Those claiming AI training on copyrighted works is "theft" misunderstand key aspects of copyright law and AI technology. Copyright protects specific expressions of ideas, not the ideas themselves. When AI systems ingest copyrighted works, they're extracting general patterns and concepts - the "Bob Dylan-ness" or "Hemingway-ness" - not copying specific text or images.
This process is akin to how humans learn by reading widely and absorbing styles and techniques, rather than memorizing and reproducing exact passages. The AI discards the original text, keeping only abstract representations in "vector space". When generating new content, the AI isn't recreating copyrighted works, but producing new expressions inspired by the concepts it's learned.
This is fundamentally different from copying a book or song. It's more like the long-standing artistic tradition of being influenced by others' work. The law has always recognized that ideas themselves can't be owned - only particular expressions of them.
Moreover, there's precedent for this kind of use being considered "transformative" and thus fair use. The Google Books project, which scanned millions of books to create a searchable index, was ruled legal despite protests from authors and publishers. AI training is arguably even more transformative.
While it's understandable that creators feel uneasy about this new technology, labeling it "theft" is both legally and technically inaccurate. We may need new ways to support and compensate creators in the AI age, but that doesn't make the current use of copyrighted works for AI training illegal or unethical.
For those interested, this argument is nicely laid out by Damien Riehl in FLOSS Weekly episode 744. https://twit.tv/shows/floss-weekly/episodes/744
That factor is relative to what is reproduced, not to what is ingested. A company is allowed to scrape the web all they want as long as they don't republish it.
I would argue that LLMs devalue the author's potential for future work, not the original work they were trained on.
Again, that's the practice of OpenAI, but not inherent to LLMs.
It's honestly absurd to try and argue that they're not transformative.
The work is reproduced in full when it's downloaded to the server used to train the AI model, and the entirety of the reproduced work is used for training. Thus, they are using the entirety of the work.
And that makes it better somehow? Aereo got sued out of existence because their model threatened the retransmission fees that broadcast TV stations were being paid by cable TV subscribers. There wasn't any devaluation of broadcasters' previous performances, the entire harm they presented was in terms of lost revenue in the future. But hey, thanks for agreeing with me?
And again, LLM training so egregiously fails two out of the four factors for judging a fair use claim that it would fail the test entirely. The only difference is that OpenAI is failing it worse than other LLMs.
It's even more absurd to claim something that is transformative automatically qualifies for fair use.
That's objectively false. It's downloaded to the server, but it should never be redistributed to anyone else in full. As a developer for instance, it's illegal for me to copy code I find in a medium article and use it in our software. I'm perfectly allowed to read that Medium article, learn from it, and then right my own similar code.
And Aero should not have lost that suit. That's an example of the US court system abjectly failing.
That's what we're debating, not a given.
Fair point, but it is objectively transformative.