Hacker News Reader

Comment by kristopolous | Hacker News Reader

"affirmed that a work of art generated by artificial intelligence without human input cannot be copyrighted under U.S. law"

Does that exist?

What would that even be? A "random2image" model?

View on HN

As a matter of law? Sure it does. Thaler said the image at issue was "autonomously created by a computer algorithm running on a machine". He's been trying to walk that back for the last couple of years though. See Thaler v. Perlmutter, 1:22-cv01564-BAH (ECF #24), D.D.C. (Aug. 18, 2023).

How about selection? If I select the good image from 1000 others? Curation is also a contribution to art.

I would say curation and editing are much more important than creating the art itself, but that might be a very unpopular opinion.

Perhaps. The argument here is that neither was done, however.

The argument about whether human selection would make the human the author of the work is irrelevant, because the human in this case isn't claiming authorship, by selection or otherwise.

Thaler built a tool that spits out images and other stuff. He wants the AI to retain ownership, and for it to grant him a sublicense to him. Its bonkers.

I wonder if I supply a random input to a fine-tuned model that can only generate what I wanted initially.

I.e. the model named "starry-night-van-gough-with-bunny" can generate only one image.

If you want to know if this would be copyrightable, just flip a coin. I don't think anyone can give you better legal advice on this example than a coin toss.

It's not a necessary test for this case, but in general I would suggest using a legal test that is AI agnostic. Imagine there is a service where you can submit a prompt and get an image in return. You might submit a prompt like, "a man in steampunk gear sitting at a table playing with poker chips".

If a human artist draws an image based on that prompt, do you share joint copyright between the two of you? Or, does the artist have full copyright over the image they drew?

If your contribution was insufficient for joint copyright in the case of the human artist, then it was also insufficient to grant you copyright in the case of the AI artist. To know whether you have a claim on the copyright of the resulting image, you only need to look at your own creative inputs.

I am not a lawyer, but that is my expectation of where this will ultimately end up.

I understand it in this way (I am not a lawyer): if you're using an AI tool to generate art, the company that's running the AI tool as SaaS can't claim copyright on the generated content. The person who uses the tool can claim copyright, as they created the content with a tool (AI). Comparable to a brush (=tool) for painting.

> Does that exist?

Yes, for the purposes of this case, because that that is an accurate description of the image in this case is not a fact in dispute between the two sides. This is a case about what the law means given that uncontroversial (between the parties) fact.

It's called "unconditional generation" so yes you supply a random input string and it generates something. StyleGAN2 is an unconditional image generation model. StyleGAN2 trained on faces from Flickr: https://thispersondoesnotexist.com/

https://en.wikipedia.org/wiki/Infinite_monkey_theorem

The plaintiff is asserting it exists. He could easily resolve the issue by listing himself, not the AI, as the creator of the work, but he's pushing the point to concretize it into law.

In this particular instance, the claim was filled out that way.

Reply on news.ycombinator.com