>the right to exclude others from making certain uses of the work: copying it, making a derivative work based on that work, distributing copies of the work to the public, and publicly performing or displaying the work.

So why would "training" "AI" on code with the intention of emitting derived works not be copyright infringement exactly?

This product is transforming copyrighted code into something that's intended to be used or sold in other works. The snippets it emits are directly derived from copyrighted code.

The most common argument against this is that humans also learn from copyrighted material. My argument against this is that CoPilot is not a human and should not be assumed to inherit rules intended for humans.

>in a field that benefits us all

As it stands currently CoPilot is proprietary and does not benefit anyone except for MicroSoft. If CoPilot was released under a FOSS license it would actually benefit us all. Most of the people against CoPilot are not against AI, but rather a proprietary AI product transforming FOSS work into other potentially proprietary works with the intention of profiting off of the completion service and hoarding the code that powers it.

The fact that you write something, doesn't automatically make that thing true.

Some uses of AI were ruled not to be infringement. This is a different case which requires a different ruling.

Well, maybe. But even if we assume that this is true, when anyone later uses the AI to reproduce a copy of the code, a copy has been made and copyright has been infringed.

I don't care if people make money using GPL'd code, but I do care if they take the code and strip the license so they can use it in non-free projects.