US appeals court rules AI generated art cannot be copyrighted

I still can't believe the guy went to Indonesia, went into the monkeys' habitat, gained their trust, set up the camera on a tripod in a way the monkeys would have access to it, adjusted the focus/exposure to capture a facial close-up -- basically engineered the entire situation specifically for that outcome, and simply because he didn't physically hit the shutter he lost credit for the photo. Meanwhile I can open my phone's camera, spin around three times, take a photo of whatever the hell happens to be in its viewfinder and somehow that is sufficient human creativity to deserve copyright protection.

Assisted work is the big clarifier I think

Is a picture edited with photoshop invalid when it uses content fill? What about a picture taken with an iphone, where AI could be part of the phone's processing pipeline or even generate details to make up for lack of optical zoom?

Does spell correction invalidate a book? what if there's AI rephrasing features at work? Where's the line?

I think as you get into those side questions, the only reasonable position becomes treating AI as tooling no different than any other piece of equipment.

> as long as it was a human that told the computer to make the image

There is the question of merit. IANAL/IIUC/etc., but I think it's necessary for a work to have merit to be copyrightable. Now, that's a somewhat vague term to me (perhaps it's clearer in a legal framework), but if I prompt "create a picture of a dog", the computer does most of the work. A prompt would have to be pretty concise, up to specifying all kinds of aspects of the image, for it to be the instructor's merit, to me (that's an important caveat).

That was one hell of a photogenic monkey

https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_disp...

Wait just so i understand it, if a single human creates an AI model and trains it, and then prompts it to create an image, is that considered "human intervention" and does that make that human the author of that image?

What if its a group of 5 humans that built the LLM and one of them prompts it?

Isn't all AI built by some of group of humans? When is AI treated like its own entity like a monkey versus a tool made by a human?

Couldn't the same argument be made for photography? You aren't making the image, the camera is doing all the work.

WRONG. The owner of the camera successfully litigated and is the copyright owner of the work! I’m not kidding about this, and for all the grief I get about being a critic of blase attitudes regarding US copyright around tech circles I’m still a huge advocate for reform.

This is very not like the monkey case, and AI firms should be grateful. Why? If this was a similar logic tree, the owners of the copyrighted material used in training would have ownership of any work produced by an AI system. As in, everything output is a “derivative work” in the eyes of the law. More cases are necessary and this is a fascinating battle to come.

https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_disp...

> Trying to assign copyright to an AI is techno-futurist bullshit by trying to give legal presence to a piece of software.

I don't quite get this argument. Companies already have legal personhood and can own copyrights, can't they? So if a company's AI creates a copyrightable artifact, who wouldn't it be intellectual property of the company?

Thaler seems to go out of his way to claim no human intervention and authorship by the AI - So yeah, that's a very specific ruling that has little to do with AI as a tool. It's really more about AI personhood.

What's potentially more of a problem is the mention of artists using Midjourney and denied copyright - and very much separate cases from Thaler.

The copyright office has already ruled recently that prompts are not enough to gain copyright no matter how detailed or how many iterations.

Furthermore, the Copyright Office stated that prompts alone do not provide sufficient human control, as AI models do not consistently follow instructions in the prompts and often "fill in the gaps" left by prompts and "generate multiple different outputs"

Copyright law: A reason that copyright trolls are less common than patent trolls is that under copyright law, works created independently are not infringing. In court, you might have to prove that you did actually create the thing independently, but I think most juries would be sympathetic to this case. "Oh, you think that the defendant combed through your giant library of millions of logos to find this one specific, rather simple looking specimen."

Also, a lot of logos are simply not "artistic" enough to be eligible for copyright. So in general, logos are more likely to be the subject of trademark litigation than copyright litigation.

Trademark law: In order to claim a trademark you must have used the mark in commerce. So a catalogue of logos not used in commerce is of no real value from a trademark perspective.

I don't think it takes that much imagination here. Not sure what good the first step is actually doing you. Might as well just AI-generate your racketeering demand letters without doing that part.

The scammers will do it anyway and simply claim the logos were all designed by humans.

This is doing it the long way round. Just set up a website that generates every combination of pixels as you scroll down it.

Or just scrape logos, barely change them, and publish them and threaten legal action.

If I prompt you to draw me a bird, I can’t claim copyright on the bird you draw. (At least not with a contract of some sort, of which you are party.)

Pretty sure this wouldn't pass the merit part unless the prompt was unusually long and precise.

the human still made the prompt

What I can guarantee, is that series of prompts itself would be copyright-able. (The series of prompts that ultimately created the image.) No matter how little they may weigh any one of those prompts in isolation. That is, assuming the EULA of the LLM doesn't require you to essentially place your prompts in the public domain.

And of course,

<s>

everyone reads the EULA. Right?

</s>

I suspect in the future we will have a jury consisting of people who are good at prompting. They will load a model that existed at a given time (e.g. when the "author" claimed they came up with the design), and then try to get similar art by just using prompting. Then a judge checks if the art looks similar, and if the prompts were simple enough.

We could have a similar approach with patents.

It's not the appeals court's job to "define what AI means," their job is to rule on the case in front of them. This particular case involved someone asserting copyright over an image that he claimed was generated by a sentient[0] AI. This image was not created by a human, and only works created by humans can be copyrighted under US law, so they ruled against him.

[0]https://thenewstack.io/stephen-thaler-claims-hes-built-a-sen...

there are a lot of ways to detect AI generated imagery with low false-positives (though false-negatives are a risk)

The practical implication is you can't copyright something that your AI generated. As the article notes, copyright applications are also being rejected in cases where a human asserts authorship over an AI generated work.

Unlicensed Human Code is 100% copyrighted and closed source.

Unlicensed AI Code is 0% copyrighted and open source and can't be closed.

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).

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.

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.

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/

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

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.

such an important topic right here. are we really going to enter an age of media that is AI generated or are we entering an age where media bifurcates into two broad categories: AI sloppish brain rot and more refined products that are hand made.