As part of the upcoming May hackathon, we'd like to showcase some AI models that tool developers might think about using in their work. While some models are unequivocally open (code is open-source; model is also explicitly open-source), other models that we are curious about have some use restrictions for the models -- e.g., non-commercial clauses (example) or ethical use restrictions such as not using for the purposes of creating disinformation (example). These models do not currently meet the OSI definitions but OSI's recent report on this indicates there is active discussion about what Open and AI should look like. While I don't think we're ready to take a stance on whether these use restrictions are acceptable for our systems, I do think they qualify as a reasonable temporary exception for the purpose of testing / exploring for the hackathon.
Why I think it qualifies for an exception:
- We still would only be using OSI-licensed code (MIT, Apache, etc.) -- this is just about the model weights which are the more debatable area.
- This is a temporary usage for purposes of the Hackathon and I'd be happy to discuss a mandatory removal date after the hackathon.
- The usage is not for supporting a tool used on-wiki at this moment etc. but for the purposes of exploring model outputs -- e.g., akin to testing compatibility with developer needs.
Miscellaneous:
- Relevant project: T332218
- Task created following documentation at https://wikitech.wikimedia.org/wiki/Wikitech:Cloud_Services_Terms_of_use#What_uses_of_Cloud_Services_do_we_not_like? and per guidance from IRC.