---
id: "workshop:what-is-public-ai-and-why-it-matters"
type: "InnovateUS workshop"
title: "What Is Public AI and Why It Matters"
url: "https://innovate-us.org/what-is-public-ai-and-why-it-matters"
date: "2026-03-11"
duration: "60 minutes"
series: "Democratic and Public AI: Practical Strategies for Buying, Building, and Governing AI"
instructors: ["Joshua Tan","David Eaves"]
moderators: ["Beth Simone Noveck"]
recording: "https://www.youtube.com/watch?v=4I2Kt9XPHxQ"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# What Is Public AI and Why It Matters

InnovateUS workshop · 11 March 2026 · 60 minutes

As artificial intelligence spreads more widely, there is concern over centralization and privatization in the ecosystem. Public AI emerges as a counter to this tendency–but what is public AI, and what does it look like in practice? Building on the procurement and market dynamics explored in Workshop 1, this session defines what “public AI” means and the arguments for how governments can take an active role in shaping it. Participants will explore the risks of outsourcing essential capabilities to private platforms, the opportunities of shared and open infrastructure, and the principles that distinguish public AI—transparency, accountability, interoperability—from purely commercial deployments.

## Learning goals

- Define “public AI” and explain how it differs from purely private or vendor-driven AI solutions.

- Connect AI procurement choices to broader questions of democratic governance, public service delivery, and rights.

- Identify core principles for public AI (e.g., openness, interoperability, accountability, human oversight).

- Begin to map how public AI could apply in participants’ own agencies or jurisdictions.

## Speakers

- **Joshua Tan**, CTO, Current AI & Public AI
- **David Eaves**, UCL, Tech Advisor UK Gov and South Africa Gov

## Moderators

- **Beth Simone Noveck**, Founder of InnovateUS and Director, Burnes Center for Change and the GovLab

Part of the series [Democratic and Public AI: Practical Strategies for Buying, Building, and Governing AI](https://innovate-us.org/workshop-series/public-ai-buying-and-building-for-civic-use).

Recording: https://www.youtube.com/watch?v=4I2Kt9XPHxQ

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/what-is-public-ai-and-why-it-matters.txt (plain text) and https://innovate-us.org/what-is-public-ai-and-why-it-matters.json (every segment with its start and end second).

**Joshua Tan**

[0:00](https://youtu.be/4I2Kt9XPHxQ?t=0) First is Beth Novick. I am founder at Innovate US and delighted to be here moderating this session today. I also am a professor at Northeastern University in my copious spare time and I'm joined today by two fantastic speakers. We have David Eves here joining us from London. David is a professor at UCL in London. He is also tech advisor to the UK government and South African government and I can tell you last week he was advising the Canadian government because I had the pleasure of being with him for those meetings. We will put David's bio, very lengthy bio, in the chat and more about him so that you can read all about his many accomplishments. In addition, he is joined by Joshua Tan. Josh leads product at Public AI. He is also a founder, co-founder at Medigov and again we will put his bio and details in the chat so you can read about both of them and their extensive expertise. Both of them have, again, significant not only knowledge but actually practical experience leading efforts to move us towards more public interest, public-minded AI. I want to talk in just a moment about what that means.

[1:27](https://youtu.be/4I2Kt9XPHxQ?t=87) Let me just say one framing note for this whole series and this is the second in six conversations about public AI. You see Luca here off camera who will be joining us for one of our future sessions, I think speaking next time. This is meant to be a very practical set of conversations around what we can do aimed mostly and most of the people in the room here are people in government, in the US, in Europe, in other countries and our goal is to focus on what we can do practically, collectively to move the ball forward toward more mission-driven, more public-interested, more sovereign, more public, more democratic AI and we'll talk about the differences in these terms. So we welcome your conversations and your experience and pushing our speakers towards those practical pieces of advice of what we can do. All right, let's dive in. Without further ado, I think we shared the poll results. I hope we did. If we didn't, we'll share them real quick so people can see who is here and you can just get a sense. But while we're doing that, let me start us off by the very simple and yet impossible question of what do we mean by public AI? When we actually use the term public AI and Josh, I'm going to start with you, what are we actually describing? Is it public ownership of models, of infrastructure? Is it open source? Is it government building for itself? Is it interoperability across vendors?

