Generative AI in Education: Image Synthesizers' Bias and Accuracy

• In Event: Virtual Poster Session
  In Virtual iPresentation Only: Division C - Section 3b: Technology-Based Environments Virtual Poster Session

Abstract

Image synthesizers are Artificial Intelligence (AI) programs that can generate an image from a text prompt. With the growth of generative AI, educators’ problem of finding an appropriate or desired image to support instruction may be solved by utilizing an image synthesizer to create an image. As AI image synthesizers emerge in education for instructional purposes, examining the images for accuracy and bias is warranted. The present study analyzed AI-generated education images through a visual content analysis. Evidence indicated an overrepresentation of stereotypical gender bias across three AI image synthesizers. Further, there was evidence that 7.9% of the images were inaccurate including eyes and hands that were malformed or disfigured, and 17.9% of the images indicated bias.

Authors

• Laurie O. Campbell, University of Central Florida

• Glenn W Lambie, University of Central Florida