The Use of AI in International Arbitration: Towards Virtual Arbitrators?

Par Matteo Cimmino 

The use of artificial intelligence in the legal sector is progressively growing and constantly brings to the interpreter’s attention interesting reflective insights. Not exempt from the consideration under review is the phenomenon of arbitration, which – especially in its international dimension – has become widely prevalent within the framework of the so-called ADR methods (especially on account of the fiduciary connotation of the relationship between adjudicator and adjudicated, as well as for the greater celerity of the related proceedings) and has recently gone through – on the occasion of the Covid-19 pandemic – an important evolutionary phase, marked by a massive introduction of technological tools in arbitration proceedings, within an overall process of their digitalization [[i]].

For what specifically concerns the subject of this article, it should be noted how manifold can be the applications of AI in international arbitration law. For classification purposes, however, it is appropriate to distinguish preliminarily – on the functional level – between “assisting” and “substituting” uses of AI systems in arbitration proceedings (what some authors portray as the distinction between “AI-assisted arbitration” and “AI-autonomous arbitration”) [[ii]].

Regarding the applications of the first type, the following cases may be considered primarily:

• Prognostic assessments about the outcome of disputes submitted to arbitration – This is the so-called “Jurimetrics”, which – for instance – could prove particularly useful for the purposes of the Third-Party Funding (TPF) mechanism [[iii]], given the high costs of international arbitration proceedings. However, if for investment arbitration some applications can already be detected [[iv]], its implementation would be much more challenging with respect to international commercial arbitration, which is characterized by a marked complexity of individual proceedings (which obviously reduces the degree of predictability of their outcome) [[v]], as well as by the general reluctance of the stakeholders to publish the awards that define them (which translates, evidently, into a significant lack of usable data for the predictive mechanism under consideration) [[vi]].

• Drafting and Correction of Arbitration Agreements – This is an application essentially aimed at reducing the risk of “pathological” arbitration agreements, which can result in the annulment of the arbitration award and the consequent overall invalidation of the related proceeding. It may happen, in fact, that errors or oversights made during the drafting of the arbitration agreement result, finally, in defects in its validity or effectiveness, which — if duly objected by the interested party — can significantly undermine the stability of the award that defines the proceeding built upon such a flawed agreement. In this regard, AI tools have recently been developed that can directly assist stakeholders in drafting ab initio arbitration clauses, as well as in the ex post correction of clauses unilaterally drafted by the parties themselves [[vii]]. For example, the American Arbitration Association has deployed online the so-called Clause-Builder, designed to assist individuals and organizations in drafting clear and effective arbitration and mediation agreements, considering the specific circumstances of the disputes involved and the parties’ preferences regarding the elements of the agreement itself [[viii]].

• Selection of the arbitrators – One of the most delicate aspects for the functioning of the arbitration system lies in the composition of the arbitral tribunal, since the fundamental principle of impartiality and third-party status of the (in this case private) judge also applies to the latter, as it does to state jurisdiction. In this regard, it should be noted primarily that, to date, no existing tool usable for selecting arbitrators can be defined stricto sensu as an “AI arbitrator appointer” [[ix]]. However, the use of tools such as Arbitrator Intelligence [[x]], which is capable of detecting any past business relations between an individual arbitrator and each of the parties involved (or their lawyers), makes it possible to more quickly and reliably assess the credibility of the individual arbitrator in terms of independence and third-party status, impartiality being, ex natura rerum, an inscrutable feature of the human mind [[xi]].

