Type: article
Abstract: Large language models have rapidly diffused into work and decision-making, yet their default operation reproduces dominant epistemologies and reshapes how users articulate problems, evaluate evidence, and reach conclusions. Existing critiques concentrate on model bias and training data, leaving the broader sociotechnical conditions of use underexamined. This paper offers a critical theory analysis of large language models conceptualized as a sociotechnical ecosystem of training data, learning algorithms, provider governance, prompts, and outputs. Four oppressive features emerge: collapsing epistemological differences, economizing deliberation, foreclosing communication moderation, and discounting substance. Together, they constrain user freedoms across the ecosystem regardless of provider intent or user awareness. Because four of the five ecosystem elements remain closed to ordinary users, the prompt is the principal site at which user agency can be exercised and the natural locus for future emancipatory design research.
Source: AIS Electronic Library (AISeL)
Authors: Tingelhoff, Fabian, Marga, Jens Joachim
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222911637
