Why Washington Gets Open Source AI Wrong | Matt White

Matt White (0:00)

In the open world like that, you can't really have a kill switch for local AI, for AI that's running on your Spark, or something, or on your laptop. And so the concept is very abstract. I don't see how that can be implemented in software, and the idea that we need a kill switch right now, I just, I'm not really in agreement.

SPEAKER_2 (0:22)

Hello everyone, and welcome back to MTS. Today, we are here with Matt White. Matt is the founding executive director at the Open Intelligence Foundation, a visiting scholar at Columbia, and the former global CTO of AI at the Linux Foundation, and CTO of the PyTorch Foundation. That's quite the stacked resume. So Matt, welcome to MTS.

Matt White (0:39)

Thanks for having me.

SPEAKER_2 (0:40)

It's a fun conversation. I know we got a chance to speak at the Open Source AI Summit in San Francisco about a month ago, but I'm excited to have you officially here at MTS, joining us in the studio. I think you have such a great background because you understand the entire Open Source ecosystem from top to bottom, and also you understand kind of the policy side to these things as well. So yeah, this should be a great conversation.

Yeah. I wanted to ask you specifically on this policy side. You were just in Washington, and you were on the Hill, you were talking to legislators about Open Source AI. What do you think they fundamentally misunderstand about Open Source?

Matt White (1:20)

I think the focus is generally on AI itself, and not looking through the lens of there are different ways to distribute AI. And so when you look at things like kill switches, and you look at putting fairly stiff regulation in place or proposing bills that would fundamentally slow down the frontier or try to identify early signs of superintelligence, there are unintended consequences that affect research within the open labs and within academia.

And so I think there's just not a really good, by and large, understanding of the technology itself, because a lot of the latest moves have been driven by what we've seen in the public headlines and around concerns around hacking of sites and cyber capabilities, but also, you know, beyond that, it's really like looking at the entire ecosystem and how open source plays a critical role in the economy, in education and academia and security research. Like, there's a lot of applications that open models touch, and that is not being looked at with by legislators, really.

SPEAKER_2 (2:40)

Yeah, I want to understand why this isn't being looked at by legislators. You say right now our legislators have an incorrect or not holistic view about what AI even is.

You shared that kill switch is a big theme that comes up within policies. What does a kill switch even mean?

Matt White (2:58)

Yeah, again, it's such an abstract concept. I think if we're looking at something that's physical, like in the worst case scenario, that's pulling a breaker or some kind of cutting power from the power going into a data center that's powering the AI that's there. But in the open world, you can't really have a kill switch for local AI, for AI that's running on your Spark or something or on your laptop. And so the concept is very abstract. I don't see how that can be implemented in software.

And the idea that we need a kill switch right now, I'm not really in agreement.

SPEAKER_2 (3:49)

Where do most of their policies and understanding about AI comes from? Because my understanding would probably be if you're a legislator, this is one of the top issues is what everyone's thinking about. What do you want to have the most holistic understanding? So just like you were at the Hill, I just want to understand how are they gaining their understanding about AI in general to then make these policy decisions?

Matt White (4:11)

Well, they generally rely on their advisors and they'll have a tech policy advisor.

But these folks, they may even have a panel and have some advisors on that panel. But by and large, there's no representation from open source and there's really not very good representation on the technical side, like the deep technical side of Brown capabilities.

And it almost feels a little unweighted, like the information that is being gathered is not sort of like neutral. It's more following the sort of narratives right now around, well, yeah, AI is just so powerful, like we need containment, we need regulation in place to slow down the frontier and pace the frontier. And these sort of narratives, which you definitely want to try and collect information from all sources. And one of the things I've been doing is putting myself out there to help educate on the importance of open source. The entire stack is PyTorch, VLM, Kubernetes, Linux, like all of these different components contribute to the end product, right? And so, and then beyond that, it's open models themselves. And so, in a world where we're approaching, intelligence fundamentally becomes a fundamental part of our lives and something akin to like broadband or the web or electricity, we want to make sure that everybody has the right and access to affordable intelligence, right? And capable intelligence. And so that's kind of a lot of my conversations we've had around that.