• Research and review of data and documents – This is probably the area in which AI is most widely used in legal matters, even beyond the borders of the arbitration phenomenon. These are “instrumental” activities with respect to the performance of legal professions, often very time-consuming, which can be significantly optimized (in terms of both time and quality) using appropriate AI systems. Consider, for example, AI tools such as KLDiscovery and NexLP, which can help stakeholders at all stages of document collection, review, and production [[xii]]. Specifically, as for the collection phase, such AI-driven tools can assist users in the gathering and analysis of documents in unstructured formats – such as chat messages, Multimedia Messaging Service (MMS), and Short Message Service (SMS) – by organizing the messages into a structured, sequential format that facilitates the subsequent review phase [[xiii]]. As for the stage of document review itself, an additional AI-based tool may also be considered. Namely, Kira Systems [[xiv]], a machine learning contract review and analysis software which can identify, extract and analyze content in contracts and documents [[xv]]. Particularly, Kira can automatically convert files into machine readable form and uses its so-called “Built-In Intelligence” to efficiently and accurately extract common clauses, provisions, and data points. Finally, regarding the phase of document drafting, reference can be made to Clio Draft [[xvi]]. This AI powered tool, which has also been presented at the Generative AI and Justice Conference held at the Laboratoire de Cyberjustice of the University of Montreal on October 1^st, 2024, can automate the whole stages of document generation and signatures collection, allowing users to save time and gain accuracy in the performance of those tasks.

Having analyzed the possible “assisting” uses of AI in international arbitration proceedings, we now come to the pivotal aspect of the present discussion, namely the question of the substitutability of the human arbitrator in the drafting of the arbitral award. In other words: can artificial intelligence independently decide disputes submitted to arbitration? On this point, it should be noted that while there are different positions of interpreters in perspective, there is instead substantial one-sidedness of doctrine with respect to a fundamental conclusion: as of now, no form of “AI-based autonomous arbitration” is conceivable. The reasons for this are various.

First, a hypothetical “automated” decision of a dispute devolved to arbitral cognition appears – to date – unthinkable, particularly due to the limitations that currently restrict the use of AI tools in the legal sector. This is especially true considering the very ontological conformation of international arbitration, which presents itself – in this sense – as a complex legal creature, governed by a patchwork of normative sources adapted from time to time by the stakeholders to the specific configuration of the matter in dispute. The flexibility that generally connotes the arbitration phenomenon allows the parties, especially with respect to transnational proceedings, to freely select the lex contractus (governing the arbitration agreement), the lex arbitri (informing the procedural aspects of the individual arbitration), and the lex causae (referring to the merits of the dispute that is the subject of the arbitration agreement). The result, then, is an extremely complex legal framework.

Considering all this, concerns must be raised – regarding the actual feasibility of the hypothesis under examination – about the various limitations identified by the doctrine with respect to the use of AI tools in performing creative, reasoning, and judgment-based tasks within the legal sector [[xvii]]. These are, in fact, obstacles that currently prevent any artificial intelligence system from truly “capturing” the complexity of international arbitration proceedings. Consider, for example, the issue defined as “lack of construct validity” [[xviii]]. Specifically, construct validity refers to the extent to which an evaluation accurately represents and measures the construct it is intended to assess. AI tools designed for the aforementioned legal applications are “trained” to resolve hypothetical legal issues based on predefined data and elements (such as those encountered in the so-called bar exam). In cases where sufficient standardization exists, even “shallow reasoning” – which is certainly within the reach of artificial intelligence – may be sufficient. However, and this is the essential point, such level of reasoning is certainly inadequate for resolving transnational arbitration disputes, which are characterized, as noted above, by a high degree of individual complexity. Furthermore, there is currently no evidence that language models can acquire the kind of in-depth reasoning skills that humans possess, and that are strictly required for the thorough decision of such disputes [[xix]].

Further highlighting the current limitations in the use of artificial intelligence in the legal sector is the so-called “Black Box” problem [[xx]], meaning – essentially – that the path the AI model takes to reach a result is not identifiable [[xxi]]. This implies that, while AI can yield results, the exact mechanisms and criteria it employs are not fully transparent. The black box issue clearly poses a substantial challenge for using AI as an arbitrator. Even if an AI model could make arbitration decisions (and, as noted above, to date it can’t), the precise factors and patterns it analyzed to arrive at them would remain unknown. This aspect would supposedly render the arbitral award incompatible with various legal systems where the principle that jurisdictional decisions must be reasoned holds significant importance [[xxii]]. Moreover, it would clearly conflict with the provisions of the 1985 UNCITRAL Model Law, which, in certain legal systems (including Canada’s), serves as the foundation for international (and domestic) arbitration regulation [[xxiii]].

Moreover, the fact that the arbitral award was rendered by AI systems could significantly hinder its circulation and recognition within domestic jurisdictions under the 1958 New York Convention. Indeed, while it is true that no domestic legal system expressly prohibits (on the ground that it is contrary to public policy) the possibility of an arbitral dispute being decided independently by an AI system, it is also true that, in fact, some legal environments require that the award be rendered by “natural” persons. For example, in France, «la mission d’arbitre ne peut être exercée que par une personne physique jouissant du plein exercice de ses droits» [[xxiv]].

Finally, and this may be considered the most significant argument, the hypothesis under consideration would betray the very spirit of the arbitration phenomenon, encapsulated in its characteristic exaltation of the fiduciary connotation of the adjudicator-adjudicated relationship. Francesco Galgano (an authoritative Italian interpreter of arbitration law) said that the ancient heart of commercial arbitration can be found depicted in a passage by Baldo degli Ubaldi, who in the 14^th century, faced with the then conformation of local procedural law, hinged on the ancient lex mercatoria, stated that «in causis mercatorum, ubi de bona fide agitur, non congruit de iuris apicibus disputare» [[xxv]]. Well, arbitration, even in its international connotation, is based first and foremost on the bona fides between the parties and the arbitrator, and this is a relationship that could not be configured if the latter role were “dehumanized,” evidently lacking any AI system of the “emotional intelligence” necessary for this purpose. In other words, there is still – on the point – an inescapable need for the “human touch” [[xxvi]].

 Image Source: All for the AI (https://allfortheai.com/wp-content/uploads/2024/04/Article-28.-Website.7.jpg).

[[i]] Suffice it to mention, in this regard, the reforms that have taken place in some of the world’s most important arbitration centers, such as the ICC or the International Center for Dispute Resolution (ICDR) of the American Arbitration Association (AAA) or the London Court of International Arbitration (LCIA). They opted to file the claim exclusively electronically with the International Centre for Settlement of Investment Disputes (IC-SID) and the Stockholm Chamber of Commerce (SCC), the latter even before the outbreak of the pandemic. Instead, the Austrian Vienna International Arbitral Centre (VIAC) and the German Deutschen Institution für Schiedsgerichtsbarkeit (DIS) have opted for hybrid forms, having the latter made it clear, however, that the «e-form is preferred».

[[ii]] See, e. g., S. Gulyamov and M. Bakhramova, “Digitalization of International Arbitration and Dispute Resolution by Artificial Intelligence,” World Bulletin of Management and Law 9 (2022): 79–85.

[[iii]] This is a mechanism in which an outside entity (the funder) provides financial resources to one of the parties involved in litigation (including arbitration) to cover legal and other costs related to the case, receiving in return a share of the proceeds or compensation that will eventually be obtained by the funded party in the event of a successful outcome.

[[iv]] C. I. Florescu, “The Interaction Between AI (Artificial Intelligence) and IA (International Arbitration): Technology as the New Partner of Arbitration,” Romanian Arbitration Journal 18 (2024): 42ff., at 61.

[[v]] A. F. Araluce, “AI in International Arbitration: Unveiling the Layers of Promise and Peril,” Iurgium (formerly Spain Arbitration Review) 49.1 (2024): 35ff., at 38.

[[vi]] K. Paisley and E. Sussman, “Artificial Intelligence: Challenges and Opportunities for International Arbitration,” New York Dispute Resolution Lawyer 11, no. 1 (2018): 35–40, at 37.

[[vii]] S. Shih and E. Chin-Ru Chang, “The Application of AI in Arbitration: How Far Away Are We From AI Arbitrators?”, Contemporary Asia Arbitration Journal 17 (2024): 69ff., at 74.

[[viii]] ClauseBuilder, American Arbitration Association, https://www.clausebuilder.org/.

