R0044/2026-03-29/Q002/SRC06/E01¶
Marvin Project: military operators trusted AI at 82% rate, with 37% reduction in brain activity in risk assessment regions from frequent AI use.
URL: https://dl.acm.org/doi/full/10.1145/3748825.3748926
Extract¶
REPORTED: "Tests of the U.S. military's 'Marvin Project' showed that operators trusted AI recommendations at an 82% rate, leading to degradation in battlefield ethical judgment capabilities."
REPORTED: "Military officers who frequently use AI-assisted decision-making systems experience a 37% reduction in brain activity in regions associated with risk assessment."
JUDGMENT: These are striking findings but should be treated with caution. The source is a conference paper citing these statistics secondarily. The 37% brain activity figure in particular requires neuroscience methodology verification. The 82% trust rate is more plausible and consistent with other automation bias research showing high deference rates.
Relevance to Hypotheses¶
| Hypothesis | Relationship | Strength |
|---|---|---|
| H1 | Supports | Documents measurable cognitive consequences of AI trust in military contexts |
| H2 | Contradicts | Quantified evidence of harm from AI over-reliance |
| H3 | Supports | The harm mechanism is operator trust pattern (automation bias), not AI system agreeableness |
Context¶
The Marvin Project is cited in multiple sources discussing military AI risks but the primary study documentation was not directly accessible in this research. The statistics should be verified against original sources before being used in publication.