Reproduce “AI Conversational Interviewing: Scaling Up Semi-Structured and In-depth Interviews” on OpenResearch

About this paper

Alexander Wuttke, Max Melchior Lang, Christopher Klamm, Quirin Würschinger, Frauke Kreuter

Researchers from LMU Munich, University of Oxford, University of Mannheim, University of Cologne, and University of Maryland developed "AI Conversational Interviewing," a method utilizing large language models to conduct semi-structured interviews at scale. This approach generates rich, open-ended textual data, providing deeper insights into public opinion and revealing diverse mental models and reasoning processes, while maintaining generally positive respondent evaluations compared to traditional surveys.

Key insights

• AI conversational interviews generate analytically valuable text data capable of revealing a broader diversity of considerations, arguments, and cognitive frames than what closed-ended survey items can capture.
• Respondents generally evaluate the AI interviewer positively for politeness, clarity, and impartiality, even when aware of interacting with an artificial agent, indicating practical viability for large-scale data collection.
• The choice of interview mode significantly influences respondent experience and data output; voice interviews yield more verbose and conversational data, while user choice of modality is crucial for overall preference and engagement.

Summary by alphaXiv.