The introduction of AI-based operator support systems in nuclear power plant main control rooms (MCRs) is being actively pursued. While AI can reduce operators' cognitive burden through functions such as accident diagnosis, procedure recommendation, and system status summarization, operator overreliance—the uncritical acceptance of AI outputs—can lead to critical consequences, including acceptance of misdiagnosis, entry into incorrect procedures, and omission of essential safety actions.
Recent studies have reported that in AI-assisted environments, users' subjective confidence remains critically decoupled from actual decision accuracy, and cross-verification behavior against primary data sources is markedly diminished. However, no design framework has systematically applied these findings to the safety-critical context of nuclear power plant MCRs.
This paper proposes Deliberate Cognitive Friction as a design principle: the systematic insertion of intentional resistance into the cognitive pathway through which operators automatically accept AI outputs, thereby promoting critical verification and autonomous judgment. The proposed framework establishes the timing of AI conclusion presentation to the operator as its primary design axis, determining where friction is applied. The mode and intensity of friction are then adjusted according to situational dimensions including plant state, information complexity, and time margin.
The framework provides a structured design direction for protecting operators' autonomous judgment in the design of MCR AI-based operator support systems.