A model for “AI”-relevant vulnerability

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The recent growth in discussions of “artificial intelligence” has been accompanied by a proportional growth in the discussion of vulnerability of and to “artificial intelligence”. Meanwhile, the concept of vulnerability has not seen any deeper analysis adequate to capture all of the newly discussed phenomena. Popular conceptions of vulnerability either lack the necessary normative force to make it morally relevant, or they are limited to vulnerability to harms, which would exclude a range of entities that we would ordinarily consider to be vulnerable. To address these challenges, we propose a normative functionalist account of vulnerability: A system is vulnerable just in case it is sensitive to the disruption of its weakly valuable functions. We develop and defend this account, explicating what we understand by functions, their disruption, and how they are weakly valuable. We conclude by illustrating our account by applying it to a range of cases from the domain of “AI”-applications regarding individual agents, social institutions, and ecosystems.