The copyright saga continues. Is AI work copyrightable? Are prompts copyrightable? Was the training data used considered copyright infringement? These questions are still ongoing and their answers may still be met with disagreement.
The Copyright Office quite literally just published yesterday a new report (Part 2) of their ongoing analysis of what constitutes copyright in age of Artificial Intelligence. What are the legal and policy issues related to copyright and artificial intelligence? So, one thing has been made clear: prompts are not copyrightable beacuse the output of a generative model is not perfectly predictable based on the prompt. For example, the generation from a well crafted prompt could be different when you use different random seeds and could change drastically with the spelling of your words.
Here are the general finds outlined in their press release:
- Copyright applies to AI-generated outputs when there is significant human creativity involved. In other words, humans have to sufficiently mix in their own expressive elements into the work.
- AI is a tool — a tool that can be used to generate human work.
- No copyright protection for pure AI output. Whether a text or images, generative AI outputs that have not been further processed by humans are not copyright protected.
- Prompts are not copyrightable. It’s not sufficient creative expression on behalf of the human.
These conclusions came from reviewing thousands of comments, attending expert meetings, listening sessions and webinars.
There will be more info coming soon for Part 3 — and this is the one that will have insights on the issue of training data. Let’s say that the conclusion is that using copyrighted works to train AI is ‘lawful’. We at Backpack think that whether something is lawful doesn’t make it good nor ethical. Backpack is made for artists by artists. You control the models that includes your work in the training and you receive compensation every time it’s used. Join the movement, join ethical AI at Backpack.
