AI in biomedicine and healthcare: Sociological perspectives on personalized HIV therapy and skin cancer detection tools

Abstract

AI-based technologies in biomedicine and healthcare are gaining more importance due to the new hype around AI and the increasing concerns about future challenges in healthcare. It is assumed that the logic within these tools, depending on which data can be digitalized in the first place, what is considered relevant, and the way data and information is used within these tools, influences knowledge constructions within the field they are used, including which information is seen as relevant. If we want to understand how medicine and healthcare might change through the usage of AI-based tools, a sociological analysis of how these tools influence constructions of knowledge in the context in which there are used is paramount. This cumulative dissertation chose two cases of AI-based tools for this endeavor: HIV treatment optimization tools as one of the first successful AI-based tools within personalized medicine, and skin cancer detection tools. These were analyzed through the lenses of sociology of categorization and classification, sociology of risk and uncertainty, and feminist science and technology studies. A key goal of the work was to explore the categories made relevant in the AI-based tools compared to categories made relevant within the context of use. The situational analysis of HIV treatment optimization tools concluded how patients, as marginalized implicated actors are discursively constructed by other social worlds. The feminist STS analysis of skin cancer detection tools identified how racial discrimination in the field led to racially biased AI-tools. The comparative analysis between the two cases studies shows how different categories of people are made relevant in the tools and which categories were regarded as relevant to solve the problem at stake. Only a minor part of the relevant categories in the field found their way into the digital tools. While social constructions inform actions in the field of HIV treatment optimization tools, the AI-based tool itself is solely based on genetic data. Differently, for skin cancer detection tools, the social category of race can become highly relevant. The dissertation concludes that AI-based tools can lead to further naturalization of social categories such as race, even more so as categorization is increasing in relevance to account for fairness through AI-based tools.