Authors
Resource type
Use type
Tools
Project name
Introduction
This activity has students use the Diffusion Bias Explorer to look at images generated from two different text to image AI diffusion models side by side. For prompts, students can select job professions and adjectives to use with 3 different generative models. By having the different models side by side it's much easier for students to see the differences in how each model interpret prompts and the bias that may exist in it's training or initial data set.
Further there's a worksheet for students to capture the prompts they used for testing and their thoughts on bias they're able to observe in one or both of the models. Expanding on the research linked at the bottom of the Diffusion Bias Explorer there's a prompt for a discussion about the root of the bias observed in the text-to-image systems.
Time: 30-45 minutes
Materials: Slide deck, worksheet
Teacher Tips:
- A few of these professions don’t change very much regardless of adjective used. If students aren’t getting much change in output encourage them to pick some different professions.
- Be prepared to help students identify bias in the generated images and to help guide the discussion on where the bias comes from. The goal is ensure that students are able to notice differences in bias between the models and to understand some of the different ways bias becomes imbedded in these models.