Reproduce “Telephone Surveys Meet Conversational AI: Evaluating a LLM-Based Telephone Survey System at Scale” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Max M. Lang, Sol Eskenazi
An LLM-based telephone survey system was developed and evaluated at scale by 60 Decibels Inc. and the University of Oxford, demonstrating the practical viability of fully automated, conversational AI for conducting complex surveys in real-world settings, including a large-scale deployment to over 2,700 participants in Peru.
Key insights
- Modern LLM-based systems can successfully manage complex structured survey logic and maintain consistent conversational flow across thousands of participants.
- While effective at guiding participants and clarifying responses for structured data, the AI currently struggles to probe for deeper, richer qualitative information compared to human interviewers.
- The integration of advanced Speech-to-Text, Large Language Models, and Text-to-Speech allows for human-like, full-duplex conversations in automated telephone surveys, expanding their practical applicability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.