AI Maps 20,000 Interactions Into Social Taxonomy
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Researchers at Carnegie Mellon and the University of Pennsylvania analyzed over 20,000 textual descriptions of two-person social interactions, using large language models to code them into a unified taxonomy of situational categories.
- Sudeep Bhatia, Associate Professor of Psychology at Penn, led the study alongside coauthor Taya R. Cohen, Professor of Organizational Behavior and Business Ethics at Carnegie Mellon's Tepper School of Business.
- The AI extracted both high-level features and core situational cues — relationships, activities, locations, and goals — then mapped them onto theoretical dimensions like conflict, power, and duty.
- Source material spanned an eclectic range: online short stories, family situations, workplace interactions, animal encounters, blog posts, novels, social media fiction, and reading-comprehension exams.
- The study replicated and extended earlier findings on situational structure at much larger scale, using a broader and more representative sample of typical adult exchanges than prior work.
- Authors flagged three limitations: reliance on short stories that likely exclude complex, nuanced situations; the known biases and constraints of current-generation LLMs; and an English-only dataset that restricts cultural scope.
- The work is published in Psychological Science (2026) under the title "The Structure of Social Situations: Insights From the Large-Scale Automated Coding of Text."
Why it matters: Psychology has long lacked a unified, empirically grounded way to describe the social situations that shape behavior, leaving researchers to work from partial and non-integrated frameworks. This study gives the field a rigorous, scalable catalog of dozens of situation classes — anchored in 20,000 real-world descriptions — that future research can use to formally test how situational features influence interpersonal behavior, goal pursuit, and personality expression.
