Drawing on Knowledge Organization (KO) theory, this paper offers a sociotechnical framework for integrating artificial intelligence (AI) into the production and classification of knowledge. We argue that the dominant AI paradigm, which frames classification as a statistical optimization task, overlooks the situated, social, and historical dimensions that give knowledge meaning. To bridge this gap, we introduce the domain-analytic framework from KO as a methodology for designing more context-aware and epistemically responsible AI systems. The paper proceeds in three parts. First, it identifies the limitations of current AI by contrasting its underlying assumptions with the core principles of domain analysis. Second, it translates these principles into a heuristic framework with actionable design questions, offering a practical path for aligning AI systems with the values and practices of specific knowledge communities. Finally, it explores the more profound philosophical challenges posed by generative AI, using the concept of semiosis to examine how machinic interpretation reshapes our understanding of documents, stability, and meaning itself. By reframing classification as both a technical and an interpretive practice, this paper provides a foundation for developing AI systems that are more reflexive and responsive to the social dynamics of human knowledge.
Published: AI & SOCIETY (2025)
https://doi.org/10.1007/s00146-025-02763-3
Authors:
Thellefsen, M. and Friedman, A.
