1

I know, that someone asked similar question, but I want to focus on one thing:

Can the weights of the AI model be covered by copyright?

What are the arguments for and what against?

Improve this question
2
  • Can you link the other question, please.
    – user35069
    Apr 12 at 19:56
  • That answer is not clear about weights itself and focus more on model and source code.
    – Maniues
    Apr 13 at 6:15

1 Answer 1

Reset to default
-1

The concept of copyright is that you can copyright the expression of an idea, but not the idea itself.

By themselves, model weights produced by automated training are not copyright-protected. Model weights represent objective facts about the world (or human language or something else), produced by analysis.

However, if tweaked by humans, they may become copyright-protected. Whether they do depends in part on whether the tweaks are necessary for performance. If the models fails without the tweaks, they are, and are thus an invention, which is patentable, but not copyrightable. If the model merely produces specific creative outputs with the tweaks, they are non-essential and might be copyright protected.

This is an evolving area of the law without clear and universal answers.

Improve this answer

Not the answer you're looking for? Browse other questions tagged .