4 ms·

> If what you said is correct then why do things like this happen? You misunderstand what I am saying. I am interested in finding a way to deploy genAI that ha

by golol 3y ago

> If what you said is correct then why do things like this happen? You misunderstand what I am saying. I am interested in finding a way to deploy genAI that has no copyright issues from the commercial deployment side. I am not saying that this is what companies are doing right now. But I want to see if it is possible to provide the service they are providing, without the copyright issues. My point is not that midjourney didn't use non-publicly viewable data in their training, it is that they did not need to. I am confident their services can be replicated without using non-publicly viewable works. Then the copyright issues that remain are those of training on publicly-viewable but not copyright-owned works, and distributing the results commercially. These I am addressing in my OP. > Just because something is publicly available doesn't mean you are permitted to use it as you see fit. Again I am not so interested in discussing the training aspects, as many people would, for example, agree that training on publicly available data for scientific reasons, or to make some internal model whose results never get shown to the public, is quite okay. It is imo the easier issue to deal with so it is not the focus of my post. > If we are going to paint with broad strokes, are you supporting this? The issue of deepfakes can be dealt with on a level that does not involve copyright. It is not relevant to genAI vs copyright for me. > I must also assume that you can see from my perspective that something being publicly available or even privately available doesn't mean that I have consented to you doing as you please. For me, morally, I am allowed to think about your work in any way I wish, and I am also allowed to use pen and paper, a drawing board, or a computer to assist me in my considerations of your work. It is just my responsibility that the products of my considerations do not reach the public eye if they are be harmful, criminal or infringing etc. If I draw profit from your publicly available work, that is not necessarily to the detriment of you. If I use your posts on reddit to train my companies internal application filtering model then I am not taking anything away from you, and I am not infringing on your copyright. If, however, I train a model that copies your very funny tweeting style, and create a bot that makes twitter posts based on that, then I am potentially taking something away from you and potentially infringing on your copyright, depending on the similarity of my posts to your posts. If I create some PR-consulting company that creates tweets for customers and claims to own and sells the right to that intellectual property to their customers, then I am quite likely infringing on copyright because my model might have produced works for which you actually own the rights.

hammyhavoc 3y ago
I don't misunderstand it whatsoever, you just choose to ignore that consent trumps any notion of commercialization and doing as you please, in both a legal and ethical context. If these things don't need to be trained on the work of others for whom they don't have consent and have not paid a license fee to, then why are they? Because it's exploitation by some of the biggest corps in the world. They should not be above the law and the utmost basic of ethics—consent. That is why it isn't "okay". I am not debating that generative AI can't be done ethically, I am pointing out that search is consensual. I can use robots.txt to opt out of being included in search results, but search engines direct traffic to the author. Authors did not consent to having their non-web content included in an LLM as an example. The problem is consent. The problem is your attitude towards consent. The primary purpose of the commercially and publicly available generative AI is to undercut human labour. Fact. What it serves up is frequently too similar to the existing work that it was trained on, and therein lies the problem on every level, not just from copyright infringement, but one of consent. You approached it from the angle of copyright, but the problem is one of consent. The reason it isn't "okay" is because consent wasn't given with these public, commercialized models. Deepfakes are part-and-parcel to copyright infringement. If you can take x and make it look more like y by applying z attributes then this is why you can make it regurgitate someone's intellectual property who did not consent to being included.
- golol 3y ago
  > If these things don't need to be trained on the work of others for whom they don't have consent and have not paid a license fee to, then why are they? You made a point about meta training on non-publicly viewable/freely available works, such as pirated movies. That's why I made the point that using non-publicly viewable data is sufficient. At the same time I said that, using data for which you have neither consent nor license is indeed necessary, and for the training on such data the copyright/fair use considerations are complicated. > I am pointing out that search is consensual. If I train a diffusion model on images from google search,and make a paid latent-space search where I just show people some pictures without any claims of selling them rights to anything, then I claim that this is just as consensual as google images. That is because I took the source images from google images where they were consensually posted, and in the worst case my model is exactly reproducing them. More likely my model is only approximately/interpolatively reproducing them. Sure, you might change your robots.txt to update google image behavior and while model won't update, but I think generally if your images has been reposted, it is quite likely that your image is still searchable on google images, even if you change the robots.txt for your personal blog. So it's not like google is very good at updating consent either. Perhaps there is a point to be made that my diffusion-search may not just reproduce your works approximately, for which I claim that consent is kind of given, but furthermore reworks or recombines them with other works. There it is perhaps true that new consent should be necessary.

by golol 3y ago

hammyhavoc 3y ago

I don't misunderstand it whatsoever, you just choose to ignore that consent trumps any notion of commercialization and doing as you please, in both a legal and ethical context. If these things don't need to be trained on the work of others for whom they don't have consent and have not paid a license fee to, then why are they? Because it's exploitation by some of the biggest corps in the world. They should not be above the law and the utmost basic of ethics—consent. That is why it isn't "okay". I am not debating that generative AI can't be done ethically, I am pointing out that search is consensual. I can use robots.txt to opt out of being included in search results, but search engines direct traffic to the author. Authors did not consent to having their non-web content included in an LLM as an example. The problem is consent. The problem is your attitude towards consent. The primary purpose of the commercially and publicly available generative AI is to undercut human labour. Fact. What it serves up is frequently too similar to the existing work that it was trained on, and therein lies the problem on every level, not just from copyright infringement, but one of consent. You approached it from the angle of copyright, but the problem is one of consent. The reason it isn't "okay" is because consent wasn't given with these public, commercialized models. Deepfakes are part-and-parcel to copyright infringement. If you can take x and make it look more like y by applying z attributes then this is why you can make it regurgitate someone's intellectual property who did not consent to being included.

golol 3y ago
> If these things don't need to be trained on the work of others for whom they don't have consent and have not paid a license fee to, then why are they? You made a point about meta training on non-publicly viewable/freely available works, such as pirated movies. That's why I made the point that using non-publicly viewable data is sufficient. At the same time I said that, using data for which you have neither consent nor license is indeed necessary, and for the training on such data the copyright/fair use considerations are complicated. > I am pointing out that search is consensual. If I train a diffusion model on images from google search,and make a paid latent-space search where I just show people some pictures without any claims of selling them rights to anything, then I claim that this is just as consensual as google images. That is because I took the source images from google images where they were consensually posted, and in the worst case my model is exactly reproducing them. More likely my model is only approximately/interpolatively reproducing them. Sure, you might change your robots.txt to update google image behavior and while model won't update, but I think generally if your images has been reposted, it is quite likely that your image is still searchable on google images, even if you change the robots.txt for your personal blog. So it's not like google is very good at updating consent either. Perhaps there is a point to be made that my diffusion-search may not just reproduce your works approximately, for which I claim that consent is kind of given, but furthermore reworks or recombines them with other works. There it is perhaps true that new consent should be necessary.