[3:03](https://youtu.be/4I2Kt9XPHxQ?t=183) What does this mean? And then David, I'm going to ask you the same question and see if you have the same answer. Thanks so much for having me here, Beth. I'm going to go out on a limb and say David and I probably have different answers, but I think it'd be interesting to dig into what exactly are the differences. Usually when I talk about public AI, I usually think about it in terms of as a form of public infrastructure. Public AI is AI as a form of public infrastructure in the sense of highways or libraries or water or electricity, or you can even think of the internet itself as a kind of public infrastructure. A lot of this is effectively operated and funded and developed by the public sector. Highways are just built by states, right? But many forms of public infrastructure are provisioned in lots of different ways.

[4:11](https://youtu.be/4I2Kt9XPHxQ?t=251) So like water and electricity, for example, depending on what jurisdiction you live in, oftentimes that looks like a private company, but regulated in a way so that the output is effectively looks like a form of public infrastructure or public good that everybody kind of gets access to. The internet is a set of public protocols operating mostly on private infrastructure. I think largely speaking though, like for me, the summary point is like some version of public infrastructure that is accessible to the broader public that is applicable to it. And finally, that is permanently public, that is in some ways serves as a kind of stable infrastructure that their people and, you know, citizens and the population at large build on top of. David, agree? Agree to disagree? And again, we put in the chat various articles where you, both of you talk about this at some length and we'll repost those as well. Tell us though, what's your short answer?

**David Eaves**

[5:22](https://youtu.be/4I2Kt9XPHxQ?t=322) It could be slightly awkward if I disagree to us, Joshua, since I'm on his advisory board, but let me say this, which is, I do think there's three distinct words that we're dealing with here. So the most important one is public. And I think that the key thing for me about what is public in kind of public infrastructure is that it's, it has a focus on inclusion and access that private does not have. And the kind of interesting thing about public is it's really about finding those leverage points where you define what is the intervention that usually the state is making to define this piece must be accessible. And that's why like people talk about the electrical grid as public infrastructure, even though actually often the provisioners of electricity are a combination of state owned companies, private sector companies, co-ops, like in the United States, you've got this crazy mix. But what is public with it is there's a regulatory regime that basically says, Hey, if I don't pay my bill, you don't cut me off right away.

[6:26](https://youtu.be/4I2Kt9XPHxQ?t=386) Because if you cut off my power, like the state comes, takes my kids away, all sorts of other bag knock on things that happen. And my ability to effectively participate in society, both as a consumer, but also as a citizen becomes deeply impeded in that moment. And so I don't know if AI needs to be public infrastructure in the same way that electricity does. But I think there is a question of like, do we want to have a public option that focuses on inclusion and access and ensures that everybody is benefiting as opposed to just those who can afford a private infrastructure? Lots to dig into here, especially as we all watch our electricity bills rise. As the Straits of Hormuz become less accessible and energy prices go up. We will see what this means in terms of procurement in terms of regulation, in terms of again, what we do to move towards this vision. So let me sort of ask you about why this conversation is happening now. And obviously, we know why this is the AI moment and why there's this focus on access to AI. But governments have long procured technology, right? We're the largest collectively governments in every market are the biggest purchasers of technology in that market. What is different now about AI than it was about digital? Or is it just that we failed with digital? We should have done this 20 years ago and we didn't. Is there something qualitatively different that makes this conversation more urgent? Or is this really more about the geopolitics of American AI about Trump that is sort of giving rise to this conversation?

**Joshua Tan**

[8:21](https://youtu.be/4I2Kt9XPHxQ?t=501) I can jump in really quickly. I think part of the narrative out there is, which is being promoted largely by the AI companies, but also by folks outside of them, that AI is a kind of exceptional technology. And it's so important and it has going to have such dramatic impacts on the economy, labor, and on government services, that it requires a different approach, right? So that is one potential motivator to say like, oh, it's so important. It's so deeply ingrained, for example, expresses of culture, governments should be involved in making sure it's intervened in a way to make sure it's really broadly accessible from the very get go. I think this kind of reasoning is very active in Europe and many countries where the models are necessarily representative of the native culture or the native language. Most models right now are coming either from the US or China. I think there's another mode and I'll try to keep my commentary that AI is a very kind of normal technology, that it is kind of like previous generations of digital and other kinds of technology. But even in these cases, I think it's still important to think about like, well, how do we make sure that if this becomes a commodity and investment drops off, that people aren't left behind, that regions, whether rural or sort of specific communities are getting access to the benefits. I mean, clearly this is a useful technology that we are sort of promoting access across broad set of communities. I think that's another rationale for it was like, okay, I in this moment, I said to David.

