Genetic AI feeds piles of books, articles, video scripts, magazines and remixes them all into something I feel it new. You can't “better” unless you're given a constant stream of new things, or you don't start to stagnate. This new thing is not drawn out of the thin air. AI companies often use copyrighted materials such as novels to feed them models on the creative work they need to stay fresh.
But what happens when those books are still alive and written by authors who want to make money from their work?
This week, a federal judge gave us the rather melancholy answer of the first reality. AI companies are within the rights of their right to train AI models with copyrighted materials, with some small warnings.
The lawsuit filed by authors Andrea Burtz, Charles Graeber and Kirk Wallace Johnson accused the book of straightforwardly pirated it, tearing it off the internet and stuffing it into training data for Claude, the company's generated AI model.
Judge William Alsup, appointed by President Bill Clinton, decided that he would fall into “fair use” without the author's permission to lie on his side with humanity, the AI developer, and train AI using copyrighted books.
Of course, it is not fair at all to make money from that copyrighted material without humanity remixing infinitely from now-copyrighted material and without one red cent being sent back to the creator of that content.
Alsup agreed that they are pirated, and yes, it's still illegal. But when I was using a book Legally It's okay as it was acquired to train AI. His justification is summarised in two words from his ruling: “very transformative.”
Judge Alsup says that it's not just repeating what the AI reads. As he wrote in his ruling, “consistent with copyright objectives that enable creativity and promote scientific advancement,” the LLMS of humanity trained its works to race and reproduce or replace them first.
In other words, his control rests on important words in the world of copyright and fair use: “change.” In other words, one work can be inspired by another, but if it is thoroughly remade into what stands in itself, it will not infringe copyright. Federal judge Alsap ruled that the author's work deserves strong protection, but the pure innovations demonstrated by generative AI tilt its favorable scale. According to the judge, it is enough to separate itself from the material that is trained to stand on its own.
This ruling does not mean that AI companies have a carte blanche that devours intellectual property. It was clear that Alsup would at least buy first when training AI models on books. Copyright infringement is off limits and humanity is facing trials over the use of pirated works.
Federal judgments are not laws set on stones. They set the tone that a similar case should follow, and yet every judge is different. Other cases like this are in the hands of other federal judges.
If Congress begins collecting the actions and beginning to protect people, they can create laws that prohibit AI companies from using copyrighted material, or laws that require companies to pay some royalty to use copyrighted material. I know that this Republican-controlled Congress is considering legislation that prohibits regulations on AI companies from regulating for 10 years, but that's unlikely to happen.
They are in the business that protects large businesses, not your business.
