The article explores the use of Artificial Intelligence (AI) in arbitral or judicial decision-making from a holistic point of view, exploring the technical aspects of AI, its practical limitations as well as its methodological and theoretical implications for decision-making as a whole. While this article takes the angle of international arbitration, it looks at examples and studies from a wide variety of legal areas and its conclusions are relevant for adjudicatory decision-making more globally. The author assesses existing studies on decision outcome prediction and concludes that the methodology and assumptions employed put into doubt the claim these models might be used for ex ante outcome predictions. The article also discusses whether AI models, which are typically based on information extracted from previous input data, are likely to follow ‘conservative’ approaches and might not be adapted to deal with important policy changes over time. The article further finds that a blind deferential attitude towards algorithmic objectivity and infallibility is misplaced and that AI models might perpetuate existing biases. It discusses the need for reasoned decisions, which is likely to be an important barrier for AI-based legal decisionmaking. Finally, looking at existing legal theories on judicial decision-making, the article concludes that the use of AI and its reliance on probabilistic inferences could constitute a significant paradigm shift. In the view of the author, AI will no doubt fundamentally affect the legal profession, including judicial decisionmaking, but its implications need to be considered carefully.
Journal of International Arbitration