**David Eaves**

[10:20](https://youtu.be/4I2Kt9XPHxQ?t=620) So the first thing is, even if this is an exceptional technology, this is not the first time that humanity has dealt with an exceptional technology. We just may not remember dealing with an exceptional technology before, but it's not that humanity has never dealt with it. I struggle to think that AI is necessarily going to be more profound than say the invention of electricity. Electricity is a deeply, deeply profound invention that completely changed every aspect of our society. We've all grown up with it, so we don't think of it that way. But there was a pre-electricity world where literally the world was dark after it got dark, unless you were wealthy enough to afford relatively expensive candles. And then there was a world where suddenly people had power and the ability to see and like could do things at night, like the whole world opened up. So this is not the first time we've been through a transformation like this. So I'm very cautious. The other thing is what I'm really cautious about, I think one of the things that makes this conversation so challenging right now is that the conversation is deeply speculative.

[11:21](https://youtu.be/4I2Kt9XPHxQ?t=681) The models for how this industry is going to play out, the models themselves are like, what is the winning approach? Is it going to be highly centralized models? Is it going to be more decentralized approaches? Is AI going to work at the edge, like on your device? Is it going to have to be centralized? Like if you knew the answers to these questions, you should not be participating in conversations like this. You should be investing money because you're going to be a billionaire. The part of the problem is like there's so much unknown. And as a result, everybody's fear has come into this conversation. And so what is AI and why is it consequential? I think we don't know. And that's actually part of what makes the conversation so difficult to have because what you end up with is people who are around the table who have different fears, all are pointing at the same thing and think they're having the same conversation. But some people are scared of big companies. Some people are scared of the United States becoming dominant. Some people are scared about a world where all sorts of jobs are going to evaporate. Some people are like, there's a thousand reasons why people are scared. And some of those things may come true, it may or will not. But because everybody's scared, we think we're having a singular conversation, but we're actually having very, very different conversations.

[12:28](https://youtu.be/4I2Kt9XPHxQ?t=748) So this is 100% right. And what I was alluding to when I sort of referenced sovereign AI, democratic AI, mission driven AI, open source AI, public as the, perhaps as the umbrella term that is subsuming all of those. But do you think that these different conversations are just a natural outgrowth of the fact that we are at the beginning of all of this and trying to figure it out? Or do you think that, let me speak specifically to the language of AI sovereignty, at least only briefly. Again, we have a lot of state and local folks here, and I want to focus on like, what are we buying and what are we doing? But I do want to call this out for a moment because there's obviously a geopolitical conversation here. There is a significant AI policy in the US is all about giving primacy to US companies and investing in US AI industry. And therefore there is a counter reaction called AI sovereignty, which is no, no, we're not going to buy American tech. We're going to buy some other tech. Is this a helpful conversation or is this a distraction? Because it just becomes competing my company over your company. And it distracts us from what we really are getting at when we talk about public.

**Joshua Tan**

[13:54](https://youtu.be/4I2Kt9XPHxQ?t=834) Distraction. I think in many ways, it's for me, like the flip side of public AI, because really within, first off, it's actually not just like a European versus America or, it's not actually not as focused. I work, for example, with many US states, for example, Maine. I've even talked to people in Utah as well. And there's a sense of like, hey, sovereignty exists also in the context of like, hey, there are communities in states in the US who don't really like just taking whatever is given to them from San Francisco. Nothing against people in San Francisco, of course, or even necessarily the big tech companies. But the point is like, people want to have their own local versions. I think that is a form of sovereignty. That's really important. Are there other kinds of sovereignty that are important as well that are requested by, desired by someone like the UK or Spain or Italy or Germany? Yes.

