‘Robot arbitrators’: current obstacles and future prospects

The concept of the 'robot judge', long the subject of debate, can no longer be dismissed as theoretical or remote. This is largely due to the emergence of platforms promoting hybrid human-artificial intelligence ('AI') models as well as fully automated decision-making systems (Kleros, Celeste, Jur, etc.). For example, the American Arbitration Association recently introduced an 'AI arbitrator' capable of drafting a proposed award, which is subsequently reviewed and issued by a human arbitrator.^[1] The use of this tool could help mitigate human factors, including cognitive biases, thereby offering a promising alternative.

Nevertheless, this article argues that a 'robot arbitrator' lacks the fundamentally human capacity to reach a fair and impartial decision based on a deep contextual understanding of legal principles, commercial realities and the parties' unique intentions.^[2] The consensus, reflected in the guidelines of the Silicon Valley Arbitration & Mediation Center ('SVAMC')^[3] and the Chartered Institute of Arbitrators ('CIArb')^[4], is that an arbitrator cannot delegate their decision-making mandate to a robot.

This article  examines the most significant obstacles to the adoption of robot arbitrators — legal (section 1), technological (section 2), and ethical (section 3) — and assesses the prospects for future.

1. Legal obstacles

This section addresses the following legal obstacles: (i) the requirement for the arbitrator to be a natural person, as enshrined in positive law; (ii) the silence of the New York Convention; and (iii) the obligation to state the reasons for the arbitral award.

i. Requirement for a natural person and the silence of the New York Convention

Legal systems are broadly divided into two groups: those that expressly provide that only natural persons may act as arbitrators (e.g. France^[5] and Peru^[6]) and those that, without expressly stating so, infer that requirement from other provisions (e.g. Italy).^[7] Regarding fundamental rights, Regulation (EU) 2024/1689, the EU Artificial Intelligence Act,^[8] also requires human oversight of certain high-risk AI systems.^[9]

Drafted before the advent of artificial intelligence, the New York Convention does not define the term 'arbitrator' and does not specify that arbitrators must be natural persons. That silence fuels doctrinal debate. Awards rendered by an AI arbitrator may nevertheless be challenged on the basis of the Model Law of the United Nations Commission on International Trade Law ('UNCITRAL Model Law')^[10] and the New York Convention^[11], if the arbitral tribunal was not composed of arbitrators agreed upon by the parties or if its composition materially departed from the law of the seat.

Therefore, neither at the national nor at the international level does the law expressly provide a legal framework for artificial arbitrators equivalent to that applicable to natural persons.

ii. Requirement for reasoned awards

The UNCITRAL Model Law^[12] and the Washington Convention^[13] require that an arbitral award state the reasons on which it is based, resolve every issue submitted to the tribunal, and set out the grounds for its decisions. Failure to comply with the obligation to state reasons may constitute a violation of international public policy.^[14] The English Arbitration Act 1996, as amended by the Arbitration Act 2025, may also be relevant where the use of AI gives rise to serious-irregularity or appeal grounds under sections 68 or 69.^[15] However, the probabilistic reasoning of a 'robot arbitrator' cannot satisfy this requirement. Some have even argued that the drafting of an award by AI would constitute 'an impermissible delegation of [arbitrators'] judicial functions' because 'drafting is an integral part of the deliberative process'.^[16] LaPaglia v. Valve Corporation,^[17] before the U.S. District Court for the Southern District of California, illustrates this point: the claimant sought to vacate the arbitral award on the grounds that the arbitrator had allegedly used generative AI to draft the award without proper supervision, which allegedly resulted in factual assertions unsupported by the record and stylistic elements characteristic of machine-generated text.

