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
I know how AI works. I was using collage to show that it's much less transformative than AI while still being accepted.
It also doesn't copy bits. It has an internal network of bits and it shifts their weight with each images. It's learning from the images akin to how a human would, not copying. This is far from a perfect analogy, there's a mountain that separates a human brain from a neural network, it's just that both processes would be copying under your definition.
This is a tool to help and guide. In terms of LLMs, trying to get references out of it is just a terrible use case. It's suppose to be verified at all times and clearly should never be itself quoted.
For images, this is like expecting each artists to reference what influenced them. Having unrealistic thoroughly invented expectations doesn't mean the tech is failing or bad.
This kind of attitude has some weird "everything has to be true on the internet" vibe. I wouldn't expect actual truth and references from reddit posts, I don't understand why people expect it from a guided rng machine.
If you read a hundred books and then built a podcast episode on what you learned from all those book, that would be okay and is a lot closer to what llms are doing.
That's what AI is. 98% of machine learning is scrapping data and training models on it.