[15:00](https://youtu.be/4I2Kt9XPHxQ?t=900) Of course, need for sort of more security in how the data gets processing, you know, how the jurisdiction, like where the data jurisdiction sort of lives. I think these are all part of a larger conversation that are not at all distraction and are quite relevant no matter what level of the stack or what level of the government deployment you are, local versus state versus national. David, you are sort of in Europe, quasi in Europe.

**David Eaves**

[15:34](https://youtu.be/4I2Kt9XPHxQ?t=934) You're in fake Europe. So, thoughts on this? I'm in the part of Europe that denies that it's part of Europe. So, I think that like for me, like a really, the interesting thing is I think there is absolutely a sovereignty question that's happening in AI in the same way that there's absolutely a sovereignty question that's happening in cloud. Like this is not a question that like restricted to just the AI conversation. There's a sovereignty question that's taking place across the entirety of the technology stack, especially if you're not inside the United States. I happen to be Canadian. The United States president keeps referring to Canada as the 51st state. Everybody in Canada is highly attuned to this fact and is deeply concerned about the dependencies the country now has on American technology in a way that that could be used against them as leverage for economic or other kinds of political concessions. And if that's true of Canada, you can understand that that's gonna be true of many other countries in the world. So, I don't think we can pretend like there's not a sovereignty question at play. I think the bigger question for me is the mistake I'm worried that we might be making is that what we're going to go to instead is a world where we're going to want to, everybody's going to want to have a national champion or a local champion of what their AI or what their cloud or what, you know, whatever technology it is their solution is going to be. And then actually, you could end up doing that in a world that leads to a thousand different flavors that are all non-interoperable and lose us the ability to connect with each other and exchange ideas and trade. And so, what I think is so important, so interesting is instead is like really focusing on what are the standards we're creating that allow for mobility or transference across vendors so that if you are not happy with the vendor, like for me, the most interesting way of dealing with sovereignty is be able to move from one provider to another. And again, I'm assessed with electricity, but that's what we did with electricity is, you know, we gave people

[17:34](https://youtu.be/4I2Kt9XPHxQ?t=1054) limited concessions, but we created multiple companies that could provision electricity, but we made them all provision that electricity in the same voltage and the same amperage so that you could move from one provider to another provider. And I think that kind of optionality is what we're really interested in. So, this idea of public AI through interoperability as opposed to ownership or, you know, government has to go out and build all its own stuff. I want to push you a little bit on just teasing out sort of the problem that that solves for us and what problem it doesn't solve for us. And maybe let me give you a very concrete example. You know, in the news recently, Maryland announced that it was signing an enterprise-wide deal with Anthropic. Massachusetts announced that it was signing an enterprise-wide deal with OpenAI. Again, these are sort of apple and oranges right now, but I want to bring us back to something really concrete. Like, what's problematical about that for you? Is it problematical? Should they have signed those? Or to push towards a more interoperable future, what should they have asked for before signing those? And then Josh, I'm going to put you on the spot.

[18:54](https://youtu.be/4I2Kt9XPHxQ?t=1134) So, I think one of the challenges right now with AI is the technology is so early and so, like, the practices are so emergent that it is actually very hard to know what to ask, which is why I'm kind of like, I would want to be engaged in a lot of experiments and very few large bets at this time. And if I do engage in large bets, what I'm not sure is possible in AI, which I think was possible in the electrical era, is that those large bets have a termination where you get to take over the service. So, one of the things that happened back in the 1850s was, you know, you got to build a facility that generated electricity and you got a mandate to run that and you got to extract rents and have a monopoly for 20 years. But at the end of it, you handed it over to me and then I got to allocate who was going to operate it next. And embedded in that was a notion of interoperability and portability that would allow me to assign different vendors to go do that or to operate it myself. And what I want to make sure is that states right now are not signing deals that are going to lock them into using a provider for the next 50 or 100 years that they're never going to be able to disentangle themselves from. Josh, over to you.

