> Does copyright law give you the exclusive right to train a neural network with your work?
Does copyright law give openai the exclusive right to exploit the output of ai trained on my data ? (the model, the weights, etc..)
Does copyright law give openai the exclusive right to exploit the output of ai trained on my data ? (the model, the weights, etc..)
Since no court has ruled on the latter question yet, let’s do a choose your own adventure:
Say that OpenAI trained a language detection model. The output of the model is simply a single vector that indicates which language that the input was written in. During training they use your copyrighted code as training data.
Would you consider that to be infringement?
1) If yes: consider instead of using fancy and scary mathematics like “neural networks”, it is just tabulating and counting keywords. Is it still infringement?
1) If no: what’s different about a neural network that outputs code, vs a neural network that outputs a single vector? Perhaps it’s the output that is infringing, not the weights?
I've seen this analogy before. A NN, especially an LLM, is very different in method and outcome. If counting keywords allowed me to replicate copyrighted material, it would probably not be fair use. IANAL, but I imagine the fact that a NN can replicate or compete with the original work makes a difference.
Perhaps it’s like learning to play guitar by watching other guitarists, then releasing songs in the same genre.
If I crank noise through a bunch of matrices and churn out the Mona Lisa because the Mona Lisa was in the training data, I've not painted the Mona Lisa. I've reproduced someone else's work through a fairly tortuous mathematical route.
The matrices don't have agency, they can't lay claim to anything.
Although I should probably say, for completeness: the matrices don't have agency yet.