasks where the use of AI was reported. (emphasis in original)).1 to the assessment of tax credits and deductions,2See Henk Snijders, Arbitration and AI, from Arbitration to ‘Robotration’ and from Human Arbitrator to Robot, 87 Arb. 223, 226 (2021).2 corporate and state actors are increasingly turning to AI to streamline various tasks that involve decision-making on the allocation of economic interests.  The rise of ADM is due to its capacity to make administrative tasks more efficient and cost-effective.3Ajunwa, supra note 7, at 361–62 (“[T]here are tangible benefits to deploying algorithmic processes for administrative decision-making . . . some of those benefits [include] efficiency, cost-savings, and the ability of algorithmic systems to discern patterns displaying historical bias in their rulemaking or adjudication.”).3  Efficiency, however, cannot be justified at the cost of fairness and justice.4Id. at 363–64 (“Cost-savings are an important consideration for the adoption of any technological tool; and even more so when the taxpayers bear that cost . . . Cost-savings, however, can never be the sole consideration for adopting automated decision-making tools.  As several legal scholars have noted, equity and justice should remain the paramount guide.”).4

ADM implies that individuals are subject to the scrutiny of machine-based systems that determine the outcome of their legal situation in relation to the allocation of economic interests or, more generally, the protection of their rights.  As might be expected, ADM raises concerns about the accuracy and soundness of AI-generated decisions and the accountability for flaws in decision-making and malfunctioning systems.5Kaminski & Urban, supra note 7, at 1957 (“These uses of AI raise a host of concerns about discrimination, accuracy, fairness, and accountability.”).5  Moreover, the fact that AI is taking on tasks hitherto performed exclusively by humans has provoked public anxiety about the possibility of machine-based systems displacing humans in activities involving decision-making on ethical and normative questions, such as the adjudication of disputes.6Huq, supra note 6, at 685 (“[M]achine decisions are not presently appropriate for decisions with ethical or normative components.  Many legal and even factual questions resolved through civil and criminal adjudication have some such component.”).6

These concerns are prompting the introduction of regulations that limit the use of AI in activities where human agency is deemed necessary.  Some of these regulations have been adopted at the domestic and regional levels; nevertheless, they have a broad personal and territorial scope of application and may be applied in the context of international arbitration.

A.     The Advent of the Right to a Human Decision:  A New Regulatory Paradigm for ADM

 In 2020, Huq drew attention to an emerging normative response to the concerns caused by the rise of ADM:  the right to a human decision.7Id. at 611 (“Machine-learning tools are perceived to be eclipsing, even extinguishing, human agency in ways that compromise important individual interests.  An emerging legal response to such worries is to assert a novel right to a human decision.”).7  Huq noted that a new regulatory paradigm was being developed on the basis of the European legal framework on data protection, specifically Article 22(1) of the General Data Protection Regulation (“GDPR”), which provides data subjects with “the right not to be subject to a decision based solely on automated processing . . . which produces legal effects concerning him or her.”

Along with Huq, many scholars have delved into the questions of the necessity and desirability of a right that guarantees a certain degree of human intervention in ADM.8Giovanni Sartor, Artificial Intelligence and Human Rights:  Between Law and Ethics, 27 Maastricht J. Eur. & Comp. L. 705 (2020); Stefan Schäferling, Governmental Automated Decision-Making and Human Rights:  Reconciling Law and Intelligent Systems 231 (2023); Yuval Shany, The Case for a New Right to a Human Decision Under International Human Rights Law (Oct. 4, 2023), https://ssrn.com/abstract=4592244; Marya Akhtar & Rikke Frank Jørgensen, A Rights-Based Approach to Automated Decision-Making in the Public Sector, in Artificial Intelligence and International Human Rights Law:  Developing Standards for a Changing World 69 (MichaÅ‚ Balcerzak & Julia KapelaÅ„ska-PrÄ™gowska eds., 2024).8  This academic debate has given rise to different formulations of this right, including “the right not to be subject to ADM”9Elena Abrusci & Richard Mackenzie-Gray Scott, The Questionable Necessity of a New Human Right Against Being Subject to Automated Decision-Making, 31 Int’l J.L. & Info. Tech. 114, 143 (2023).9 and “the right to contest AI.”10Kaminski & Urban, supra note 7, at 1988.10  The aim of this section is not to comment on the specific contours and nomenclature of each of these formulations, but rather to identify those that are relevant for outlining the possible scope and characteristics of the right to a human decision in international arbitration.

B.     Formulations of the Right to a Human Decision Relevant to Its Implementation in International Arbitration

Michael Cheng puts forward a threefold classification of the “different formulations of the right to human decision,”11Michael Cheng, The Right to Human Decision:  Analyzing Policies, Ethics, and Implementation 2–3 (Jan. 31, 2024), https://openreview.net/forum?id=VrxtvBw0ly.11 which is useful for analysing the alternatives for the implementation of this right in international arbitration.  According to Cheng, the right to a human decision can be formulated as:  (i) a right to contest AI-generated decisions, which entitles individuals to challenge AI-generated decisions before a “human reviewer;” (ii) a right to opt out for a human alternative, this formulation enables individuals subject to ADM by default to request a human decision-maker; and (iii) a total or partial exclusion of AI from decision-making, a total exclusion takes place when there is an express prohibition on the use of AI in decision-making, while a partial exclusion occurs when a decision cannot be based solely on ADM, meaning that ADM is permitted, but a certain degree of human intervention is required.

The formulation of the right to a human decision as a right to opt out for a human alternative is not consistent with the existing legal framework of international arbitration, insofar as the parties are not subject to ADM by default.  In fact, the bulk of curial laws and arbitration rules presuppose that arbitrators are human beings, as will be explained in section IV.  

This leaves the characterisation of the right to a human decision in international arbitration either as a right to challenge AI-generated decisions before a human reviewer or as an exclusion of AI from decision-making.  While the former seems more appealing, as it would allow the parties to use AI-driven arbitration and have recourse to a human reviewer; the latter is the only possible formulation in jurisdictions that proscribe the use of AI in decision-making by explicitly requiring arbitrators to be natural persons.  The scope and characteristics of these formulations will be discussed in detail in section V.

C.     Normative Manifestations of the Right to a Human Decision:  Regulations Limiting ADM and Their Application in International Arbitral Proceedings

As noted before, the GDPR established a new paradigm for ADM that laid the foundation for the right to a human decision.12Huq, supra note 6, at 621–22; Kaminski & Urban, supra note 7, at 2012–14; Abrusci & Mackenzie-Gray Scott, supra note 35, at 121; Shany, supra note 34, at 20.12  Indeed, Article 22(1) of the GDPR imposes a limitation on ADM by granting data subjects the “right not to be subject to a decision based solely on automated processing . . . which produces legal effects concerning him or her.”

Other regional and domestic instruments also contain provisions that reflect different formulations of the right to a human decision.  These instruments regulate ADM and establish privacy and human rights standards that may be applicable in the context of international arbitration, provided that factors such as the seat of arbitration or the residence, nationality or domicile of the parties trigger their application, according to their territorial and personal scope.

1.     Regional and Domestic Regulations Limiting ADM and Their Application in International Arbitral Proceedings

In addition to the GDPR, there are other regulatory frameworks that set limits on ADM by granting rights to individuals.  These rights not only reflect the different formulations of the right to a human decision described above, but may also have an impact on arbitral proceedings.

For instance, the African Union Convention on Cyber Security and Personal Data Protection provides in Article 14(5) that individuals shall not be subject to decisions “based solely on automated processing,” along the lines of the GDPR.  The former and the latter envisage a partial exclusion of ADM, to the extent that both regulatory frameworks state that a decision cannot be based solely on ADM, which implies that ADM is not prohibited as such, but requires a certain degree of human intervention.

Also at the regional level, there is an instrument that establishes a standard of human rights protection for individuals subject to ADM:  the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (the so-called “Framework Convention on AI”), which states in Article 14(2) that its Contracting Parties shall adopt measures that provide individuals with effective remedies in case of human rights violations caused by AI systems, including “an effective possibility for persons concerned to lodge a complaint to competent authorities.”13Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law art. 14(2)(c), Sept. 5, 2024, C.E.T.S. No. 225.13  This is a notable manifestation of the right to a human decision as a right to contest AI-generated decisions.

At the domestic level, an illustrative example of the “right to opt out for a human alternative” can be found in the U.S. Blueprint for an AI Bill of Rights published by the White House Office of Science and Technology Policy.  This instrument embodies the principle of “Human Alternatives, Consideration and Fallback” whereby individuals “should be able to opt out from automated systems in favor of a human alternative.”

These regional and domestic instruments show a general trend in the current regulatory frameworks on AI:  limiting the deployment of AI in decision-making by granting rights to individuals subject to ADM.

That said, despite the adoption of different formulations of the right to a human decision in the regional and domestic instruments mentioned above, such instruments cannot serve as the sole basis for the articulation of the right to a human decision in international arbitration due to their material and territorial scope.  Those instruments deal with data protection and human rights and are applicable in the regional or state jurisdictions where they have been adopted. 

While it is true that arbitral tribunals may be bound by data protection and human rights standards set out in regional or domestic instruments, depending on factors such as the nationality, residence or domicile of the parties and the seat of the arbitration,14Martin Zahariev, Mission (Im)possible:  Where GDPR Meets Commercial Arbitration, in Austrian Yearbook on International Arbitration 2020 3 (Christian Klausegger et al. eds., 2020) (“In other words, by virtue of either the establishment criterion . . . or the targeting criterion . . . arbitral institutions and arbitrators might fall within the territorial scope of the (GDPR).”); id. at 9–10 (“For commercial arbitration the hypothesis where arbitrators and arbitral institutions located outside EU provide dispute resolution services to individuals located in the EU seems to be of practical importance.”); Wei Gao, The ECHR in Action:  Its Applicability and Relevance for Arbitration, 26 Int’l J. Hum. Rts. 1608, 1626–27 (2022) (“[P]rivate parties choosing arbitration as a dispute resolution mechanism . . . must be aware of the rights that they enjoy under the ECHR . . . Such awareness may be crucial for the parties to rationally exercise their informed judgment in an arbitration, e.g. when they select the seat of arbitration within or outside an ECHR Contracting State.”).14 the limits imposed on ADM would apply exclusively to the processing of personal information disclosed in the context of arbitral proceedings or possible cases of human rights violations resulting from non-compliance with the standards of protection laid down in the instruments concerned.

On the other hand, questions relating to the conduct of arbitral proceedings, including the composition of the tribunal and the making of awards, are governed by other legal instruments, namely curial laws, arbitration rules and the parties’ agreement.

2.     The EU AI Act and Its Applicability to International Arbitration with regard to the Deployment of AI in Arbitral Decision-Making

Among the regional instruments that regulate ADM, there is a piece of EU legislation particularly relevant to the articulation of the right to a human decision in international arbitration:  the European Regulation on Artificial Intelligence (“EU AI Act”).15Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act), 2024 O.J. (L 1689) [hereinafter EU AI Act].15  This Regulation establishes a general framework for the use of AI following a risk-based approach.16Recital 26 EU AI Act reads:  “In order to introduce a proportionate and effective set of binding rules for AI systems, a clearly defined risk-based approach should be followed.  That approach should tailor the type and content of such rules to the intensity and scope of the risks that AI systems can generate.  It is therefore necessary to prohibit certain unacceptable AI practices, to lay down requirements for high-risk AI systems and obligations for the relevant operators, and to lay down transparency obligations for certain AI systems.”  Id. recital 26.16  More specifically, it classifies AI systems in risk levels according to the systems’ likelihood to cause harm,17Mireille M. Caruana & Roxanne Meilak Borg, Regulating Artificial Intelligence in the European Union, in The EU Internal Market in the Next Decade – Quo Vadis? 108, 117–18 (Ivan Sammut & Ivan Mifsud eds., 2025) (“The AI Act adopts a ‘risk-based’ approach and lays down a ‘horizontal’ (as opposed to sector-specific) legal framework for AI that aims to ensure legal certainty . . . The Act . . . precludes AI systems that pose an unacceptable risk, listing several prohibited practices.  Thus, it identifies and heavily regulates AI systems with high risks and briefly mentions systems with limited risks.  It also includes a new Chapter V on general-purpose AI (‘GPAI’) models, which speaks of GPAI models with ‘systemic risk’ and ones ‘released under a free and open-source licence.’”).17 namely unacceptable risk,18EU AI Act, supra note 41, art. 5.18 high risk19Id. art. 6, annex III.19 and risks that entail transparency obligations.20Id. art. 50.20  The obligations that the EU AI Act imposes on “providers” and “deployers” depend on the risk level.

