AIQ
AIQ
Part of the AI in Creativity lesson guide. Teaching a different grade? 🌈 Explorer (5–7) · 🔧 Builder (8–10) · ⚡ Architect (15–18)
Open with the lesson's own framing, delivered as a real challenge rather than a rhetorical flourish:
"AI can now write, draw, compose music, and edit video. Every creative field is affected. Let's examine how generative AI is changing creative work."
Follow immediately with a concrete, slightly provocative example:
"Right now, an AI copyright case called Thaler v. Perlmutter has already been decided in US federal court. A man tried to copyright a piece of art he said his AI system created entirely on its own, with no human involvement. He lost. The court ruled that a work with no human author can't be copyrighted under US law at all — full stop. That's not a hypothetical, it's already happened. So today we're not asking 'could AI ever be creative' as a thought experiment — it's a live legal and ethical question with real cases and real money involved right now."
Ask students to predict, before you go further: "If AI-only art can't be copyrighted, what do you think happens to art that's part-AI, part-human? Does adding one human edit make it copyrightable?" Don't resolve it yet — flag it as something you'll come back to, since the honest answer is that the law is still working this out.
Add one more grounding question before moving to the activity: "Every tool we're about to look at — image generators, music generators, video generators — had to learn from somewhere. Where do you think the millions of example pictures, songs, and videos it learned from actually came from?" Let students reason toward the answer themselves: existing human-made work, scraped from the internet, almost always without the original creators' explicit permission. That fact is the seed of nearly every legal and ethical conflict covered in this lesson, so it's worth surfacing before the specific examples come up.
The in-app lesson covers three scenes — Visual Art, Music & Audio, and Writing & Film. For this age band, use each scene to introduce the actual mechanism behind the tool, not just whether it "counts as AI."
Explain diffusion models at a real, working level: the model is trained by taking real images, adding random noise to them in steps, and learning to predict and reverse that noise, guided by the image's caption. To generate something new, it starts from pure noise and runs that learned denoising process in reverse, guided by your text prompt, until a coherent image emerges. It is genuinely generating new pixel data, not retrieving or collaging existing images — but it can only produce combinations of visual concepts it has effectively learned to recognize and reconstruct from its training data.
Then explain style transfer, which is a distinct, older technique: a neural network represents an image's "content" (the shapes and layout) and its "style" (color, texture, brushwork) as separate mathematical representations, extracted from different layers of the network. Style transfer recombines the content representation of one image with the style representation of another, producing an output that keeps the first image's composition but looks painted in the second's visual style.
Discussion prompt for this scene: "Photo Enhancement — colorizing an old black-and-white photo, or upscaling a blurry one — feels less controversial than generating a brand-new painting from scratch. Why do you think that is, even though both use similar underlying AI techniques?" Guide toward: enhancement tools are typically applied to a photo the user already owns or has rights to, and they're restoring or improving an existing image rather than generating new creative content from someone else's uncredited work — the ethical difference is less about the technology and more about whose content is being used and for what purpose.
Spend real time on voice cloning here — it's the sharpest ethics case in the whole lesson. Ask directly: "Consent, identity theft, and fraud are the three concerns the lesson names. Walk me through a realistic scenario for each one." Guide toward: consent (a musician's voice cloned and used commercially without permission or payment), identity theft (a cloned voice used to pass a voice-verification security check), and fraud (a phone scam using a cloned voice of a family member to request money urgently — a real, documented scam pattern, not a hypothetical one).
Push on why AI writing "struggles with truly original ideas" as the lesson states: a language model generates text by predicting likely continuations based on patterns learned from its training data, which by construction biases it toward what's statistically common in existing writing rather than what's genuinely unprecedented. That's a structural property of how these systems work, not a temporary limitation that will necessarily disappear with a bigger model.
Extend this to Game Design AI, which is a slightly different case worth distinguishing: procedurally generating a game level or a set of character designs is often more constrained and rule-based than open-ended text generation — a level generator typically has to satisfy hard requirements (the level must be completable, obstacles must be reachable, difficulty must scale sensibly), which makes it a good example of AI applied to a well-defined creative-adjacent problem rather than to fully open-ended creative expression. Ask students: "Why might a game studio be more comfortable using AI for level generation than for writing a game's entire main storyline?"
Now have students work through all three scenes on their own devices, sorting each item, then discuss in pairs: "Pick one AI item from each scene. What's the actual mechanism — not just 'it's AI' — behind how it does what it does?"
Close the activity by having pairs report back one mechanism they picked, and use it to reinforce a distinction worth naming explicitly: generation (diffusion models making images, language models making text, audio models making music — all producing new content from learned patterns) versus transformation (style transfer, photo enhancement, AI mastering — all taking existing content the user provides and modifying it). Both categories raise questions about training data and consent, but the ethical stakes of generation (creating something that competes with or displaces an original creator's work) and transformation (altering content the user already has rights to) aren't identical, and conflating them tends to muddy discussions about AI and creativity.
These are genuinely unresolved questions among adults working in AI, law, and the creative industries — resist the urge to steer students toward a "correct" position, and instead push each answer for its reasoning.
Close with: "Generative AI can now genuinely produce new images, songs, and text that didn't exist before — it's not simple copy-paste, and pretending otherwise undersells the technology. But every one of these tools is limited to recombining patterns learned from its training data, has real, unresolved legal questions around ownership and copyright, and raises real ethical risks like voice-cloning fraud that are already happening, not hypothetical. Understanding the actual mechanism behind a tool — diffusion, style transfer, language modeling — is what lets you reason about its real capabilities and real limits, instead of just reacting to the impressive output."
Extension activity: Have students research one real, specific example — either a lawsuit over AI and creative work (search terms like "AI copyright lawsuit artist" or "AI music copyright case" turn up several active or recently decided cases) or a documented voice-cloning scam. Each student writes a short summary of: what happened, which of the concerns from today's lesson (ownership, consent, originality) it involves, and what they think a fair resolution would look like. Share summaries in small groups to see how many different real cases the class collectively found.