To address these legal limitations, amending the New York Convention and revising national and international instruments to expressly authorise (or prohibit) the use of AI arbitrators and the recognition of awards rendered by such arbitrators would constitute an essential first step.^[18] The proposal to establish an AI Ethics Review Committee ('AI-ERC') would help determine whether the use of AI has shifted from permissible assistance to an impermissible delegation of adjudicative power.^[19]

2. Technological obstacles

This section addresses the following technological obstacles: (i) the inability to replicate genuine legal reasoning; (ii) the scarcity and fragility of arbitration data; and (iii) the 'black box' and 'hallucination' phenomena.

i. Legal reasoning

Existing technologies do not enable machines to engage in genuine legal reasoning. AI, designed to solve problems using mechanistic formulas based on pattern recognition and statistical probabilities calculated from previously decided cases, does not employ deductive or logical reasoning to apply known legal rules. Assessing the parties' conflicting statements and evidence exceeds the current capabilities of these algorithms.^[20]

ii. Scarcity and fragility of arbitration data

The availability of large amounts of data is offset by the relatively small number of arbitration cases worldwide and their wide variation. Even the leading institutions receive no more than a thousand new cases per year. In 2025, the International Chamber of Commerce ('ICC') received 841 cases, the Singapore International Arbitration Centre ('SIAC') 625, and the London Court of International Arbitration ('LCIA') 362, which does little to satisfy the 'hunger for data' within the arbitration community. Furthermore, the confidentiality of arbitral awards and the rapid evolution of legal practice — rendering older data obsolete — further exacerbate this shortage.

     iii.  'Black box' and 'hallucination' phenomena

Even when used as a simple tool, AI raises questions about the causal link between the input data and the final decision. It is difficult to identify the factors that influenced a given result and to detect errors in the algorithm. This phenomenon, otherwise known as the 'black box', hinders human oversight of the accuracy of decisions or predictions produced by the AI tool and makes it more difficult to provide transparent and truthful justifications for specific decisions.^[21]

Added to this is the risk of 'hallucinations': lacking the necessary knowledge to provide an exact answer to a particular question, the system relies on computational probabilities and may generate unreliable — or even misleading — results and fabricated or incorrect sources.

To mitigate these obstacles, the data used to train AI arbitration software should be regulated to minimise bias and regularly updated to reflect the latest precedents and legal developments. For example, the UNCITRAL Model Law on Automated Contracting could provide a useful reference point for automated and potentially autonomous arbitration.^[22]

3. Ethical obstacles

This section addresses the following ethical obstacles: (i) algorithmic bias; (ii) privacy and cybersecurity issues; and (iii) the risk of dehumanising arbitral justice.

i. Algorithmic bias

As early as 2016, a ProPublica investigation found that an algorithm used to assess the likelihood of reoffending in the United States disproportionately assigned erroneous 'high-risk' scores to Black defendants.^[23] Once an established pattern favours a dominant group in a branch of arbitration, a 'robot arbitrator' could rule in favour of that group in any future dispute without examining the merits of the case, based on the data provided to it. Applying this observation to investment arbitration, where investors have more often than not prevailed against host states, an AI arbitrator could systematically rule in their favour.^[24]

ii. Confidentiality and cybersecurity

The use of a 'robot arbitrator' would lead to requests for the publication of arbitral awards, calling into question the interests of the parties and the associated confidentiality. The rise in cybercrime also raises ethical concerns regarding the security of these tools, as hacking, malware and other criminal attacks could allow decisions to be manipulated.^[25]

     iii. The dehumanisation of justice

Whilst a 'robot arbitrator' may surpass humans in terms of rationality and information processing, emotional intelligence eludes it. Human arbitrators, however, combine the application of the law with considerations of fairness, requiring a degree of discretion that AI — lacking empathy and intrapersonal and interpersonal skills — cannot exercise. Some go as far as to say that AI brings “the risk of dehumanisation and a loss of meaning in a society where we no longer look at the essence of what we do, but solely at processes and their efficiency". The risks are particularly high for arbitration and the administration of justice, as these rely on the exercise of 'moral, ethical and human values' that machines may never be able to adequately replicate.^[26]

