Hacker News Reader

Comment by GMoromisato | Hacker News Reader

I don't quite get it. Can Pixar copyright its GPU-rendered films or can it only copyright the human-created models?

How will the USPTO draw a legal distinction between AI-generated and GPU-generated?

If you put something that is copyrighted through a mechanical transformation process it has the same copyright. For instance a video of a movie retains the copyright status of the movie (as well as probably being an illegal copy itself) even if you added camera shake and audience coughing sounds. So there's not an issue with things like this.

Well OK, so what if I put my copyrighted prompt through the mechanical transformation process of the AI model?

... then all elements of the prompt that come from your human creativity and that survive into the final result remain under your copyright. Which in most cases should be almost as good as owning the full copyright, although it might influence fair use defenses if even a full copy infringes on only a minor portion of the work.

If your prompt is a novel, it's probably copyright but you don't have copyright over someone else's thing they made using a different novel-length prompt, even if the thing is similar to yours, since they didn't copy the prompt. If your prompt is a sentence, it's probably too minimal to qualify for copyright.

I think the point is to decide where to draw the line. An AI model is a stochastic process, you don't get the same results from the same prompt (even at temperature 0 because the GPU will do multiplications in a different order and errors compound). It's not really the same as a digital compositing pipeline and a lighting engine.

Also you have to account for the PB of training data which consists of copyrighted works that goes into the model and is interpolated to produce the result. Personally this tips the balance to AI output being new work that's not copyrightable for me because otherwise all of the people who own copyright on the training data should have joint copyright of the output too because in a sense all of the training data is part of the prompt too.

Your prompt doesn't really contain all of the decisions that were made in the final output most of the time. When you use a rendering engine it does contain only your decisions and you have a license to use the work of all the people who created the software and other assets you're using to make the film.

> How will the USPTO draw a legal distinction between AI-generated and GPU-generated?

Patents and trademarks are not copyrights. The U.S. Copyright office handles copyrights, the USPTO handles the other two.

In the spirit of your question, though, the concern predates AI.

Programs that you can use a stylus to draw with - in a similar manner as pen/ink and paper - have been around a long time, and are getting better.

Let the definition of human art be human effort put into creating something. If I put "etymology of the word art" into Google I get "The word art comes from the Latin noun ars (meaning practical skill, craft, or work of art)."

If I spend an hour drawing something in a program - strokes in the same exact manner as physical paint and canvas, but digitally adjustable and editable afterward - this is still art. But it's hard to prove that from just the output file alone. It's even worse for 3d model creation - someone who painstakingly molds/creates a 3d model in Blender is definitely doing art, even if there is no physical artifact until it's printed or projected.

Perhaps art programs will need to start authenticating input devices and movement sensors and recording their stream during the artistic process, and including that as part of the output file. This will only be worse if you want to create art but not train AI. The only solution to that is not create digital art.

Makes me think of Worth Enough, by Radoxist[0]. I think he spent about a year on it.

[0] https://www.deviantart.com/radoxist/art/Worth-enough-7324787...

Reply on news.ycombinator.com