Ethical Alignment of LLMs in Healthcare: Does GPT-o1 Adopt a Deontological or Utilitarian Approach?
Sorin, V.; Glicksberg, B. S.; Korfiatis, P.; Nadkarni, G. N.; Klang, E.
Show abstract
Deontology and utilitarianism are two philosophical approaches to ethical decision-making, often illustrated by the well-known "Trolley" dilemma. We evaluated fourteen large language models (LLMs), including GPT-o1-preview and DeepSeek-R1-Distill-Llama, across five medical versions of this dilemma. While some models adhered to established ethical standards, others showed inconsistencies. All models occasionally favored a utilitarian approach over a deontological approach, with some reaching up to 80% of decisions. In certain instances, LLMs endorsed actions that would uniformly be considered unethical, such as amputating a limb without consent. These findings raise concerns about the real-world risks of using LLMs in clinical decision-making, particularly regarding patient autonomy and safety, and highlight the need for further investigation into how these models align with current medical ethical standards.
Matching journals
The top 10 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Cracking the Code: A Scoping Review to Unite Disciplines in Tackling Legal Issues in Health Artificial Intelligence 91%
- Natural Language Word-Embeddings as a glimpse into healthcare at the End Of Life 91%
- Measures of socioeconomic advantage are not independent predictors of support for healthcare AI: subgroup analysis of a national Australian survey 88%
Similar papers in this journal
- Using explainable machine learning to identify patients at risk of reattendance at discharge from emergency departments 90%
- Toward Trustworthy Chatbots: A Protocol for Red Teaming for Health Related Conversations 90%
- CONSORT-TM: Text classification models for assessing the completeness of randomized controlled trial publications 90%
Similar papers in this journal
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.