Regarding the deployment of AI in decision making, Recital 61 of the EU AI Act and Paragraph 8(a) of its Annex III qualify as high-risk “AI systems intended to be used by a judicial authority . . . in researching and interpreting facts and the law and in applying the law to a concrete set of facts, or to be used in a similar way in alternative dispute resolution.”21Id. recital 61; annex III, ¶ 8.21

Although the EU AI Act does not explicitly refer to arbitration, a literal and contextual construction of its provisions reveals that arbitration does fall within its material scope.  Indeed, the last sentence of the aforementioned provisions refers to “alternative dispute resolution,” and arbitration is an adjudicative type of alternative dispute resolution.  Likewise, as part of their mandate, arbitrators are responsible for “researching and interpreting facts and the law” and “applying the law to a concrete set of facts.”  Therefore, it is clear that the EU AI Act applies to arbitration.  Maxi Scherer, Ole Jensen and Russell Childree subscribe to this interpretation, observing that “[g]iven the Act’s references to the use of artificial intelligence systems by ‘a judicial authority’ and ‘alternative dispute resolution,’ there can be little doubt that the Act applies to arbitral proceedings.”22Maxi Scherer, Ole Jensen & Russell Childree, Regulating the Use of Artificial Intelligence in International Arbitration:  The EU AI Act and Beyond, 2024 Cahiers de l’Arbitrage / Paris J. Int’l Arb. 653, 665 (2024).22

These authors also claim that arbitrators and arbitral institutions are to be considered “deployers,” under Article 3(4) of the EU AI Act,23Id. at 665–66 (“[A] ‘deployer’ is ‘any natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity.’  This broad definition readily applies to arbitrators given that they may use AI systems in the exercise of their arbitral mandate, for instance by asking an AI system to summarise the parties’ positions or draft parts of correspondence, procedural orders and awards.  Arbitral institutions likewise exercise authority over parties’ disputes, including by appointing arbitrators, deciding on requests for consolidation and scrutinising draft awards.  They too would therefore qualify as ‘deployers.’”).23 and that the Act applies to arbitrations if (i) the arbitral tribunal has its “place of establishment” in the EU—i.e., the seat of the arbitration is in one of the EU Member States;24Id. at 666–67 (“Pursuant to Article 2(1)(b), the EU AI Act applies to deployers ‘that have their place of establishment or are located within the Union’.  A literal interpretation of this provision results in the territorial applicability of the EU AI Act in a number of scenarios . . . Arbitral tribunals have their ‘place of establishment’ in the EU if the seat of the arbitration is in one of the Member States, irrespective of where the arbitrators are otherwise located.”).24 or (ii) the arbitrators or the arbitral institution have their place of establishment or are located within the EU, pursuant to Article 2(1)(b).25Id. (“Article 2(1)(b) applies more readily to arbitral institutions that are headquartered in a Member State.  In that case, the institution has both its ‘place of establishment’ and ‘location’ in the EU.  The same applies to institutions that originate from a third country but have an office location in the EU.  In that case, at least the regional office would be ‘located within the Union’ and thus subject to the EU AI Act.”).25  In these circumstances, it is likely that many international arbitrations fall under the Act’s scope.  Then, the question is, what legal effects would the application of the EU AI Act have on arbitration?

As previously noted, the deployment of AI systems “in researching and interpreting facts and the law and in applying the law to a concrete set of facts” is categorised as high-risk, meaning that the requirements laid down in Section II, Chapter II of the EU AI Act must be met.  In particular, the Act imposes on deployers a set of obligations that include:  (i) taking appropriate measures to ensure that the AI systems are used according to their intended purpose—Article 26(1) EU AI Act; and (ii) assigning human oversight of the AI systems “to natural persons who have the necessary competence, training and authority”—Article 26(2) EU AI Act.

It is important to highlight the deployers’ obligation to assign and maintain human oversight of high-risk AI systems.  The human oversight must be assigned “to natural persons who have the necessary competence, training and authority,” meaning that the designated persons shall possess the specialised knowledge and license to carry out this task.  In fact, deployers shall be AI-literate themselves, under article 4 of the Act.  It is also worth noting that human oversight “shall aim to prevent or minimise the risks to . . . fundamental rights that may emerge when a high-risk AI system is used in accordance with its intended purpose or under conditions of reasonably foreseeable misuse.”26EU AI Act, supra note 41, art. 14(2).26

In this manner, although the deployment of AI systems to examine and interpret the facts and the law and to apply the law to the facts of the case is not prohibited as such, it does entail compliance with the obligations mentioned above.  These obligations include assigning and maintaining human oversight of such systems.  Human oversight must be entrusted to natural persons with the necessary competence, training and authority, and must be aimed at preventing or minimising risks to human rights.

Normative and Legitimacy Concerns Raised by the Deployment of AI in Arbitral Decision-Making

Before the recent rise of AI, it was a given that arbitrators were natural persons.  Various curial laws and arbitration rules address the appointment, replacement and challenge of arbitrators on the basis that arbitrators are human beings.  Some curial laws even prescribe that arbitrators must be natural persons.  Additionally, the provisions on the form and content of awards require that they be signed by the arbitrators and that they state the reasons on which they are based.  These requirements pose difficulties for AI-generated awards considering that AI systems cannot sign documents or assent to the content of the decisions they make, not to mention the impossibility of understanding how some AI systems reach decisions due to the opacity of their internal processes.

In terms of legitimacy, leaving the adjudication of disputes to AI systems may have a negative impact on the perception of arbitration.  The good standing of arbitration is largely due to the ability of the parties to select arbitrators.  In selecting the members of the tribunal, the parties not only consider individuals who have the knowledge and experience necessary to resolve a particular dispute, but also seek individuals that possess excellent interpersonal skills and high ethical standards, which instils confidence in the parties and enhances the legitimacy of the arbitration.

A.     Normative Concerns About the Deployment of AI in Arbitral Decision-Making

This subsection examines the provisions relating to the composition of the arbitral tribunal and the making of awards in curial laws and arbitrations rules.  Special attention is paid to the UNCITRAL Model Law on International Commercial Arbitration (“MAL”), as it has influenced the adoption of curial laws in 93 states.1U.N. Comm’n on Int’l Trade L., Status:  UNCITRAL Model Law on International Commercial Arbitration (1985), with amendments as adopted in 2006, https://uncitral.un.org/en/texts/arbitration/modellaw/commercial_arbitration/status.   1  Additionally, the analysis of curial laws is complemented by references to the lex arbitri of London,2Arbitration Act 1996, c. 23, as amended by Arbitration Act 2025, c. 4 (Eng.).2 Beijing3Arbitration Law of the People’s Republic of China, adopted Aug. 31, 1994, effective Sept. 1, 1995, as amended Sept. 1, 2017, effective Jan. 1, 2018 (China) [hereinafter Chinese Arbitration Law 1994].3 and Paris,4References to the French lex arbitri only correspond to the provisions applicable in international arbitration pursuant to Title II, Book IV Code de procédure civile.4 three of the most preferred seats for international arbitration5Queen Mary University of London & White & Case, supra note 24, at 5 (“The five most preferred seats for arbitration are London, Singapore, Hong Kong, Beijing and Paris.”).5 whose arbitration laws are not based on the MAL.

Regarding the arbitration rules analysed in this subsection, the focus is on the UNCITRAL Arbitration Rules and the Rules of the ICC,6ICC Rules of Arbitration (2021).6 the SIAC7SIAC Arbitration Rules (2025).7 and the LCIA,8LCIA Arbitration Rules (2020).8 four of the most widely used sets of arbitration rules in international arbitration.9Queen Mary University of London & White & Case, supra note 24, at 5 (“The five most preferred sets of arbitral rules are the ICC Rules, HKIAC Rules, SIAC Rules, LCIA Rules and UNCITRAL Rules.”).9

Similarly, to have a complete view of the normative landscape of international arbitration regarding the composition of the tribunal, this subsection refers to the rules on the appointment and replacement of arbitrators in the ICSID Convention10Convention on the Settlement of Investment Disputes Between States and Nationals of Other States, opened for signature Mar. 18, 1965, 17 U.S.T. 1270, 575 U.N.T.S. 159 [hereinafter ICSID Convention].10 and the ICSID Arbitration Rules,11ICSID Rules of Procedure for Arbitration Proceedings (Mar. 21, 2022).11 two of the most widely applied instruments in treaty-based investment arbitration.

1.     Rules on the Appointment, Replacement and Challenge of Arbitrators in Curial Laws

Chapter III of the MAL sets out rules applicable to the composition of the tribunal, including the appointment and challenge of arbitrators.  Article 11(1) provides that, “[n]o person shall be precluded by reason of his nationality from acting as an arbitrator.”  Likewise, under Article 12(2), “[a]n arbitrator may be challenged only if circumstances exist that give rise to justifiable doubts as to his impartiality and independence.”

The references to the “nationality” and the “independence and impartiality” of arbitrators are clear indications that these provisions are directed at individuals.12Annabelle O. Onyefulu, Artificial Intelligence in International Arbitration: A Step Too Far?, 89 ARB. 56, 59 (2023) (“Article 11 of the (MAL) impliedly provides for arbitrators to be natural persons. It provides that a person shall not be estopped from representing parties as an arbitrator by virtue of his/her nationality. It is in doubt if an intelligent machine can have a nationality.”).12  Although nationality is an attribute that both natural and legal persons possess, independence and impartiality are quintessential human qualities, especially in the context of dispute settlement.

While machine-based systems may produce outputs that reflect biases, depending on aspects such as the information fed into the systems,13Ana Fernández Araluce, AI in International Arbitration: Unveiling the Layers of Promise and Peril, 49 IURGIUM 35, 40 (2024) (“AI is based on the way it was coded and programmed; accordingly, if the data is biased or incomplete, AI will inevitably mirror these flaws.”).13 the circumstances falling within the scope of Article 12(2) MAL are conflicts of interest, that is, situations where arbitrators have a personal or economic interest vis-à-vis the dispute or the parties.14Kasap, supra note, 3, at 240 (“[A]n arbitrator must be both independent and impartial with respect to the parties and the dispute . . . The former concerns the arbitrator’s potential financial, professional, or personal ties to a party or a dispute that might influence their decision, whereas the latter addresses any favoritism or prejudice shown by an arbitrator toward a party or the matter in dispute.”).14  The IBA Guidelines on Conflicts of Interest in International Arbitration illustrate this by stating that the wording of Article 12 MAL incorporates the so-called “reasonable third party test,” which is precisely aimed at assessing conflicts of interest.15INT’L BAR ASS’N, GUIDELINES ON CONFLICTS OF INTEREST IN INTERNATIONAL ARBITRATION, Explanation to General Standard 2(b) (2024).15  Also, the legislative history of Article 12 MAL confirms that this provision was intended to deal with conflicts of interest.16Report of the Secretary-General: Possible Features of a Model Law on International Commercial Arbitration, U.N. Doc. A/CN.9/207, ¶ 65, in XII Y.B. UNCITRAL 75, 85 (1981) (“As regards challenge of arbitrators, the issues to be considered are on what grounds an arbitrator may be challenged . . . The reasons relate to the dispute at hand such as a financial interest or previous involvement in the subject-matter under dispute or a certain relation to one of the parties.”).16

From the reading of Articles 11(1) and 12(2) MAL, and considering that 93 states have adopted legislation based on the MAL, it is fair to say that, as a rule, the bulk of curial laws presuppose that arbitrators are human beings.  This claim is reinforced by the fact that the curial laws of some of the most preferred seats for international arbitration also contain provisions indicating that arbitrators are natural persons. 

