Faculty of Science, Technology and Medicine
Faculty of Law, Economics and Finance
Faculty of Humanities, Education and Social Sciences
Interdisciplinary Centre for Security, Reliability and Trust
Luxembourg Centre for Systems Biomedicine
Luxembourg Centre for Contemporary and Digital History
Luxembourg Centre for European Law
Luxembourg Centre for Socio-Environmental Systems
The Doctoral School in Science and Engineering is happy to invite you to Emre KOCYIGIT’s defence entitled
Mitigating Deceptive Design: An Interdisciplinary Approach to AI-driven Detection
Supervisor: Assoc. Prof Gabriele LENZINI
Deceptive designs (dark patterns) are strategies that manipulate users into decisions against their best interests, causing harms such as privacy violations and financial loss. Effective detection is essential to mitigate these risks, yet several challenges persist.
First, despite taxonomies categorizing dark patterns, their assessment lacks objective criteria and measurement instruments. Second, dataset limitations—including limited size, diversity, and labeling inconsistencies—hinder the development of robust detection tools and reliable evaluation. Third, traditional detection approaches struggle with the evolving, diverse, and multimodal nature of dark patterns, while offering limited transparency and explainability.
This thesis addresses these challenges through a computer science approach with interdisciplinary considerations, including human-computer interaction, privacy, and user experience design. To mitigate subjectivity, measurable features are proposed within a systematic framework for cookie consent processes, enabling more objective detection. The thesis also examines existing datasets, identifying critical quality issues such as limited representativeness and noisy labeling. In response, a benchmark dataset is built, annotated by experts and aligned with a unifying taxonomy. Additionally, a multi-agent framework using Large Language Models for data augmentation is developed and validated through improved detection performance. The central contribution focuses on Multimodal Large Language Model-based detection tools, leveraging techniques such as Retrieval Augmented Generation and Chain-of-Thought while integrating the proposed measurable features to enhance both accuracy and explainability. The detection strategies were evaluated through quantitative analyses, expert interviews, and empirical comparisons. The study also conducts the first empirical evaluation of open-source Multimodal LLMs for dark pattern detection, comparing their effectiveness with proprietary models. The thesis concludes with a discussion of remaining challenges and open problems regarding objective detection of dark patterns.