Generative ai and sociology: analyzing virtual respondent technology

The article analyzes the use of virtual respondents and large language models (LLM) in sociological research, including a review of domestic and foreign examples. The technical aspects of survey simulation with the help of LLM, including the choice of models, as well as practical recommendations for their application are considered. The advantages of this technology, such as the possibility of rapid data collection and adaptability, as well as disadvantages, including ethical risks and limitations in representativeness, are analyzed. Based on two pilot studies, it is found that to improve the effectiveness of the virtual respondent, it is recommended to first conduct a pilot survey with real participants, upload the results to the LLM and create a generalized virtual respondent trained on real responses to generate more reliable responses to the questions of interest.