Yoan Hermstrüwer (University of Zurich): The Judicial Mind of Large Language Models

About the Speaker:
Yoan Hermstrüwer is Professor for Legal Tech, Law and Economics as well as Public Law at the University of Zurich. He is also a Research Affiliate at the Max Planck Institute for Research on Collective Goods in Bonn.

About the Topic:
This study investigates whether legal doctrine can constrain AI decision aids from pursuing independent political agendas, similar to how it constrains human judges. We experimentally test five large language models by having them predict European Court of Human Rights decisions using post-training cases to avoid memorization. We compare a baseline condition (using zero-shot instructions) against three treatments that provide different forms of doctrinal access through retrieval augmented generation (RAG): simple case law access (vector-based RAG), structured decision trees (graph-based RAG), and both combined. We measure prediction accuracy, reasoning quality, contextual attention, and doctrinal structure utilization, and find that a combination of vector-based and graph-based RAG yields the best results.

About the Speaker Series:
The Speaker Series of the new Minerva Fast Track Research Group “Artificial Justice” is organised by Katharina Isabel Schmidt. The Series invites guest speakers who work at the intersection between law, computer science, and the humanities. Neither technical nor juristic knowledge is a prerequisite for participation — the Series is aimed at anyone with an interest in critical and interdisciplinary perspectives on “Law and AI”. The event takes place on Zoom and is scheduled to last one hour.

The virtual lecture will be held as a video conference via Zoom. Please register no later than Tuesday, 14 October 2025 using this LINK.

You will receive the login details on Tuesday afternoon. If you do not receive an email containing the login data, please check your spam folder as well.

Please confirm upon registration that you agree to the use of Zoom and that you will not record the event. By attending the event you confirm that you have read and agreed to Zoom’s Terms of Service and Privacy Policy.