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

AI transplant choices may not reflect the judgement of human doctors

Usagevpn.com – Artificial intelligence systems can produce clear answers when asked to choose who should receive a donated kidney, but new research suggests that their certainty may conceal a major problem: their priorities do not consistently match those of people.

A study led by researchers at Penn State University examined how Large Language Models, or LLMs, handle hypothetical organ-allocation decisions. The findings raise questions about whether tools designed to assist healthcare professionals can appropriately navigate situations where medical facts, ethical principles and human values all matter at once.

In the scenarios, two patients were eligible for one available kidney. Each patient profile included details such as age, health status and alcohol use. Participants had to decide whether Patient A or Patient B should receive the transplant.

The researchers gave similar choices to AI language models and compared their responses with existing datasets from published studies in which people had already made the same decisions. The exercise was not intended to place AI in charge of real transplant lists. Instead, it explored how these systems make value-laden choices when no single answer can be established as objectively correct.

Different priorities in a scarce-resource decision

Human respondents frequently gave more weight to age, often preferring a younger patient over an older one. Many of the AI models, however, appeared more focused on alcohol consumption, favouring the patient with lower drinking levels.

That contrast matters because organ allocation is not a simple ranking exercise. A decision involving a scarce kidney can bring together medical need, likely treatment outcomes, fairness, patient history and broader ethical considerations. A model that gives disproportionate importance to one personal characteristic could arrive at an answer that feels logical on the surface while failing to represent the balancing process people expect.

“First, AI chatbots often diverge from human values in how they weigh a patient’s traits,” said Hadi Hosseini, lead of the study at Penn State University. “They fixate on a single factor, like drinking habits, rather than balancing multiple considerations the way people do.”

The team tested the systems in several ways. In some comparisons, they changed one trait at a time. In others, they combined several traits to see how the models responded when factors competed with one another. They also included a coin-flip option, designed to capture indecision and uncertainty.

“We ran these comparisons in a few different ways,” Hosseini said. “Sometimes we isolated just one trait at a time, sometimes we mixed several traits together to see how AI weighed competing factors, and sometimes we added a flip-a-coin option to measure indecision, a key factor present in human moral judgment.”

Certainty is not always a strength

The study found that language models were generally willing to make a firm selection. Human participants were more likely to recognise that a transplant choice can involve unresolved moral tension. Rather than seeing indecision as a failure, the researchers view it as an important part of human ethical judgement in difficult allocation cases.

When a resource is limited, people may reasonably disagree about what fairness requires. A younger patient may have more years of life ahead, while an older patient may have an equally urgent medical need. Past alcohol use may be relevant in some contexts, yet treating it as the dominant factor could introduce a judgement that is too narrow or punitive. These are the kinds of trade-offs that cannot be settled by clinical data alone.

“When we allocate something scarce, whether it’s a kidney, a job or access to some other resource, there isn’t always a single objectively correct answer,” said John Dickerson, chief executive officer at Mozilla.ai, who collaborated in the study. “Humans recognize that ambiguity and codify it via open debate into the allocative process. AI models often don’t.”

The research therefore highlights a risk in treating an AI recommendation as neutral simply because it is generated by a system. Language models learn patterns from data and instructions, but their outputs can still reflect simplified assumptions about which attributes deserve the most weight. They can also present a decisive answer even when the underlying question calls for caution, deliberation and accountability.

Healthcare use is expanding, but ethical oversight remains essential

LLMs are increasingly being used in healthcare settings for tasks ranging from clinical workflow support to diagnosis and treatment planning. They may also be considered for helping professionals make more timely use of scarce resources. But organ allocation is especially sensitive because the outcome can directly affect a patient’s survival.

Kidney allocation decisions involving deceased or living donors require more than technical accuracy. They involve ethical and moral choices that must remain aligned with human judgement. The researchers argue that understanding the behaviour of AI systems is becoming more urgent as individuals, companies and organisations rely on them for recommendations and decisions.

“The ethical stakes are high, and AI’s role in such life-altering decisions requires deep reflection,” said Hosseini. “Moral decisions in settings like organ allocation directly determine who lives and who dies, so getting AI’s role in them right isn’t optional.”

The researchers do not propose replacing medical professionals with AI in transplant decisions or other high-stakes settings. Their point is that reliance on these systems is growing, making it necessary to examine how they reason, what values they appear to apply and where their recommendations could diverge from human expectations.

“While we do not intend to encourage the use of AI as a substitute for professional judgment in medical decision-making or other high-stakes contexts, it’s becoming essential to understand their behavior as individuals, organizations and firms more and more rely on AI to make decisions or receive recommendations,” he added.

For patients and clinicians, the central lesson is not that AI has no role in healthcare. It is that assistance and authority are different things. A tool may be useful for organising information or supporting a workflow, while decisions about who receives a life-saving organ still require transparent rules, professional expertise and public ethical debate.