Department of Mathematics and Computer Science

The use of chatbots based on large language models (LLMs) poses potential risks to user privacy (Weidinger et al., 2022). Studies indicate that there are concerns regarding the inappropriate use of communication data (Belen-Saglam et al., 2022). There is an urgent need for privacy-conforming interaction concepts and explanation mechanisms for language models to address users' privacy concerns.

The goal of this master’s thesis is to contribute to this important field of research through a scoping review and a subsequent study. The goal of the review is to identify existing or proposed privacy-conforming interaction concepts and explanation mechanisms for LLMs to address users' privacy concerns.

Based on the results of the scoping review, a study should then be conducted to further explore privacy-conforming interactions with LLMs. The most promising concepts found in the scoping review should be realized in a user interface and tested in a user study. This study could involve both quantitative and qualitative approaches to gain insights into user behaviors regarding privacy in LLM interactions while identifying effective interaction design to meet users‘ privacy concerns.

References

Laura Weidinger, Jonathan Uesato, Maribeth Rauh, Conor Griffin, Po-Sen Huang, John Mellor, Amelia Glaese & others. 2022. Taxonomy of Risks posed by Language Models. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT '22). Association for Computing Machinery, New York, NY, USA, 214–229. https://doi.org/10.1145/3531146.3533088

Belen-Saglam R, Nurse JRC and Hodges D. 2022. An Investigation Into the Sensitivity of Personal Information and Implications for Disclosure: A UK Perspective. Front. Comput. Sci. 4:908245. doi: 10.3389/fcomp.2022.908245

S. Zhang et al. Adanonymizer: Interactively Navigating and Balancing the Duality of Privacy and Output Performance in Human-LLM Interaction. 2024. https://arxiv.org/abs/2410.15044