Interviewing Silicon Experts: A Persona-Based LLM Interview Pipeline in Automated Driving

Abstract

Human expert interviews are valuable but often limited by slow recruitment, scarce expertise, and network-based sampling, especially for abstract human-automation topics requiring multidisciplinary input. This paper presents a persona-based LLM expert-interview pipeline for structuring expert knowledge before human involvement. We constructed 28 discipline-specific personas and purposively selected seven differentiated “silicon experts” across human factors, UX design, algorithm engineering, accident investigation, regulation, accessibility, and product management. Using drivers’ minimum mental model (MMM) as a demonstration topic, each persona completed a structured interview protocol and returned JSON-formatted responses. The pipeline generated 59 candidate items, producing concrete, traceable, and auditable outputs. Results showed both convergence and persona-specific divergence: shared priorities included driver responsibility, operational design domain, and permitted non-driving-related tasks, while divergent contributions highlighted OTA updates, sensor limitations, and warning perceivability. The pipeline is positioned as a preliminary scoping tool for preparing subsequent human expert validation.