Original title: Mapping the Emerging Social Science of Large Language Models
Yi Yang, Xiao Jia, Zeyun Dong, Chenzhang Wang +1 · September 2026
Large language models, or LLMs, are quickly reshaping work, school, and daily choices, but social science research studying them is scattered. Researchers organized this growing field by analyzing tens of thousands of papers with automated computer tools and careful reading. They grouped the research into three main areas. The first area, called LLM as Social Minds, studies whether models show human-like traits, such as reasoning or bias. The second area, LLM Societies, looks at how groups of AI agents interact and coordinate. The third area, LLM-Human Interactions, examines how people actually use, trust, and respond to these tools. This third area makes up most of the overall research. However, studies on artificial minds and multi-agent groups dominate top conference citations, while human-focused studies lead in top journals. Overall, this map helps researchers track how model behaviors, multi-agent systems, and human institutions shape the social impacts of AI.
You are organizing a massive library with no labels. You start grouping books by the topics they cover, but some books could fit into multiple groups. In this paper, the researchers used computer methods to group thousands of research papers about artificial intelligence into three distinct categories: studies of the AI itself, studies of multiple AIs together, and studies of how people use AIs. They found that these groups are stable and that different groups get more attention in academic conferences versus journals.
This taxonomy helps researchers, funders, and policymakers see the whole landscape of social-science work on LLMs. It shows that studies of model behavior and agent groups dominate top conferences, while studies of human use dominate journals. Understanding this split can guide interdisciplinary collaboration and publication strategy.
The abstract doesn't state explicit limitations, but the taxonomy is based on papers from five databases and a curated set of 198 papers; it may not capture work in other languages or non-indexed venues. The agreement metrics (77.78% and 73.83%) indicate that some papers do not fit neatly into one domain. The venue analysis is limited to selected conference and journal subsets.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Mapping the Emerging Social Science of Large Language Models — Yi Yang, Xiao Jia, Zeyun Dong et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.