How social theory missed a Copernican moment with large language models

This commentary argues that social theory missed a Copernican moment with large language models (LLMs) not because it lacked relevant insight, but because it rarely entered the recent LLM wave as an explicit, design-shaping interlocutor. Through three cases (Actor–Network Theory and nonhuman agency, educational theory and attention, and Luhmannian systems theory and self-referential communication), it shows how concepts long elaborated in social thought could have informed the anticipation, interpretation, and design questions surrounding LLMs. The claim is not that these traditions offered literal blueprints for model architectures, but that they identified dynamics that became technically salient. The commentary also situates this missed encounter in the institutional organization of contemporary artificial intelligence (AI), where commercial agendas, benchmark cultures, and optimization-oriented research selectively privilege some forms of social knowledge over others. At the same time, it notes that social theory has influenced machine learning more visibly through ethical evaluation and governance than through architectural imagination. It concludes by reframing prediction as a modest, revisable, and accountable practice, and by proposing co-owned problems, instrumentalization without capitulation, and bilingual training as conditions for a more contributive relation between social theory and AI.