Human-AI Deliberation: An LLM-Based Approach to Privacy Decision-Making
Requirements
-
Required: Proficiency in programming to set up and adapt an LLM-based prototype system.
- Desirable: Completion of the lecture on "Human-Computer Interaction"
Contents
Recent work by Ma et al. (2025) has introduced an LLM-based approach to support human decision-making through iterative and structured deliberation between humans and AI.
The interaction does not present the user with a proposed decision-outcome, the system presents multiple perspectives on a decision problem. This design shifts the role of AI from being a decision-maker to being a deliberation partner, helping users arrive at deliberate and well-reflected decisions.
In sensitive contexts—such as deciding whether to share health data for research purposes—it is essential that people maintain autonomy to make deliberate decisions about how their data is used.
However, data-sharing decisions in complex ecosystems are often difficult to navigate. As a result, people frequently rely on heuristics (e.g., “just accept” or “always decline”) instead of engaging in careful reflection. This tendency can undermine informed consent and reduce individuals’ sense of control. Human-AI deliberation has the potential to counter this by structuring the decision process and presenting different perspectives, thereby supporting users in making more reflected and autonomous choices.
Research Aim
The aim of this thesis is to design, implement, and evaluate an LLM-based *Human-AI Deliberation* system for privacy decisions.
The specific objectives are:
- To develop a prototype system that uses an LLM to facilitate structured deliberation for privacy-related decisions.
- To conduct a qualitative user study evaluating how participants perceive the deliberation process and the decisions they arrive at.
- To assess whether users perceive privacy decisions made using this system to be deliberate and under their control.
References:
- Conceptualization and technical realization of Human-AI Deliberation:
Shuai Ma, Qiaoyi Chen, Xinru Wang, Chengbo Zheng, Zhenhui Peng, Ming Yin, and Xiaojuan Ma. 2025. Towards Human-AI Deliberation: Design and Evaluation of LLM-Empowered Deliberative AI for AI-Assisted Decision-Making. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25). Association for Computing Machinery, New York, NY, USA, Article 261, 1–23.
https://doi.org/10.1145/3706598.3713423
-Role of Reflection:
David Leimstädtner, Peter Sörries, and Claudia Müller-Birn. 2025. Designing Value-Centered Consent Interfaces: A Mixed-Methods Approach to Support Patient Values in Data-Sharing Decisions.
https://arxiv.org/abs/2407.03808
- Privacy Decisions in Complex Data-Ecosystems:
Qiurong Song, Yanlai Wu, Rie Helene (Lindy) Hernandez, Yao Li, Yubo Kou, and Xinning Gui. 2025. Understanding Users' Perception of Personally Identifiable Information. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25). Association for Computing Machinery, New York, NY, USA, Article 240, 1–24.
https://doi.org/10.1145/3706598.3713783