Reproduce “In Silico Sociology: Forecasting COVID-19 Polarization with Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Austin C. Kozlowski, Hyunku Kwon, James A. Evans
Researchers from the University of Chicago utilized GPT-3, trained on pre-October 2019 data, to simulate liberal and conservative responses to future COVID-19 issues, accurately forecasting real-world partisan polarization in 84% of cases. This methodology, termed "in silico sociology," revealed that established ideological patterns in American discourse largely prefigured the pandemic's politicization, driven by differing reliance on institutional trust versus individual freedom.
Key insights
- Pre-existing ideological structures and belief systems within American discourse significantly prefigured the politicization of COVID-19, suggesting that ideologies possess an internal logic that channels the reception of new issues.
- Large Language Models can serve as "digital doubles" or empirically realistic agents for social simulations, enabling a new paradigm of "in silico sociology" to explore complex sociological questions.
- Qualitative analysis of LLM-generated justifications elucidated core ideological fault lines, such as differential trust in institutions and the prioritization of public safety versus personal freedom, driving simulated partisan divides.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.