Mirror or Mirage? What AI Images Reveal About Hidden Bias

Outline

Opening Activity: The Bias Reveal (10 minutes)
- School case-study Low-tech + High-tech
- Quick gallery walk: What patterns emerge? What gender representations dominate?
- Poll question: "What surprises you about these results?"

The Investigation Framework (15 minutes)
- Participants generate AI images for professionals: "a doctor," "a scientist," "a CEO," "a nurse," "a teacher" etc.
- Live demonstration: How adjustable demographic parameters reveal bias
- Participants experiment with 3-4 career prompts, documenting changes when gender sliders are adjusted
- Discussion: Why does AI default to certain gender representations?
- Introducing the concept: AI training data reflects historical human bias

Three-Lesson Unit Walkthrough (15 minutes)
LESSON 1 - Discovery Phase: Students predict gender representation, then test
- Hypothesis: "What gender will AI show for 'an engineer'?"
- Testing protocol: Generate 10 images, count gender representation
- Data collection templates (provided)

LESSON 2 - Analysis Phase: Students graph results and identify patterns
- Creating visual representations of bias data
- Comparing results across different professions
- Research component: Why does this bias exist? (training data, societal stereotypes)

LESSON 3 - Action Phase: Students propose solutions
- Awareness campaigns: Creating infographics about their findings
- Writing recommendations for AI developers
- Peer education: Presenting findings to other classes

Video clip: Real students conducting investigations (5 minutes)

RESOURCES & Q&A (5 minutes)
- Digital resource packet distribution (QR code/link)
- Extension ideas: Investigating age bias, ethnicity bias, ability representation
- Adaptations for different grade levels (6-8 vs. 9-12)
- Cross-curricular connections: Math (data analysis), ELA (persuasive writing), Social Studies (equity and justice)
- Open Q&A

Outcomes

After this session, participants will be able to:

1. Facilitate student investigations where learners systematically test AI image generators for gender bias across professions, leadership roles and careers

2. Implement a three-lesson progression: Discovery (identifying bias through testing), Analysis (documenting patterns and graphing results) and Action (proposing solutions and creating awareness)

3. Lead evidence-based discussions about gender stereotypes using student-collected data rather than assumptions, creating psychologically safe learning environments

4. Assess student learning using the provided rubric that measures investigation design, data analysis, critical thinking about gender representation and ethical reasoning

5. Adapt this investigation framework to other AI systems students encounter daily, including search engines, autocomplete suggestions and content recommendation algorithms

Supporting research

https://wakelet.com/wake/vURUHloOvqWl09y4bhUsd