I’ll set aside the “Documentation Requirement” for now, as it seems to be part of the ongoing discussion about data information. However, I do agree with the claim that the current OSAID is both “unimplementable” and “unenforceable”. While I acknowledge that the wording of the definition has been gradually evolving in a positive direction, when OSAID 1.0 is released at the end of the month, what exactly will OSI and the volunteers supporting OSI be expected to do?
It’s clear that reviewing each AI system one by one will be too much to ask of the existing volunteer effort, and as Samj-san pointed out, I don’t believe that these decisions can be made objectively. If OSI continues to only review licenses (legal conditions) as before, there could be significant opportunities for providers of AI systems to engage in misrepresentation. Additionally, relying solely on a process that has historically focused on copyright and license reviews could place a heavy burden on the existing review process.
I assume that Stefan-san and the board are aware of these issues, but at the very least, the general OSI membership, including myself, have not yet been informed about any plans. The first priority should be releasing the checklist aligned with RC1’s contents. However, even if a valid OSAID 1.0 and checklist are completed, simply publishing them won’t give those in the Open AI/ML industry a clear idea of how to implement or apply them. This is what concerns me. While it is not my place to decide if OSAID 1.0 should be released on schedule, I remain unsure if this is the right course of action given the current state of affairs. I worry about divisions within the community.
That said, part of me also believes we should try releasing it.
When the “Open Source Definition” was created, it was derived from the lessons learned during the early days of the Debian Project about what constituted free software and what the conditions for freedom were. In other words, there were already various established practices within the community, and the OSD was a collection of those experiences.
In contrast, the concept of “Free and Open” within the AI world hasn’t had nearly the same accumulation of experience as we had in 1998. Furthermore, related legal systems, interpretations of copyright law, and laws around publicity and privacy vary widely between countries. In such a context, if we don’t release the definition soon, there may be no industry-wide push to adhere to OSAID’s four freedoms, nor the motivation to work towards a complete definition within the AI/ML sector.
Open Source Group Japan, the guardian of the term and trademark “Open Source” in Japan, and I support OSI’s ambitious challenge to define “Open Source AI”. However, even if OSAID 1.0 is released, we will not immediately fully accept it. Instead, we will continue to question whether it is a valid definition of a free and open AI system for Japan’s AI/ML community. It may take years before we decide whether this definition should be accepted, and there is a chance that OSAID could eventually be deemed ineffective. Nonetheless…, if by that time we have managed to prevent open-washing, I would consider it a success.