Artificial Intelligence-Generated Synthetic Data in Healthcare: A False Promise for Underserved Populations?

Area of Science:

• Medical AI Ethics
• Health Equity
• Data Science

Background:

• AI-generated synthetic data is proposed to solve underrepresentation of underserved groups in medical AI.
• Proponents claim it can fill data gaps, enhance fairness, and reduce bias without privacy risks or costly collection.
• This view is challenged by potential amplification of existing biases and consent complexities.

Purpose of the Study:

• To critically evaluate the ethical implications of AI-generated synthetic data in medical AI.
• To examine how synthetic data generation may perpetuate or amplify biases.
• To analyze the challenges synthetic data poses to informed consent for vulnerable populations.

Main Methods:

• Ethical analysis of AI-generated synthetic data generation processes.
• Examination of opacity in synthetic data to identify bias reproduction.
• Assessment of informed consent complexities introduced by synthetic data.

Main Results:

• AI-generated synthetic data can reproduce and amplify discriminatory patterns from source data due to inherent opacity.
• Synthetic data complicates informed consent, potentially harming underserved populations.
• The technology fails to address, and may exacerbate, data disparities.

Conclusions:

• AI-generated synthetic data presents significant ethical challenges, particularly for underserved populations.
• Synthetic data does not solve, and may worsen, issues of bias and underrepresentation in medical AI.
• Focus should shift to dismantling structural barriers for authentic inclusion in medical research.