LLMs for Explainable Automated Vehicles: Supporting Indonesia’s Vision for 2045

Abstract

Indonesia’s Vision 2045 aspires to achieve developed-country status by emphasizing human development and technological advancement. Automated vehicles (AVs) are among the frontier technologies that could contribute to this ambition, yet Indonesia’s heterogeneous traffic conditions pose unique challenges for decision-making. Current AI-based AV controllers act consistently but lack transparency, limiting trust in their reasoning. This study investigates whether Large Language Models (LLMs) can enhance explainability in AV decision-making by generating human-understandable justifications. Using OpenAI models as a case study, we tested ethically challenging traffic scenarios and prompted models with 12 categories of human reasons identified in prior work. Results show that integrating these reasons improved alignment with expert judgments and supported socially acceptable maneuvers, though response times and reliability remain critical limitations. We conclude that LLMs are better suited as explanatory modules within Advanced Driver Assistance Systems than as primary controllers, offering pathways to safer and more trustworthy AV adoption in Indonesia.