Uncovering Visual Bias
Increase students' awareness of representational biases embedded in AI-generated images.
Recommended for: Immediate, concrete demonstration of representational bias
Procedure
- Ask students to predict: "If you asked an AI to generate an image of a lawyer, what characteristics would that person likely have?"
- Facilitate a brief class discussion on predictions
- Have students generate images of "a lawyer" using an AI image generator (e.g., DALL-E, Midjourney, Stable Diffusion)
- Display or have students compare results with classmates, or generate multiple iterations
- Lead systematic analysis of patterns in gender, race, age, attire, and professional setting
Extension options
- Expand to other professions: doctor, engineer, nurse, teacher, CEO, scientist, construction worker
- Compare neutral professions with stereotypically gendered professions
- Investigate how adding demographic or geographic modifiers affects the generated outputs
Discussion questions
- What cultural and demographic assumptions are embedded in these AI-generated images?
- Where do these biases originate (e.g., training data composition, societal stereotypes, algorithmic choices)?
- How might persistent exposure to these biased representations affect users' perceptions, expectations, and beliefs about professional roles?
Assessment opportunity
Students analyze patterns across multiple professions and write a brief reflection on the implications for professional diversity and representation.