In the case of London, Section 26(1) of the English Arbitration Act provides that “[t]he authority of an arbitrator is personal and ceases on his death.”  Similarly, the lex arbitri of Beijing, contains references clearly directed at individuals when laying down the rules on independence and impartiality of arbitrators.  Indeed, pursuant to Article 34 of the Arbitration Law of the People's Republic of China, the parties have the right to challenge an arbitrator—and they must withdraw—if “the arbitrator has a personal interest in the case,”17Chinese Arbitration Law 1994, supra note 55, art. 34(2).17 “the arbitrator has other relationship with a party . . . which may affect (their) impartiality,”18Id. art. 34(3).18 or “the arbitrator has privately met with a party . . . or accepted an invitation to entertainment or gift from a party.”19Id. art. 34(4).19

The case of Paris is particularly important since its lex arbitri expressly requires that arbitrators must be natural persons.  Article 1450 of the French Code of Civil Procedure (Code de procédure civile) stipulates that “the role of the arbitrator can only be exercised by a natural person.”20“La mission d’arbitre ne peut être exercée que par une personne physique.” Code de Procédure Civile, art. 1450 (Fr.).20

In jurisdictions that incorporate an express requirement for arbitrators to be natural persons, such as Paris (and France in general),21The lex arbitri of France—i.e., the provision on international arbitration envisaged in the French Code de procédure civile—applies nationwide.21 the deployment of AI in decision-making is especially problematic due to (i) the arbitrators’ duty not to delegate their mandate, including their decision-making tasks; and (ii) the lack of legal certainty regarding the validity of AI-generated awards and agreements to refer disputes to AI-driven arbitration, which makes such awards and agreements prone to challenges.

It is noteworthy that other important jurisdictions for international arbitration, such as Singapore and Hong Kong, the second and third most preferred seats for international arbitration,22Singapore and Hong Kong are the second and third most preferred seats for international arbitration, respectively. Queen Mary University of London & White & Case, supra note 24, at 7.22 respectively, have enacted the MAL into their curial laws,23In the case of Singapore, Section 3 of the International Arbitration Act 1994 (Cap. 143A) (Sing.) expressly provides that the MAL has the force of law, subject to the modifications contained in the Act. The MAL is also the First Schedule of the Singapore International Arbitration Act 1994. As for Hong Kong, the Arbitration Ordinance (Cap. 609) (H.K.) states in Section 4 that the MAL has the force of law, subject to the modifications and supplements provided for in the Ordinance. Likewise, the MAL is contained in Schedule 1 of the Arbitration Ordinance.23 and therefore the considerations made above regarding the MAL apply to those jurisdictions.

2.     Rules on the Appointment and Replacement of Arbitrators in Arbitration Rules

To a great extent, arbitration rules also presuppose that arbitrators are human beings.  As with curial laws, this is evidenced by the language of the rules on the appointment and replacement of arbitrators.

Article 6 of the UNCITRAL Arbitration Rules requires the appointing authority to consider the independence, impartiality and nationality of prospective arbitrators, attributes which, in the context of dispute settlement, unambiguously allude to individuals.24Abhilasha Vij, Arbitrator-Robot:  Is A(I)DR the Future?, 39 ASA Bulletin 123, 134 (2021). (“The ‘Appointment of Arbitrator’ as provided under Article 6 of the UNCITRAL Arbitration Rules does not envisage the possibility of referring a dispute to a machine . . . The rule contemplates a natural person who could be independent and impartial as an arbitrator, and would belong to certain nationality.”).24

Leading institutions like ICC and SIAC also specify in their rules that arbitrators are natural persons, as evidenced by their rules on the replacement of arbitrators.  Article 15(1) of the ICC Arbitration Rules stipulates that “[a]n arbitrator shall be replaced upon death” and, in a similar vein, Section 30(1) of the SIAC Rules envisages the “death” of an arbitrator as a ground for replacement.

Similarly, the LCIA Arbitration Rules contain language from which it can be inferred that arbitrators are natural persons.  When laying down the rules on the formation of the tribunal, Article 5(4) stipulates that, before appointment, each candidate must provide a written summary of “his or her qualifications and professional positions” and sign a written declaration stating if there are circumstances likely to give rise to justifiable doubts as to “his or her impartiality or independence.”

3.     Rules on the Appointment and Replacement of Arbitrators in the ICSID Convention and the ICSID Arbitration Rules

Under the ICSID instruments, the appointment of a human arbitrator is mandatory.  Article 14(1) of the ICSID Convention makes it clear that arbitrators must be natural persons by stipulating that “[p]ersons designated to serve on the Panels shall be persons of high moral character and recognized competence in the fields of law, commerce, industry or finance.”  Qualities such as “high moral character” or “recognized competence” can only be attributed to individuals.  Likewise, Article 15(1) of the ICSID Convention envisages the “death” of a member of a panel as a ground for replacement, leaving no doubt that panels must be composed of natural persons.

As for the ICSID Arbitration Rules, references directed at individuals can also be found in the provisions relating to the establishment of the tribunal.  For instance, Rule 13(3) states that the parties may not appoint “an arbitrator who is a national of the State party to the dispute or the State whose national is a party to the dispute,” unless both parties agree on such an appointment; similarly, Rule 13(4) expresses that “a person previously involved in the resolution of the dispute as a conciliator, judge, mediator or in a similar capacity may be appointed as an arbitrator only by agreement of the parties.”

4.     Provisions Relating to the Making of Awards:  The Challenges Posed by the Lack of Legal Personality of AI Systems and the Black Box Problem

In addition to the rules on the appointment, replacement and challenge of arbitrators, there is another set of rules that pose major difficulties for the deployment of AI in arbitral decision-making:  the provisions relating to the making of awards.  These provisions establish two requirements that are particularly problematic in the case of AI-generated awards, namely (i) the signing of the award and (ii) the provision of reasons for the award.

These two requirements can be found in Articles 31(1) and 31(2) of the MAL, which stipulate that awards must be signed and reasoned, respectively.  In a similar vein, Articles 48(2) and 48(3) of the ICSID Convention provide that the award “shall be signed” and “shall state the reason on which it is based,” respectively.

The fulfilment of these requirements, in turn, raises two issues related to AI-generated awards:  (i) AI systems lack legal personality, which implies that AI systems cannot legally sign the decisions they make; and (ii) the so-called “black box problem,” which compromises the rendering of a reasoned award.25Boris Praštalo, Arbitration Tech Toolbox:  AI as an Arbitrator:  Overcoming the “BlackBox” Challenge?, Kluwer Arb. Blog (Aug. 23, 2024), https://legalblogs.wolterskluwer.com/arbitration-blog/arbitration-tech-toolbox-ai-as-an-arbitrator-overcoming-the-black-box-challenge/ (“The ‘black box’ problem poses a significant challenge to having AI serve as an arbitrator.  Even if an AI model could be designed to render decisions in arbitration cases, the exact factors and patterns it considered to reach its conclusions would remain unknown.”).25

As for the signing of awards, some authors suggest that only humans can meet this requirement26Snijders, supra note 28, at 234 (“[A]rbitrators must sign the arbitral award . . . It is true that this signature can be placed in the form of a qualified electronic signature . . . However, such a signature is . . . of human origin.”).26 or that AI systems, as such, are incapable of signing.27Irene Ng & Valeria Benedetti del Rio, When the Tribunal Is an Algorithm:  Complexities of Enforcing Orders Determined by a Software under the New York Convention, in 60 Years of the New York Convention:  Key Issues and Future Challenges 121, 125 (Katia Fach Gómez & Ana Mercedes López Rodríguez eds., 2019) (“[T]he algorithmic arbitrator might not be able to sign its final decision.  This is generally the case when the decision is made by a software . . . In this scenario, the decision is taken automatically by a computer, without any specific formal requirements such as signatures.  The final award, therefore, may lack the necessary characteristic that is generally under scrutiny when the authenticity of a document needs to be confirmed.”).27  This requirement serves two purposes:  identifying the signatory and attesting the signatory’s knowledge of and assent to the content of the award.28Snijders, supra note 28, at 234 (“‘The function of the signature is always the same, i.e., the signatory indicates that he is aware of the content of the document and expresses his will with his signature’.  Such a will cannot be expressed by a robot.”).28  Additionally, pursuant to Article IV(1)(a) New York Convention (NYC), the enforcement of awards requires the production of a “duly authenticated original award,” provision that has been construed as requiring the signature of the arbitrators.29Eidenmüller & Varesis, supra note 28, at 31 (“Article IV . . . requires a duly authenticated original . . . This provision has been interpreted to require the signatures of arbitrators as natural human beings.’”).29  Since machine-based systems lack legal personality, they cannot identify themselves, let alone express their assent to the content of the decision they produce.

As regards the black box problem, it essentially denotes the inability of human beings to understand how the machine makes the decision.30See Praštalo, supra note 77.30  Should the parties not understand how an AI system would resolve their dispute, the requirement for a reasoned award would be compromised.  The CIArb Guideline 18 illustrates this by stating in Section 2.6, “the ‘black box’ nature of many AI Tools can encumber the obligation to furnish transparent and truthful justifications for specific decisions.”

B.     Legitimacy Concerns about the Use of AI in Arbitral Decision-Making

Arbitration is the preferred method of resolving cross-border disputes,31Queen Mary University of London & White & Case LLP, 2021 International Arbitration Survey: Adapting Arbitration to a Changing World (2021), https://www.qmul.ac.uk/arbitration/media/arbitration/docs/LON0320037-QMUL-International-Arbitration-Survey-2021_19_WEB.pdf (“International arbitration is the preferred method of resolving cross-border disputes for 90% of respondents, either on a standalone basis (31%) or in conjunction with ADR (59%).”).31 and this is largely due to the parties’ ability to select the members of the tribunal.32Queen Mary University of London & White & Case, supra note 23, at 3 (“The primary advantages of international arbitration for resolving disputes involving public interest issues include the ability to select arbitrators with relevant experience or knowledge (47%).”).32  In addition to the professional qualifications and experience of arbitrators, their reputation and interpersonal skills play a decisive role in the parties’ decision on the composition of the tribunal.33Emilia Onyema, Empirically Determined Factors in Appointing Arbitrators in International Commercial Arbitration, 73 Arb. 199, 205 (2007) (“Interviewees were asked, ‘What factors affect your choice of arbitrators?’ . . . The following factors were ranked topmost:  reputation, expertise, common sense, knowledge of applicable law and knowledge of relevant language.”).33

The potential implementation of AI-driven arbitration might undermine the perception of arbitration as a fair and reliable ADR option, as the parties could perceive an AI-generated award as less legitimate than an award rendered by human arbitrators.  The absence of recognised professionals in the tribunal coupled with the fact that AI systems lack the human abilities to empathise and discern may lead to the belief that AI-driven arbitrations are less legitimate than arbitrations conducted by human arbitrators.

1.     The Legitimacy of Arbitration Viewed from the Perspective of the Parties’ Ability to Select Arbitrators and the Public Confidence in Human Arbitrators

“An arbitration is only as good as the arbitrator(s).”34Christophe Seraglini, Who Are the Arbitrators?  Myths, Reality and Challenges, in Legitimacy:  Myths, Realities, Challenges ICCA Congress Series No. 18, at 589, 594 (Albert Jan van den Berg ed., 2015) (“The selection of arbitrators is determinative of the quality of the decision considering that ‘an arbitration is only as good as its arbitrators.’”).34  This maxim reflects not only the importance of the persons that serve as arbitrators but also one of the main reasons behind the good standing of arbitration:  the parties’ ability to select the members of the tribunal.

The legitimacy of arbitration largely rests on the trust that parties place in arbitrators.35Rusinque, supra note 5, at 118 (“[M]uch of the confidence placed in the tribunal is due to the parties’ ability to nominate its members . . . [T]he parties not only bear in mind certain knowledge or skills relevant to solve the dispute, but also the personal reputation and professional background of the candidate.”).35  Under the principle of party autonomy, parties are at liberty to set out the rules for the composition of the tribunal, including the requirements that its members must meet as to their experience and qualifications.36Id. at 108 (“Party autonomy is the bedrock of arbitration, and the appointment of arbitrators is not the exception.  By virtue of this principle . . . parties have the power to establish the number of arbitrators, the qualifications they must possess, the method of nomination and replacement, and so forth.”).36  The fact that parties decide on the composition of the tribunal, in many cases by directly appointing its members, gives the parties a positive perception of the tribunal and, ultimately, lends legitimacy to the proceedings and the award.

