Abstract
Purpose :
In the past few years, there has been remarkable progress in accessibility of open-source artificial intelligence (AI) image generators developed to help humans understand how AI sees our world. Here we characterize perception of racial and sex diversity in ophthalmology by AI.
Methods :
OpenAI’s open-source DALL-E 2 AI was queried for ophthalmologists possessing various features which all included “american” and “portrait photo.” The first 40 faces for each search term were categorized on the basis of race (White, Black, Asian, Hispanic, Other) and sex (Male, Female) by two independent reviewers. If race or sex was not agreed upon, a third reviewer independently provided a classification or if still indeterminate the image was labeled as such. Duplicate images and images that did not adequately show facial features were excluded from categorization.
Results :
1560 DALL-E 2 images were included in analysis. Control search queries specifying ophthalmologist sex and gender outputted accurate images validating the tool. The query “american ophthalmologist, portrait photo” portrayed the majority of ophthalmologists as White (75%) and male (77.5%). Vision scientists were depicted as 65% White and 77.5% male. Glaucoma specialists were portrayed with the highest levels of non-White racial diversity (47.5%) and oculoplastics had the most female representation (55%). Young/inexperienced/amateur ophthalmologists were perceived to have more non-White racial diversity (27.5%) and female representation (28.3%) relative to old/experienced/mature ophthalmologists (23.3% non-White and 18.3% female). Ophthalmology department chairs (25%) had slightly more racial diversity compared to residents (22.5%), but residents had greater female representation (30%) compared to chairs (15%). The United States South was associated with the highest percentage of both non-White (47.5%) and female (25%) ophthalmologists.
Conclusions :
Our results suggest the DALL-E 2 AI may perceive a trend of increasing racial and sex diversity in younger, newer ophthalmologists compared to more senior ophthalmologists. Future investigations should attempt to validate how AI may be used as a tool to evaluate ophthalmology’s progress towards becoming more inclusive of increasingly diverse ophthalmologists.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.