Cognitive Load-Aware Edge AI: Reducing Acute Operator Fatigue in Human-AI IoT System

As the number of AI-driven IoT systems cooperating with human operators grows, system performance is no longer limited solely by computational resources. It also depends on the operator's cognitive capacity. The paper discusses the newly introduced Cognitive Load-Aware Edge AI (CLAE) framework, in which AI agents dynamically adjust their behavior based on the inferred load on the operator's cognitive capacity. The proposed framework leverages a range of interactions and observations, including alert response latencies, physiological indicators, and workload context, to assess immediate mental stress levels. By doing this, the system sets the alert frequency, assigns decisions, and determines the automation depth for both edge and IoT deployments. This paper models cognitive capacity as a limited resource and uses a Kalman filter to combine behavioral and physiological signals. Edge orchestration layer applies adaptive regulations that switch system operation across various timescales, such as milliseconds (alert filtering), seconds (automation adjustments), and minutes (decision-making delegation). Testing of this cognitive-aware orchestration in a simulated industrial monitoring facility shows a 41% decrease in operator error rates and a 29% improvement in incident resolution when compared to available state-of-the-art methods. The ablation studies verify that the combination of multimodal sensing and multi-timescale adaptation is the key factor for improved performance. The re-conceptualization of human-AI collaboration in this paper is achieved by treating cognitive capacity as a primary system resource, thereby enabling more sustainable and effective operation of large IoT systems.