Towards a pragmatist dealing with algorithmic bias in medical machine learning

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Bioethics

Background:

• Machine learning (ML) offers transformative potential for medical diagnostics, therapeutics, and prognostics.
• The integration of ML in healthcare introduces significant ethical considerations, particularly concerning algorithmic bias.
• Algorithmic discrimination stemming from biased training data poses a critical challenge to equitable healthcare delivery.

Purpose of the Study:

• To address the challenge of discriminatory algorithmic judgments in medical machine learning.
• To propose a novel framework for evaluating the ethical implications of ML applications in medicine.
• To advocate for a pragmatic approach to managing algorithmic bias in healthcare settings.

Main Methods:

• Analysis of the ethical challenges posed by biased training data in medical ML.
• Critique of the distinction between justified differential treatment and unjustified bias.
• Application of a reformulated pragmatist philosophy, drawing on William James's concept of truth.

Main Results:

• The distinction between justified and unjustified algorithmic bias is difficult to implement in practice due to data complexities.
• A purely objective assessment of algorithmic fairness in medicine is often unattainable.
• The proposed pragmatist approach shifts focus from objective truth to practical utility.

Conclusions:

• A pragmatist approach, prioritizing outcome-based therapeutic usefulness, should guide the assessment of ML in medicine.
• This approach offers a more practical solution to managing algorithmic bias than striving for unattainable objective truth.
• Adopting a focus on therapeutic utility can lead to more equitable and effective ML applications in healthcare.