AI Health Biases Surface

AI-driven medical tools exhibit biases that can lead to inadequate healthcare recommendations for women and ethnic minorities. These large language models (LLMs) sometimes minimise or misinterpret symptoms in female, Black, and Asian patients.

These biases stem from underrepresentation and misrepresentation in the datasets used to train the AI. This can result in misdiagnoses, particularly for gender and ethnic minorities, and perpetuate existing inequalities in healthcare. AI models may rely on demographic shortcuts, leading to incorrect results for specific groups.

To mitigate these biases, open science practices, inclusive data standards, and participant-centred development of AI algorithms are essential. Thorough evaluation of AI models across diverse demographic groups is necessary to ensure fair and accurate diagnoses for all patients.