Redaction and Exhibition: Enacting and Communicating Privacy by Exploring Identifiability and Artistic Practice

Human-centered privacy empowers individuals to understand and control their data. While often situated in human-computer interaction (HCI) and usable privacy and security, research in this field benefits from cross-disciplinary perspectives. This thesis develops a dialogue between computer science and the arts and humanities to examine both how privacy is enacted in practice and how it is communicated to users. We study privacy and identifiability in unstructured text data, such as conversation transcripts collected by social and behavioral science researchers, focusing on redaction as a key mechanism for protecting identity. We evaluate the capabilities of existing redaction tools and large language models (LLMs) for automating this process and, based on researchers' needs, design a semi-automated, human-in-the-loop workflow. This approach operationalizes context-dependent definitions of identifiability while leveraging LLMs to flag sensitive information in a graphical format, reducing the overhead of traditional methods while maintaining researcher control. In parallel, we investigate how privacy is made legible and engaging, drawing on a large corpus of privacy-related artworks we collected to analyze how artists communicate privacy concepts. We translate these insights into HCI by prototyping designs and outlining potential studies on effects of incorporating artistic mechanisms into privacy communication methods.