ethique_TM

Review of Bioethics and Brains: A Disciplined and Principled Neuroethics by John R. Shook and James Giordano

Fri, 27 Jun 2025 00:33:19 -0400

Walking Backward to Ensure Risk Management of Large Language Models in Medicine

Thu, 26 Jun 2025 20:00:00 -0400

J Law Med Ethics. 2025 Jun 27:1-11. doi: 10.1017/jme.2025.10132. Online ahead of print.

ABSTRACT

This paper examines in what way providers of specialized Large Language Models (LLM) pre-trained and/or fine-tuned on medical data, conduct risk management, define, estimate, mitigate and monitor safety risks under the EU Medical Device Regulation (MDR). Using the example of an Artificial Intelligence (AI)-based medical device for lung cancer detection, we review the current risk management process in the MDR entailing a "forward-walking" approach for providers articulating the medical device's clear intended use, and moving on sequentially along the definition, mitigation, and monitoring of risks. We note that the forward-walking approach clashes with the MDR requirement for articulating an intended use, as well as circumvents providers reasoning around the risks of specialised LLMs. The forward-walking approach inadvertently introduces different intended users, new hazards for risk control and use cases, producing unclear and incomplete risk management for the safety of LLMs. Our contribution is that the MDR risk management framework requires a backward-walking logic. This concept, similar to the notion of "backward-reasoning" in computer science, entails sub-goals for providers to examine a system's intended user(s), risks of new hazards and different use cases and then reason around the task-specific options, inherent risks at scale and trade-offs for risk management.

PMID:40574504 | DOI:10.1017/jme.2025.10132

May, Joshua. (2023). Neuroethics: Agency in the Age of Brain Science. New York, US: Oxford University Press

Wed, 25 Jun 2025 23:15:13 -0400

Parental Authority and the Weight of Assent: Navigating Moral Dilemmas in Adolescent End of Life Care

Wed, 25 Jun 2025 20:00:00 -0400

Though the idea of assent emerged in the 1980s and adaptation in paediatrics has become more common practice, adolescent end-of-life care adds a multi-faceted layer of increased complexity to the assent process. In the current era of moral pluralism, medical decisions must account for much more than clinical considerations: they must abide by legal standards of decision-making which usually prioritize parental preferences. In the case of an adolescent at the end of life, there remains much room for improvement when disagreements arise. Conflict abounds between the parent or other legal decision-maker and healthcare team in situations where the parent prefers not to disclose the severity of the adolescent’s illness or the healthcare team believes continuing lifesaving therapies are not in the adolescent’s best interest. These dilemmas challenge how we solicit adolescents’ preferences (assent) and their options for refusal (dissent). Disagreements create tension and weaken communication between the family and providers. Healthcare workers struggle to fulfil their professional obligations and retain their professional identities when values collide. We explore the moral distress healthcare workers face navigating conflict amongst various stakeholders involved in the adolescent’s end-of-life care. Strategies to minimize moral distress are also provided.

It Is Not About AI, It’s About Humans. Responsibility Gaps and Medical AI

Wed, 25 Jun 2025 20:00:00 -0400

A lot of the language we use to refer to AI, including in healthcare, uses terminology that originally and literally applies to humans and human relationships. Such terminology includes both non-evaluative terms, like “learning,” “memory,” or “intelligence,” and evaluative terms, like “trust” or “responsibility.” In this article I focus on the latter type and the way it is applied specifically to the case of medical AI. Focusing on the discussion of “responsibility gaps” that, according to some, AI generates, I will suggest that such terminology is revealing of the nature of healthcare professional obligations and responsibility prior to and independently of the assessment of the use of AI tools in healthcare. The point I make is generalizable to AI as used and discussed more broadly: the language used to refer to AI often tells more about humans and human relationships than about AI itself and our relationship with it. In healthcare, whatever else AI will allow us to do, it can prompt us to reflect more thoroughly on professional responsibility and professional obligations.

The Ethics of Humanlikeness in AI Therapy Chatbots