Moreover, the public’s confidence in arbitrators also contributes to the legitimacy of arbitration.  AI systems are unlikely to reach the level of public esteem of human arbitrators, not only because machine-based systems will never possess the expertise, professional recognition and interpersonal skills of an experienced arbitrator, but more importantly, because public esteem is earned by a career of distinction conducted with decorum and integrity.  Arbitrators are permanently subject to public scrutiny,37Flavia Marisi, Independence and Impartiality:  The Role of Soft Law in International Arbitration, 85 Arb. 326, 343 (2019) (“A further mechanism to encourage arbitrators to issue carefully reasoned and impartial awards is the scrutiny they are subject to by the fellow arbitrators’ community, scholars, and the civil society.”).37 and it is arbitration users and arbitral institutions who have the final say on the persons who serve as arbitrators.38Rusinque, supra note 5, at 120 (“Who can be in a better position to determine whether an arbitrator acts with probity than the users of arbitration and the arbitration community?  No party or institution would nominate an arbitrator whose integrity and fair judgment are questionable.”).38

2.     AI Systems Lack the Human Abilities to Empathise and Discern

The deployment of machine-based systems in arbitral decision-making also causes disquiet due to AI’s lack of capacity to empathise39Scherer et al., supra note 48, at 541 (“When considering AI for arbitral decision-making, some have speculated about the feasibility of ‘robot-arbitrators’ . . . Authors typically either assert that AI is inevitable in the future, or express scepticism, mainly on the assumption that some “human factor” would be necessary to ensure empathy and emotional justice.”).39—i.e., to understand people’s emotions and their situation in relation to a problem40Daniel Goleman, Emotional Intelligence:  Why It Can Matter More Than IQ 74 (1996) (“[T]he root of caring, stems from emotional attunement, from the capacity for empathy.  That capacity—the ability to know how another feels—comes into play in a vast array of life arenas, from sales and management to romance and parenting, to compassion and political action.”).40—and to discern—i.e., to exercise good judgment based on previous experiences.41Id. at 34 (“[T]he core of interpersonal intelligence includes the ‘capacities to discern and respond appropriately to the moods, temperaments, motivations, and desires of other people.’”).41  These human abilities are key when conducting fact-finding enquiries and dealing with unmeritorious claims.

Indeed, adjudicating disputes requires more than applying the law to a given set of facts—as AI would do mechanically.42Kasap, supra note 3, at 235 (“Human arbitrators do not apply the law purely as a matter of logic and legal syllogism or based on patterns and probabilities, which is how AI functions.  An arbitrator’s personality, as well as their understanding of the world, is not only important for whatever decisions are rendered but also for the parties who appoint the arbitrators in the first place.”).42  Emotional intelligence and interpersonal skills are considered key for an efficient and smooth arbitration.43Id. at 235 (“Practitioners emphasize that arbitrators’ ability to understand and interact with the parties can be reframed as ‘juridical open-minded-ness’ something that sharply delimits the success of the arbitration.  In line with this claim, a 2010 survey found that arbitration users attached importance to the soft skills of arbitrators and consider these skills as having an impact on the efficiency and cost of a case.  It is thus not surprising that parties search for an arbitrator who has effective communicative skills and can show empathy, both of which require emotional intelligence.”).43  The ability to empathise is particularly important for arbitrators to fully grasp the views of the parties and identify the interests at stake in the dispute.  Furthermore, arbitrators conduct fact-finding enquiries where the perception of emotions holds considerable value,44Florescu, supra note 2, at 52 (“[E]motions, feelings, moods, weaknesses . . . impact human judgement . . . Reflecting the need for a nuanced analysis of facts that might and perhaps should influence a decision, emotions are still needed to give sense and purpose to the legal decision-making process.”); Goleman, supra note 92, at 74 (“People’s emotions are rarely put into words; far more often they are expressed through other cues.  The key to intuiting another’s feelings is in the ability to read nonverbal channels:  tone of voice, gesture, facial expression, and the like.”).44 like the examination of witnesses.45Reed, supra note 11, at 87 (“At heart, the art of arbitral decision-making lies in intuiting whether and when witnesses are telling the truth, in perceiving the human stories underlying a business dispute, in crafting an award with the right reasoning and the right amount of reasoning.”).45

Similarly, based on their experience and discernment, arbitrators can detect and prevent bad faith manoeuvres, such as frivolous claims and defences, ensuring that the proceedings are conducted without undue delays.46Seraglini, supra note 86, at 594 (“Experience indeed has an important role to play in ensuring a high quality of justice regarding the conduct of the proceedings as well as the merits of a case.  An experienced arbitrator:  (a) knows how to efficiently conduct arbitral proceedings; and (b) has the proper skills to adequately resolve the matter in dispute.”).46  More importantly, arbitrators weigh up the possible outcomes of their decisions and exercise discretion to arrive at the most appropriate solution for each specific case.  As Gizem Halis observes:

An AI arbitrator that can excel in logical—mathematical intelligence will process a numeric input and deploy a series of algorithms to generate an output.  When doing so, however, the AI arbitrator will not be aware of the award that it will enter, nor will it understand why it has entered such an award.  Since an AI arbitrator does not have a sense of self, it has no intrapersonal skills to gauge the success of its performance . . . This may cause grave mistakes in arbitration, like the example of the AI algorithm that decided to sacrifice the life of the pilot in a flight simulation because it determined that crashing the plane was the optimal path to obtaining the highest landing score.47Kasap, supra note 3, at 234.47

The Right to a Human Decision in International Arbitration

The right to a human decision is an appealing solution to address the concerns raised by the deployment of AI in arbitral decision-making.  Its different formulations offer versatility and, more importantly, it aligns international arbitration with the regulatory framework on ADM.  This section outlines the possible formulations, scope and legal bases of the right to a human decision in international arbitration.

The first subsection explores the limits of party autonomy and the prospects for the use of AI in arbitral decision-making.  In particular, the first subsection discusses whether and, if so, to what extent the parties are entitled to contract out of the provisions identifying arbitrators as natural persons.

Subsequently, the second and third subsections analyse the possible formulations of the right to a human decision in international arbitration.  The second subsection focuses on the right to a human decision as an exclusion of AI from arbitral decision-making, which is the most plausible formulation in jurisdictions that require the appointment of human arbitrators.  In turn, the third subsection considers the right to a human decision as a right to contest AI-generated decisions.  This formulation seeks to ensure that AI-driven arbitration and AI-generated awards are compatible with the regulatory framework for the use of AI in decision-making.

A.     Exploring the Limits of Party Autonomy and the Prospects for the Use of AI in Arbitral Decision-Making

As explained in section IV, the deployment of AI in arbitral decision-making raises concerns about the compatibility of AI-driven arbitration and AI-generated awards with the provisions relating to the composition of the arbitral tribunal and the making of awards.  Similarly, the good standing and legitimacy of arbitration might be also compromised not only because AI is incapable of empathising and discerning, but also because AI is unlikely to achieve the level of public trust enjoyed by human arbitrators.

However, the one-million-dollar question is, can those concerns be overcome?  Put differently, is there an alternative approach that leaves it up to the parties to decide whether to refer their dispute to AI-driven arbitration, without compromising arbitration’s perception and without contravening the legal framework of international arbitration and the regulatory standards on ADM?

Addressing these questions implies analysing the limits of party autonomy, as well as approaching the legitimacy of arbitration from the perspective of one of its most notable characteristics:  procedural flexibility.  In this manner, the first question to be answered is, can the parties opt out of the provisions that identify arbitrators as natural persons? 

1.     Can the Parties Opt Out of the Provisions that Identify Arbitrators as Natural Persons?

As noted in section IV, the bulk of curial laws and arbitration rules presuppose that arbitrators are natural persons.  The question is, can the parties contract out of these provisions?  In principle, the parties are at liberty to modify the provisions on the appointment, replacement and challenge of arbitrators, insofar as these provisions contain default rules, that is, rules that supplement the agreement of the parties when they do not estipulate the way arbitrators are to be appointed, replaced or challenged.  This reasoning applies to the MAL, whose Articles 11(2) and 13(1) state that, “[t]he parties are free to agree on a procedure of appointing the arbitrator” and “[t]he parties are free to agree on a procedure for challenging an arbitrator,” respectively.

Nonetheless, a few curial laws and international instruments mandatorily express that arbitrators must be natural persons.  Some examples are the French Code of Civil Procedure,1Code de Procédure Civile, art. 1450 (Fr.).1 the Dutch Code of Civil Procedure (Wetboek van burgerlijke rechtsvordering)2Snijders, supra note 28, at 234 (“[T]he first sentence of Article 1023 of the Dutch Code of Civil Procedure provides that only natural persons of legal capacity can be appointed as arbitrators.”).2 and the ICSID Convention.3ICSID Convention, supra note 62, art. 14(1).3  Where the lex arbitri or the instrument applicable provides that arbitrators must be natural persons, neither agreements to refer disputes to AI-driven arbitration nor AI-generated awards are likely to be enforced.4Onyefulu, supra note 64, at 63 (“It has been contended that an arbitral award rendered by an AI arbitrator may contravene several public policy grounds.  For example, where the national legislation of the seat of arbitration expressly provides that an arbitrator must be a natural person, any award rendered by an AI arbitrator is in contravention of the public policy of the national legislation.”).4

Moreover, even if the seat of the arbitration is a jurisdiction where the parties may opt out of the rules providing for the appointment of natural persons as arbitrators, but enforcement of an AI-generated award is sought in a jurisdiction where only natural persons can serve as arbitrators, such enforcement may be refused on grounds of the public policy exception envisaged in Article V(2)(b) of the NYC.5Ng & Benedetti del Rio, supra note 79, at 131 (“If a party intends to use Article V.2(b) to invalidate an arbitral award by an algorithmic arbitrator, such party must prove that the use of an algorithmic arbitrator would be against the public policy of that country . . . [T]here are a few countries such as France that effectively do so by limiting the definition of an arbitrator to that of a natural person.”).5

This distinction between instruments that mandatorily require arbitrators to be natural persons and instruments that, while identifying arbitrators as human beings, allow the parties to establish the rules on the composition of the arbitral tribunal is crucial to understand why the right to a human decision is susceptible of two formulations in international arbitration, namely:  (i) as an exclusion of AI from arbitral decision-making, if only natural persons can serve as arbitrators; and (ii) as a right to contest AI-generated decisions, in cases where the parties are entitled to opt out of the provisions that identify arbitrators as natural persons.

2.     The Legitimacy of Arbitration Viewed from the Perspective of Its Procedural Flexibility and the Pre-Eminence of Party Autonomy

Two of the most important features of arbitration are procedural flexibility and party autonomy.6Karl-Heinz Böckstiegel, The Role of Party Autonomy in International Arbitration, 52 Disp. Resol. J. 24 (1997) (“The principle of party autonomy is fundamental to arbitration in general and to international arbitration in particular.”).6  Thanks to them, the parties are entitled to tailor the arbitration to their needs and expectations.7Queen Mary University of London & White & Case, supra note 83, at 1 (“The field of international arbitration is dynamic by nature.  Its hallmarks of flexibility and party autonomy allow it to develop and adapt in response to the needs of its users.”).7  There is no doubt that the good standing of arbitration is also due to these key principles.  Additionally, party autonomy has been, and continues to be, the cornerstone of arbitration.8Report of the Secretary-General, supra note 68, ¶¶ 17, 78 (“Probably the most important principle on which the model law should be based is the freedom of the parties in order to facilitate the proper functioning of international commercial arbitration according to their expectations.  This would allow them to freely submit their disputes to arbitration and to tailor the ‘rules of the game’ to their specific needs.”).8  Other principles and features of arbitration, including procedural flexibility and the parties’ ability to select arbitrators, hinge upon party autonomy.

In this context and given the growing demand for expeditious and cost-effective proceedings, the implementation of AI-driven arbitration and AI-generated awards could foster the perception of arbitration as an efficient and flexible ADR option.9Queen Mary University of London & White & Case, supra note 24, at 28–29 (“We asked what respondents considered to be the principal drivers for greater use of AI in international arbitration. . . . The most selected reason was the potential to save party and counsel time (54%) . . . Reduction of costs was the second most selected driver, with 44% of respondents citing this benefit.”).9  After all, arbitration is a service industry10Alan Shilston, No Hired Guns, 64 Arb. 1 (1998) (“It should never be forgotten that . . . arbitration is a service activity in the promotion of international trade dedicated to providing a very real option to the state’s procedural methods of resolving disputes in public.”).10 and, as such, it develops and adapts in line with the changing demands of its users.11Niyati Ahuja & Lucila Marchini, New York Arbitration Week 2021 Redux:  Options, Opportunities, Pivots, and Challenges in an International Arbitration Career, Kluwer Arb. Blog (Dec. 6, 2021) https://legalblogs.wolterskluwer.com/arbitration-blog/new-york-arbitration-week-2021-redux-options-opportunities-pivots-and-challenges-in-an-international-arbitration-career/ (“Arbitration is a service industry which adapts to the needs and demands of its users.  Those needs and demands have changed over time.  Economies have become more globalized, and this globalization has led to the development of the arbitration practice in all parts of the world.”).11

Moreover, the fact that arbitration users have the last word on the implementation of AI-driven arbitration and AI-generated awards would enhance the legitimacy of arbitration.  Lengthy and costly proceedings constitute a major concern for many arbitration users, and the seemingly indifferent attitude of arbitrators and arbitral institutions towards this situation has raised doubts about the legitimacy of arbitration.  In the words of Stephan Schill:

[T]he legitimacy of arbitration may be questioned by disputing parties, in light of increasing costs and length of arbitration proceedings, as regards the efficiency and expediency of the process; the legitimacy of arbitration is affected in this context as arbitration institutions and arbitrators may be seen by some as having a financial interest in the number and duration of arbitrations.12Stephan Schill, Developing a Framework for the Legitimacy of International Arbitration, in Legitimacy:  Myths, Realities, Challenges ICCA Congress Series No. 18 789, 800 (Albert Jan van den Berg ed., 2015).12

B.     The Right to a Human Decision as an Exclusion of AI from Arbitral Decision-Making

The right to a human decision already exists in the jurisdictions and legal instruments that require arbitrators to be natural persons.  As noted above, France and the Netherlands provide in their curial laws that the role of arbitrators can only be performed by natural persons.  In the same vein, the ICSID Convention states that, in order to serve on arbitration panels, persons must have “high moral character” and “recognized competence.”  In these cases, the right to a human decision is formulated as an exclusion of AI from decision-making.

