Ivan is a Big Tech and India specialist. He also covers security, platforms, AI, security, and apps for TNW. It's a lot of stuff. He loves to say "Bleh." Ivan is a journalist covering Big Tech, India and policy. He also covers security, platforms, and apps to TNW. It's a lot of stuff. He loves to say "Bleh."
Last week, GitHub released Copilot, an AI-powered tool that suggests code snippets automatically.
OpenAI was used to train the tool by using publicly available source code from different projects. This looks like any other AI project training method. However, many people took to Twitter to criticize GitHub's decision and call it a violation of copyright.
I am leaving GitHub as my copilot uses OpenSource code to train me. It is so odd. Anyone can fork it, and GitHub allows anyone to feed OpenSource code to it from anywhere. This is allowed by US copyright law. I'm also certain that we shouldn't strengthen copyright laws Armin Roacher (@mitsuhiko), July 3, 2021
If @GitHub (Microsoft), truly believes that copilot isn’t infringing upon anyone’s work, I want them to give me a chance to show it. I’ll donate $50k (or @EFF if you can’t agree) if they release a Copilot software that is solely based on Windows kernel source. 1/ https://t.co/WMWD6FTcR2 Jake Williams, @MalwareJake July 3, 2021
Julia Reda, a former Member of European Parliament and researcher, argues on her blog that GitHub's tool does not violate copyrights.
Banning it would be a bigger mistake. To stop this practice, you would need to enforce stricter copyright laws. This would make copyleft less viable than Githubs tool. Although the idea that a large corporation like Microsoft could use public code may seem to be against copyleft, banning it would prove even more costly. This practice could be prevented by tighter copyright laws. It would also undermine copyleft more than Githubs tool.
She added that text mining and data mining are not in violation of copyright laws. Machine-generated work is also not considered derivative work. The Copilot tool generates code snippets that aren't covered by intellectual property rules.
The argument that GitHub Copilot's outputs are derivative works from the training data rests on the assumption that machines can produce works. This assumption is false and counterproductive. Copyright law only applies to intellectual creations. Machine-generated code such as GitHub Copilot does not constitute a work, and therefore is not considered a work under copyright laws.
The debate surrounding IP-related policies in relation to machine-generated work is heated. However, it will take some time before these arguments are settled. You will have to continue tweeting your frustrations.