Training AI does not violate copyright, says GitHub CEO – 5/3/2023 – Tech

The chief executive behind the artificial intelligence (AI) accelerating program development GitHub Copilot, Thomas Dohmke, says training AIs is in line with copyright, as it’s not about reproduction, it’s about research.

The company is the target of a lawsuit on charges of disrespecting the intellectual property of programmers who published their code on the GitHub digital library. The technology, created in partnership with OpenAI (from ChatGPT), automatically completes snippets of programming scripts.

“Artificial intelligence will allow computing professionals to be more creative and will not replace programmers”, says Thomas Dohmke in an interview with the report this Wednesday (3), on the third day of the Web Summit technology event in Rio de Janeiro.

Dohmke promised that he will develop a game in 18 minutes with the help of Copilot, but did not provide further details on how the program will be.

In March, GitHub announced Copilot X, an updated version of OpenAI’s most powerful AI-built programming helper, GPT-4. This update, in addition to completing snippets of programming code, allows the user to interact with the platform through text and audio.

The Copilot individual plan subscription costs US$ 10 (about R$ 50.30) and already writes 46% of the programming codes, according to data from GitHub. The company offers free access to the technology to high school and college students and also to participants in large open source projects.

The platform informs that developers who reach the criteria for free subscription are automatically informed by the platform.

Programmers from all over the world have been sharing their code as well as explanations of their projects on GitHub since its founding in 2008. This information helps other developers get their programs off the ground. “The platform has become what we call the home of open source over the past 15 years,” says Dohmke.

According to Dohmke, he joined GitHub as a user in 2008 to develop his own app creation platform, HockeyApp, acquired by Microsoft in 2014.

Two people filed an unidentified class action lawsuit against GitHub, Microsoft and OpenAI, with allegations that the Copilot training represented “unprecedented piracy.”

GitHub and Microsoft lawyers invoke the fair use doctrine, which defines ways to use data without disrespecting intellectual property. In the United States, for example, doing research does not infringe copyright.

The new applications developed by OpenAI and Microsoft are in limbo, according to InternetLab’s culture and knowledge coordinator, Alice Lana. Artificial intelligence training is classified as research, but the technology is used in products such as Copilot, whose subscription is paid for by the general public.

GitHub’s programming assistant also has an option to automatically check for patent infringement when Copilot suggests codes with more than 150 characters or if that snippet is similar to another one stored on GitHub. “This prevents the programmer from accidentally using this knowledge without attributing authorship,” says Dohmke.

“We work with other open source entities such as the Linux, Cloud Native and Apache foundations and public policy makers in many countries, including Europe, to find solutions on the use of AI and intellectual property”, he adds.

The open source community has dealt with intellectual property lawsuits since its inception. Red Hat was sued under allegations of using a competitor’s code in the Linux operating system, a kind of Windows without commercial attribution. The process only came to an end with the bankruptcy of SCO (Santa Cruz Operation), which developed the Unix operating system.

For Dohmke, the collaborative open source development process today won the competition with proprietary programs. “No company has been able to beat the competition against these millions of developers who effectively play the greatest team sport on earth, working to build software.”

Google, for example, maintains an open source version of the Chrome browser called Chromium. Changes made by contributors to this project can be implemented in Chrome.

Large Brazilian companies such as digital banks Nubank and Neon and retailer Via are GitHub customers.

“Based on our last report in October, which is our annual balance by country, Brazil is our third largest market outside the United States, behind China and India. We are, in part, at the Web Summit to talk to Brazilian programmers, and then in São Paulo we will meet with clients”, says Dohmke.

---

THOMAS DOHMKE, 44

He is chief executive of GitHub, a site where software developers share code, acquired by Microsoft in 2018. He founded the application creation platform HockeyApp, also acquired by Microsoft in 2014. He holds a PhD in mechanical engineering from the University of Glasgow, in UK.