It can also be argued that the right to a human decision exists in the jurisdictions whose curial laws provide for human arbitrators by default.  As discussed earlier, curial laws, especially those based on the MAL, contain default rules that identify arbitrators as natural persons.  Thus, if the parties’ agreement stipulates that the seat of the arbitration is in a jurisdiction whose curial laws presuppose that arbitrators are human beings, the right to a human decision as an exclusion of AI from decision-making comes into existence, unless the parties decide to opt out of the provisions designating natural persons as arbitrators.

1.     The Arbitrators’ Duty Not to Delegate Their Mandate:  The Influence of the CIArb Guideline 18 and the SVAMC Guidelines on the Articulation of the Right to a Human Decision

As described in section II, the arbitrators’ primary obligation consists in resolving the dispute between the parties.  This obligation, in turn, entails specific duties for arbitrators, including the duty to personally complete their mandate and not to delegate their decision-making tasks.  Gary Born describes this duty as follows:

Most fundamentally, an arbitrator cannot delegate the duty of deciding a case, attending hearings or deliberations, or evaluating the parties’ submissions and evidence to others:  these are the essence of the arbitrator’s adjudicative function and they are personal, non-delegable duties.  The arbitrator alone must fulfill those obligations, just as a judge alone most fulfill these responsibilities in most legal systems.13Born, supra note 12, § 13.04[A][8].13

The recently adopted CIArb Guideline 18 confirms the necessity of preserving human agency in arbitral decision-making, while leaving the door open for arbitrators to employ AI tools that contribute to enhancing the accuracy and efficiency of processing information disclosed in the proceedings.  Indeed, this instrument states in Section 8.2 that, “Arbitrators should not relinquish their decision-making powers to AI but may use AI to support more accurate and efficient processing of submitted information, always ensuring independent judgment.”  Similarly, the SVAMC Guidelines provides that:

An arbitrator shall not delegate any part of their personal mandate to any AI tool.  This principle shall particularly apply to the arbitrator’s decision-making process.  The use of AI tools by arbitrators shall not replace their independent analysis of the facts, the law, and the evidence.14SVAMC Guidelines, supra note 19, Guideline 6.14

Both the CIArb Guideline 18 and the SVAMC Guidelines stress the necessity for arbitrators to maintain an “independent judgment,” which implies that, regardless of the use of AI tools, arbitrators remain responsible and accountable for the execution of their duties and decision-making tasks.  In other words, the deployment of AI “to support more accurate and efficient processing of submitted information” does not exempt arbitrators from personally performing their decision-making role, which includes evaluating the parties’ submissions, evaluating the admissibility and relevance of evidence, and interpreting and applying the law to the facts of the case and deciding the case.

The importance of these instruments should not be overlooked.  They reflect best practices and can be used in the construction of other sources of law applicable in international arbitration.15Rusinque, supra note 5, at 122 (“Soft-law instruments are relevant for the construction of . . . provisions . . . Said instruments compiled the best practices in international arbitration, depicting the so-called procedural lex mercatoria, understood as a set of transnational principles . . . relevant for conducting arbitral proceedings.”).15  In particular, the CIArb Guideline 18 and the SVAMC Guidelines begin to set the tone for the general recognition of the right to a human decision formulated as an exclusion of AI from arbitral decision-making.  Their express references to the arbitrators’ duty not to delegate their mandate to AI constitutes a clear limit on the deployment of AI in arbitral decision-making.

C.     The Right to a Human Decision as a Right to Contest AI-Generated Decisions

Given the importance of party autonomy and the flexibility of arbitration, coupled with the fact that the appointment of human arbitrators is not mandatory under several curial laws—including those based on the MAL, ADR service providers are likely to offer and administer AI-driven arbitrations.  In fact, the AAA-ICDR recently announced the launch of an “AI arbitrator” intended for use in construction-related disputes.16AAA-ICDR, supra note 25.16  In these circumstances, the right to a human decision would continue to apply, but in the form of a right to contest AI-generated decisions.

This subsection explains how AI-driven arbitration can be implemented and, specifically, how the legal requirements relating to the signing and reasoning of awards can be met.  Likewise, this subsection examines the scope of the two mechanisms that form part of the right to contest AI-generated decision, namely human oversight of AI-driven arbitration and review of AI-generated decisions.

1.     How Can AI-Driven Arbitration Be Implemented?  Signing and Reasoning of AI-Generated Awards

As explained in section IV, the rules on the making of awards establish two requirements that pose difficulties for AI-generated awards, namely (i) the signing of the award and (ii) the provision of reasons for the award.  These requirements can be met if AI-driven arbitration is administered by ADR service providers.

For reasons of practicability and objectivity, the parties are in no position to equip themselves with the AI system for the resolution of their disputes.  AI-driven arbitration is to be conducted under the auspices of independent ADR service providers.

Typically, ADR service providers are nonprofit or trade and commerce organisations with legal personality.  In this manner, the parties, by virtue of their party autonomy and ability to select arbitrators, can designate an ADR service provider as arbitrator, which will then be responsible for conducting the proceedings and adjudicating the dispute through AI-driven arbitration.  

The possibility of designating a legal person as an arbitrator is not new and is generally accepted in many jurisdictions.17Born, supra note 12, § 12.04[B][1] (“[M]ost developed jurisdictions permit juridical persons to serve as arbitrators, although in practice the appointment of juridical persons as arbitrators is very unusual.”).17  Although the bulk of curial laws contain provisions that identify arbitrators as natural persons, most of such provisions are default rules that can be modified by the parties.  Moreover, Article I(2) NYC states that the definition of “arbitral awards” encompasses “those made by permanent arbitral bodies to which the parties have submitted.”18João Moreira and Riccardo Segate, The “It” Arbitrator:  Why Do Corporations Not Act as Arbitrators?, 12 J. Int’l Disp. Settlement 525, 537 (2021) (“[T]he New York Convention does not impose any requirements that an arbitrator is a natural person and explicitly admits arbitral awards made by permanent arbitral bodies.”).18

In these circumstances, the requirement to sign the award, which implies that the signatory identifies itself and expresses its assent to the content of the award, can readily be fulfilled by ADR service providers administering AI-driven arbitration.

As regards the requirement for a reasoned award, the black box problem can also be overcome.  Indeed, technological advancements in AI now enable the so-called explainable AI to set out the reasons for its decisions.19Praštalo, supra note 77 (“A potential solution to this conundrum has been devised in the form of the so-called ‘Explainable AI’ which not only provides desired outputs, but also explains how it generated those outputs.”).19  Furthermore, AI-generated awards will be subject to the scrutiny of ADR service providers, which ultimately will assess the soundness and legal basis of the decisions contained in the award. 

It is of the utmost importance that, given the lack of legal personality of AI, ADR service providers assume responsibility and accountability for the performance of the tasks and duties involved in arbitral decision-making.  The implementation of AI-driven arbitration by ADR service providers cannot circumvent the obligations incumbent on arbitrators.

2.     Contestation Mechanisms:  Human Oversight of AI-Driven Arbitration and Review of AI-Generated Decisions

The most suitable formulation of the right to a human decision for AI-driven arbitration is the right to contest AI-generated decisions.  An effective contestation of AI-generated decisions requires at least two mechanisms:  (i) human oversight of AI-driven arbitration, which aims at averting and mitigating risks posed by ADM; and (ii) review of AI-generated decisions, which enables the parties to require human review of decisions made by AI.  These two mechanisms have their legal bases in both the instruments applicable in international arbitration and the instruments regulating the use of AI in decision-making.   

Arbitrators, including ADR service providers administering AI-driven arbitration, have the duty to render valid and enforceable awards.20Born, supra note 12, § 13.04[A][4] (“One of the arbitrator’s most significant obligations is to render an award that is enforceable.  This duty is frequently expressed, as an objective, in institutional arbitration rules.”).20  To this end, they shall preserve the integrity of the proceedings driven by AI, ensuring that the system treats the parties with equality and gives them the opportunity to present their case—Articles V(1)(b) NYC and 18 MAL.  It is also crucial to ensure that the agreement submitting the dispute to AI-driven arbitration is valid under its applicable law—Articles V(1)(a) NYC and 34(2)(a)(i) MAL.  Ideally, the parties’ agreement or the institutional rules of the ADR service provider should designate a system of law in which the appointment of legal persons as arbitrators is possible as the proper law of the agreement; otherwise there is a risk that the lex arbitri or the substantive law of the underlying contract, which could govern the validity-related aspects of the arbitration agreement in the absence of an express choice-of-law provision for the agreement itself,21Renato Nazzini, The Law Applicable to the Arbitration Agreement:  Towards Transnational Principles, 65 Int’l & Comp. L.Q. 681 (2016) (“There are currently diverging views as to whether the law applicable to the arbitration agreement should be the law chosen by the parties to govern their substantive legal relationship or the law of the seat of the arbitration.”).21 require the appointment of human arbitrators, compromising the validity of the agreement.  

Additionally, ADR service providers administering AI-driven arbitrations must also consider the applicability of regulation on ADM.  As explained in section III, international arbitrations may fall within the scope of regional instruments limiting ADM, in which case the standards on privacy and human rights established therein must be observed to preserve the integrity of the proceedings and prevent challenges to AI-generated awards.  More importantly, the EU AI Act establishes the obligation for deployers to assign human oversight of high-risk AI systems, including those intended for use “in researching and interpreting facts and the law and in applying the law to a concrete set of facts.”

Against this backdrop, human oversight of AI-driven arbitration is necessary to prevent technical malfunctions and flaws in AI decision-making that may compromise the integrity of the proceedings or the validity and enforcement of AI-generated awards.

It is worth highlighting that human oversight of AI-driven arbitration must be limited to verifying the correct functioning of the system.  A full-service and well-calibrated AI-arbitration system should be capable of conducting the proceeding and adjudicating the dispute without human intervention.  The potential for AI-driven arbitration lies in disputes characterised by intensive document production and, particularly, in arbitration sectors with sufficient publicly available information, so that AI systems can be trained on how to reason and decide as a human arbitrator would.22Paul Bennett Marrow, Mansi Karol & Steven Kuyan, Artificial Intelligence and Arbitration:  The Computer as an Arbitrator—Are We There Yet?, 74 Disp. Resol. J. 35–36 (2020) (“Computer scientists posit that with enough data and properly designed algorithms, the ‘well-trained’ computer should soon be able to produce an acceptable arbitration award.  The necessary data set would hopefully include thousands of transcripts from actual arbitration proceedings; hundreds of thousands of actual awards; all known reported judicial opinions issued by courts (and) all relevant statutes and rules used by judges, lawyers, arbitrators, and the administrators of the arbitration process.”).22

As regards the review of AI-generated decisions, foreseeing the possible grounds and scope of this contestation mechanism, including the questions of whether it should only cover awards or extend to procedural orders, is not an easy endeavour.  One might expect that, since AI-driven arbitration aims to streamline arbitral proceedings and make them cost-effective, a de novo review of AI-generated awards in the form of an appeal is off the table.  If the human reviewer had to go over the claims and defences of the parties and decide on the merits of the case for a second time, AI-driven arbitration would make little sense.

