Large language models can predict the results of social science experiments
A new study shows that large language models (LLMs) can accurately predict results from social science experiments, suggesting they could serve as powerful tools for hypothesis generation and research design in the social sciences.
Background
Large language models (LLMs) like GPT-4 are AI systems trained on vast amounts of text to predict and generate human-like language. Social science experiments study how people behave in controlled settings (e.g., moral dilemmas, economic games). This paper, published in *Nature*, shows that LLMs can simulate these experiments, accurately predicting human responses without running the actual study. This matters because it could let researchers test hypotheses cheaply and quickly before costly real-world trials, but also raises concerns about AI "standing in" for human subjects — potentially missing cultural or individual nuance. The findings build on a growing debate about whether LLMs are just statistical text predictors or actually develop some form of social or causal understanding.