Artists have complained about their artwork being stolen, people are arguing about threads.net stealing their data on despite this being a public forum, Reddit, Twitter, Github and other platforms are putting up walls to to stop AI bots from scraping everything.
However generative AI and large language models have been been spitting out their training data including copyright notices and other stuff verbatim. “poem poem poem to get personal data from ChatGPT”.
So, instead of providing all our comments for free to LLMs, how about adding a copyright notice to everything we write?
I propose the Attribution-NonCommercial-ShareAlike 4.0 International license? Basically, if somebody uses your comment, they have to attribute you, but they may not use it for commercial purposes.
This license requires that reusers give credit to the creator. It allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, for noncommercial purposes only. If others modify or adapt the material, they must license the modified material under identical terms.
All you’d need to do is add this text CC BY-NC-SA 4.0 Deed anywhere in your comment or post.
When I post something here, I usually have the hope that someone else benefits from it in some small, tiny way. Either because it makes them laugh or because it gives them valuable information.
I don’t understand why anyone would post if they can’t bear the idea that someone else might get a tiny, little something for free.
It really bothers me how normalized the “screw everyone else, I want to get paid!” Attitude has become. OP is just assuming that everyone agrees with it. I want AI to be well trained. If self-interest must be assumed then consider how useful AI is for you.
And copyright doesn’t even apply here in any event. Training an AI is not copying, it is learning.
I think a lot of the concern here, for me if noone else, is them taking the data and then turning it around into a closed for-sale product. If AI is going to be trained, it should be trained well, but if the result of doing so is them turning around and charging [me/us/everyone, as applicable] an ass load for the privilege of its use then I want no part of it.
AI trained on public data should be public. So if adding boilerplate is the solution to this problem, let it be infectious licensing which forces opening of the resultant model to the public.
Even private for-pay AI is useful to me. Even ones I don’t pay for myself, since other AI developers have been making heavy use of existing AI models to generate data for training their own new models.
In any event, as I said, copyright doesn’t even apply here. Adding a CC license does nothing, it’s not “infectious” to AI models trained off of it.
Yeah, CC doesn’t cover it in any case. Any attempt would probably need some sort of bespoke license to specifically target the training use case while still allowing comments to be used like normal.
And a Microsoft-sized pile of money to fight it out in court.
If copyright doesn’t apply to AI training, then no license of any sort will “work” because the trainers could simply refuse it.
Copyleft works because if you refuse the license you’re left with no rights to use the work in question. But if you don’t need copyright permission to train, that’s fine.
It’s not the solution. It looks more like the problem to me.
As has been said, licensing doesn’t work. When you write something, you automatically have the copyright. With a license, you allow others to copy it, which otherwise would be illegal. For this to work, copyright would have to be extended to cover learning. That’s obviously terrible. But let’s assume for a moment that copyright is only extended to cover machine learning (ML).
You would not be allowed to use anything on the internet for ML, unless there was a license allowing it. It would basically outlaw scraping the net for training data. You’d have to sift through everything to find stuff, with a permissive license. Of course, no for-profit enterprise would pick anything up with an infectious license.
AI companies would have to pay for training licenses. Non-profits that cannot pay would be limited to public domain data: Stuff that is so very old that it is out-of-copyright, some government publications and, of course, your infectiously licensed posts. Sound good?
The for-profit stuff would be more expensive to generate a steady cash-flow to “rights-holders”, like streaming does today.
Well, maybe that’s what you want. For some people, this is simply a matter of ideology. They feel that (intellectual) property is supposed to work like that, and damn the consequences. I’m going to assume you are not like that.
We’d create a new, steady flow of money to property owners, who have to do nothing in return. It’s nice to be rich. I don’t think we need to make it nicer, but that would be the result.
It would be great for corporations like Microsoft that already have a lot of intellectual property for training. It would also be great for the likes of Meta, that can just amend their TOS to get a license from their users. Traditional publishers would likely also see a nice windfall profit, as they’d be able to sell all their old newspapers, magazines and books.
To me, this just seems crazy. It’s doubling down on everything that’s already going wrong.
I’m guessing that that is not the outcome you want. So, the question would be, how you came to support a policy that would lead to it.
CC BY-NC-SA 4.0
Power nor wealth doesn’t come from a single individual, but from a great many.
Just because you vote, say or do something, doesn’t mean it’ll change much of anything. But if hundreds, thousands, millions, or billions do, then we have change. It’s exactly the same with data. A single data-point is not worth very much.
If we all work together, we can make sure that none of us can benefit from the other? How does that even make sense?
Look. I am unable to understand why this bothers you. I like feeling I have a positive effect on the world. I like knowing, EG, that my taxes help the less fortunate. What you are saying seems completely absurd to me.
Your comment assumes everyone agrees ChatGPT or other LMMs/etc are a net benefit to society.
I’m on the fence about it myself.
There’s no such assumption. The people who use it must feel it benefits them, or they wouldn’t use it. By the same logic, the makers must be getting something out of it, as well.
If there is some net negative for society that I am not seeing, then I don’t see why I should even be allowed to offer my work for AI training. Normally, harming other people is a big “NO”.
ETA: You’re not allowed to shoot people with a patented gun, except in special circumstances. Whether or not you have permission of the patent-holder of the gun is immaterial.
Are you being willfully obtuse or do you really not comprehend? It feels like you’re trolling.
So, instead of providing all our comments for free to LLMs, how about adding a copyright notice to everything we write?