The review of AI-generated awards should provide the parties with the opportunity to flag technical flaws that affect the system’s decision-making.  Two specific issues on which this contestation mechanism should focus are hallucinations and systemic algorithmic bias, which are characteristic risks posed by ADM.23Stefanie G. Efstathiou & Mihaela Apostol, Arbitration Tech Toolbox:  ChatGPT–Arbitral Assistant or Fourth Arbitrator?, Kluwer Arb. Blog (July 22, 2023) https://arbitrationblog.kluwerarbitration.com/2023/07/22/arbitration-tech-toolbox-chatgpt-arbitral-assistant-or-fourth-arbitrator/.23  When making decisions, AI systems might refer to non-existent or inapplicable legal sources, rely on evidence not produced in the proceedings or make erroneous assumptions based on the information they have previously processed.

Conclusion

The right to a human decision is gaining momentum and arbitration can benefit from it.  Its versatility makes it suitable for instruments that require the designation of human arbitrators and those that allow the parties to resort to AI-driven arbitration through ADR service providers.  Most importantly, the right to a human decision aligns international arbitration with the regulatory standards on ADM.

As reliance on ADM increases, new regulatory frameworks are established, and the right to a human decision is an appropriate channel for their enforcement in international arbitration.  Compliance with privacy and human rights standards is key to preserve the integrity of arbitral proceedings, especially if AI is used in decision-making.

The CIArb Guideline 18 and the SVAMC Guidelines provide a sound legal basis for the recognition of this right as an exclusion of AI from decision-making, since such instruments make it clear that human arbitrators “should not relinquish their decision-making powers to AI.”  Additionally, where parties are entitled to designate ADR service providers administering AI-driven arbitration, this right is formulated as a right to contest AI-generated decisions, facilitating compliance with the regulatory standards on ADM.

Lastly, the right to a human decision enhances the good standing and legitimacy of arbitration as an efficient, cost-effective, flexible and user-oriented ADR alternative, leaving up to the parties the possibility of utilising AI in decision-making.