**Joshua Tan**

[20:14](https://youtu.be/4I2Kt9XPHxQ?t=1214) Okay. I think it kind of goes back to the earlier point about locality, the way I would kind of motivate this response is like, ultimately, if you're, let's say, an official, and I think I saw the poll, like most of the folks here are operating at a state slash provincial level. And, you know, what you care about is like, you know, taking care of the citizens in your state, making sure businesses are successful, figure out what are like growth strategies in these different regions. I think it's really, really, what I kind of try to emphasize here is like, if you want to do that, I'm not going to judge you if you, you know, chatty BT, or anthropic or whatever it is, because I don't know your local concerns. But I do know that when I talk to a lot of the provinces and states in the countries that do care about what I think of as like, local control, sovereignty, benefits occurring to local businesses and individuals rather than, you know, leaving it, you know, a big part of the conversation is around optionality.

[21:28](https://youtu.be/4I2Kt9XPHxQ?t=1288) How do I make sure that I have the ability, like, if I do this small experiment here, that doesn't lock me in forever, like 20 years down the line? How do I make sure that I think interoperability is a big piece of this, right? But, and what I usually sort of the answer there, and this is why I think like, we actually do kind of know how these systems work already, David, I feel like we can start like making very concrete recommendations in terms of like, you know, you should control your own orchestration, right? If you're a, like a state, or an agency, you should really try quite hard to avoid getting locked into a particular user interface where you're like, buying from a business development person. And then like, it comes like a fully packaged thing that you can then use and deploy, like, to the degree possible, you should try to just be, you should, you should just be consuming an API. I'm like, you can, you can call like, the traffic API, you can call the anthropic API, it doesn't really matter. But if you're control, if you're consuming these companies resources at that technical level, you're preserving optionality. And this works at different parts of the stack for different kinds of tools.

[22:43](https://youtu.be/4I2Kt9XPHxQ?t=1363) But again, it's like control how you orchestrate your own kind of AI solutions. That's a big part of this, I think, answer. So we're getting a lot of comments in the chat and questions in the Q&A, reminder questions in the Q&A, and then stuff in the chat, you should feel free to answer one another. David, you're doing yeoman's work here having two conversations at the same time. So I want to bring some of that expertise back into the out loud conversation as well. And let me, I'm just to be really tough on you. Jorge Valenz asked the question, Jorge leads AI in Miami. And digital, I think, Jorge, forgive me if I'm butchering your title, but all things techie in Miami for Miami-Dade County. And he's, you know, asking the question, like, what do I do tomorrow, essentially? You know, love the description of what public AI is. And how do you see AI branching away from our very market driven approaches? But what do you see as the first steps? And so I want to talk to you both about the technical side of this, the procurement side of this. And then Josh, you've also written a lot about institutional stewardship and what the institutional and governance side of this looks like. So let's talk for a moment first about the technical side. I want to come back to him for a moment to the, I know we're talking about interoperability and standards, and we're going to ask of commercial vendors, but there's also a lot happening in the kind of training building open source space. So first, let me just tell us what you've seen, what seems to be working, what's concrete, is there something Miami can do tomorrow on the building side?

**David Eaves**

[24:30](https://youtu.be/4I2Kt9XPHxQ?t=1470) So building training tech? Yeah, let me let me kind of say a few things, some of which may not Okay, yes, David first, then fine, then Josh, like, sorry, may not be completely popular. But the first is, I'm like, I'm actually not gung ho on my government's like moving unbelievably quickly on new technologies. Like there is a lot that is unknown in this space, that the benefits of using an AI, at this point, must be very significant for you, in order for you to justify moving this quickly and potentially locking in on models and approaches that don't necessarily make sense. Like, I, I'm always very interested in like, okay, well, how's the private sector working this? How is it making it? How is it successfully using it? Because it's going to adapt much more quickly than the than the public sector is. And I like struggle, like even in the public sector, I like struggle, like even in the private sector being like, what are the really concrete institutional level ways that people are using AI, I get that there's a lot of people who are using it as individuals making their individual jobs more effective, but like actual institutional wise, here's how we reorganize here, we restructure ourselves. And here's how we read our business processes. There's not a lot of that yet. And I get quite nervous when my governments are like making those types of bets, because the companies want to use you to invest your capital to figure out what those bets, the companies win when everybody's making lots of bets to figure out how how these things will work. Because they don't need all of those bets to work, they just need one or two of them to work. And then they can go to the world and be like, oh, here's the bet that works. And all the people who made the wrong bets, those companies, the AI companies don't suffer the consequences of that the people who made those bets do. And I don't know the degree to which I want my public sector players being part of making those bets as opposed to consuming the lessons of those bets. But what I do think we can do is there's a lot of muscles that you could be building right now that will be good for you, whether you're going to use AI a lot in the future or not.

