From Debian to Open Source AI
What does open mean in AI? A timely topic, the concept of open AI just roiled the leading AI organization in the world, leading to the ousting of San Altman as CEO of OpenAI, his possible move to Microsoft, and now the news he is going back to OpenAI as CEO with a super board shuffle!
What we know about the split speaks to a shift in the technology community not seen since the early days of open source, dating back 50 years ago.
All the debate today, in many respects, revolves around the idea that AI is far different than open-source software- so much so that we need a new definition for what is meant by “open” and “AI.”
Stefano Maffulli, executive director of the Open Source Initiative (OSI), sat for an interview for The New Stack Makers at the Open Source Summit in Bilbao, Spain, earlier this Fall to discuss an effort underway to define open source AI. An initial list of principles, now in the works, will be released by the year’s end.
“It’s a lot of complexity, a lot of new things,” Maffulli said about the rise of AI. “It really reminds me a lot of the early days in the seventies, when software started to come out of the laboratories. It was a new artifact, a new creation of human creativity, the creative minds, And we needed to think about what to do with this thing. You know, the software, the source code and the binary code? Do we want to apply copyright? Do we want to apply patent law? Do we want to do a whole new thing, and slowly, it started to consolidate, the whole social norms started to consolidate behind using copyright. We are at the same sort of stage in history where we have these large language models, these new transformers that require a lot of data that gets fed into a machine and creates a model with weights and parameters.
“All these new parameters don’t fully squarely fit into the copyright framework, because they’re not technically a creation of the human mind. They’re the result of an elaboration. Some scholars, some legal experts think that they’re not covered by copyright even. But there is a really big complexity down there.”
But we don’t have 20 years to develop a new definition.
“What we do have is a lot of experience in creating definitions and listening to communities,” Maffulli said. “The interesting thing about the Open Source definition is that it didn’t come out of the vacuum, right, that it was already in place. Actually, the Debian free software guidelines were the result of the consultation from the Debian community, and those are the ones that became the Free Software, the Open Source definition. We do have a way and this practice of putting together the most brilliant minds that are thinking about the problem of what open means in the AI systems in this AI space.”
Even the hallmarks of open source become reasons why transparency becomes a paradox. A model that does not reveal its sources has far less chance of getting sued for copyright infringement than a model that imbues a more transparent approach.
And in the balance? Altman getting ousted shows the differences in belief systems. A “safetyist,” community emerged at OpenAI with Altman on the other side of the battle, wanting more to try it first, see if it breaks, and then move on.
The history of open source shows us one thing that we can turn to when observing these mega battles. The preservers of trust become far more valuable for commerce than the wanton quest for technical superiority. We can see that much in the past week’s events when it comes time to define what we mean by “open” and “AI.”