Title:What's Privacy Good for? Measuring Privacy as a Shield from Harms due to Personal Data Use

Abstract:We propose a harm-centric conceptualization of privacy that asks: What harms from personal data use can privacy prevent? The motivation behind this research is limitations in existing privacy frameworks (e.g., Contextual Integrity) to capture or categorize many of the harms that arise from modern technology's use of personal data. We operationalize this conceptualization in an online study with 400 college and university students. Study participants indicated their perceptions of different harms (e.g., manipulation, discrimination, and harassment) that may arise when artificial intelligence-based algorithms infer personal data (e.g., demographics, personality traits, and cognitive disability) and use it to identify students who are likely to drop out of a course or the best job candidate. The study includes 14 harms and six types of personal data selected based on an extensive literature review.
Comprehensive statistical analyses of the study data show that the 14 harms are internally consistent and collectively represent a general notion of privacy harms. The study data also surfaces nuanced perceptions of harms, both across the contexts and participants' demographic factors. Based on these results, we discuss how privacy can be improved equitably. Thus, this research not only contributes to enhancing the understanding of privacy as a concept but also provides practical guidance to improve privacy in the context of education and employment.