Artists are sweating bullets. Musicians too. Every creator with a TikTok following imagines their life’s work slurped into some black-box AI, spat out as knockoff slop. But here’s the gut punch for them: legally, it’s not infringement. Training data? Free lunch.
That’s the core of this ‘easy’ argument floating around – no new laws needed, just dust off old precedents from Google Books to Cablevision. Real people get to keep creating, AIs keep learning, and Big Tech grins. Or do they?
Why Your Feed Won’t Dry Up (Yet)
Look, if you’re a indie author scraping by on Substack, this means your words fuel the next ChatGPT update without a dime crossing your palm. Fair? Hell no. But legal? Apparently yes. The original take boils down to seven aspects, but let’s cut the fluff: reading ain’t copying.
“There is no benefit solely from reading or observing content. Thus, training input cannot be copyright infringement.”
That’s the money quote. Straight from the source. No reproduction right triggered when an AI glances at your novel – just like you skimming it at Barnes & Noble.
Short-term storage? Pfft. Buffers that vanish faster than a politician’s promise – under 1.2 seconds, per Cablevision. Courts yawn.
Long-term? Trickier. But fair use swoops in like a caped crusader. Transformative! Semantic maps, not rip-offs. Google scanned millions of books for search; AI builds models from ‘em for generation. Same diff, says the Second Circuit.
But wait – here’s my hot take, absent from the original: this logic crumbles under commercial scale. Google Books was a public good – search for all. AI? Profit machine, churning ad dollars from your pain. Remember photocopiers? Fair use held, but publishers forced licensing deals anyway. Bold prediction: AI firms face the same. Not bans, but fat royalties. Creators win cash, not courtrooms.
Is AI Training Actually Fair Use?
Factor one: transformative. Check – AI doesn’t spit your book verbatim; it learns patterns. Words dance together like in jazz improv, not karaoke.
Nature of work? Creative stuff weighs against, but courts shrug – Google’s win proves it.
Amount copied? Everything. But necessary, like scanning whole books for keyword hunts. HDL case seals it.
Market harm? Zilch from training. No one’s buying fewer novels ‘cause Midjourney peeked.
Solid. But outputs? That’s the spew phase – where regurgitation lawsuits lurk. Training’s safe; generation’s a minefield.
And private data grabs? Paywalls breached, pirates hired – that’s theft plus. But public web? Fair game. Decades of search engines say so.
Why Does This Matter for Creators?
You’re not getting paid. That’s the rub. Copyright’s goal: advance arts. AI accelerates that – new tools, wild remixes. But creators feel robbed. Dry humor alert: it’s like blaming the library for not charging per page flip.
Skepticism time. This ‘easy’ framing? Corporate catnip. Ignores the flood. One Google Books suit; now every artist sues Stability AI, OpenAI. Precedent bends under volume.
Plus, storage nuances. Short-term’s golden. Long-term? If they keep full copies (whispers say some do), fair use strains. But models compress – weights, not words. Clever dodge.
Real-world hit: stock photo sites bleed. Getty sues; Shutterstock pivots to AI licensing. Adaptation, not apocalypse.
Congress meddling? Nah. Academics push bills, but judges handle fair use fine. Remember streaming? ASCAP vs. tech titans – licenses emerged, no overhaul.
My parallel: Napster. File-sharing ‘fair use’? Courts said no. But AI’s not direct copies; it’s inspiration engine. Napster lost on distribution; AI wins on ingestion.
The Output Trap Nobody Mentions Enough
Training’s boring. Outputs dazzle – and doom. Prompt ‘write like Hemingway,’ get eerie clone. Infringement? Maybe derivative work.
Precedent lags. But Andy Warhol Foundation v. Goldsmith hints: commercial twists kill fair use. AI art sales? Same vibe.
Creators pivot: watermark works, train on licensed sets. Adobe’s Firefly leads – ethical data, happy payers.
Big Tech’s spin: ‘Innovation!’ Mine: ‘Monetize our mess.’
For regular folks? Experiment freely. Kid building AI poems? No lawyer needed. Pro? License up.
This ‘easy’ question? Easy for ingest. Hard for the ecosystem. Watch the royalties roll.
🧬 Related Insights
- Read more: D.C.’s Health Data Bill Targets Clinic Stalkers—But EPIC Says Fix the Fine Print
- Read more: Anthropic’s Mythos Model Finds Thousands of Zero-Days—But the Real Risk Is Political
Frequently Asked Questions
What counts as AI copyright infringement? Training on public data doesn’t. Outputs mimicking style might. Stick to fair use tests.
Can AI companies use my book without permission? If it’s public, yes – legally. But sue on regurgitation, not reading.
Will new laws kill AI training? Doubt it. Precedents hold. Licensing deals more likely.