Within the European sphere, the European Commission established a high-level expert group on AI in 2018, tasked with assessing the ethical issues arising from the use of AI and drawing up guidelines centred on the principle of justice and the need for judgments to be free from bias or stigmatisation.^[27] Furthermore, countries lacking the means to access these tools should be assisted so as to ensure more uniform access, as the absence of such access could lead to challenges based on a lack of transparency and non-discrimination.^[28]

Conclusion

In the short term, the relationship between predictive machines and humans is complementary; as the cost of prediction falls, predictive machines will assist with more sophisticated tasks. Although AI can already support arbitrators effectively across many areas, it is doubtful that it will be able to replace human arbitrators in the near future.

_____________

^[1].  T. Jones, 'AAA unveils AI simulation tool', Global Arbitration Review, 5 March 2026.

^[2].  C. Morgan and M.P. King, 'A Human Story: AI, Arbitration and the Importance of Judgment' in S. Nappert and F. Carvalho Dias de Oliveira Silva (eds), AI and Arbitration (2026).

^[3].  Silicon Valley Arbitration & Mediation Center (SVAMC), 'Guidelines on the Use of AI in Arbitration' (April 2024), Guideline 6.

^[4].  Chartered Institute of Arbitrators (CIArb), 'Guideline on the Use of AI in Arbitration' (April 2025), Part IV, ss 8, 9.

^[5].  French Code of Civil Procedure, Art 1450.

^[6].  Peruvian Arbitration Law (2008), Art 20.

^[7].  Italian Code of Civil Procedure (1940), Art 812.

^[8].  Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (AI Act) [2024] OJ L 1689, Art 14(4) and 14(5).

^[9].  A. Al Mahdouri, 'Public Policy and the Enforcement of AI-Generated Arbitral Awards' (2025) 42(6) Journal of International Arbitration, pp 759-790.

^[10].  UNCITRAL Model Law on International Commercial Arbitration, Art 34(2)(a)(iv

^[11].  New York Convention on the Recognition and Enforcement of Foreign Arbitral Awards (1958), Art V(1)(d).

^[12].  UNCITRAL Model Law on International Commercial Arbitration, Art 31(2).

^[13].  ICSID Convention (1965), Art 48(3).

^[14].  New York Convention (1958), Art V(2)(b).

^[15].  See supra note 2.

^[16].  A. Ross, 'Mourre on the dehumanisation of arbitration', Global Arbitration Review, 25 April 2025.

^[17].  LaPaglia v Valve Corporation, US District Court, Southern District of California, 8 April 2025 ('John LaPaglia v. Valve Corporation').

^[18].  A.O. Onyefulu, 'Artificial Intelligence in International Arbitration: A Step Too Far?' (2023) 89(1) Arbitration: The International Journal of Arbitration, Mediation and Dispute Management 56, p 77.

^[19].  H. Kaur, 'AI Ethic Review Committees and the Future of Responsible AI in International Arbitration', Kluwer Arbitration Blog, 9 March 2026.

^[20].  O.F. Cabrera Colorado, 'The Future of International Arbitration in the Age of Artificial Intelligence' (2023) 40(3) Journal of International Arbitration, pp 301–342.

^[21].  S. Moody, 'Embrace AI to reduce bias in arbitral decision-making, says Astigarraga', Global Arbitration Review, 13 March 2024.

^[22].  N. Gielen, D.I. Tan and Ö. Yazar, '2025 in Review: Technology', Kluwer Arbitration Blog, 10 February 2026.

^[23].  P.B. Marrow et al., 'Artificial Intelligence and Arbitration: The Computer as an Arbitrator - Are We There Yet?' (2019) 74(4) Dispute Resolution Journal, pp 35–76.

^[24].  See supra note 17.

^[25].  See supra note 17.

^[26].  See supra note 14.

^[27].  See supra note 17.

^[28].  Ibid.