Beyond the stereotypes: Artificial Intelligence image generation and diversity in anesthesiology

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Medical Education

Background:

• Artificial Intelligence (AI) is increasingly used in anesthesiology for patient safety and efficiency.
• Understanding AI's impact on professional representation is crucial.

Purpose of the Study:

• To evaluate AI-generated images for demographic accuracy in anesthesiology.
• To identify social biases in AI depictions of the anesthesiology workforce.

Main Methods:

• Cross-sectional analysis of AI-generated images (January-February 2024).
• Demographic data from American Society of Anesthesiologists (ASA) and European Society of Anesthesiology and Intensive Care (ESAIC).
• Assessment of 1,200 images from ChatGPT DALL-E 2 and Midjourney by independent reviewers for sex, race/ethnicity, age, and emotional traits.

Main Results:

• AI models significantly underrepresented demographic diversity compared to ASA/ESAIC data.
• Predominant portrayal of anesthesiologists as White (64.2%-83.0%).
• Biased associations between gender and race/ethnicity, and underrepresentation of younger professionals.

Conclusions:

• AI models exhibit notable biases in gender, race/ethnicity, and age representation.
• Current AI images do not reflect the actual diversity of the anesthesiologist workforce.
• Need for diverse training datasets and bias mitigation strategies in AI for accurate medical representation.