Beyond consent: Reconstructing ethical justification in medical adaptive machine learning systems.

Area of Science:

• Medical Ethics
• Artificial Intelligence in Healthcare
• Machine Learning Governance

Background:

• Medical Adaptive Machine Learning Systems (MAMLS) continuously learn from clinical data, blurring lines between therapy and research.
• This necessitates re-evaluating their ethical classification and the applicability of informed consent requirements.

Purpose of the Study:

• To analyze the ethical challenges posed by MAMLS, particularly concerning informed consent.
• To explore the limitations of consent-centered ethics in the context of adaptive machine learning systems.
• To propose an alternative ethical framework for governing MAMLS.

Main Methods:

• Conceptual analysis of ethical frameworks for adaptive machine learning in clinical settings.
• Examination of the implications of irreversibility and data bias in MAMLS on consent and autonomy.
• Critique of traditional research ethics and proposal for a public health ethics approach.

Main Results:

• Informed consent faces structural limitations due to the irreversibility of deep learning models and the "autonomy dilemma" caused by data bias.
• Withdrawability, a key aspect of consent, is undermined, potentially transforming consent into institutional delegation.
• Individual consent can paradoxically harm collective autonomy when data is unrepresentative or biased.

Conclusions:

• Ethical justification for MAMLS requires moving beyond research ethics to a public health ethics framework.
• Transparency, accountability, and participatory governance are crucial for building public trust in MAMLS.
• Addressing MAMLS ethical challenges necessitates a broader societal approach beyond individual consent.