How Ageism Contributes to AI Harms: Findings from the AI and Aging Harms Project

Background and Purpose: Researchers have documented age bias in AI systems and drawn attention to how older adults have been marginalized in AI harms and policy discourses. The concepts of digital ageism, AI ageism, and generative ageism have been developed from that work. What is still missing is an understanding of the role of ageism itself in actual AI harms experienced by older adults. This gap is striking given the relatively long-standing research on how racism, sexism, and ableism get embedded and reinforced through AI in practice. The goal of this overdue analysis is to move beyond theoretical critiques and descriptions of age bias embedded in systems to understanding the role of ageism in documented harms.

Methods: This analysis was part of the larger Aging and AI Harms Project. An international team of six gerontology scholars qualitatively analyzed the 80 verified incidents across three AI harms databases that directly harmed older adults. The data for analysis were descriptions of the incidents provided in the databases and at least two of the media articles reporting on each incident and linked within each databases’ report. After all members coded the same 6 incidents to develop the coding scheme, the team of six analyzed incident content written in English, German, Korean, Polish, Portuguese, and Spanish, relying on translation into English for other languages. As the larger Project identified harms experienced by older adults and not only those that could be attributed to ageism or age discrimination, we analyzed these data for evidence of various levels of ageism at play in the harm. We used an inductive coding process where two analysts separately coded each incident report before meeting with the team to discuss and reach consensus.

Results: Approximately two-thirds of the 80 incident reports that involved harms to older adults could be associated with ageism as a contributing underlying factor. Ageism could be identified as embedded in intentional or unintentional design of algorithms, biased systems, and AI-facilitated scams directed at older persons. These included ageism-involved AI harms such as deaths, misrepresentation, and loss of coverage for critical health and long-term care services. These resulted from a range of AI uses, from care staffing allocation systems and automated decision making systems to scams.

Conclusions and Implications: This is the first study to analyze the role of ageism in actual AI harms experienced by older adults, providing an evidence base for future research, advocacy, and policy. By qualitatively analyzing evidence of ageism in reports of 80 verified incident reports in 25 countries, we demonstrate that ageism contributes to harms across diverse applications of AI. It is important to highlight that this research identified and analyzed incidents that rose to the level of media attention. Future studies should engage civil society groups that represent older adults and those focused on AI and civil rights to identify harms to older adults that are underreported. The current study can inform efforts such as the UN Convention on the Rights of Older Persons to address the scaling of ageism.