Comment by kmeisthax | Hacker News Reader

Comment by kmeisthax | Hacker News Reader

Parent Full thread

Creative Commons as a license basically says "if copyright says you need permission to do something, you can do it, subject to these conditions". If training is fair use then Creative Commons explicitly does not restrict AI training, period. If it's not fair use then the model authors have to provide proper attribution and a CC-BY-SA license declaration on their model.

For billions of text posts. Good luck. With that.

On the surface Creative Commons licenses are "AI-friendly" - in that they exist to opt-out of parts of copyright protection - but in practice AI goes against the spirit of Creative Commons so heavily that actually complying with it is impossible. If your attribution string requires its own search engine, maybe it's not effective attribution?

For the record I don't think training is going to be fair use. The AI model you're training is usurping the market for new art/writing/etc, and there's a clear chain of provenance from the training data to the final model and specific outputs (in at least some cases, e.g. "draw a 1970s scifi robot" giving you C3PO).

I think a explicit AI-hostile / AI non-friendly license is needed which explicitly denies work being trained on. For the small minority who do not wish to feed the LLM, shortsighted maybe but I can see a group of people who are all for source available but not for AI.

I imagine most licenses that are AI restrictive indirectly can potentially be circumvented.

Edit: looks like there already is https://github.com/non-ai-licenses/non-ai-licenses

The problem is that AI is making use of the Fair Use exception. They don't need to follow the license - fair use means they can just ignore it!

i think the bigger problem folks have is these ai tools charge a premium for data which was consumed from publicly open websites, where folks originally contributed their knowledge with the belief that others could access the information as-is (paraphrased or what have you)- without being charged a fee

cost has always been an issue, from bandwidth to storage to system maint, to electricity - the bills gotta get paid somehow; its not like websites are gov funded to keep those 9's happy

Licenses can strip you of your fair use rights. In fact, it's common practice to take away fair use even for really weak licenses (e.g. D&D's OGL 1.1, a license to do nothing). Creative Commons has to have an explicit clause stating that any fair use is allowed regardless of the other clauses of the license just to not be interpreted as a restriction of fair use.

Whether we should be stripping people of their free speech rights[0] to stop a parasitic enclosure of the commons that might sidestep existing copyright law, which is being cheered on by the proprietary world, is another question. Copyright law as it currently stands does not and cannot protect artists against AI that is going to salami slice tiny bits of statistical data from their work and mix it with hundreds of other slices to form a new image. That sort of harm is difficult to prosecute even if I personally think it's Not Fair Use. So adding a NoAI clause to Creative Commons would mainly serve to add a bunch of new incompatible licenses, and argumentation over whether or not this is Free Culture.

The correct way to stop generative AI is with collective action, not individual action, because the harm is to the whole class of artists, not a single artist. There is no "collective copyright" that would allow regulation or prohibition of generative AI.

[0] Under US law, fair use is the projection of the 1st Amendment into copyright law.

Is creative commons viral? If not, I'm guessing that around now a lot of people are repenting and wishing they had listened to RMS et al.

That's how I, NAL, read "share-alike": derivative work needs to be shared again instead of walled off.

In which case they can't stop people using the outputs of their LLMs to create new ones. But I guess it'll need to be decided in court, and their plan is to achieve dominance (or enough of a lead to guarantee it, which is the same thing) before others start trying.

Creative Commons isn't one license, it's six. There's a handful of clauses that can be mixed and matched to form a license:

* BY: Attribution requirement, every CC license has this

* SA: Share-alike, aka "copyleft", RMS-style "virality"[0] but without a source code requirement since that is impractical outside of software.

* NC: NonCommercial, license only covers not-for-profit usecases and filesharing

* ND: License does not allow derivative works

These can be combined like so:

* CC-BY

* CC-BY-SA <- This one was used by StackOverflow!

* CC-BY-ND

* CC-BY-NC

* CC-BY-NC-SA

* CC-BY-NC-ND

Only the first two are "RMS compatible" licenses.

[0] Copyright is already viral, share-alike/copyleft clauses just patch the virus to attack itself. It's the "treating your leukemia with HIV" of licensing.

> without a source code requirement since that is impractical outside of software.

Preferred form for editing is practical for any content that is edited in digital form; if it is impractical at all, its for works where editing is done to some non-digital artifact which is then digitized, and where the norm for further modifications would be the same, and where the digital form is itself not renderable back into the artifact. But I can't think of any real examples.

Thank you!

Reply on news.ycombinator.com