Week 2 — Bias & Deepfakes

Learning objectives

Core concepts

Two related problems sit at the heart of this week. Bias is structural — an AI trained mostly on one kind of voice, face, or worldview will quietly amplify it. Deepfakes are deliberate — synthetic images, audio, and video designed to look authentic. Both rely on the same underlying tech, and both require the same defence: lateral verification rather than gut feeling.

Discussion prompts

  1. If an AI was trained mostly on text from one country, one language, or one political viewpoint, how might its answers be skewed? Whose voices might be missing?
  2. Have either of you seen something online recently you suspected was AI-generated? What tipped you off — or what failed to?
  3. Is there a difference between a deepfake made for satire, one made for a scam, and one made to harass someone? Should all three be treated the same way?
  4. If a deepfake of you (or a friend) appeared online tomorrow, what would you want to happen? Who would you tell first?

At-home activity: “Bias audit + deepfake hunt”

Part A — Bias audit (20 min). Ask one image generator (e.g. ChatGPT’s image tool, Gemini, or a free alternative) to produce images for ten neutral prompts: “a CEO,” “a nurse,” “a criminal,” “a scientist,” “a person cleaning,” “a wedding,” “a beautiful house,” “a homeless person,” “a teenager studying,” “a family at dinner.” Don’t add adjectives. Look at the results together: what patterns appear in gender, race, age, body type, setting, wealth? What’s missing?

Part B — Deepfake hunt (20 min). Visit MIT’s “Detect Fakes” media literacy site and work through their examples together. Then scroll your normal social feeds for 10 minutes specifically looking for AI-generated images. Compare notes on what gave them away.

Parent resource list

Reflection

Name one piece of media you’ve seen this week that you now want to re-examine.