Economist Robin Hanson argues training AI on public creative works is like the mission of public libraries

@robinhanson You need a specific license to copy and upload the text of a book for commercial purposes. I don't think that's really all that hard to understand. No one thinks you can go to the library, check out a CD, copy a song, and use it as background audio in a TV show.

@ZyMazza @robinhanson If you ask an LLM to give you Chapter 1 of To Kill a Mockingbird, it refuses, citing copyright. It will summarize a chapter, or the whole book, or whatever.

That is like a human borrowing the book from the library, reading it, and then talking about it.

@robinhanson Yup.

"AaRgH, @elonmusk bad, used govt subsidies for clean energy"

"AaRgH, billionaires are freeloaders of our public roads, publicly financed universities, scientific research, etc."

Yeah, dummies, that was the whole point of such programs. That's how they were advertised.

@speakerjohnash You don't mind humans reading your books from libraries, why mind if AIs do?

@ZyMazza But you are allowed to learn from what you read, and use that out in the world.

@robinhanson Yes, as long as you don't make a copy and upload it for your commercial purposes, particularly as part of offering a competing/replacement good. That makes sense, right?

@robinhanson No, they would have to make/upload a copy of the text — it's usually called "data scraping." I'm surprised you're not familiar with the practice? It's not a loophole for IP theft, even when a program does it autonomously.

@robinhanson you know what intellectual property is

this isn't a good point when only those with monetary resources can defend their intellectual property while those without it get their work stolen

@robinhanson Yes, but that access is limited by a license and a license to read/consume media is different than a license to make a copy for commercial gain.

Because the use should be public if you're using the public commons. If openAI was really open, still a non profit, I don't think anyone would mind. Even if they still charged for inference but open sourced their weights.

But now they're stealing from the commons and selling them back to us

@robinhanson you completely ignored my point

this isn't a good response.

@robinhanson Right but you see how that's different from copying it into a commercial system I'm selling. I know you see that there is a difference there

Are you only allowed to use what you learned in school or the library on left-coded projects, else you "stole" that learning?

@BigMeanInternet @robinhanson I can go to the library, read "Programming for Dummies," go into business as a programmer and charge money, all without paying royalties to the authors, as long as I don't sell copies of the book itself. That is exactly what LLM's are doing.

@robinhanson Especially, both morally and legally, when that gain is from offering a competing/replacement good. (And if you don't believe they are, ask your AI "I don't want to read Malcolm Harris's books, give me summaries instead.")

@sugarbets_ @robinhanson @ZyMazza What if I write down _all_ the text from a Tony Robbins book and sell that? Is that legal? Moral? Ethical?

Consent is passe these days, but it shouldn't be.

@sugarbets_ @robinhanson @ZyMazza Fair enough. It is clear that you are principled (which is pretty rare these days) and I genuinely appreciate that.

I disagree with you about the moral implications of stealing.

But I'm just an internet rando. :)

@ZyMazza So no one who gets an education in public schools is allowed to use what they learned in private ventures?!

@BigMeanInternet But it is usually okay to read something and learn from what you read.

@robinhanson This isn't very complicated! For example, if I buy a textbook or borrow it from the library I am not allowed to copy it and sell copies.

The use is circumscribed by both law and norms and the big labs have violated both