Prompting Bias: Assessing representation and accuracy in AI-generated images

This research article presents the results of an experiment with AI-generated images, using methods drawn from both UX testing and visual analysis, in which we compare images generated from two different applications (DALL-E and FireFly), using various prompting strategies and analyze them individually and collectively for biased representations and factual inaccuracies; we discuss our findings in terms of a hypothetical scenario in which we are using generative AI applications to produce an image for a website homepage. Based on the extensive instances of both biased representations and physical inaccuracies, we found the generated images unsuitable for the purpose we described. Finally, we discuss these findings in terms of the larger questions of AI use and whether technical and professional communication can be automated away.