Endnotes

1Mahnoor Waqar, The Use of AI in Arbitral Proceedings, 37 Ohio St. J. on Disp. Resol. 343, 351 (2022) (“[The] introduction of AI systems which uses Natural Language Processing (‘NLP’) . . . would significantly reduce the tasks of a lawyer engaging in arbitration.”); Layan Al Fatayri, AI in International Arbitration:  What Is the Big Deal?, Blog of the Am. Rev. Int’l Arb. (Oct. 22, 2024) https://aria.law.columbia.edu/ai-in-international-arbitration-what-is-the-big-deal/ (“Through natural language processing (NLP) and machine learning algorithms, AI can efficiently sift through large volumes of documents to identify relevant information, reducing the time and costs associated with manual review.”).
2Maroof Rafique, Why Artificial Intelligence Is a Compatible Match for Arbitration, 88 Arb. 310, 315 (2022) (“[A]utomation is increasing its share in the market for legal services.  According to some reports, 80% of leading law firms are already using applications of AI.  Arbitrators are using AI for e-discovery, procedural automation, and drafting of awards.”); Cristina Florescu, The Interaction Between AI (Artificial Intelligence) and IA (International Arbitration):  Technology as the New Partner of Arbitration, 18 Revista Română de Arbitraj 42 (2024) (“Nowadays, nearly all phases of an international arbitration are facilitated by technology, with tools broadly accessible for:  selection of arbitrators, case management, discovery, legal research, document review and hearings.”).
3Gizem Halis Kasap, Can Artificial Intelligence (“AI”) Replace Human Arbitrators?  Technological Concerns and Legal Implications, 2021 J. Disp. Resol.209, 238–39 (2021)(“[Some] international arbitration laws . . . do not explicitly state that arbitrators must be human beings but envisage them as such by requiring standards that only a human can meet or attribute certain characteristics that are exclusive to human beings.”).
4Id. at 238 (“In [some] national laws, the arbitrator is explicitly required to be a natural person with full capacity, as opposed to being a legal person.  The most notable standards are set by the French Code of Civil Procedure, the Dutch Code of Civil Procedure, and the Portuguese Voluntary Arbitration Law.  All of which stipulate that only ‘a natural person’ or ‘individuals’ can act as an arbitrator.”).
5Felipe Rusinque, The System of Party-Arbitration:  The Legitimacy of Party-Appointed Arbitrators and the Right to Make Unilateral Nominations, 27 Int’l Arb. L. Rev. 102, 125 (2024) (“[T]he ability to select arbitrators . . . gives the parties control over the composition of the tribunal, so they not only ensure that arbitrators possess the knowledge and skills required to solve the dispute, but also bring to the tribunal persons held in high esteem for their professional career and integrity.”).
6Aziz Z. Huq, A Right to a Human Decision, 106 Va. L. Rev. 611 (2020).
7Margot E. Kaminski & Jennifer M. Urban, The Right to Contest AI, 121 Colum. L. Rev. 1957, 1960 (2021) (“[T]he use of AI in decision-making has increased.  The government and private sector now use algorithms to decide how to distribute welfare benefits, whether to hire or fire a person, whether expressive material should be removed from online platforms, whether to keep people in prison, and more.”); Ifeoma Ajunwa, Automated Governance, 101 N.C. L. Rev. 355, 357 (2023) (“In many ways, as consumers, we have already acquiesced to our algorithmic overlords.  We have increasingly delegated consequential business decisions about insurance, home purchase, or employment opportunity to automated algorithmic processes.”).
8Huq, supra note 6, at 613–14 (“[M]any people feel a loss of control over key life decisions.  Machines, they fear, resolve questions of critical importance on grounds that are beyond individuals’ ken or control.  Many individuals experience a loss of elementary human agency and a corresponding vulnerability to an inhuman and inhumane machine logic.  For some, ‘the very idea of an algorithmic system making an important decision on the basis of past data seem[s] unfair.’”).
9Nigel Blackaby, Constantine Partasides and Alan Redfern point out the following “key elements of an international arbitration:”  (i) the agreement to arbitrate, (ii) the place of arbitration, (iii) the commencement of an arbitration, (iv) the appointment of the arbitral tribunal, (v) arbitral proceedings, (vi) the award of the tribunal and (vii) the enforcement of the award.  Nigel Blackaby, Constantine Partasides & Alan Redfern, Redfern and Hunter on International Arbitration ¶ 1.19 (7th ed. 2023).
10Mary Mitsi, The Decision-Making Process of Investor-State Arbitration Tribunals 21 (2019) (“The decision-making process signifies the specific process in which arbitrators engage in order to resolve a specific dispute.”).
11Lucy Reed, The 2013 Hong Kong International Arbitration Centre Kaplan Lecture – Arbitral Decision-Making: Art, Science or Sport?, 30 J. Int’l Arb. 85, 87 (2013) (“The science of arbitral decision-making lies in rigorously assessing the evidence, methodically finding the relevant material facts, identifying the governing law [and] applying that law to the facts.”).
12Gary Born, International Commercial Arbitration § 13.04[A][6] (3d ed. 2021) (“The arbitrator exercises authority conferred by law, to resolve the parties’ disputes in a legally-binding manner.  As noted above, ‘the services of an arbitrator are principally and habitually those of settling a dispute between two or more parties.’ . . . [T]his is an adjudicative function and authority, which consists in hearing the parties’ submissions and evidence in fair, objective proceedings and rendering an impartial, reasoned decision that finally decides their rights on the basis of their submissions and evidence.”).
13Id. § 13.04[A] (“The arbitrator’s central obligation is to resolve the parties’ dispute in an adjudicatory manner. . . . [T]his is the basic object of any arbitration agreement and it is the essential reason that the parties appoint an arbitrator and present their cases to him or her.”).
14Id. § 13.04[A][2] (“Essential elements of the arbitrator’s obligation to resolve the parties’ dispute in an adjudicative manner are the related duties to be and remain impartial and independent and to conduct the arbitration impartially and independently.”).
15Blackaby et al., supra note 9, ¶ 5.56 (“It seems appropriate to expect the same standard of professional care from a lawyer . . . who is serving as an arbitrator as would be expected in the course of his or her other professional work.  The parties to an arbitration entrust an arbitral tribunal with an important task, for which they are prepared to pay—often generously.  They therefore expect the arbitrator to perform the task with due care.”).
16Born, supra note 12, § 13.04[E] (“[A]n arbitrator’s acceptance of his or her appointment entails an implied undertaking to complete that mandate, by issuing a final award.”).
17Id. § 13.04[A][4] (“The arbitrator’s obligations include deciding all of the disputes which are presented to him or her. . . . [A]n arbitrator’s failure to decide questions that are submitted to the tribunal constitutes infra petita, and may be grounds for annulling or refusing to recognize the award.”).
18Id. § 13.04[A][8] (“An arbitrator’s obligations include the duty not to delegate his or her responsibilities or tasks to third parties.”).
19Silicon Valley Arb. & Mediation Ctr., Guidelines on the Use of Artificial Intelligence in Arbitration Guideline 6 (2024) [hereinafter SVAMC Guidelines]; Chartered Inst. of Arbitrators, Guideline on the Use of AI in Arbitration art. 8.2 (2025).
20Born, supra note 12, § 13.04[A][8] (“Arbitrators obviously use clerical assistance, which can readily extend beyond typing, organizing files and the like, to dealing with administrative matters; these tasks will often be conducted either by secretarial or similar staff, but sometimes by junior lawyers or interns. . . . [I]n some instances arbitrators will appoint a ‘secretary’ to assist the tribunal.”).
21Id. (“[I]t is common practice for secretaries (or other junior lawyers) to conduct legal research for tribunals, to organize files and witness statements (sometimes going beyond merely administrative work) and sometimes to draft administrative communications, procedural orders, or portions of awards.”).
22Danna Zhang, Integrating New Technologies in International Arbitration:  Thoughts on the Future of Our Profession, in Stockholm Arbitration Yearbook 2024, 100, 110–11 (Patrik Schöldström & Christer Danielsson eds., 2024) (“In the very near future, arbitrators will also likely be able to acquire personal AI secretaries who are fully tailored and trained in accordance with their individual preferences . . . .  Many lawyers will recognise these tasks as typical junior-level work:  drafting documents or email replies and preparing summaries for seniors to review.  And in many cases, this is not too different from the tasks typically performed by tribunal secretaries in international arbitration proceedings.”).
23Queen Mary University of London & White & Case LLP, 2025 International Arbitration Survey:  The Path Forward:  Realities and Opportunities in Arbitration (2025) https://www.whitecase.com/sites/default/files/2025-06/white-case-qmul-2025-international-arbitration-survey-report-web.pdf (“Arbitration is commonly praised for its flexibility and ability to meet the needs of its users.  Yet concerns persist about delays and inefficiencies throughout the arbitral process.  Time and cost of the proceedings consistently arise among the most significant drawbacks for users.”).
24According to the 2025 International Arbitration Survey of Queen Mary University and White & Case, arbitration users selected “saving time” and “reducing costs” as the two main reasons for the use of AI in arbitral proceedings.  Id. at 28–29.
25The AAA-ICDR announced the launch of an “AI arbitrator” intended for use in construction-related disputes and indicated that AI will also support other industries in 2026.  Am. Arbitration Ass’n-Int’l Ctr. for Dispute Resol., AAA-ICDR to Launch AI-Native Arbitrator, Transforming Dispute Resolution (Sept. 17, 2025), https://www.adr.org/press-releases/aaa-icdr-to-launch-ai-native-arbitrator-transforming-dispute-resolution/.
26Horst Eidenmüller & Faidon Varesis, What is an Arbitration?  Artificial Intelligence and the Vanishing Human Arbitrator, 17 N.Y.U. J.L. & Bus. 49, 62 (2020) (“A fully automated ‘robot’ AI arbitrator system, which would substitute human arbitrators, requires performance of a variety of tasks . . . A full-service AI arbitrator system would require a fully automated case management process of accepting cases, performing conflict checks, managing multiple appointments, cases, and hearings at the same time, without the involvement of human input.  In addition, it would need to perform factual analysis of a given case, including e-discovery of facts and document analysis . . . Based on this analysis, the system would have to render a final and binding decision as well as give the reasons for reaching this decision.”).
27David Freeman Engstrom, Daniel E. Ho, Catherine M. Sharkey & Mariano-Florentino Cuéllar, Government by Algorithm:  Artificial Intelligence in Federal Administrative Agencies 6 (Feb. 2020) (report to the Admin. Conf. of the U.S.), available at https://law.stanford.edu/wp-content/uploads/2020/02/ACUS-AI-Report.pdf (“The government’s AI toolkit is diverse and spans the federal administrative state.  Nearly half of the federal agencies studied (45%) have experimented with AI and related machine learning (ML) tools.”  “Adjudicating government benefits and privileges, from disability benefits to intellectual property rights” was one of the governmental tasks where the use of AI was reported. (emphasis in original)).
28See Henk Snijders, Arbitration and AI, from Arbitration to ‘Robotration’ and from Human Arbitrator to Robot, 87 Arb. 223, 226 (2021).
29Ajunwa, supra note 7, at 361–62 (“[T]here are tangible benefits to deploying algorithmic processes for administrative decision-making . . . some of those benefits [include] efficiency, cost-savings, and the ability of algorithmic systems to discern patterns displaying historical bias in their rulemaking or adjudication.”).
30Id. at 363–64 (“Cost-savings are an important consideration for the adoption of any technological tool; and even more so when the taxpayers bear that cost . . . Cost-savings, however, can never be the sole consideration for adopting automated decision-making tools.  As several legal scholars have noted, equity and justice should remain the paramount guide.”).
31Kaminski & Urban, supra note 7, at 1957 (“These uses of AI raise a host of concerns about discrimination, accuracy, fairness, and accountability.”).
32Huq, supra note 6, at 685 (“[M]achine decisions are not presently appropriate for decisions with ethical or normative components.  Many legal and even factual questions resolved through civil and criminal adjudication have some such component.”).
33Id. at 611 (“Machine-learning tools are perceived to be eclipsing, even extinguishing, human agency in ways that compromise important individual interests.  An emerging legal response to such worries is to assert a novel right to a human decision.”).
34Giovanni Sartor, Artificial Intelligence and Human Rights:  Between Law and Ethics, 27 Maastricht J. Eur. & Comp. L. 705 (2020); Stefan Schäferling, Governmental Automated Decision-Making and Human Rights:  Reconciling Law and Intelligent Systems 231 (2023); Yuval Shany, The Case for a New Right to a Human Decision Under International Human Rights Law (Oct. 4, 2023), https://ssrn.com/abstract=4592244; Marya Akhtar & Rikke Frank Jørgensen, A Rights-Based Approach to Automated Decision-Making in the Public Sector, in Artificial Intelligence and International Human Rights Law:  Developing Standards for a Changing World 69 (Michał Balcerzak & Julia Kapelańska-Pręgowska eds., 2024).
35Elena Abrusci & Richard Mackenzie-Gray Scott, The Questionable Necessity of a New Human Right Against Being Subject to Automated Decision-Making, 31 Int’l J.L. & Info. Tech. 114, 143 (2023).
36Kaminski & Urban, supra note 7, at 1988.
37Michael Cheng, The Right to Human Decision:  Analyzing Policies, Ethics, and Implementation 2–3 (Jan. 31, 2024), https://openreview.net/forum?id=VrxtvBw0ly.
38Huq, supra note 6, at 621–22; Kaminski & Urban, supra note 7, at 2012–14; Abrusci & Mackenzie-Gray Scott, supra note 35, at 121; Shany, supra note 34, at 20.
39Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law art. 14(2)(c), Sept. 5, 2024, C.E.T.S. No. 225.
40Martin Zahariev, Mission (Im)possible:  Where GDPR Meets Commercial Arbitration, in Austrian Yearbook on International Arbitration 2020 3 (Christian Klausegger et al. eds., 2020) (“In other words, by virtue of either the establishment criterion . . . or the targeting criterion . . . arbitral institutions and arbitrators might fall within the territorial scope of the (GDPR).”); id. at 9–10 (“For commercial arbitration the hypothesis where arbitrators and arbitral institutions located outside EU provide dispute resolution services to individuals located in the EU seems to be of practical importance.”); Wei Gao, The ECHR in Action:  Its Applicability and Relevance for Arbitration, 26 Int’l J. Hum. Rts. 1608, 1626–27 (2022) (“[P]rivate parties choosing arbitration as a dispute resolution mechanism . . . must be aware of the rights that they enjoy under the ECHR . . . Such awareness may be crucial for the parties to rationally exercise their informed judgment in an arbitration, e.g. when they select the seat of arbitration within or outside an ECHR Contracting State.”).
41Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act), 2024 O.J. (L 1689) [hereinafter EU AI Act].
42Recital 26 EU AI Act reads:  “In order to introduce a proportionate and effective set of binding rules for AI systems, a clearly defined risk-based approach should be followed.  That approach should tailor the type and content of such rules to the intensity and scope of the risks that AI systems can generate.  It is therefore necessary to prohibit certain unacceptable AI practices, to lay down requirements for high-risk AI systems and obligations for the relevant operators, and to lay down transparency obligations for certain AI systems.”  Id. recital 26.
43Mireille M. Caruana & Roxanne Meilak Borg, Regulating Artificial Intelligence in the European Union, in The EU Internal Market in the Next Decade – Quo Vadis? 108, 117–18 (Ivan Sammut & Ivan Mifsud eds., 2025) (“The AI Act adopts a ‘risk-based’ approach and lays down a ‘horizontal’ (as opposed to sector-specific) legal framework for AI that aims to ensure legal certainty . . . The Act . . . precludes AI systems that pose an unacceptable risk, listing several prohibited practices.  Thus, it identifies and heavily regulates AI systems with high risks and briefly mentions systems with limited risks.  It also includes a new Chapter V on general-purpose AI (‘GPAI’) models, which speaks of GPAI models with ‘systemic risk’ and ones ‘released under a free and open-source licence.’”).
44EU AI Act, supra note 41, art. 5.
45Id. art. 6, annex III.
46Id. art. 50.
47Id. recital 61; annex III, ¶ 8.
48Maxi Scherer, Ole Jensen & Russell Childree, Regulating the Use of Artificial Intelligence in International Arbitration:  The EU AI Act and Beyond, 2024 Cahiers de l’Arbitrage / Paris J. Int’l Arb. 653, 665 (2024).
49Id. at 665–66 (“[A] ‘deployer’ is ‘any natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity.’  This broad definition readily applies to arbitrators given that they may use AI systems in the exercise of their arbitral mandate, for instance by asking an AI system to summarise the parties’ positions or draft parts of correspondence, procedural orders and awards.  Arbitral institutions likewise exercise authority over parties’ disputes, including by appointing arbitrators, deciding on requests for consolidation and scrutinising draft awards.  They too would therefore qualify as ‘deployers.’”).
50Id. at 666–67 (“Pursuant to Article 2(1)(b), the EU AI Act applies to deployers ‘that have their place of establishment or are located within the Union’.  A literal interpretation of this provision results in the territorial applicability of the EU AI Act in a number of scenarios . . . Arbitral tribunals have their ‘place of establishment’ in the EU if the seat of the arbitration is in one of the Member States, irrespective of where the arbitrators are otherwise located.”).
51Id. (“Article 2(1)(b) applies more readily to arbitral institutions that are headquartered in a Member State.  In that case, the institution has both its ‘place of establishment’ and ‘location’ in the EU.  The same applies to institutions that originate from a third country but have an office location in the EU.  In that case, at least the regional office would be ‘located within the Union’ and thus subject to the EU AI Act.”).
52EU AI Act, supra note 41, art. 14(2).
