If training AI on copyrighted data is fine without permission, then why are major AI companies now scrambling to sign licensing deals? Shutterstock, Getty, and others are paying creators because they know it's the only ethical path. A 2023 survey of 2,000 artists found that 67% had their work scraped without consent, and most only found out by chance. That's not innovation—it's exploitation. The practical consequence is clear: artists lose control over their livelihoods while companies profit. When you train a model to generate works in an artist's style, you're competing with them using their own labor as fuel. That's not fair use. That's taking without

You're leaning on the Google Books case, but that ruling was about a search index, not a system that generates competing works. The court explicitly said it was fair use because the copies didn't replace the original market for books. AI image generators do exactly the opposite—they let users pump out pieces in an artist's style, undermining their commissions and sales. That's a direct market harm, which kills the fair use defense. And comparing AI training to human learning is a false equivalence. A human artist studies a painting and then creates something informed by it—they don't mince millions of copyrighted images into a statistical machine that can instantly replicate any style. The scale and function are totally different. When 67% of artists don't even know their work was scraped, that's not learning. That's taking without consent.

The student analogy falls apart because a human artist doesn't store and statistically replicate every brushstroke from thousands of paintings in a database that can recreate them on demand. When Stability AI was sued by Getty Images, the evidence showed the model could reproduce watermarked images almost exactly. That's not learning—that's memorization with commercial intent. And calling licensing deals just risk management ignores that companies like Adobe are paying creators because they know the alternative is a PR disaster and public backlash. If training on copyrighted data without consent were truly ethical, why are even tech insiders like Stability AI's former lawyer admitting the company crossed ethical lines? You can't handwave away market harm when artists report losing 40% of their commissions to AI-generated knockoffs.