[26:30](https://youtu.be/4I2Kt9XPHxQ?t=1590) Like so for example, organizing your data, making it making it accessible and easy to access, like these are things that most governments are still failing to do pre AI. And the people who are going to move really quickly on AI are the people who got a whole bunch of basic things right, you have to get all these basic muscles right before you can use those really big muscles. And people want to cheat and go use the big muscles. So my like, my thing was like, get these things like get unique identifiers for your citizens, get your data organized, get a data exchange set in place, have a single sign on system, have a way where you can move data around portably. So you can actually if you get AI agents, you could use that to serve them well. And short of that, at least your own 311 and 911 operators can move data can access data and serve your citizens well. So let's get a whole bunch of basic things right, and that will position you to be successful. Josh, let me bring you into this. But I but I also want the government folks here to push back because I want to just be very clear, like some governor wakes up, some mayor wakes up and goes, why are we behind on AI sign this thing? So, Josh, over to you. But I will let people weigh in on what they need to tell their politician.

**Joshua Tan**

[27:38](https://youtu.be/4I2Kt9XPHxQ?t=1658) I'm gonna, I'm gonna say, like, push back a little bit about David, to be fair, I don't think, David, correct me if I'm wrong. I don't think you're saying like, you know, like, let's not do public entrepreneurship. You know, like, there needs to be innovation in the public sector. But you know, there are certain kinds of innovation that like, where like, makes more sense, you know, you shouldn't be like building your own agentic frameworks and releasing them and trying to like get market share, you should be like, figuring out like, yeah, doing stuff on data, doing new stuff on how ownership could be allocated or figure out ways to empower like local companies, newbie vendors, instead of just taking the big options, right? I think there's all sorts of ways we should be innovating. To be clear, though, I think there is product innovation that can't happen within governments or within the public sector or adjacent. I personally work with a lot of fairly radical public servants, they're usually called librarians, but they are kind of crazy. And you don't tend to go out there. And it's been quite awesome. So we do this, I'll just get like, give an example, right? In terms of like, tomorrow, what could you potentially do?

[28:42](https://youtu.be/4I2Kt9XPHxQ?t=1722) This is like a pilot that we're literally doing. I'll just put a quick link in here. It's a project in combination with the Library of Congress, the Utah State Library, the New Jersey State Libraries, but there's also recent Georgia, Massachusetts, etc. Where we're basically like trying to innovate, co-design these, you know, accessibility platforms, we call them like sort of drop in AI makerspaces with librarians and dropping into libraries, right? You know, reusing and repurposing existing public infrastructure, libraries, as the kind of deployment vehicle for new kinds of AI services. And to be clear, actually, one of the key things here is like, we are not trying to invent all the new AI services, we are distributing and orchestrating access to these tools. And we can always swap them out, we can swap, you know, like the replet out for a different coding service. And we decide like, it's the new ones better, or cheaper, or, you know, better for the public, right? So these are the kinds of tests and pilots that can kind of get started immediately. But also, so I just kind of briefly emphasize, I think there's also a lot of work that we could be doing. Day one type things on building more support for public compute. Now, that's like a much longer conversation. Maybe we're gonna talk about that at some later point. I think A big part of this public AI idea is that, you know, what is infrastructure? It is stuff that other people use, your population, your businesses, your citizens use to build really good things or to do really cool things with. And I think if you can offer, you know, compute is one thing, there are other like data is another. If you can provide these as assets, you can really unlock a huge amount of creativity and productivity from, you know, your community, that I think can really be powerful in a way that, you know, sometimes only you as governments can actually do.

