Today, the values of Open Source are largely foreign in both mobile and cloud. That’s a problem. I’d like to propose how the Open Source community can prepare for the age of AI.
To set the stage, let’s briefly look back to 2016 when the Italian government launched a Department of Digital Transformation. It created a three-year plan to build the nation’s public, digital infrastructure with Open Source software.
The Italian government is known for many things – porn-star politicians, a revolving door of power, huge public debt – and, yeah, most of them aren’t good. This initiative was, by contrast, a massive achievement. They built a portal aimed at IT departments, with a catalog of software platforms, APIs, publishing tools, data and guidelines. They followed the principles advocated for years by the Open Source Initiative, Linux Foundation, Open Infrastructure Foundation and others. It certainly looked like an example to follow.
A beat later, I sighed in disappointment after squinting at millions of lines of Open Source code and documentation. While it was Open Source, there were intrinsic ties to proprietary cloud functionalities, rendering it not very “open” at all. But we can’t blame the government. Over the last decade, Open Source hasn’t widely addressed the challenges of mobile and cloud computing. That means we don’t have an easy way to judge what Open Source means in those contexts.
As we stand at the precipice of a new era of artificial intelligence and machine learning, we can’t afford to make the same mistake.
Open Source’s lost decade
This dependency on proprietary cloud functionalities is a massive problem in today’s geopolitical environment. The European Union has been talking about strategic sovereignty but depending on large American companies for running crucial infrastructure like what the Italian government created, is in conflict with those objectives.
Twenty years ago, when the public administration was commissioning and deploying applications that only ran on HP-UX or SCO Unix, I made myself hoarse advocating for an alternative that worked: “Run it on GNU/Linux!” There was a large community developing code and shipping solutions; we were equal parts cheerleaders/finger waggers.
Today, how could we advise the Italian government? ‘Don’t use those serverless functions, ignore the convenience of compute and storage on demand, let go of the practical advantages of not having to run your own datacenter and invest instead in learning how to do things differently. And ignore that it will probably take you years to learn how to do it, you’ll get there…’ This doesn’t really sound like digital transformation.
Don’t get me wrong: That’s exactly the right thing to say. But how did we get to the point where we don’t have alternatives to suggest? Why did we run out of options?
It’s the result of the Open Source movement ignoring the introduction of two technologies: cloud and mobile. The values of Open Source are largely foreign in these technologies today. Cloud and mobile introduced new ways to develop, distribute and execute software that were not available in the 1980s when the GNU Manifesto on free software was published as a means to protect freedom. That definition is just not easily adaptable to modern computing. OSI is the defender of the Open Source Definition, which has been evolving since the early 2000s, but with the swift rise in the development and adoption of AI technologies, Open Source standards must be established by the community without haste.
Where to go from here
In some ways, we’re back to when GNU software ran only on proprietary Unix workstations and we had to tolerate proprietary software to co-exist with GNU and Linux. The community solved this problem by developing more Open Source software, replacing proprietary components one by one on all hardware architectures (Debian officially supports 10 different architectures and more, counting the unsupported ones.) Over a couple of decades, and thanks to Intel’s “open architecture,” Open Source took over the proprietary systems of DEC Alpha, Silicon Graphics, Sun, IBM, HP and others.
Can we again develop open alternatives for mobile and cloud? The scale of the effort is much larger this time. Besides developing portable code we’d also need to develop the new “open platforms” to run such code on. It’s a daunting proposition: think of how you’d replace the iPhones/Androids or AWS and the thousands of web services the modern devices rely on.
Given that most of those environments already are built on Open Source, we’d also need to create new frameworks to evaluate the openness of these new platforms. We must be realistic about what it will take to protect true Open Source in a much more complex arena.
Catching the next wave: Artificial intelligence
We’ve relied for decades on copyright and copyleft but this approach is showing its limitations with modern technologies. AI and ML are posing an even greater challenge to the Open Source Definition than cloud and mobile did.
AI and ML blend the boundaries of software and data. AI systems introduce new artifacts for which the applicability of copyright law is questionable. Generative AI systems also pose new and intricate legal challenges to the many established understandings of patents and trade secrets. The large quantities of data required to build functional ML systems also attract other laws – from privacy protection to security to non-discrimination and accessibility laws – all the way to basic human rights protections. Much of the existing OSI Approved Licenses are ineffective in these contexts.
Despite the popularity of the term “Open Source AI,” there’s no shared and agreed definition. And despite the popularity of software licenses applied to ML models, not everyone agrees about the applicability of their terms. Case in point, Meta released pieces of its LLaMA model under a very permissive license (the GPLv3) but later sent cease-and-desist letters to distributors of a modified version of that model, contrary to the terms of the license they used.
To bring the principles of Open Source with us for the tech that comes next, we must think hard, carefully and quickly about how to adapt the guiding principles of “open” to the AI/ML field. OSI is calling for stakeholders to join the global drafting process of a definition of “Open Source” applied to AI/ML.
Here’s how you can help:
- Become an individual member of OSI: This will show the wide interest of the community, not just the larger corporate donors. There are three tiers: Supporting and Professional, if you can afford a donation, or you can join for free, if $4/month is too much.
- Donate to OSI: The donations help us increase our independence from corporate donors.
- Follow the AI drafting process: OSI will hold meetings online and in person during the next months, releasing new drafts of the Definition. Leave comments and spread the word.
- Sign up for the Deep Dive webinar series: Hear experts’ thoughts on “Open Source” in the AI world.
Join us for the next chapter by engaging with the Open Source community. Make your voice heard and help us avoid a repeat of the lost decade of Open Source.