53U.N. Comm’n on Int’l Trade L., Status:  UNCITRAL Model Law on International Commercial Arbitration (1985), with amendments as adopted in 2006, https://uncitral.un.org/en/texts/arbitration/modellaw/commercial_arbitration/status.  
54Arbitration Act 1996, c. 23, as amended by Arbitration Act 2025, c. 4 (Eng.).
55Arbitration Law of the People’s Republic of China, adopted Aug. 31, 1994, effective Sept. 1, 1995, as amended Sept. 1, 2017, effective Jan. 1, 2018 (China) [hereinafter Chinese Arbitration Law 1994].
56References to the French lex arbitri only correspond to the provisions applicable in international arbitration pursuant to Title II, Book IV Code de procédure civile.
57Queen Mary University of London & White & Case, supra note 24, at 5 (“The five most preferred seats for arbitration are London, Singapore, Hong Kong, Beijing and Paris.”).
58ICC Rules of Arbitration (2021).
59SIAC Arbitration Rules (2025).
60LCIA Arbitration Rules (2020).
61Queen Mary University of London & White & Case, supra note 24, at 5 (“The five most preferred sets of arbitral rules are the ICC Rules, HKIAC Rules, SIAC Rules, LCIA Rules and UNCITRAL Rules.”).
62Convention on the Settlement of Investment Disputes Between States and Nationals of Other States, opened for signature Mar. 18, 1965, 17 U.S.T. 1270, 575 U.N.T.S. 159 [hereinafter ICSID Convention].
63ICSID Rules of Procedure for Arbitration Proceedings (Mar. 21, 2022).
64Annabelle O. Onyefulu, Artificial Intelligence in International Arbitration: A Step Too Far?, 89 ARB. 56, 59 (2023) (“Article 11 of the (MAL) impliedly provides for arbitrators to be natural persons. It provides that a person shall not be estopped from representing parties as an arbitrator by virtue of his/her nationality. It is in doubt if an intelligent machine can have a nationality.”).
65Ana Fernández Araluce, AI in International Arbitration: Unveiling the Layers of Promise and Peril, 49 IURGIUM 35, 40 (2024) (“AI is based on the way it was coded and programmed; accordingly, if the data is biased or incomplete, AI will inevitably mirror these flaws.”).
66Kasap, supra note, 3, at 240 (“[A]n arbitrator must be both independent and impartial with respect to the parties and the dispute . . . The former concerns the arbitrator’s potential financial, professional, or personal ties to a party or a dispute that might influence their decision, whereas the latter addresses any favoritism or prejudice shown by an arbitrator toward a party or the matter in dispute.”).
67INT’L BAR ASS’N, GUIDELINES ON CONFLICTS OF INTEREST IN INTERNATIONAL ARBITRATION, Explanation to General Standard 2(b) (2024).
68Report of the Secretary-General: Possible Features of a Model Law on International Commercial Arbitration, U.N. Doc. A/CN.9/207, ¶ 65, in XII Y.B. UNCITRAL 75, 85 (1981) (“As regards challenge of arbitrators, the issues to be considered are on what grounds an arbitrator may be challenged . . . The reasons relate to the dispute at hand such as a financial interest or previous involvement in the subject-matter under dispute or a certain relation to one of the parties.”).
69Chinese Arbitration Law 1994, supra note 55, art. 34(2).
70Id. art. 34(3).
71Id. art. 34(4).
72“La mission d’arbitre ne peut être exercée que par une personne physique.” Code de Procédure Civile, art. 1450 (Fr.).
73The lex arbitri of France—i.e., the provision on international arbitration envisaged in the French Code de procédure civile—applies nationwide.
74Singapore and Hong Kong are the second and third most preferred seats for international arbitration, respectively. Queen Mary University of London & White & Case, supra note 24, at 7.
75In the case of Singapore, Section 3 of the International Arbitration Act 1994 (Cap. 143A) (Sing.) expressly provides that the MAL has the force of law, subject to the modifications contained in the Act. The MAL is also the First Schedule of the Singapore International Arbitration Act 1994. As for Hong Kong, the Arbitration Ordinance (Cap. 609) (H.K.) states in Section 4 that the MAL has the force of law, subject to the modifications and supplements provided for in the Ordinance. Likewise, the MAL is contained in Schedule 1 of the Arbitration Ordinance.
76Abhilasha Vij, Arbitrator-Robot:  Is A(I)DR the Future?, 39 ASA Bulletin 123, 134 (2021). (“The ‘Appointment of Arbitrator’ as provided under Article 6 of the UNCITRAL Arbitration Rules does not envisage the possibility of referring a dispute to a machine . . . The rule contemplates a natural person who could be independent and impartial as an arbitrator, and would belong to certain nationality.”).
77Boris Praštalo, Arbitration Tech Toolbox:  AI as an Arbitrator:  Overcoming the “BlackBox” Challenge?, Kluwer Arb. Blog (Aug. 23, 2024), https://legalblogs.wolterskluwer.com/arbitration-blog/arbitration-tech-toolbox-ai-as-an-arbitrator-overcoming-the-black-box-challenge/ (“The ‘black box’ problem poses a significant challenge to having AI serve as an arbitrator.  Even if an AI model could be designed to render decisions in arbitration cases, the exact factors and patterns it considered to reach its conclusions would remain unknown.”).
78Snijders, supra note 28, at 234 (“[A]rbitrators must sign the arbitral award . . . It is true that this signature can be placed in the form of a qualified electronic signature . . . However, such a signature is . . . of human origin.”).
79Irene Ng & Valeria Benedetti del Rio, When the Tribunal Is an Algorithm:  Complexities of Enforcing Orders Determined by a Software under the New York Convention, in 60 Years of the New York Convention:  Key Issues and Future Challenges 121, 125 (Katia Fach Gómez & Ana Mercedes López Rodríguez eds., 2019) (“[T]he algorithmic arbitrator might not be able to sign its final decision.  This is generally the case when the decision is made by a software . . . In this scenario, the decision is taken automatically by a computer, without any specific formal requirements such as signatures.  The final award, therefore, may lack the necessary characteristic that is generally under scrutiny when the authenticity of a document needs to be confirmed.”).
80Snijders, supra note 28, at 234 (“‘The function of the signature is always the same, i.e., the signatory indicates that he is aware of the content of the document and expresses his will with his signature’.  Such a will cannot be expressed by a robot.”).
81Eidenmüller & Varesis, supra note 28, at 31 (“Article IV . . . requires a duly authenticated original . . . This provision has been interpreted to require the signatures of arbitrators as natural human beings.’”).
82See Praštalo, supra note 77.
83Queen Mary University of London & White & Case LLP, 2021 International Arbitration Survey: Adapting Arbitration to a Changing World (2021), https://www.qmul.ac.uk/arbitration/media/arbitration/docs/LON0320037-QMUL-International-Arbitration-Survey-2021_19_WEB.pdf (“International arbitration is the preferred method of resolving cross-border disputes for 90% of respondents, either on a standalone basis (31%) or in conjunction with ADR (59%).”).
84Queen Mary University of London & White & Case, supra note 23, at 3 (“The primary advantages of international arbitration for resolving disputes involving public interest issues include the ability to select arbitrators with relevant experience or knowledge (47%).”).
85Emilia Onyema, Empirically Determined Factors in Appointing Arbitrators in International Commercial Arbitration, 73 Arb. 199, 205 (2007) (“Interviewees were asked, ‘What factors affect your choice of arbitrators?’ . . . The following factors were ranked topmost:  reputation, expertise, common sense, knowledge of applicable law and knowledge of relevant language.”).
86Christophe Seraglini, Who Are the Arbitrators?  Myths, Reality and Challenges, in Legitimacy:  Myths, Realities, Challenges ICCA Congress Series No. 18, at 589, 594 (Albert Jan van den Berg ed., 2015) (“The selection of arbitrators is determinative of the quality of the decision considering that ‘an arbitration is only as good as its arbitrators.’”).
87Rusinque, supra note 5, at 118 (“[M]uch of the confidence placed in the tribunal is due to the parties’ ability to nominate its members . . . [T]he parties not only bear in mind certain knowledge or skills relevant to solve the dispute, but also the personal reputation and professional background of the candidate.”).
88Id. at 108 (“Party autonomy is the bedrock of arbitration, and the appointment of arbitrators is not the exception.  By virtue of this principle . . . parties have the power to establish the number of arbitrators, the qualifications they must possess, the method of nomination and replacement, and so forth.”).
89Flavia Marisi, Independence and Impartiality:  The Role of Soft Law in International Arbitration, 85 Arb. 326, 343 (2019) (“A further mechanism to encourage arbitrators to issue carefully reasoned and impartial awards is the scrutiny they are subject to by the fellow arbitrators’ community, scholars, and the civil society.”).
90Rusinque, supra note 5, at 120 (“Who can be in a better position to determine whether an arbitrator acts with probity than the users of arbitration and the arbitration community?  No party or institution would nominate an arbitrator whose integrity and fair judgment are questionable.”).
91Scherer et al., supra note 48, at 541 (“When considering AI for arbitral decision-making, some have speculated about the feasibility of ‘robot-arbitrators’ . . . Authors typically either assert that AI is inevitable in the future, or express scepticism, mainly on the assumption that some “human factor” would be necessary to ensure empathy and emotional justice.”).
92Daniel Goleman, Emotional Intelligence:  Why It Can Matter More Than IQ 74 (1996) (“[T]he root of caring, stems from emotional attunement, from the capacity for empathy.  That capacity—the ability to know how another feels—comes into play in a vast array of life arenas, from sales and management to romance and parenting, to compassion and political action.”).
93Id. at 34 (“[T]he core of interpersonal intelligence includes the ‘capacities to discern and respond appropriately to the moods, temperaments, motivations, and desires of other people.’”).
94Kasap, supra note 3, at 235 (“Human arbitrators do not apply the law purely as a matter of logic and legal syllogism or based on patterns and probabilities, which is how AI functions.  An arbitrator’s personality, as well as their understanding of the world, is not only important for whatever decisions are rendered but also for the parties who appoint the arbitrators in the first place.”).
95Id. at 235 (“Practitioners emphasize that arbitrators’ ability to understand and interact with the parties can be reframed as ‘juridical open-minded-ness’ something that sharply delimits the success of the arbitration.  In line with this claim, a 2010 survey found that arbitration users attached importance to the soft skills of arbitrators and consider these skills as having an impact on the efficiency and cost of a case.  It is thus not surprising that parties search for an arbitrator who has effective communicative skills and can show empathy, both of which require emotional intelligence.”).
96Florescu, supra note 2, at 52 (“[E]motions, feelings, moods, weaknesses . . . impact human judgement . . . Reflecting the need for a nuanced analysis of facts that might and perhaps should influence a decision, emotions are still needed to give sense and purpose to the legal decision-making process.”); Goleman, supra note 92, at 74 (“People’s emotions are rarely put into words; far more often they are expressed through other cues.  The key to intuiting another’s feelings is in the ability to read nonverbal channels:  tone of voice, gesture, facial expression, and the like.”).
97Reed, supra note 11, at 87 (“At heart, the art of arbitral decision-making lies in intuiting whether and when witnesses are telling the truth, in perceiving the human stories underlying a business dispute, in crafting an award with the right reasoning and the right amount of reasoning.”).
98Seraglini, supra note 86, at 594 (“Experience indeed has an important role to play in ensuring a high quality of justice regarding the conduct of the proceedings as well as the merits of a case.  An experienced arbitrator:  (a) knows how to efficiently conduct arbitral proceedings; and (b) has the proper skills to adequately resolve the matter in dispute.”).
99Kasap, supra note 3, at 234.
100Code de Procédure Civile, art. 1450 (Fr.).
101Snijders, supra note 28, at 234 (“[T]he first sentence of Article 1023 of the Dutch Code of Civil Procedure provides that only natural persons of legal capacity can be appointed as arbitrators.”).
102ICSID Convention, supra note 62, art. 14(1).
103Onyefulu, supra note 64, at 63 (“It has been contended that an arbitral award rendered by an AI arbitrator may contravene several public policy grounds.  For example, where the national legislation of the seat of arbitration expressly provides that an arbitrator must be a natural person, any award rendered by an AI arbitrator is in contravention of the public policy of the national legislation.”).
104Ng & Benedetti del Rio, supra note 79, at 131 (“If a party intends to use Article V.2(b) to invalidate an arbitral award by an algorithmic arbitrator, such party must prove that the use of an algorithmic arbitrator would be against the public policy of that country . . . [T]here are a few countries such as France that effectively do so by limiting the definition of an arbitrator to that of a natural person.”).
105Karl-Heinz Böckstiegel, The Role of Party Autonomy in International Arbitration, 52 Disp. Resol. J. 24 (1997) (“The principle of party autonomy is fundamental to arbitration in general and to international arbitration in particular.”).
106Queen Mary University of London & White & Case, supra note 83, at 1 (“The field of international arbitration is dynamic by nature.  Its hallmarks of flexibility and party autonomy allow it to develop and adapt in response to the needs of its users.”).
107Report of the Secretary-General, supra note 68, ¶¶ 17, 78 (“Probably the most important principle on which the model law should be based is the freedom of the parties in order to facilitate the proper functioning of international commercial arbitration according to their expectations.  This would allow them to freely submit their disputes to arbitration and to tailor the ‘rules of the game’ to their specific needs.”).
108Queen Mary University of London & White & Case, supra note 24, at 28–29 (“We asked what respondents considered to be the principal drivers for greater use of AI in international arbitration. . . . The most selected reason was the potential to save party and counsel time (54%) . . . Reduction of costs was the second most selected driver, with 44% of respondents citing this benefit.”).
109Alan Shilston, No Hired Guns, 64 Arb. 1 (1998) (“It should never be forgotten that . . . arbitration is a service activity in the promotion of international trade dedicated to providing a very real option to the state’s procedural methods of resolving disputes in public.”).
110Niyati Ahuja & Lucila Marchini, New York Arbitration Week 2021 Redux:  Options, Opportunities, Pivots, and Challenges in an International Arbitration Career, Kluwer Arb. Blog (Dec. 6, 2021) https://legalblogs.wolterskluwer.com/arbitration-blog/new-york-arbitration-week-2021-redux-options-opportunities-pivots-and-challenges-in-an-international-arbitration-career/ (“Arbitration is a service industry which adapts to the needs and demands of its users.  Those needs and demands have changed over time.  Economies have become more globalized, and this globalization has led to the development of the arbitration practice in all parts of the world.”).
111Stephan Schill, Developing a Framework for the Legitimacy of International Arbitration, in Legitimacy:  Myths, Realities, Challenges ICCA Congress Series No. 18 789, 800 (Albert Jan van den Berg ed., 2015).
112Born, supra note 12, § 13.04[A][8].
113SVAMC Guidelines, supra note 19, Guideline 6.
114Rusinque, supra note 5, at 122 (“Soft-law instruments are relevant for the construction of . . . provisions . . . Said instruments compiled the best practices in international arbitration, depicting the so-called procedural lex mercatoria, understood as a set of transnational principles . . . relevant for conducting arbitral proceedings.”).
115AAA-ICDR, supra note 25.
116Born, supra note 12, § 12.04[B][1] (“[M]ost developed jurisdictions permit juridical persons to serve as arbitrators, although in practice the appointment of juridical persons as arbitrators is very unusual.”).
117João Moreira and Riccardo Segate, The “It” Arbitrator:  Why Do Corporations Not Act as Arbitrators?, 12 J. Int’l Disp. Settlement 525, 537 (2021) (“[T]he New York Convention does not impose any requirements that an arbitrator is a natural person and explicitly admits arbitral awards made by permanent arbitral bodies.”).
118Praštalo, supra note 77 (“A potential solution to this conundrum has been devised in the form of the so-called ‘Explainable AI’ which not only provides desired outputs, but also explains how it generated those outputs.”).
119Born, supra note 12, § 13.04[A][4] (“One of the arbitrator’s most significant obligations is to render an award that is enforceable.  This duty is frequently expressed, as an objective, in institutional arbitration rules.”).
120Renato Nazzini, The Law Applicable to the Arbitration Agreement:  Towards Transnational Principles, 65 Int’l & Comp. L.Q. 681 (2016) (“There are currently diverging views as to whether the law applicable to the arbitration agreement should be the law chosen by the parties to govern their substantive legal relationship or the law of the seat of the arbitration.”).
121Paul Bennett Marrow, Mansi Karol & Steven Kuyan, Artificial Intelligence and Arbitration:  The Computer as an Arbitrator—Are We There Yet?, 74 Disp. Resol. J. 35–36 (2020) (“Computer scientists posit that with enough data and properly designed algorithms, the ‘well-trained’ computer should soon be able to produce an acceptable arbitration award.  The necessary data set would hopefully include thousands of transcripts from actual arbitration proceedings; hundreds of thousands of actual awards; all known reported judicial opinions issued by courts (and) all relevant statutes and rules used by judges, lawyers, arbitrators, and the administrators of the arbitration process.”).
122Stefanie G. Efstathiou & Mihaela Apostol, Arbitration Tech Toolbox:  ChatGPT–Arbitral Assistant or Fourth Arbitrator?, Kluwer Arb. Blog (July 22, 2023) https://arbitrationblog.kluwerarbitration.com/2023/07/22/arbitration-tech-toolbox-chatgpt-arbitral-assistant-or-fourth-arbitrator/.
Back to Top
About the Contributor
generic image

SJD Candidate

Felipe Rusinque is a Colombian-qualified lawyer and SJD candidate at the Central European University, Vienna, Austria. He holds an LLM in International Business Law (Ghent University) and PgDip in Contract Law (Rosario University). Felipe also has a 5-year track record as a legal assistant in infrastructure arbitration and experience in the legislative and research work of UNCITRAL and UNIDROIT, with a focus on arbitration and ISDS.