Explainable and reproducible AI: culturally responsive AI for health equity in minoritized groups

Area of Science:

• Healthcare technology
• Artificial intelligence
• Health equity

Background:

• Artificial intelligence (AI) offers transformative potential in healthcare, enhancing diagnostics, personalizing treatments, and improving operational efficiency.
• AI excels at identifying complex data patterns, potentially aiding clinical decision-making for faster diagnoses and tailored treatments.
• However, AI model consistency, reliability, and validation across diverse populations and settings are crucial for equitable healthcare delivery.

Purpose of the Study:

• To explore key considerations for ensuring AI promotes health equity in marginalized communities.
• To address the challenge of AI model interpretability, often described as "black boxes" in complex applications.
• To build on existing work in humanitarian and climate AI contexts regarding anticipatory health action.

Main Methods:

• This commentary explores the critical need for explainability and reproducibility in clinical AI models.
• It emphasizes embedding these features within culturally responsive frameworks to mitigate bias.
• The discussion draws parallels with anticipatory health action strategies in humanitarian and climate AI.

Main Results:

• Explainable AI (XAI) provides crucial insights into AI model predictions, enhancing trust and understanding for medical professionals.
• Physicians generally prefer XAI over non-explainable AI models due to increased transparency.
• Addressing historical and structural bias is essential for equitable AI implementation.

Conclusions:

• Achieving health equity through AI necessitates embedding explainability and reproducibility.
• Culturally responsive frameworks are vital to counteract bias in AI algorithms and their application.
• XAI is a critical component for trustworthy and equitable AI in diverse healthcare settings.