Integrating large language models and affective computing for human-machine symbiosis in intelligent driving

The evolution of Driver Assistance Systems (DAS) is shifting focus from mere safety to integrating emotional and psychological well-being, transforming intelligent connected vehicles (ICVs) from passive tools into cognitive partners that require complex, bidirectional interaction.1 Affective computing (AC), which enables machines to recognize and interpret human emotions, provides a crucial foundation for this shift. Large Language Models (LLMs) can significantly advance AC by processing multimodal data, enabling a transition from functional execution to empathetic human-machine interaction. Despite early applications like mandated fatigue monitoring, current systems are limited by passive responsiveness and opacity.4 While LLM-enhanced AC promises to address these issues, this integration creates a Collingridge's Dilemma (Figure 1). This commentary examines this paradox, focusing on the technical potential, limitations of LLM-empowered AC and the associated governance complexities, aiming to foster discussion on responsible innovation in next-generation intelligent driving.

This commentary examines the paradoxical challenges from the convergence of these three domains: (1) the transformative potential and technical limitations of LLM-empowered AC and (2) the governance complexities surrounding responsible innovation in driving assistance. We seek to promote discussion on balancing technological progress with ethical implementation2,5 in developing next-generation intelligent driving systems.