New project will develop the use of artificial intelligence for interviews

When social scientists want to explore people's views on various issues, they often face a difficult choice.

Should they use surveys, which can gather responses from many people to pre-defined questions relatively quickly and cheaply? Or is it better to conduct qualitative interviews with individuals, which are much more in-depth but also labour-intensive and difficult to scale up?

Researchers behind a new project supported by the Villum Foundation (see box) will now develop and test a new method that uses artificial intelligence to combine some of the advantages of scientific surveys and semi-structured interviews.

The aim is to develop an AI model, a so-called AI interviewer, that can not only answer questions like ChatGPT, but also ask qualified questions and meet formal requirements such as data security.

According to project leader and Associate Professor Hjalmar Bang Carlsen from the Copenhagen Centre for Social Data Science (SODAS), this would be a rare methodological breakthrough. He believes that AInterviewer can become a completely new tool in social science research:

"The ambition is that generative AI will enable us to collect qualitative data at the population level. This will allow us to representatively analyse the experiences and attitudes of population groups in a way that leaves more room for the respondent's perspective and experience horizon than the standardised survey," says Hjalmar Bang Carlsen, who is leading the project in collaboration with Anna Rogers from the IT University of Copenhagen.

"At the same time, it is important to emphasise that AInterviewer will not be able to replace the semi-structured interview, which functions as an interaction between the interviewer and the interviewee. AInterviewer is a complement that will probably qualify the traditional survey and thus create a third way of collecting interview data in the social sciences".

Green transition as a case study

To develop and test the AI model, the researchers are conducting a survey of Danes' views and experiences of the green transition. It will be carried out in three versions:

850 participants will conduct AI interviews, once a first version of AInterviewer has been developed. 50 will participate in normal qualitative interviews, while 400 will complete a survey.

This will allow us to compare the quality of AI interviews with two known methods and to further develop the model. As a bonus, it will also provide new insights into Danes' views on the green transition,' says Hjalmar Bang Carlsen.

In practice, AInterviewer will be designed as a chatbot that can conduct interviews in both Danish and English. The model will be open source and thus available to other researchers, and the project will allow AInterviewer to be developed further through new projects.

Read more about the Villum Foundation's grant round: 47 Million DKK Awarded for Interdisciplinary Research