Scientific Text as Field: Tracking Normative Shifts in Science with a Human–AI Workflow

• In Event: Science, Knowledge & Technology Section Roundtables
  In Refereed Roundtable (60 minute): Table 1: AI and Changing Social Norms

Abstract

Scientific fields increasingly frame their work in terms of societal relevance, responsibility, and utility, yet we lack systematic evidence on how these norms evolve and who advances them. My research treats scientific text as field, using abstracts as performances of legitimate norms. I develop a human–AI workflow that formalizes iterative inductive–deductive movement between theory and evidence, yielding reusable sentence-level measures of “socialized” norms (explicit societal/clinical/policy/industrial relevance and normative constraints) alongside internal epistemic ideals. With this measurement infrastructure, I conduct a cross-field, longitudinal analysis to test whether responsibility frames displace or layer onto epistemic ideals and to identify the actors driving change (subfields, institutions, regions, continuing authors vs. new entrants). This study applies KOB decomposition to separate composition effects from within-group shifts driving normative change. Robustness checks assess alternative periodizations, baseline definitions, and AI model families. Contributions are threefold: (1) a portable codebook and sentence-level measures adaptable to other norm-detection tasks; (2) cross-field, longitudinal evidence on the “normative turn” in science; and (3) an interpretive bridge linking textual dynamics to theories of scientific professionalism.

Author

• Cai Siman, The Chinese University of Hong Kong