WHEN INTERFACES ADAPT: DESIGN PRINCIPLES FOR GENERATIVE UI IN SAFETY-CRITICAL HMIS

Abstract

In safety-critical process industries such as oil and gas, operators depend on Human-Machine Interfaces (HMIs) to monitor and control complex systems. As automation increases, the operator’s role shifts from direct control to supervision; however, maintaining Situational Awareness (SA) and Cognitive Load (CL) remains a central challenge. Adaptive HMIs offer the potential to address this by tailoring information presentation to the operational context. This thesis explores how Generative AI (GenAI) can be configured to support adaptive HMIs in safety-critical environments, using the oil and gas domain as a centered example. Through interviews with eight operators and a literature review examining the intersection of SA, CL, adaptive HMIs, and safety-critical environments, six design principles were developed. These principles emphasize constrained, predictable adaptation rather than fully dynamic interface behavior, ensuring operators retain understanding and control over both the process and the interface. The principles were evaluated through a workshop with six designers, providing initial validation of their applicability, particularly for familiar domains such as alarm handling. Taken together, the results point to the need for adaptive HMIs to balance cognitive support with spatial stability and operator authority, aligning with human-centered AI foundations of reliability, safety, and trustworthiness. The study contributes to understanding how GenAI can enable context-aware, adaptive support while maintaining transparency and human oversight essential to safety-critical operations. Future work should focus on implementing these principles in functional prototypes and evaluating them in realistic operational settings.