[[ix]] See E. Chan, M. G. Bel, and B. Malek, “SVAMC Draft Guidelines on Using AI in Arbitration: A Focus on the Selection of Arbitrators and Arbitrators’ Use of AI.” NYSBA New York Dispute Resolution Lawyer 17, no. 1 (2024): 12-15, at 13, who, in this regard, state that «the existing tools that assist parties and counsel in selecting and appointing arbitrators provide collated data and information on the arbitrators’ profiles to the parties, but the parties still must analyze the data to select an arbitrator».

[[x]] Arbitrator Intelligence, https://arbitratorintelligence.vercel.app/. The tool under consideration, however, cannot be defined as an AI tool in the traditional sense. Instead, it combines machine learning, data aggregation, and analysis to generate reports on arbitrators’ decision-making processes and case management. These reports are based on data collected from surveys and publicly available information. The platform uses AI-like technologies to enhance its data processing, but its primary function is more focused on data-driven decision-making rather than artificial intelligence in the conventional sense. So, while AI features may be involved in analyzing large data sets, the tool itself is more of a specialized platform for arbitration-related intelligence rather than a full-fledged AI tool.

[[xi]] P. I. Barbirotto and F. Marrella, “Arbitrato internazionale e piattaforme telematiche: alcune riflessioni sul ruolo dell’intelligenza artificiale,” Rivista dell’Arbitrato 1 (2023): 191.

[[xii]] Technology Resources for Arbitration Practitioners – Document Collection, Review and Production, INT’L BAR ASS’N, https://www.ibanet.org/technology-resources-for-arbitration-documents.

[[xiii]] S. Shih and E. Chin-Ru Chang, “The Application of AI in Arbitration …” supra note 7, at 75.

[[xiv]] Kira Systems, https://www.kirasystems.com/.

[[xv]] Gülüm Bayraktaroğlu-Özçelik and Ş. Barış Özçelik. “Use of AI-Based Technologies in International Commercial Arbitration.” European Journal of Law and Technology 12, no. 1 (2021): 1-21, at 4.

[[xvi]] Clio Draft, https://www.clio.com/draft/?cta=top-nav-na.

[[xvii]] See S. Kapoor, P. Henderson, and A. Narayanan, “Promises and Pitfalls of Artificial Intelligence for Legal Applications.” Journal of Cross-disciplinary Research in Computational Law (2024): 1-13, at 4-8.

[[xviii]] Id., at 5.

[[xix]] This is all the more true in the event that the parties choose as the law applicable to the merits of the dispute a non-state body of law (e.g., the lex mercatoria) or opt for the so-called ex aequo et bono arbitration, as AI certainly cannot be expected to embrace concepts such as justice and fairness, since they derive from “common feeling” and are, as a result, extremely variable (which also poses the so-called issue of “contamination”).

[[xx]] See B. Praštalo, “Arbitration Tech Toolbox: AI as an Arbitrator – Overcoming the Black-Box Challenge.” Kluwer Arbitration Blog, August 23, 2024.

[[xxi]] See L. Blouin, “AI’s Mysterious ‘Black Box’ Problem, Explained.” University of Michigan-Dearborn News, March 6, 2023. https://umdearborn.edu/news/ais-mysterious-black-box-problem-explained.

[[xxii]] For example, in Italy, the article 111, c. 6 of the Constitution, states: «Tutti i provvedimenti giurisdizionali devono essere motivati».

[[xxiii]] See UNCITRAL Mode Law, Article 31(2), which clearly provides: «The award shall state the reasons upon which it is based (…) ».

[[xxiv]] Art. 1450.1, French Code de Procédure Civile.

[[xxv]] P. I. Barbirotto and F. Marrella, op. cit., p. 198.

[[xxvi]] D. H. Lindquist and Y. Dautaj, “AI in International Arbitration: Need for the Human Touch,” Journal of Dispute Resolution (2021): 39ff.

Ce contenu a été mis à jour le 19 novembre 2024 à 15 h 01 min.