責任の拡散：生成AIを活用した臨床的意思決定に対する患者の道徳的判断

As Generative Artificial Intelligence (GAI) becomes increasingly embedded into complex medical decision-making, it is anticipated that patients' psychological and behavioral responses will diverge significantly from those observed in traditional healthcare settings. This study, therefore, aims to investigate a key challenge for AI governance: how does GAI involvement in therapeutic decisions influence patients' moral judgment of the physician when adverse outcomes occur?

A between-subjects vignette experiment was designed. A sample of 331 participants was recruited to evaluate a clinical scenario in which a physician prescribed a treatment leading to an adverse outcome. The experimental design orthogonally manipulated two variables critical for healthcare governance: (a) physician adherence to GAI-generated advice, and (b) the nature of the recommended regimen (personalized versus standard-of-care). Moral blame directed toward the physician served as the primary dependent variable. The data were analyzed using PROCESS Model 1 (v4.1) in SPSS 28 for moderated regression analysis.

The analysis shows a significant diffusion of responsibility i.e., when an adverse outcome occurs, physicians who adhere to GAI advice incur significantly less moral blame than those who do not (p < 0.001). Furthermore, the type of GAI-endorsed treatment moderates this relationship (p < 0.05). The blame mitigation effect is substantially stronger when the regimen is personalized (p < 0.001) and attenuated, though remaining significant, under the standard-of-care condition (p < 0.05). Finally, perceived patient harm strongly and positively predicts moral blame (p < 0.001).

The findings demonstrate that following an adverse outcome, patients ascribe less moral culpability to physicians who incorporate GAI recommendations, an effect most pronounced for personalized treatment plans. This highlights a diffusion of responsibility in AI-mediated care. Ethical and sustainable GAI integration will, therefore, require the creation of clear frameworks for accountability, comprehensive physician training, and transparent patient communication.

Not Applicable.