Protect Me from What I’m Not! Personal Data Accuracy in Large Language Models

  • Workshop
  • Living Room
  • Wednesday 20.05 — 10:30 - 11:45

Organising Institution

University of Münster

Germany

The University of Münster is one of the largest universities in Germany with a rich and time-honoured tradition. It enjoys an outstanding reputation in the region and far beyond. Fifteen faculties with 120 degree programmes and some 30 research centres comprise the institutional backbone of the University. Some 42,500 students and 5,000 academics appreciate the University's excellent research opportunities, high-quality teaching, promotion of junior researchers, and the advantages of living in the city of Münster. Our slogan sums it up best: "living.knowledge". The University of Münster is signatory of the Magna Charta Universitatum.
Due to “hallucinations” and uncurated training data, Large Language Models (LLMs) frequently output false information concerning natural persons. As technical privacy research on LLMs is still at an early stage, there is currently little work concerning these personal data inaccuracies. They are also underexplored in legal discussion. In this workshop, participants gain a deeper understanding of this important issue, both from a legal and technical side. Martin Bernklau, who is a prominent victim of personal data inaccuracies generated by Bing Copilot, will discuss his experience with the participants. We then provide a taxonomy of personal data inaccuracies. The participants are asked to challenge this taxonomy for completeness and precision in interactive discussion and to come up with their own examples. We will incorporate this feedback into future versions of the taxonomy.

Host

Paulina Jo Pesch

FAU Erlangen-Nürnberg - Germany

Paulina Jo Pesch is Assistant Professor of Civil Law, Law of Digitalisation, Data Protection Law and Law of Artificial Intelligence at the Friedrich-Alexander university of Erlangen. Her research focuses on the legal challenges of technologies that challenge privacy and regulation. Her current work focuses on the legal implications of Large Language Models (LLMs), AI image generators and automated decision-making systems. Paulina’s research spans several areas of law, with a particular focus on data protection, copyright, AI regulation, and fundamental and human rights.

Host

Kristina Magnussen

University of Münster - Germany

Kristina Magnussen is a computer scientist and research assistant at the University of Münster. She holds a Master's degree from the University of Innsbruck. In her research, she focuses on data protection issues in Large Language Models (LLMs), especially personal data accuracy. She is part of the SMARD-GOV project, funded by the German federal government, which investigates how LLMs can be utilised in public authorities while complying with data protection laws.