A Methodology to Assess Human-Automated System Conflict from Safety Perspective

Assessment of Situation Awareness Conflict Risk between Human and AI in Process System Operation

The conflict between human and artificial intelligence is a critical issue, which has recently been introduced in Process System Engineering, capturing the observation and action conflicts. Interpretation conflict is another source of potential conflict that can cause serious concern for process safety as it is often perceived as confusion, surprise, or a mistake. It is intangible and associated with situation awareness. However, interpretation conflict has not been studied with the required emphasis. The current work proposes a novel methodology to quantify interpretation conflict probability and risk. The methodology is demonstrated, tested, and validated on a two-phase separator. The results show that interpretation conflict is usually hidden, mixed, or covered by traditional faults, and noises in observation and interpretation, including sensor faults, logic errors, cyberattacks, human mistakes, and misunderstandings, may easily trigger interpretation conflict. The proposed methodology will serve as a mechanism to develop strategies to manage interpretation conflict.

Towards pilot's situation awareness enhancement: A framework of adaptive interaction system and its realization

Adaptive interaction system in flight control always aims to enhance the pilot's situation awareness (SA) to achieve human-in-the-loop control. Most adaptive interaction systems are always activated according to the pilot's current workload state. However, the pilot may already lose important information during a high workload, and thus the corresponding reaction of the adaptive interaction system would lag. Moreover, most adaptive interaction systems adopt the expert's knowledge as a reference to generate information. Still, the tacit knowledge that reveals the information interrelationship is seldom studied, despite being the foundation of the interactive information display. To solve the above problems, we proposed an adaptive interaction system architecture with three subsystems. Firstly, we developed a workload level prediction subsystem, where physiological parameters are used to predict future workload levels, thus avoiding interaction system lag; Secondly, we developed a tacit expert knowledge mining subsystem to discover the interrelationship hidden in the expert's perceived information, which will guide the interactive information interface. Thirdly, we developed a tips information inference subsystem to provide the lost SA information based on expert knowledge and the pilot's online perceived information. The effectiveness of the proposed system is verified via a comparative experiment utilizing the control interface of a remotely piloted aircraft.