Step Further Towards Automated Social Science: An AI-Powered Interview Platform

Abstract

This paper introduces MimiTalk, an AI-powered automated interview platform designed to address the challenges of traditional qualitative research methods in social sciences. By leveraging Large Language Models and advanced system architecture, the platform aims to reduce resource requirements while maintaining high-quality data collection standards. We conducted comprehensive experiments with 20 participants recruited through Prolific, evaluating the platform's performance through both quantitative metrics (information entropy, NLP analysis, and semantic coherence) and qualitative feedback. Results demonstrate that AI-conducted interviews achieve comparable information entropy (around 4.1) to human-led interviews, with high semantic coherence scores (mean 0.8170) and positive user experience feedback. While the platform successfully addresses geographical constraints and interviewer bias, limitations in emotional cue detection and cultural sensitivity were identified. This research contributes to the evolving landscape of automated qualitative research methods, suggesting that while AI cannot fully replace human interviewers, it can significantly enhance research efficiency and scalability. The findings lay groundwork for future developments in AI-assisted qualitative research methodologies.

Keywords: Artificial Intelligence, Qualitative Research, Large Language Models, Automated Interviews, Research Methodology, Human-AI Interaction