Abstract
Generative AI provides synthetic simulation of existing societal data. We hypothesized that Generative AI output may be used to evaluate diversity and stereotypes amongst healthcare providers. Dall-E 3, a text-to-image generator, was used to generate a total of 360 images based on pre-defined healthcare provider terms. Consensus scoring was performed to evaluate diversity parameters in images. Google Vision was used to generate image labels that were then categorized to analyze differences among race and sex cohorts. Sex and race diversity for various doctor and nurse terms was modest: 3.2 and 2.8, respectively, on a qualitative 5 point scale (where 5 represents equal diversity). These results are consistent with recently reported statistics, demonstrating that Generative AI reflects real-world data. We also identified stereotypes related to appearance, facial expressions, and clothing associated with sex and race. Our study, which is the first of its kind, provides a unique framework incorporating Generative AI and ML tools to quantify diversity and societal perceptions of healthcare providers. The proposed framework provides real-time intelligence on biases in the healthcare workforce.