Constrained Generative UI for the Cockpit: A Comparison of LLM-Assembled and Pre-Defined Interfaces

Abstract

In-car infotainment systems are growing in complexity, making traditional design and engineering approaches costly and difficult to scale. Recent advances in large language models (LLMs) offer context-aware, dynamic capabilities, enabling interfaces to adapt in real-time and reducing the need for manual design. We propose a scalable framework for Constrained AI-Assembled Graphical User Interfaces (GUIs) that dynamically assembles the interface at runtime from a pre-validated library of components using an LLM. We evaluated this approach in a comparative user study (N = 30) in a simulated driving environment, measuring usability, as well as driver’s cognitive load, distraction, and situational awareness. Quantitative results showed no significant differences between AI-assembled and conventional pre-designed interfaces regarding usability and driver’s mental state. Our qualitative feedback highlighted occasional challenges in information prioritization. We conclude that AI-assembled GUIs have the potential to achieve performance similar to conventional pre-defined interfaces and provide design recommendations to guide future work.