Copyright infringement in AI art • Hacker News Board

Re: Copyright infringement in AI art

#151

Post by 6gvONxR4sf7o » Wed, Aug 17, 2022, 9:49 AM UTC

visarga wrote: post #63

6gvONxR4sf7o wrote: post #56

I think it’s clear that what’s happening is (mostly) legal, but the better question is whether it should continue to be. It will be a bad outcome if a few people profit off of a huge group of people’s labor without their consent and without compensating them, in order to put that large group out of work. Technological advances change the labor market all the time, but it’s absolutely different to exploit your own lab…

It's not just the model publishers that profit from generative models, the users gain something as well. And the access barrier being lowered, now many more people can benefit. Even the artists can benefit. Let's not forget the contribution of the prompt to the final result. The users have a finger in there too. That means they are directly involved in the creative process, they deserve a claim to the copyright.

If the artists benefiting too is a reason to allow this, then the artists having a say in how their data is used shouldn’t be a problem.

Re: Copyright infringement in AI art

#152

Post by visarga » Wed, Aug 17, 2022, 9:05 PM UTC

6gvONxR4sf7o wrote: post #151

visarga wrote: post #63

Earlier quoted context omitted.

It's not just the model publishers that profit from generative models, the users gain something as well. And the access barrier being lowered, now many more people can benefit. Even the artists can benefit. Let's not forget the contribution of the prompt to the final result. The users have a finger in there too. That means they are directly involved in the creative process, they deserve a claim to the copyright.

If the artists benefiting too is a reason to allow this, then the artists having a say in how their data is used shouldn’t be a problem.

Training neural nets is a novel transformative process - it only takes gradients for each training example, it doesn't save any image, just take the gradients and sums them up overlapping the influence of all the examples, mixing up everything in a way we can't tell which feature comes from which training example. More like inverting the dataset into abstract principles and then generating top-down novel examples.

It's like learning, but differs in that humans can't possibly see so much data. It's not like copying, but it can generate copies of training examples as an edge case. It feels like stealing, but you can't assign who is the victim, maybe it draws a bit from multiple sources of inspiration. It's very creative and inspires all sorts of idea juxtapositions, but some people consider it's only parroting known styles.

In the end I think we will want to use the new toys so we will allow training on anything. Maybe a "Do-not-use-for-training-neural-nets" flag will be observed by the conscientious labs.

Re: Copyright infringement in AI art

#153

Post by antioppressor » Fri, Aug 19, 2022, 12:34 PM UTC

jacobsievers wrote: post #128

antioppressor wrote: post #126

Yeah, but what happens if you explicitly state that your art/images must not be included in AI training?

If I've legitimately obtained a copy of your work, then I'm entitled to use it, read it, even make my own copies for personal use. Any new work I attempt to distribute would be measured against your original to determine infringement.

So if you use part of my image, which has watermark embedded via noise and it is found in your image, then you are fucked. Good.

Re: Copyright infringement in AI art

#154

Post by stemlord » Sat, Aug 20, 2022, 3:35 AM UTC

thriftwy wrote: post #104

stemlord wrote: post #23

Earlier quoted context omitted.

I'm dreaming of an AI that can determine whether a song was AI-generated or is truly one degree from human hands

Why? The only thing I care about is whether a song is good

Because its more critical than ever that we have a voice

Re: Copyright infringement in AI art

#155

Post by quitit » Fri, Aug 26, 2022, 5:24 PM UTC

ChadNauseam wrote: post #87

6gvONxR4sf7o wrote: post #56

I think it’s clear that what’s happening is (mostly) legal, but the better question is whether it should continue to be. It will be a bad outcome if a few people profit off of a huge group of people’s labor without their consent and without compensating them, in order to put that large group out of work. Technological advances change the labor market all the time, but it’s absolutely different to exploit your own lab…

Cat’s out of the bag. Stable-diffusion runs on an m1 mac, is competitive with dall-e, and will soon be open source. Midjourney was made independently from dall-e and is arguably better. This is the most revolutionary artistic tool to ever exist and anyone with a few thousand bucks can download a script from github and train it themselves. My advice to creatives is to figure out how to thrive in this new environment b…

The creative process, like many kinds of fields applicable to business is about problem-solving. AI text-to-image generation doesn't replace that function, however it does form an excellent tool, especially when it comes to rapid conceptualisation.

This will allow more people to be creative problem solvers without needing to possess technical skills in image creation. Much in the same way that graphics apps allowed people to make image without needing to learn studio or art skills. Or DTP tools allowed more people to publish without the tedium and high set up costs.

I will still be hiring illustrators and designers, and this may be one of their tools, and it would be their responsibility to be experts in it, but it doesn't replace them - it makes them better illustrators and better artists.

The right way to think about this is not that it shrinks the field, rather it opens it and accelerates it - for that it's a very welcome addition.

No creative is scared of this - they're looking forward to the next-generation approach, and it's clear that 2D images are not the end point. Soon we'll have 3D (already in progress), soon we'll have music, soon we'll have this for animation and programming.

One can easily imagine all sorts of ways this technology can be implemented, take a simple example of a video game: Imagine a horror genre that hones into what scares the viewer and delivers further upon that.