Correct Legal Reasoning in Artificial Intelligence 

Partners: Utrecht University & Tilburg University
Type: Postdoc
Duration: 2026—2027

AI has shown potential in the legal domain, and its wider deployment could increase access to justice by reducing workloads and lowering systematic barriers that prevent citizens from receiving the justice they deserve. However, how can we ensure that the legal reasoning used by AI systems is correct? Incorrect reasoning can have serious consequences, particularly when AI systems are used in high-stakes situations to support decisions that affect people’s lives.

Many contemporary AI systems operate as black boxes, and their reasoning capabilities remain poorly understood. At the same time, legal reasoning is complex and multi-faceted. This project investigates what it means for legal reasoning in AI systems to be sound, how this can be evaluated, and how AI systems can be designed to meet the resulting requirements.

Within AlgoSoc’s justice sector, the project aims to understand which properties the legal reasoning capabilities of AI systems should have. By investigating contemporary large language models and comparing their reasoning with relevant standards for legal reasoning, the project contributes to a deeper understanding of how AI systems reason in the legal domain. It also explores the design of AI systems that combine large language models with explicit knowledge representations grounded in legal theory, with the aim of making their reasoning more explainable and responsible.

External supervisor: prof. mr. dr. Johan Wolswinkel, Tilburg University