Hacker News Reader

Comment by Quanttek | Hacker News Reader

This is fantastic. If companies want to create AI models, they should license the content they use for the training data. As long as there are not sufficient legal protections and the EU/Congress do not act, tools like these can serve as a stopgap and maybe help increase pressure on policymakers

View on HN

It's going to be interesting to see how the lawsuits against OpenAI by content creators plays out. If the courts rule that AI generated content is a derivative work of all the content it was trained on it could really flip the entire gen AI movement on its head.

If it were a derivative work[1] (and sufficiently transformational) then it's allowed under current copyright law and might not be the slam dunk ruling you were hoping for.

[1] https://en.wikipedia.org/wiki/Derivative_work

"sufficiently transformational" is carrying a lot of water here. At minimum it would cloud the issue and might expose anyone using AI to lawsuits where they'd potentially have to defend each generated image.

Sufficiently transformational only applies to copyrightability, but AI works are not copyrightable under current US law, so it's a non-issue.

Oh, interesting, I didn't realize that's how it worked. Thanks for the additional context around this. Guess it's not as upending as I thought it could be.

Not if it is AI generated. So far only humans can be original enough to warrant copyrights, at least in the US .

BTW, the right to prepare derivative works belongs to the copyright holder of the reference work.

I doubt that many AI works are in fact derivative works. Sure, some bear enough similarity, but a gross majority likely doesn't.

My biggest fear is that the big players will drop a few billion dollars to silence the copyright holders with power go away, and new rules are put in place that will make open-source models that can't do the same essentially illegal.

…then I'll keep enjoying my Stable Diffusion and pirated models.

If the courts do rule that way, I would expect a legislative race between different countries to amend the relevant laws. Visual generative AI is just too lucrative a thing.

> they should license the content they use for the training data

You mean like OpenAI and Adobe ?

Only the free and open source models didn't licensed any content for the training data.

Adobe is training off of images stored in their cloud systems, per their Terms of Service.

OpenAI has provided no such documentation or legal guarantees, and it is still quite possible they scraped all sorts of copyright materials.

There is in fact, an extreme amount of circumstantial evidence that they intentionally and knowingly violated copyright en mass. It’s been quite a popular subject in tech news the past couple weeks.

> OpenAI has provided no such documentation

OpenAI and Shutterstocks publicly announced their collaboration, Shutterstocks sells AI generated images, generated with OpenAI models.

Google scrapes copyrighted material every day and then presents that material to users in the form of excerpts, images, and entire book pages. This has been ruled OK by the courts. Scraping copyrighted information is not illegal or we couldn't have search engines.

Google is not presently selling "we trained an AI on people's art without permission, and you can type their name in along with a prompt to generate a knockoff of their art, and we charge you money for this". So it's not really a 1:1 comparison, since there are companies selling the thing I described right now.

That pretty clearly would fall under transformative work. It is not illegal for a human to paint a painting in the style of, say, Banksy, and then sell the resulting painting.

Humans and AI are not the same thing, legally or physically. The law does not currently grant AI rights of any kind.

If a human isn't violating the law when doing that thing, then how is the machine violating the law when it cannot even hold copyright itself?

I'm not sure how to explain this any clearer: Humans and machines are legally distinct. Machines don't have the rights that humans have.

Fair Use is the relevant protection and is not specific to manual creation. Traditional algorithms (e.g: the snippets, caching, and thumbnailing done by search engines) are already covered by it.

In some locales sitting on the street writing down a list of people coming and going is legal, but leaving a camera pointed at the street isn't. Legislation like that makes a distinction between an action by a person (which has bounds on scalability) and mechanized actions (that do not).

What's not prohibited is allowed, at least in the US.

Scraping is only legal if it's temporary and transformational. If Google started selling the scrapped images it would be a different story.

What is not transformational for generative AI ?

No they are not. They train their models on Adobe Stock content. They do not train on user content.

https://helpx.adobe.com/manage-account/using/machine-learnin...

"The insights obtained through content analysis will not be used to re-create your content or lead to identifying any personal information."

"For Adobe Firefly, the first model is trained on Adobe Stock images, openly licensed content, and public domain content where the copyright has expired."

(I work for Adobe)

There is a small difference between any and all. OpenAI certainly didn't licence all of the image they use for training.

source for OpenAI paying anyone a dime? don't you think that would set a precedent that everyone else deserves their cut?

Reply on news.ycombinator.com