ChatGPT as a fourth arbitrator? The ethics and risks of using large language models in arbitration

Abstract Large language models (LLMs) like ChatGPT have the potential to significantly change how arbitral proceedings are conducted by aiding arbitrators in performing case analyses, drafting decisions, and undertaking legal research; this may enable arbitrators to render high-quality decisions more quickly. However, the use of LLMs in arbitration comes with potential risks due to a lack of reliability and confidentiality concerns, which, to an extent, can be mitigated. Beyond this, the use of LLMs in arbitration raises a more specific concern: whether the use of these models is compatible with the scope of the personal mandate of arbitrators. Given that arbitrators are bound to solve disputes themselves and are generally not allowed to delegate this duty to third parties, the arbitral community must establish clear standards regarding the use of LLMs. This article proposes evaluating the acceptability of using LLMs in arbitration based on the particular task of the arbitral tribunal in question. An arbitral tribunal must handle its core tasks of materially deciding a dispute without recourse to these models. However, ancillary tasks that may influence the outcome may be accomplished with the support of LLMs, provided that disclosure and consent of parties are ensured. Tasks that have no bearing on the outcome may be accomplished with the support of an LLM at the arbitral tribunal’s discretion.