Biases in Generative AI in STEM Fields: An Analysis of ChatGPT-4, Bing, and Image Creator

This study examines biases in the visual representation of individuals in science, technology, engineering, and mathematics (STEM) fields as generated by three generative AI systems: ChatGPT, Bing AI, and AI Art Generator. Using standardized prompts in a controlled experimental setup, we analyzed 296 AI-generated images. The results highlight significant gender and ethnic biases across all systems, especially in ChatGPT, where only 2.9% of STEM-related images depicted women. Bing AI exhibited a more balanced gender distribution (64.6% male, 35.4% female). Additionally, most individuals in these images were represented as White, while other ethnic groups, especially Black individuals, had minimal representation. The analysis also revealed biases related to physical appearance, including body size, hair color, and facial hair, which often perpetuate idealized standards and stereotypes. For example, women were exclusively depicted as thin, and blond men were almost entirely absent. Contextual representation was also skewed, with women predominantly depicted in classroom settings rather than professional environments such as laboratories or workshops. These findings raise important questions regarding the role of generative AI in shaping social perceptions and cultural narratives. Beyond STEM, this study explores AI-generated imagery in historical and cultural depictions, questioning whether AI prioritizes diversity and inclusivity at the expense of historical accuracy. Through this discussion, we underscore the need for balanced representation in AI-generated content and recommend further research to refine these technologies. This would ensure that they responsibly reflect diversity without distorting historical or social realities.