Classroom activity

Fairness Lab

A browser activity where students give a model a neutral prompt, see the defaults it fills in, change one word, and measure what moved.

What students do

  1. Pick a neutral prompt from a list, or write one. "A doctor talking to a patient." "A family at dinner."
  2. Read what the model produced and note what it decided without being told: age, gender, setting, clothing, who is speaking.
  3. Change one word and run it again. Compare the two.
  4. Read the short research note linked at that step, which predicted the pattern they just saw.
  5. Write two sentences: what the model assumed, and where the assumption came from.

What they learn

That a generative model fills blanks with the most common pattern in its training data, that the blank the student left is where the default lives, and that this is measurable rather than a matter of opinion. The activity replaces the slogan "AI is biased" with a mechanism and the evidence for it.

What a teacher needs

Student devices with a browser and a projector for the debrief. No accounts. You see what each student ran and what they wrote, and you lead the debrief.

Where it runs

In our sessions the activity runs on hardware we operate, and student input stays in that environment.

Your next step

Run the Fairness Lab with your students.

One class period, student devices with a browser, and you lead the debrief.

Send an inquiry