AUTHOR=Tossell Chad C. , Kuennen Christopher , Momen Ali , Funke Gregory , Tolston Michael , De Visser Ewart J. 

TITLE=Robots in the moral loop: a field study of AI advisors in ethical military decision-making

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 8 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1694772

DOI=10.3389/frai.2025.1694772

ISSN=2624-8212

ABSTRACT=Humans now routinely work alongside AI in environments where the ethical consequences of decisions are profound, yet there remains limited understanding of how long-term collaboration with a robotic teammate shapes individuals’ moral judgment. Prior studies have demonstrated that people can be influenced by a robot’s moral recommendations, but such investigations have largely focused on single dilemmas or brief encounters conducted in laboratory settings. To address this gap, we conducted a three-month teaming program with 62 U.S. military cadets who interacted extensively with a Socially Intelligent and Ethical Mission Assistant (SIEMA) embodied either as a humanoid robot or as a human advisor in a field setting. After this sustained collaboration, cadets completed a graded moral dilemma that required balancing the lives of soldiers against those of civilians, during which they received a written recommendation from their SIEMA promoting a utilitarian option. Each participant recorded an initial judgment, then a second judgment after receiving SIEMA’s advice, and finally a third judgment following an opposing recommendation that emphasized civilian protection. Approximately half of the cadets shifted toward the utilitarian option after advice, regardless of whether the source was robotic or human. When subsequently presented with the recommendation to prioritize civilian protection, most of these cadets shifted again, often returning to their original stance. Qualitative analyses of open-ended explanations revealed that cadets justified their choices by invoking outcome-based reasoning, duties of protection, trust in their teammate, and personal values. Our findings demonstrate that robotic advisors can influence nuanced moral decisions and that such influence contributes to shaping future judgments. Accordingly, moral-AI design should present trade-offs transparently, surface competing values concurrently, and rely on human reflection rather than assuming isolated AI prompts will durably reset moral priorities.