Tagged for OEG Connect: The Open Source AI Definition - HackMD

What’s of interest? The Open Source AI Definition - HackMD

Tell me more!

An Open Source AI is an AI system made available under terms that grant the freedoms to:

  • Use the system for any purpose and without having to ask for permission.

  • Study how the system works and inspect its components.

  • Modify the system for any purpose, including to change its output.

  • Share the system for others to use with or without modifications, for any purpose.

  • Precondition to exercise these freedoms is to have access to the preferred form to make modifications to the system.

h/t Martin Dougiamas: "The draft definition of Open Source AI is really …" - Open EdTech

Where is it?: The Open Source AI Definition - HackMD

This is one among many items I will regularly tag in Pinboard as oegconnect, and automatically post tagged as #OEGConnect to Mastodon. Do you know of something else we should share like this? Just reply below and we will check it out.

1 Like

As noted above, this came my way via a Mastodon post by OE Global board chair @moodler The reference link is a part of and comes from the Open Source Initiative, who is championing this drive for definitions of Open AI (the concept, not the company that just uses the name!).

Also, note the posts reference to a paper on a Model Openness Framework paper:

I’m not deeply in the know here, just treading water to be informed, but defiintely interested, and am curious about what AI efforts now fall into following this definition? And while this is for technologies, is there room for working out attributes, if not definitions, for what openness with AI in education might look like - e.g. the outputs from UNESCO).

These are broad, are there ways to make them understandable, approachable to practitioners?

Hey hey - I am more in the know on this, in my guise as a member of the Board of the Open Source Initiative. Happy to field any questions. We also say “Open Source AI” to distinguish from a rather poorly named commercial entity who are anything but…

The way that I tend to see this is that we have a bunch of licenses, based on underpinning definitions, that enable a whole host of open education sharing practices. We are able to easily share software because we have open source licenses (thank you @moodler amongst many others), and we are able to easily share OERs because we have Creative Commons licenses. I see a clear definition of open source AI as the underpinning enabler to being able to share open source AI things in the future.

I think there’s a lot of alignment between this work and UNESCO thinking in the AI space, especially in education and the need to have more transparency and ownership of edtech as digital public goods. OSI and UNESCO are both involved with the Digital Public Goods Alliance and workshops around the development of the Open Source AI Definition were held at their 2023 meeting in Ethiopia.

This report from Stanford might also be interesting in terms of the use of open models (taking into account that in the absence of an agreed definition of Open Source AI, their current methodology and OSI work might not perfectly align) https://aiindex.stanford.edu/wp-content/uploads/2023/04/HAI_AI-Index-Report_2023.pdf

It’s an interesting picture nonetheless. Very happy to talk about this stuff more, and of course there’s loads of info along with the current version on the OSI website.

1 Like

Thanks, Anne-Marie, you are my guru here and glad to see your response.

It seems that the licensing efforts are on the platforms, which is of course so important, but I imagine many here are trying to sort out the complex licensing of content that comes out. Or that goes into the bins.

So I wonder if/where there is room to develop a similar set of freedoms for what educators, students, creators can do with the “stuff”?

It’s messy, but intriguing, right?

I think the big problem with the output of genAI (and I am no lawyer here) is that it is unclear whether they are covered by existing IP law with the balance of opinion tending towards they are not. IP law in many countries was written prior to this tech existing and so ideas about human authors and what counts as “original” reflect this. Any efforts in working out how to use and share the outputs of genAI need this bigger legal picture to shake out first.

I’m also not sure that we need new licenses. You’ll note that the Open Source AI Definition doesn’t specify new licenses. Simply that OSI approved / OSD compatible licenses must be used. This reflects that lots of existing licenses are fit for purpose.

I’d be keen to see whether once the bigger legals issues shake out, whether CC licenses will continue to do the heavy lifting in any AI generated OER space.

(There’s also a point in here about if there’s no copyright then potentially there’s less impediment to sharing, but I hesitate to make statements like that because there’s a whole other minefield of issues around what goes in and comes out of genAI).