Legally everything you write is already copyrighted, and no notice is required. Creative Commons licenses are a way to reduce the restrictions on what people can do with your content. They don’t impose any extra obligations beyond what would exist without any copyright notice at all.
What you say is true, but how are you going to prove anything is copied or scraped and used for commercial purposes? Untraceable copyright violations make copyright worthless.
My point is that adding the CC notice doesn’t make any violations traceable or even less likely. Your comments are just as likely to be scraped with it versus without it. You’re not adding any restrictions on the use of the comments; you’re just selectively removing some restrictions.
I think you’re missing the “Non Commercial Share Alike” part of the license. Also, when the text is revealed to have a licence, then steps can possibly be taken. If nothing you write has a licence, nothing can possibly be done AFAIK. IANAL though 🤷
To me, it’s not much work to hit Ctrl+V at the end of writing a comment for the eventuality that a training set be forced to be made open.
How does your proposition help with this?
However generative AI and large language models have been been spitting out their training data including copyright notices and other stuff verbatim. “poem poem poem to get personal data from ChatGPT”.
In that case just append a unique string to your comments.
As others have already said, including a Creative Commons license does not make a difference.
Are any of them lawyers? Are you a copyright lawyer?
The courts haven’t even decided if LLM creations counts as reproductions or unique works under the legal framework.
So, instead of providing all our comments for free to LLMs, how about adding a copyright notice to everything we write?
Per your own example, the LLMs are trained with some possibly copyrighted content, so why would adding a license now matter if it didn’t previously?
CC BY-NC-SA 4.0
Per your own example, the LLMs are trained with some possibly copyrighted content, so why would adding a license now matter if it didn’t previously?
You’re answering your own question.
The courts haven’t even decided if LLM creations counts as reproductions or unique works under the legal framework.
And how would this be enforced?
That’s the thing, it has to be detected first. One of the only ways to do so is by letting the model reveal its training data and finding copyrighted stuff in there - the smoking gun. As one of my links describes, that has already been done (see the poem poem poem link).
You would need a place to take that smoking gun first.
Aren’t they called courts?
Which ones take these cases? If they are in a different country how does that work?
I’m not a legal expert 🤷 CC BY-NC-SA 4.0
If I ever get to the point where I find out my words have been regurgitated by an AI and I care enough to want to sue them for it, I’ll find a legal expert to help me with it.
Right now it would be a waste of time. There’s no legal reason to bother because it isn’t a violation of fair use yet. It could be. There’s no precedent covering this kind of thing.
Once there is, it wouldn’t be useful unless you’re an author anyway.
we are all authors of content. you are the author of your comment, are you not?
CC BY-NC-SA 4.0
Ehhhh, if you want to parse the word usage, sure.
But copyright around your stuff when you write for a living, or even a hobby, is a different thing entirely. This is doubly true with LLM “AI”. See, the machine learning we currently have just isn’t going to be able to intimate you in particular, or me in particular with typical comment lengths.
But once you get into longer forms of writing, be it poetry or prose, then you get something that the machine can learn to imitate directly. That’s where there’s a difference that matters. Not because it’s inherently okay to use anyone’s writing of any length or type without even the courtesy of attributing. It isn’t, but that doesn’t mean that there’s any harm or loss involved, and that’s where copyright laws will need to be reviewed and possibly changed.
Generally, an author is only different from anyone else by dint of putting in effort to construct their words and edit them carefully. Not that we netizens can’t put that effort into casual comments, nor does that mean every author actually puts in the work. But there is s difference between the concepts in usage, if not in every dictionary.
Tbh, it’s largely a matter of value, which isn’t exactly a clearly quantifiable thing. I mean, “first” is obviously without value beyond a split second of eye rolling. And I’ve seen some serious time put into some seriously valueless writing that was much longer than some books. We all know how much value there can be in a good comment because that’s what reddit tried to trade on; the accumulated value of user writing. But most of this kind of thing is very low value, low effort drivel.
Copyright doesn’t matter for AI training data because that AI is considered a derivitive work therefor using whatever content they find for training data is fair use under current copyright law. People are literally training AI on Pixar content without copyright being an issue.
Also if you don’t want people using your stuff then why are you posting it in the open in a public board that basically everyone has access to? If you want to protect something then the first step would be not handing it to everyone and everything with an internet connection.
I don’t know if a verdict has already been on the topic, nor by whom, but it’s my impression that the issue of copyrighted material used to train LLMs or GPTs is still hotly debated. IIRC Microsoft was sued because of Copilot on the grounds of copyright infringement. Especially with stuff like GPLv3 where derivative works must be of the same licence.
Personally, it’s not much of an issue to me whether my public musings will be scraped. I’d rather they are scraped and used for opensource stuff, but there’s no way for me to enforce that. However, if a model starts spitting out “CC BY-NC-SA 4.0”, I’d at least find that funny and maybe it might even help bring a case on the grounds of breaching the “Non Commercial” clause of the licence.
This reminds me of the posts people used to make on FaceBook where they’d post a paragraph of text specifically denying FB/Meta the right to do anything with their personal data. It’s well-intentioned, but it may not persuade a bot to ignore what you write.
That’s not the point. I write bots. I know they don’t care about what you feel.
It’s all already logged and timestamped and what not… If you ever truly need to make a copyright claim, all you really need is proof you had the idea first which should be as easy as providing the comment/post in question, as well as proof that you are indeed the user that made it.
