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
You made a lot of points here. Many I agree with, some I don't, but I specifically want to address this because it seems to be such a common misconception.
AI stores original works like a dictionary does. All the words are there, but the order and meaning is completely gone. An original work is possible to recreate by randomly selecting words from the dictionary, but it's unlikely.
The thing that makes AI useful is that it understands the patterns words are typically used in. It orders words in the right way far more often than random chance. It knows "It was the best of" has a lot of likely options for the next word, but if it selects "times" as the next word, it's far more likely to continue with, "it was the worst of times." Because that sequence of words is so ubiquitous due to references to the classic story. But over the course of following these word patterns, it will quickly glom onto a different pattern and create a wholly new work from the original "prompt."
There are only two cases in which an original work should be duplicated: either the training data is far too small and the model is overtrained on that particular work, or the work is the most derivative text imaginable lacking any flair or originality.
Adding more training data makes it less likely to recreate any original works.
I am aware of examples where it was claimed an LLM reproduced entirely code functions including original comments. That is either a case of overtraining, or far too many people were already copying that code verbatim into their own, thus making that work very over represented in the training data (same thing, but it was infringing developers who poisoned the data, not researchers using bad training data).
Bottom line: when created with enough data, no original works are stored in any way that allows faithful reproduction other than by chance so random that it's similar to rolling dice over a dictionary.
None of this means AI can do no wrong, I just don't find the copyright claim compelling.