Friends of Open Source,
Last week saw the launch over the strenuous objections of censored community members of the OSI’s Open Source AI Definition (OSAID), which conflicts with the 25-year-old Open Source Definition for any software that “infers, from the input it receives, how to generate outputs” (i.e., virtually all software).
A past OSI board declared “a process that is not open cannot be trusted to produce a product that can be considered open”, yet when my statistical analysis revealed serious voting irregularities involving Meta’s lawyers, OSI leadership admitted to it and apologised only to uphold the flawed decision to exclude training data (which we believe was made last year in a since-deleted private Google Group). You can see the details and data in my presentation to Kwaai on the subject: Lies, Damned Lies, and Statistics: The path to a meaningful Open Source AI Definition
It has already drawn heavy fire from all directions for demonstrably failing to protect the four freedoms by not requiring the source (i.e., data). Just from within our own industry (with many others on my blog):
- OSD author and OSI co-founder Bruce Perens has declared “the [OSAID] is less than Open Source”, that “the training data is ‘source code’”, and that “you can apply the original Open Source Definition to machine learning” (in response to a resigned board member’s questioning the mandate of membership and requesting a referendum on it).
- Software Freedom Conservancy writes Open Source AI Definition Erodes the Meaning of “Open Source”, demands the OSAID be rescinded as “the moral and ethical way out of this bad situation”, and promises to run a ticket for the OSI board to do exactly that.
- Debian GNU/Linux — who already refer to models validated by OSAID as ToxicCandy — are preparing a General Resolution declaring OSAID “unacceptable”, observing it’s “like not releasing the source code at all and claiming that it’s still free software because ‘a skilled person’ could rewrite it” (their description is clearer and language more colourful).
- FSF is working on freedom in machine learning applications and have determined that “we cannot say a ML application is free unless all its training data and the related scripts for processing it respect all users, following the four freedoms”.
In order to protect our community and those dependent on it against what is potentially a dangerous fork — and anticipated future “harmonisation” — of the Open Source Definition, we have copied it and all approved licenses to a new stable home at https://opensourcedefinition.org along with lost prior versions from internet archives, and a Git repository for same. We encourage you to clone and publish it yourself, and refer specifically to v1.9 until the community comes to clear consensus on future updates (if any were to achieve a high bar akin to a constitutional amendment).
We have also just launched the Open Source Declaration at https://opensourcedeclaration.org and would love you to join the dozens of individuals and organisations who have already signed this simple, clear, and neutral statement:
We declare that Open Source is defined solely by the Open Source Definition (OSD) version 1.9.
Any amendments or new definitions shall only be recognized with clear community consensus via an open and transparent process.
I will likely be permanently banned from the [OSI] community for drawing the matter to your attention as others have already been threatened, so don’t reply [t]here. We look forward to welcoming you on the other side at https://discuss.opensourcedefinition.org/ where we are hosting an uncensored Discourse server just like this one and can have an open exchange on the future of Open Source.
This is particularly pertinent given today’s news from the US, and we are now on the clock to preserve the powerful differentiation of Open Source.
Kind regards,
Sam Johnston
Kwaai Open Source AI Lab