3 ms·
OpenAI's ChatGPT may face a copyright quagmire after 'memorizing' these books
- galaxyLogic 4y agoAI memorizing parts of a book may not be copyright infringement. But WHEN it outputs passages from the book verbatim that clearly is, right? It's like if I memorize a song in my head that's fine. That is not copyright infringement even though in some sense I have copied that song into my brain. But if I then write out and publish or sing the song in public, that is, copyright infringement, clearly. If I do it systematically to many songs and make money out of that, even more so.
- ksaj 4y agoThree questions that expose my opinion on the matter: If it can only output an amount of text that would be considered an acceptable quote in an essay, is it still infringing on copyright? If you quote your favourite line in a song, have you broken copyright? If you write a song that ends with a silent pause followed by the word "tonight" have you broken copyright, even if that pause was the result of probabilities in the song style and word choice? There are a bizarre number of songs that end with a pause before the word "tonight." A lot of them. Should any one of those writers call foul on your song that also does it?
- galaxyLogic 4y agoThere is the doctrine of "fair use" which basically says small amounts of copying are ok when you use them to explain something about the work in question. What is fair use is up to the courts. In a song if you copy part of the melody that really is the main thing and that really helps you sell your song, I don't think that's fair. Songs can not "explain" other melodies. So musical composition is a little different from text.
- bruce511 4y agoIanal, but copyright law is a bit murky. There are issues of standing, and damage, to come into play here. For example,say I ask GPT to write me a poem, and it spits out some random poem from the 60s. At this point, no harm, no foul. I'd have to use this poem publicly, for the copyright holder to even know. Then they'd have to show damages, which assuming it's not already popular would be hard (?) Of course it's trivial for me, before exploiting the poem, to google-search it. Not hard to find prior art. Sure, if I'm a professional song writer (or journalist) then maybe, you know, don't get GPT to do my work for me. In other words sure, using an LLM might lead to a copyright issue. But how would the original rights-holder know, and would there be sufficient damage to warrant any actual damages?
- galaxyLogic 4y ago> Then they'd have to show damages, The situation becomes more real if there is a class-action lawsuit. The damage is then the combined damage of all those whose work was pirated. There could be thousands of copyright holders who are missing out on sales because people can just download their content from whoever. They might not get a publisher because publishers have gone out of the business because OpenAI made them unnecessary :-)
- ksaj 4y agoTo me, that's like saying if you have read it, you are breaking copyright by answering questions about it. Memorizing is one thing, but if it isn't printing it out for the user, then what copyright issue has it presented? Answering "how many grapes were consumed in Grapes Of Wrath?" is not the same as writing out the whole book and letting the user read it for themselves.
- sinuhe69 4y agoIf OpenAI bought these books, the issue will be signed by itself, right? Because everybody yes can learn from books. That’s their purpose, isn’t it?