Psychologists have long known that social context has a profound effect on human behavior, but lacked a unified, empirically based way to explain it. New research addresses this problem by using generative AI to systematically classify thousands of everyday social interactions. In the new study, researchers analyzed thousands of textual descriptions of social interactions between two people and used generative artificial intelligence (AI) to code the interactions by characteristics, resulting in a taxonomy of categories of social interactions. We then related these groups to variables such as conflict, power, and obligation, providing a comprehensive data-driven framework for quantifying the structure of interactions.
The study, “The Structure of Social Situations: Insights from Large-Scale Automatic Coding of Text,” by researchers from Carnegie Mellon University and the University of Pennsylvania (Pennsylvania), was published in the journal Psychological Science. “Researchers have proposed many frameworks for representing social situations, but due to the diversity and complexity of real-world situations, many are partial, unintegrated, and do not map to situations encountered in everyday life,” says study co-author Taya R. Cohen, professor of organizational behavior and business ethics at Carnegie Mellon University’s Tepper School of Business. “Our research advances the study of social cognition and behavior by using AI to create a more comprehensive framework for the structure of social situations.”
Because social situations have a profound influence on human behavior and mental life, understanding the structure and aspects of such situations has been a major topic of psychological research for decades. However, gaps remain, leaving the field without a rigorous understanding of how the most important characteristics relate to commonly encountered social interactions.
In this study, researchers analyzed more than 20,000 detailed text descriptions of social interactions between two people. They used a large dataset of short stories describing social interactions in daily life (e.g., family situations, interactions at work, interactions with animals, pet accidents, etc.) as well as short situation descriptions from other sources (e.g., blogs, novels, fiction published on social media, reading comprehension tests).
In this study, we combined large-scale language modeling (LLM) techniques to extract high-level situational characteristics from the dataset and core situational cues such as relationships, activities, locations, and goals (who, what, where, and why) that constitute the observable aspects of each situation.
“A central challenge in psychology is understanding the structure of social situations – the patterns and psychological characteristics that shape how people think, feel, and behave in social situations,” explains Sudeep Bhatia, associate professor of psychology at the University of Pennsylvania, who led the study. “Our research provides a rigorous, integrative framework for mapping everyday social situations and relating them to key theoretical aspects of psychology.”
This study found systematic associations between situational characteristics and observable cues as suggested by existing taxonomies, replicating and extending findings from previous studies, but on a larger scale. In particular, this study utilized a broader and more representative group of typical interactions experienced by adults.
“Our study provides a rich descriptive catalog of dozens of situation classes for researchers to test and refine their theories. It can be used not only to model the distributional structure of situations, as we have done, but also to formally study the effects of situations on interpersonal behavior, situation perception, goal pursuit, and the interaction between situation and personality,” Bhatia added.
Among the study’s limitations, the authors note that their analysis relies on short stories. Short stories are similar to the short autobiographical narratives used in previous studies, but more complex and nuanced situations are likely excluded. Moreover, their findings relied on analyzes conducted using the current generation of LLM, which has biases and limitations. Finally, this study only investigated English narratives, limiting the cultural scope of the conclusions.
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
