AIQ
AIQ
Everything a teacher needs to deliver this lesson — pick your grade's script below once you've read the background.
This lesson runs across all four AIQ age modes, so the depth changes a lot from Explorer to Architect, but every version is building toward the same core idea: AI can now produce art, music, writing, and video that looks and sounds genuinely impressive, and it does that by learning patterns from enormous amounts of existing human-made work and recombining them — not by having ideas, feelings, or lived experience the way a person does. By the end of the lesson, a student should be able to:
You don't need an art or computer science background to teach this lesson well — you need one mental model, and the lesson gives it to you directly: every one of these tools was trained by showing a computer program millions of examples of existing human-made images, songs, or text, and having it learn the statistical patterns that connect a description ("a cat riding a dragon in watercolor style") to the kind of output that fits it. Once trained, the tool doesn't retrieve a matching picture from a library — it generates a brand-new arrangement of pixels, notes, or words that fits the patterns it learned. That's the difference between "copying" and what these systems actually do, and it's the single most useful thing to get across at every age band.
For image generation specifically — the technology behind tools like Midjourney and DALL-E — the general approach used by most modern systems is called a diffusion model. The plain-language version: training teaches the model what "noise being removed from a picture" looks like, guided by a caption, across millions of examples. To generate a new image, the system starts from pure random static and, step by step, removes noise in a way that's guided by your text description, until a coherent image emerges. It's less like a artist starting with a blank canvas and more like slowly bringing an image into focus out of TV static, with the text prompt steering which image comes into focus. A related but distinct technique, style transfer, takes one image's content (the shapes and layout) and repaints it using another image's style (brushwork, color palette) by separating the two mathematically and recombining them — this is what lets an app repaint your photo "in the style of Van Gogh."
Text and video generation work on a related idea but with different training data. Large language models learn patterns across enormous amounts of written text and generate new text one likely next word at a time, which is why they can write in almost any style or genre convincingly but — as the lesson itself notes — tend to struggle with truly original ideas, since by construction they're producing something statistically similar to what's come before rather than something unprecedented. Video tools like Sora extend a related idea to sequences of video frames, generating footage of scenes that never existed from a text description alone.
Music and voice tools raise the most immediate ethical questions, and it's worth being ready for them. Tools like Suno and Udio can generate a complete song — melody, instrumentation, and sung vocals — from a short text description, by having learned patterns across huge libraries of existing music. Voice cloning is a related but separate technology: given as little as a few seconds of clean audio of someone's voice, a model can generate new speech in that same voice saying things the person never actually said. The lesson names the real concern directly — consent, identity theft, and fraud — because unlike a stylized painting, a cloned voice can be used to impersonate a specific, real, identifiable person without their knowledge.
Finally, there's a genuine, ongoing debate worth presenting honestly rather than resolving for students: is an AI system that recombines learned patterns from millions of human works actually being "creative," or is it an extremely sophisticated remix machine? Reasonable people disagree, and part of the disagreement is that human creativity is also shaped by everything an artist has seen, heard, and been influenced by — so the line between "influenced by" and "trained on" isn't as clean as it first seems. What is settled, as of this lesson's writing, is the legal question in the United States: in Thaler v. Perlmutter, a federal appeals court held that a work generated entirely by AI, with no human author, cannot receive US copyright protection. Works that combine meaningful human creative choices with AI-generated elements sit in a much greyer, still-developing area of the law — worth naming for Hacker and Architect students, without overstating how settled it is.
This is a light-prep lesson. You need: