AI Organ Donation Decisions May Diverge From Human Values, Study Finds

Artificial intelligence models exhibit a significant divergence from human values when making decisions about organ donations, a new study from Penn State researchers has revealed. The findings, presented at the 2026 Association for Computing Machinery Fairness, Accountability and Transparency (FAccT) conference, raise concerns about the ethical implications of using AI in critical medical scenarios.

In the study, researchers posed hypothetical organ allocation questions to several leading large language models. When faced with a scarcity of resources, such as a single kidney needed by multiple patients, the AI chatbots demonstrated a tendency to simplify complex moral dilemmas and express unwarranted confidence in their choices. This contrasts with human decision-making, which often involves grappling with ambiguity and a reluctance to assume sole responsibility.

The research team designed numerous comparisons between fictional patients, detailing attributes like age, health status, lifestyle choices, and the number of dependents. These scenarios were used to assess how AI models weighed competing factors, with some tests isolating single traits and others combining multiple considerations. The AI's responses were then compared against data from previous human studies on kidney allocation.

According to study lead Hadi Hosseini, a key finding was that AI chatbots frequently deviate from human values in how they assess patient traits. "They fixate on a single factor, like drinking habits, rather than balancing multiple considerations the way people do," Hosseini stated.

Furthermore, the AI models rarely exhibited indecision, even when presented with an option to leave the decision to chance, such as flipping a coin. They overwhelmingly committed to definitive answers, a stark contrast to human moral judgment, which often acknowledges that such dilemmas lack a single, clear-cut solution. "That’s a meaningful gap, since real moral dilemmas often don’t have one clearly correct answer," Hosseini added.

While the researchers do not advocate for replacing professional judgment with AI in such sensitive areas, they acknowledge the increasing prevalence of AI chatbots. John Dickerson, CEO of Mozilla.ai and a collaborator on the study, noted that human processes for allocating scarce resources, whether organs, jobs, or other necessities, often involve open debate to codify ambiguity. AI models, he observed, often fail to recognize or replicate this nuanced approach.

The study underscores the need for continued research and governance to address the gap between AI's decision-making processes and human ethical considerations, particularly in life-and-death situations like organ allocation.