When should a computer decide? Judicial decision-making in the age of automation, algorithms, and generative artificial intelligence

Abstract

This thesis interrogates what it is about the essenLally human activity of judging disputes that cannot be reliably or safely recreated through AI systems. To conduct this inquiry, two key ‘ingredients’ of the judicial decision-making process that exemplify this innately social practice are selected for analysis. The ability to participate in judicial decision-making, and the exercise of judicial discretion, are manifestations of the law as it exists within a wider social context, which shapes and condiLons it. Contrary to some analysis, the law is not applied through the mechanical selection of ‘correct’ precedents to cut and dried facts. The quotidian world of legal practice shows us that the law and its application are cultivated iteratively through a variety of legal actors. In this sense, it is malleable, contingent on context, and not necessarily computable in AI systems, including in the context of sophisticated Large Language Models. Indeed, it is this interpretation of the judicial role – as it exists within this wider social context – that prompts consideration of the features of judicial decision-making that either: most clearly exhibit the social and dynamic nature of law; and/or cater to it. This leads to the conclusion that the ability to participate effectively, and the exercise of judicial discretion, cannot be reliably or safely captured by AI systems, including Large Language Models.

Date of Award	Jul 2024
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Leverhulme Interdisciplinary Network on Cybersecurity and Society (LINCS) & Northern Ireland Department for the Economy
Supervisor	Niall McLaughlin (Supervisor) & John Morison (Supervisor)