Artificial intelligence can make faster and more confident choices than people, but a new study suggests that AI may approach difficult medical decisions differently from humans.
Researchers from Penn State University examined how large language models, or LLMs, would decide which patient should receive a life-saving kidney when only one organ was available.
The findings showed that AI kidney transplant decisions often relied on fewer factors than human decisions.
The researchers said the difference raises important questions about using AI in situations that involve complex ethical judgments.
Researchers Compare AI With Human Decisions
For the study, researchers presented LLMs with hypothetical kidney allocation scenarios.
The scenarios came from existing datasets based on earlier research involving human participants. Each scenario presented two eligible patients competing for one available kidney.
The patients differed in characteristics such as age, health and alcohol consumption.
The AI systems then had to choose which patient should receive the available organ.
Researchers tested the models in several ways. Sometimes they changed one characteristic at a time. In other cases, they combined multiple characteristics to see how the systems handled competing factors.
They also gave some models a coin-flip option to examine whether AI would acknowledge uncertainty.
AI Focused on Different Patient Factors
The study found an important difference between human participants and AI models.
Human respondents generally placed greater importance on age. They tended to favor younger patients over older patients when making allocation decisions.
However, several AI models placed more emphasis on alcohol consumption.
According to lead researcher Hadi Hosseini, the models sometimes focused heavily on one characteristic rather than balancing several factors.
“AI chatbots often diverge from human values in how they weigh a patient’s traits,” Hosseini said.
He added that the systems could fixate on one factor instead of considering the broader context.
AI Shows Less Uncertainty Than Humans
Another difference involved uncertainty.
Human participants often recognized that difficult allocation decisions do not always have one clearly correct answer.
The AI systems, however, generally selected one patient without showing the same level of hesitation.
Researchers said this matters because decisions involving scarce resources often require discussion about competing ethical principles.
John Dickerson, chief executive officer of Mozilla.ai and a collaborator on the study, said humans can recognize ambiguity and use debate to shape allocation processes.
AI models, he said, often do not handle that ambiguity in the same way.
AI Raises Questions About Medical Ethics
The findings come as AI systems become increasingly involved in healthcare.
Large language models are already being explored for tasks involving clinical workflows, diagnosis, treatment planning and other medical processes.
Kidney allocation presents an especially difficult challenge because doctors and healthcare systems must consider both medical factors and ethical principles.
Therefore, AI kidney transplant decisions cannot be evaluated only by asking whether a model produces a quick or consistent answer.
Researchers argue that developers also need to examine whether AI systems reflect human values and appropriately handle uncertainty.
AI Should Not Replace Medical Judgment
The researchers stressed that their findings should not be interpreted as support for replacing doctors with AI in high-stakes medical decisions.
Instead, they said understanding how AI systems behave is becoming increasingly important as organizations use these tools for recommendations and decision-making.
Hosseini said the ethical stakes are particularly high when AI influences decisions that can affect whether people receive life-saving treatment.
He argued that AI’s role in such situations requires careful consideration of its values, limitations and decision-making behavior.
What the Study Means for the Future of AI
The study highlights a broader challenge facing artificial intelligence.
AI systems can process information quickly and provide confident answers. However, speed and confidence do not necessarily mean that a decision reflects the complex reasoning humans use in morally difficult situations.
As healthcare organizations explore AI applications, researchers say they must consider more than accuracy.
They also need to examine how these systems weigh competing factors, handle uncertainty and respond to decisions involving human values.
For AI kidney transplant decisions, those questions are especially important because organ allocation involves a scarce resource and consequences that can be life-changing.
