My Stance on AI-generated Code

I recently added this to Shoelace's contribution guidelines, which sums up my position on AI-generated code.

As an open source maintainer, I respectfully ask that you refrain from using AI-generated code when contributing to this project. This includes code generated by tools such as GitHub Copilot, even if you make alterations to it afterwards. While some of Copilot's features are indeed convenient, the ethics surrounding which codebases the AI has been trained on and their corresponding software licenses remain very questionable and have yet to be tested in a legal context.

I realize that one cannot reasonably enforce this any more than one can enforce not copying licensed code from other codebases, nor do I wish to expend energy policing contributors. I would, however, like to avoid all ethical and legal challenges that result from using AI-generated code. As such, I respectfully ask that you refrain from using such tools when contributing to this project. At this time, I will not knowingly accept any code that has been generated in such a manner.

Also let it be known that I would have zero problem with people using Copilot et al if software licensing and attribution were respected. But it means nothing if a computer "learns" from licensed and copyrighted code and spits it back out in your project like it's your own.

As humans, we learn from each other's code and we produce new code based off those learnings.

"What's the difference then?"

The difference is that humans can't store copies of every available codebase in their brain. We remember concepts. Computers remember by copying.