An explanation of why there is no “degree of open”

Open Source AI
1

In the conversations around Open Source AI there has always been someone invoking a shortcut to resolve “the data conundrum”: admit that openness is a spectrum. This is a flawed argument that the OSI unequivocally refutes and that the whole Open Source community should forcefully push against.

Freedom is a binary concept: you’re either free or you’re not. You can’t tell a prisoner that they’re free because they’re not chained to a wall and they can move around in their cell or walk in the jail yard one hour per day. That’s not freedom. Those are degrees of deprivation. A prisoner is free only once they get out of prison. Only then can they enjoy freedom in any way shape or form they like.

For Open Source it’s the same: you can have software projects using licenses and terms of use on a range of freedom-depriving conditions and those are NOT considered Open Source. The ethical licenses, the licenses with commercial restrictions, the time-shifted licenses, etc. offer degrees of deprivation, they’re not Open Source because they don’t grant the basic freedoms.

Now think of the variety of Open Source projects released with an OSI-Approved License®. There are projects like SQLite, published without a public roadmap, few committers, tiny or no community around them. Those are Open Source projects.

There are also projects like OpenStack, Eclipse IDE or Kubernetes, organized around strict community rules, hierarchical chain of command, and promises to their users: these are Open Source, too.

There are projects that are Open Source and designed to further the progress of the United Nations sustainable development goals: they’re also Open Source.

All these projects are “Open Source”, they’re not “more” Open Source than others. Some of them may add more promises (like OpenStack’s 4 Opens) or requirements (“be designed to do good for humanity”, like the DPGs.) As a user you’re free to decide to give more value to an open community than a single-vendor project or one that is designed not to harm people. That’s part of the freedoms you’re granted.

For AI systems it’s exactly the same concept: you either have the freedoms to use, study, modify and share or you don’t. There is a gate the AI system, like software, must pass to grant freedoms: before that there is the gradient of deprivation.

I used this image to illustrate this concept. Think of being Open Source as the prisoner passing that gate: what they do with their project’s freedom is their choice.

960×540 316 KB

I’ve been brainstorming with @webmink (his is the prisoner metaphor) and the rest of the team on how to explain this. I’m sharing our current approach here (after using it in recent presentations) to get more comments before posting this more widely.

What do you think of this argument? How would you change the image to be more clear?

[posting this in the AI category because this topic pops up often in AI discussions but as it’s a general topic, really, not only an AI-related one]

3 Likes
2

During last year’s discussions about OSAID, there were times when I felt truly torn. But now I can say it clearly: you are correct. Open Source is a standard that demonstrates freedom, and by its very nature, that is a binary concept.

Before the term “Open Source” existed and everyone was using the phrase “Free Software,” many of us—myself included—watched with dismay as the FSF attacked the X Consortium and the BSD community. I loved copyleft, but I could never agree with the FSF’s attitude toward other communities of freedom. I wanted to regard the GPL, X11, and BSD licenses all equally as free.

It was around that time that the term “Open Source” was born. Honestly, at first I thought it was a bit of a strange name. But while the FSF’s brand of Free Software seemed somewhat doctrinal and philosophical, the “Open Source Definition” was remarkably straightforward: any license that met the OSD was recognized as an equally free license. Because of that, it didn’t take me long to become a supporter of Open Source.

Open Source succeeded precisely because it focused on one extremely simple principle: it defines the criteria for whether a license qualifies as free. If the concept had been something like “half-free” or “80% free,” no one would have paid attention. And I believe it will be the same for AI.

1 Like
4

Freedom is a binary concept is so cool. Open source has been a key driver of new industries and economies, enabling sector-based innovation. However, we often see cloud providers like AWS, Azure, GCP, and Alibaba Cloud forking open-source projects and offering them as services with minimal modifications—sometimes even lagging behind by a version or two. This is a clear misuse of open source and an unfair practice.We should expect meaningful contributions back to the open-source community from such giants. While I strongly believe OSIAD will help shape an open AI economy, I also worry that, too.

5

That’s a big debate we (collectively) haven’t had at the depth it deserves: Does Google misuse the Linux kernel to run its ad business? Is it unfair how GitHub uses git? And how about Amazon’s use of projects made available with non-reciprocal licenses (like the MIT license) to power its very proprietary services and give back peanuts, if at all? Is that fair? Fair for whom? How much is due to a misunderstanding of the license by the original authors? What’s the role of venture capital (as opposed to other forms of capital) in all this?

I call this lack of debate the lost decade of Open Source. BTW, OSI is now being harshly criticized because we pushed to have the AI debate now and avoid ending up in the same dramatic void that surrounds Open Source for cloud and mobile.

1 Like