Introduction to the Special Issue on Ethical Performance Aspects of AI-Enabled Systems in High-Consequence Domains

It is with great pleasure that we introduce this Special Issue that focused on one of the most pertinent topics of our time—the ethical considerations of using AI systems in high-consequence domains. It is estimated that 2 billion humans have used AI and that hundreds of millions use AI on a daily basis within their jobs and daily lives (Porwal, 2026). Ethical issues associated with AI-based systems, particularly as they relate to Human-Machine Teaming (HMT), represent an emerging, necessary, and impactful research area for the community (Pflanzer et al., 2023). High-consequence domains represent a decidedly pertinent context under which research is needed to better understand the factors that shape attitudes toward AI and to parse out the consequences of using AI in these contexts. The current special issue centers on high-consequence domains ranging from medical triage, to lethal autonomous weapons (LAWS). It involves topics ranging from the use of AI as a decision aid, AI alignment, methodological considerations including evaluation methods as well as understanding the maturity of the literature, and the use of AI as a bi-directional attention cue.

AI systems may serve in different roles when teaming with humans, perhaps even in an advisory capacity. Schelble and colleagues (2026) examined human reactions to AI-based advisors during ethically hard decision-making scenarios. They found that relative to human advisors, advice from an AI was associated with lower stress and less perceived responsibility during ethical decision making. Yet, AI advisors were believed to have lower performance relative to human advisors. These results motivate a potentially paradoxical situation wherein humans experience greater ease but potentially poorer performance with AI-based advisors relative to human advisors. The study also examined how different levels of influence shape human attitudes toward AI advice.

Summerville and colleagues (2026) examined the role of AI alignment in contexts where there are no absolute correct decisions. In ethical dilemmas, behavioral choices may be subjectively favored rather than being objectively correct, and this causes ambiguity and creates challenges for AI developers. What do designers program if there is no “right” answer? The key innovation discussed by Summerville and colleagues herein is the idea of an alignable AI trained on the attributes that impact decision making rather than training/adapting it to human behaviors. They completed two studies, the first of which demonstrated that the degree of alignment between decision makers was predictive of trust across six attributes in a mass casualty scenario. Study 2 expanded the medical realism of study 1 and added two types of AI-based decision algorithms into the scenario. Consistent with study 1, they found that the AI’s alignment predicted trust and willingness to use the AI.

Morey and colleagues (2026) offer a narrative review of pre-deployment methodologies for AI evaluation. They center on Cognitive Systems Engineering (CSE) and Naturalistic Decision Making (NDM) as methods that could be applied to the evaluation of AI systems. Both CSE and NDM attempt to capture the complexity of real work and the joint activity between humans and machines which is critical for understanding ethical decision making. These methods emphasize considering the relevant environment, relevant expertise, and representative cognitive challenges associated with the AI use in situ. This paper details a compelling rationale for why CSE and NDM represent effective methods for evaluating AI systems.

Shekh et al. (2026) provide a scoping review to examine factors shaping policy, governance, design, and development of lethal autonomous weapons (LAWS). They found that very little research covers operational decisions to deploy LAWS, noting a critical gap in the literature. The scoping review included discussions of regulatory and moral requirements, organizational culture, training, system characteristics, and operational demands. Clearly, there is currently a dearth of literature to support some research questions involving LAWS, which is both an opportunity and a concern for the community to address.

Perhaps the quintessential exemplar of a high-consequence domain is fighter operations (i.e., dogfighting) wherein pilots must divide their attention among a dynamic set of aviation-based tasks and mission-centric demands all while facing an adversary with hostile intent. Smith and Schnell (2026) used an applied aviation context to examine when and how a machine can monitor pilot psychological states within a cockpit environment and use intuitive cues to direct attention. They used an innovative cockpit interface to integrate eye tracking, head position, real-time workload assessment, and task performance during a human-agent teaming context using a proxy fighter aircraft and operational pilots as subjects. Their integrated system was designed to dynamically understand the pilots’ attention and performance within a dogfight, and cue attention in a task-centric, performance-enhancing manner. Overall, they found benefits of the system in the form of increased performance, higher trust, and reduced workload. This highlights the potential benefits of machines operating as bi-directional teammates in high-consequence activities like dogfighting.

We hope that you enjoy this Special Issue and that it helps you in your AI journey.