As artificial intelligence (AI) is gradually being integrated into various industries, human-AI collaboration has become a subject of significant concern. A joint study conducted by the Department of Psychology at Lingnan University found that while AI helps maintain employee performance during surges in workload, allowing it to autonomously decide when to take over and relinquish control may make it difficult for users to regain situational awareness, thereby increasing mental fatigue and even leading to human-machine conflict. Academics point out that enhancing the transparency of intelligent systems to help users understand AI decision-making logic, while ensuring humans retain control over task allocation, is essential for maintaining long-term performance and sound judgement. These findings have been published in the renowned interdisciplinary academic journal, International Journal of Human-Computer Interaction.
Lingnan research team uses the aviation industry as their research background, employing flight simulators to test different modes of human-machine collaboration. The team utilises the multi-attribute task battery (MATB), developed by NASA, to mimic work scenarios where operators have to multitask, and analyses how the distribution of decision-making and control authority between humans and intelligent systems affects safety, work efficiency, and user experience. Photo source: Lingnan University
Led by Prof Jie (Jay) Xu, Director of the Lingnan University Cognitive Science Research Centre and Associate Professor of the Department of Psychology, the research team used the aviation industry for their research background, employing flight simulators to test different modes of human-machine collaboration. The team used the multi-attribute task battery (MATB), developed by NASA, to mimic work scenarios where operators have to multitask, and analysed how the distribution of decision-making and control authority between humans and intelligent systems affects safety, work efficiency, and user experience.
The research team recruited 92 university students to participate in two rounds of practical testing, during which they were required to perform three tasks simultaneously within the simulation platform: tracking involving joystick manipulation, system monitoring, and resource management.
The first round of the experiment was designed to determine participants' baseline performance. When the workload surged, their performance in the tracking task, which requires continuous control and focus, dropped significantly, with accuracy falling from 76 per cent to 55 per cent. Meanwhile, other tasks, including system monitoring and resource management, remained largely unaffected. The team noted that these results suggest that, under high-pressure work conditions, employees allocate greater priority to discrete response-based tasks than to continuous monitoring tasks, leading to a noticeable decline in tracking performance.
In the second round, the team compared the impact of two different authority allocation strategies on work efficiency. The first was the human-led authority allocation (HLAA), where the operator decides when to delegate tasks to the intelligent system and when to retake control. The second was the shared authority allocation (ShAA), where the intelligent system automatically detects workload changes and proactively takes over or returns certain tasks.
The results showed that when the workload suddenly peaked or emergencies occurred, the ShAA helped alleviate human pressure, particularly by assisting with the tracking task, thereby preventing operators from being overwhelmed by the surge in workload and maintaining stable performance.
However, when the workload decreased, the HLAA resulted in superior performance. The team pointed out that when an intelligent system takes over autonomously, the operator may become detached from the actual operation, and when the system subsequently returns control, they may not be able to regain situational awareness immediately, leading to a temporary decline in performance. Allowing humans to decide for themselves when to hand over or reclaim control helps maintain control over the work situation and reduces unnecessary human-machine authority conflict.
The study also found that in long-duration work scenarios, participants using the ShAA experienced higher levels of fatigue in the latter stages of the experiment. Their fatigue scores were considerably higher than those in the human-dominant group, indicating that a higher degree of automation does not necessarily equate to a lower mental workload.
Prof Jay Xu Jie, Director of the Lingnan University Cognitive Science Research Centre and Associate Professor of the Department of Psychology, Photo source: Lingnan University
Prof Jie (Jay) Xu pointed out that while the intention of introducing intelligent systems is to assist humans in efficiently processing vast amounts of data and handling complex tasks, this study reveals that it is a misconception to assume that higher levels of automation will always make work easier, more efficient, or lead to better performance. This is especially true when AI determines when to take over, as operators must continuously monitor system operations, understand its decision-making logic, and anticipate its next moves. These additional cognitive demands may increase mental effort. Furthermore, the root cause of human-machine conflict may not be related to AI capability.
Prof Xu stated, "If an intelligent system fails to explain the rationale behind authority transfer to the user clearly, it is easy to create a cognitive gap between the human and the machine, which impairs collaboration. Future intelligent system design should place greater emphasis on transparency and human-machine communication mechanisms. This would allow users to understand system status and decision-making reasons, and adjust the authority allocation mechanism flexibly, based on the work context. This balance between automation efficiency and human authority is key to reducing human-machine conflict."
He also noted that these findings provide vital reference value for various industries, including in future flight deck design, remote operations, intelligent transportation, and human-machine collaboration scenarios involving high-risk decision-making.
Co-authors of the paper are Dr Ge Xianliang and Prof Song Xiaolei from Shaanxi Normal University; Dr Xu Hanlin, Dr Zhang Ke, and Dr Hao Ni from Zhejiang University.
For the full research paper, please visit: https://www.tandfonline.com/doi/full/10.1080/10447318.2026.2647133#abstract
