We embarked on a process, promising at the beginning of the year that we’d make a first announcement at All Things Open, kickstarting a public conversation. We’ve delivered, thanks to contributions of many experts and sponsors. But it’s only the starting point. There is a lot more to do.
After two community reviews in person and a first pass at online comments, we released a new draft version 0.0.3.
The base of the conversation is a preamble to explain “why Open Source AI,” followed by the beginning of a formal definition: the document will get longer. Open Source experts will recognize the heavy borrowing from the free software definition and the structure of the GNU Manifesto: it’s not a mistake. We believe that the consensus on a Definition of Open Source AI will emerge after the stakeholders will have made a similar journey that led to the Open Source Definition. The OSD is basically a checklist that appeared after decades of free software development, when developers, users, business leaders, lawyers and policymakers had time to learn what freedom meant in the context of software. We don’t have decades to wait for AI but we can accelerate by building on top of what many of us already know and reach out to diverse communities to join the conversation.
That’s what the OSI is doing with these Deep Dive: AI cycles: inviting multiple stakeholders to learn and share their knowledge as we all make progress together towards a common understanding of AI systems.
What’s in draft v.0.0.3
The four freedoms have received a bit of wordsmithing for consistency and clarity, making them shorter compared to previous drafts. I removed the words “without any limitation” from the Use and Share principles as recommended by Chestek, and because a question about copyleft also came up at the workshop in Monterey.
The current version reflects the consensus of the suggestions emerged from the workshops in Raleigh and Monterey, and the online comments to v.0.0.2.
In addition to those changes, I did some cleanup of the word soup, removing all instances of the most loaded concepts like trustworthy, reliable, fair, etc. from the preamble: they only appear in the “Out of scope” section.
Enjoy and comment on draft 0.0.3.
Known issues and next steps
There is no consensus on what definition of AI system to use. The draft 0.0.3 still uses the definition introduced by OECD in 2019 for lack of a better option. We’ll continue the conversation.
We have two more in-person workshops scheduled before the end of the year: Nov 15 at the DPGA annual summit in Addis Ababa; and Dec 12-13 at the Linux Foundation AI.Dev conference in San Jose. These were not planned at the beginning of the year when we announced the 2023 series but they’re extremely important to reach African tech leaders and policy makers and AI developers.
At this point we want to close DDAI 2023 thanking the sponsors Google, Amazon, GitHub, OSS Capital, GitLab, Weaviate and Sourcegraph; the Linux Foundation for their travel grants; and individual donors, because we couldn’t have hosted the webinar series and run three in-person meetings without them.
We’re working on a plan for 2024 that includes expanding our reach to other communities with an eye on reaching consensus on a 1.0 release of the Open Source AI Definition in the quickest amount of time.