AI-Augmented Surveys: Leveraging Large Language Models and Surveys for Opinion Prediction
Researchers have developed a framework using large language models (LLMs) to predict public opinion trends from survey data. This AI-augmented approach can retrodict missing opinions in historical surveys and predict opinions for years when specific questions were not asked. The models, tested on General Social Surveys from 1972-2021, showed strong performance in recovering trends, such as the rise in support for same-sex marriage, though predicting entirely unasked opinions remains challenging. The study highlights how LLMs can enhance survey research by filling data gaps and how surveys can help calibrate LLMs for simulating human opinions. AI
IMPACT Enhances survey research by enabling prediction of historical and unasked opinion trends, potentially improving understanding of societal shifts.