ScholarWorks@성균관대학교

초록

The proliferation of social networking services and generative artificial intelligence has intensified the tension between the desire for social connection and the growing risks to facial privacy. Although technical methods for facial de-identification have been developed, they often fail to reflect how users experience privacy and how it is shaped by the social contexts in which the images are shared. This study explores how users’ experiences vary depending on the degree of generative AI-based facial de-identification and the relationship between the viewer and the person in the image. We conducted a user study with 36 participants who evaluated modified versions of their own photos in neutral or negative social scenarios. Greater facial de-identification yielded stronger privacy control and a reduced sense of risk. However, when imagining an adversarial social connection, it increased privacy concerns and reduced trust, indicating that privacy judgment is highly contextual. Participants desired more control over how their image could be modified, as well as more customization options, while defining user groups based on social relationships. Based on these suggestions, we redesigned our system to offer more flexible and socially responsive privacy options. This study challenges a purely technical view of privacy, arguing that it is a social, interpretive, and dynamic process negotiated within specific relationships. Ultimately, our findings provide both empirical insights and practical design guidance for creating privacy-preserving systems that are more meaningfully aligned with the lived realities of users.

키워드