“AI” as gaslighting – Considering the affordances of stochastic generative systems in relation to anthropo-centered epistemic operations

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).

This paper mobilizes the notion of ‘AI’ as a carcinogen, as developed by Berman, to then inquire and analyze what is the possible human impact that results from the democratization of the use of Large Language Models and Generative Pretrained Transformers, from the perspective of how epistemic frameworks are mobilized as a core component of sociability.It then advances the proposal that we understand ‘AI’ as a case of the ‘Thomas Theorem’, and that, as such, LLMs and GPTs can effectively operate as systems of non-targeted, non-intentional, gaslighting.From that approach, the paper follows up with a materially-grounded analysis of how the operations performed on data for the creation and usage of LLMs and GPTs mean that these technologies have the affordance of being episteme-fixative, and that such an affordance (as defined here) is specific to these technologies, in a way that allows for non-essentialist and non-deterministic analysis that considers both changes and continuities in both media systems and our everyday engagement with them.It concludes by suggesting venues of inquiry that are made easier by the previous conclusions, and returns to capitalism’s uptake of technology, rather than to any immanent aspect of technology, as a central aspect to consider in evaluating the epistemic harms that might come out of these affordances.