"Open" has lost its power in the age of AI

Why a taxonomy of access won't democratize AI.

I won't spend time acknowledging the hard work happening in the "What is Open AI?" conversation, because I think self-congratulatory nature of the work, to-date has become part of the problem. The groups defining "open" are increasingly the same, siloed groups talking to one another while the people most affected by AI are left outside the room. Every new definition of "open" seems to reduce the agency of ordinary people rather than increase it. It feels like we're waiting for Moses to come down from the mountain with our instructions.

Call me a curmudgeon, if you will.

This recent paper specifically, just felt like an uninspiring death knell for openness as a term:

"Openness takes multiple forms. Beyond the binary of “open” or “closed,” openness in AI systems manifests through varying degrees of access to data, code, weights, documentation, and governance. " - https://cacm.acm.org/research/unpacking-open-source-artificial-intelligence-toward-a-framework-for-openness-in-foundation-models/

It's similar to what the G7 writes:

The openness of an AI is not binary. It exists on a spectrum, ranging from models that share only weights under restricted licenses, to models that make all elements fully available under open licenses. G7 members and stakeholders are encouraged to recognize and communicate which degree of openness applies to a given AI, rather than using the term "open" without further qualification.

What's striking is that these frameworks mostly measure access to artifacts - code, weights, data, and documentation. They say much less about whether ordinary people can meaningfully influence the technology they use. That's what humans need to know right now.

I do understand the intention here. I have done my own investigation, and learning to understand what openness looks like in the Age of AI, and it's incredibly hard, if not impossible to break down AI as being open - or not. Maybe because it isn't open, generally speaking.

Breaking openness into layers, components, and spectra makes perfect academic sense. But in a moment where a handful of companies are rapidly consolidating AI infrastructure and influence, that framework doesn't answer the question ordinary people are asking: How do I have a say?

If ordinary people cannot meaningfully understand, modify, influence, or govern a system, calling it "open" stretches the word beyond usefulness.

The G7 also argues that AI openness should be community-driven. But which community? And how does that community exercise influence? Today, the conversation remains largely academic, corporate, and siloed, leaving the people most affected by these systems on the outside looking in. We already had an enormous gap in equality and equity in open source, this is magnifying it.

What I have found helpful in the discussion is alignment through governance - measuring success based on how regular people want to, and have the ability to influence technology they use, build and contribute-to. Governance standards (yes, governance is mentioned in the earlier article as a positive) and tools to enable alignment are what we should be building.

Perhaps Governance in service of alignment will merge with the new "open" to advance democracy overall - but that seems a ways down the line. I will be at FOSSY with some peers to talk about this later this week.