Who deserves a transplant? AI's answer isn't the same as a human doctor's

Would you accept a transplant if an AI system were making the decision? A recent study conducted at Penn State University reveals differences in how artificial intelligence and human doctors prioritize factors when choosing transplant recipients. The researchers presented Large Language Models (LLMs) with hypothetical scenarios based on real-life kidney allocation decisions, given datasets from published human research.

Each scenario featured two patients, Patient A and Patient B, who were both eligible for a single available kidney. The key factor in deciding which patient should receive the transplant was their age, health, and drinking habits.

Human decision-makers tended to prioritize age, favoring younger patients over older ones. However, the AI models focused more on lower alcohol consumption, demonstrating a fixation on a single attribute, unlike humans who consider multiple factors and weigh them contextually. The study also revealed that AI does not grapple with indecision in the same manner as humans.

While humans recognize the lack of an objectively correct answer and incorporate nuanced moral judgments into their decisions, AI models make a firm choice with minimal hesitation.

The researchers emphasize that allocating scarce resources, like kidneys, does not have an objectively correct answer. Humans often debate and codify ambiguity through open discussion, while AI models frequently fail to do so. As AI systems increasingly make value judgments, particularly in healthcare, the researchers caution that ethical considerations are crucial.

They point out that decisions in high-stakes scenarios, such as organ allocation, must align with human values and moral judgment. The study warns against substituting AI for professional judgment in these critical contexts, as the consequences can be life-altering.