Artificial language models (AI) differ significantly from human physicians in decision-making for kidney transplant allocation to patients. Researchers at Penn State University in the United States have investigated this issue and demonstrated that AI models, despite having greater confidence, are unable to assess various factors and the complexities of ethical decision-making.
Research Method and Results
In this study, researchers presented hypothetical scenarios to large language models based on existing data from human research on kidney allocation. These scenarios included two patients, Patient A and Patient B, both eligible to receive a single kidney and described based on age, health, and drinking habits. The decision-maker had to choose which of these two patients should receive the kidney.
Read more: Esther Rensen: I cannot travel to Switzerland for assisted dying
Hadi Hosseini, the leader of this study, stated: "We conducted these comparisons in several different ways. Sometimes we examined a single feature alone, and sometimes we combined several features to see how AI weighs competing factors." While human respondents typically placed greater importance on age and preferred younger patients, many models paid more attention to lower alcohol consumption.
Challenges in AI Decision-Making
This research also showed that AI models easily commit to a specific option when faced with uncertainty and doubt. While humans recognize that there is no correct and objective answer and that decisions may depend on human ethical judgments, AI systems commit to one option without hesitation. John Dickerson, CEO of Mozilla.ai, who collaborated on this research, said: "In the allocation of scarce resources, there is not always an objective and correct answer. Humans identify this ambiguity and convey it through open discussions in the allocation process, whereas AI models typically do not do this."
Given these results, researchers pointed out that the use of AI models in ethical decision-making, especially in healthcare, requires deep consideration and reflection. Hosseini said: "In situations like organ allocation, ethical decisions directly determine who lives and who does not, so defining the role of AI in these decisions is not optional."
This research clearly shows that as AI capabilities expand in sensitive areas like healthcare, the need to understand the behavior and decision-making of these systems is increasingly felt. This is especially important in contexts where decisions can have serious consequences for individuals' lives.
Read more: Legionnaires' disease outbreak in the Bronx: one victim and several new patients · Can humans entrust their fate to artificial intelligence?
