A Penn State University study reveals that AI chatbots make faster but less nuanced decisions than doctors when it comes to allocating a kidney. Faced with scenarios involving two eligible patients, the models often favor a single criterion (notably alcohol consumption), whereas humans weigh several factors, including age, and acknowledge the moral ambiguity of these choices.
The researchers found that the AI shows virtually no hesitation, unlike human participants, who incorporate uncertainty into their judgment. According to Hadi Hosseini, the systems frequently deviate from human values by simplifying complex dilemmas. John Dickerson of Mozilla.ai points out that the allocation of scarce resources is based on a moral debate that the models fail to replicate.
While life-learning models (LLMs) are increasingly used in healthcare, particularly to optimize care pathways and resource management, the authors emphasize that organ allocation decisions require strict alignment with human values. They believe that the role of AI in these life-saving choices must be carefully defined, the goal not being to replace medical judgment, but to understand the biases and limitations of these systems.
 Pascal Lemontel
|
21/09/2026 | 
Anglais
Arabic
Deutsch
Français
Español
Italien
Português
Russian