[30:41](https://youtu.be/4I2Kt9XPHxQ?t=1841) So let me push you now on the sort of institutional stewardship question or conversation, sort of in order to achieve some of those benefits, what have you seen in terms of, or what are you advocating for in terms of institutional or governance guardrails? Again, institutional configurations, way to think about how you're organizing talent, doing training, anything that you're seeing that sort of can promote this idea of institutional stewardship, so that governments are actually in the position to manage public resources like this, or to be better consumers of consumer resources. Josh, why don't you go ahead and then we'll go over to David. I can jump in a little bit on this. I suspect I'm going to be give a weirder answer than David.

[31:43](https://youtu.be/4I2Kt9XPHxQ?t=1903) So there's a one particular version of a solution to this that we've been kind of building. So for those of you who don't know, public AI, really public AI is not actually an organization. It's best thought of as like kind of the open source movement. We build, you know, but we also do a lot of work, you know, advocacy around the world. But in particular, we do a lot of work with the country of Switzerland. Among other things, we, the primary deployers of the Swiss, you know, national model to Switzerland itself, as well as to the world. And one of the things that I think, you know, really interesting developments here is we've recently started, we actually just like a few weeks ago launched a cooperative, like a Swiss cooperative to manage and govern this wider infrastructure, like this infrastructure that we're already built, that's deploying the national model. That's like a kind of a preview of the sites. You can click through and learn more about it there.

[32:43](https://youtu.be/4I2Kt9XPHxQ?t=1963) But the, I think the larger idea here is that there are like different kinds of public private partnerships, where we can sort of say, like, there's a state, there are set of governments, and Switzerland's quite decentralized, which are sort of building the actual large language model. But then there's a separate set of governance for decisions that need to happen at the kind of deployment level about like, how do people get access? And how do we involve a larger set of people to make, you know, to sort of like directly govern this? So we're using this kind of like cooperative infrastructure to kind of bridge that gap. I think there's also different creative ways that we can figure out how do we make sure that this infrastructure or these services are well governed, in a way that is truly redounds the public benefit, as opposed to, you know, private capture. David, before you jump in, I just want to put another call to everybody for questions, because we're going to shift to your questions momentarily. We'll paste the instructions again in the chat for how to find the q&a if you can't find it.

**David Eaves**

[33:48](https://youtu.be/4I2Kt9XPHxQ?t=2028) I think for me, the questions of governance are really interesting, because they come back to this point I'm trying to make about us being so early, like the practices are that the tools themselves are so emergent. And then the practices lag behind the tools. And so they're like, like, not even emergent, like, we're like, like, almost in the dark people stumbling around. Like, I don't know, is your relationship with AI going to be like that with your utility company, where you're just buying electricity and applying AI in different places of your organization? Is your relationship AI going to be more like that with a on prem software product where you're buying consuming, but you're provisioning the electricity and you're servicing it? Or is it going to be like a cloud provider, where you're connecting it to various different tools through microservices? Or is your relationship with AI going to be like that of a product where AI is going to be baked into every single product the cities buy? And like, you're actually using all the AI providers, because you're really going to be dependent on whichever AI provider is serving Granicus on this product, or, you know, I got Tyler technologies on that product. And so, like, the governance of how you would think about that is going to look really different depending on which model those those look like. A model where it's more centralized, there may be there's like a high priest committee somewhere in the city that's really determining like, who, who is the trusted AI that we want to rely on? And what's our strategy look like? And then everybody flows from that. But if it's like AI is embedded in every single software you you buy, and, and it's being provisioned by lots of different people, you're going to have almost no agency, and you're going to be basically reliant on whatever federal state and other type of, you know, regulatory and like best practices are constraining those actors behavior, or you're driving a hard line, you're saying we're just not going to buy any product that's serviced by that has that company embedded in it, which I find hard to imagine. So I don't know what the government like, like, I would just

[35:48](https://youtu.be/4I2Kt9XPHxQ?t=2148) I want to read it to people all the time, like, people who tell you that they know what the future looks like on this are wrong. I do not know what the future looks like. What I'm interested in is what are the range of options that are in front of us? And what's the muscle we can build today that will position us to be in a better place to capitalize on the opportunities tomorrow. So the future is what we make of it. So given that I've done a terrible job moderating, and have not gotten to half my questions, I want to shift momentarily to the questions that are in the q&a. So I'm going to ask you one more, giving each of you an opportunity also to read what's come in. And then we will sort of ask you which ones you want to take first. And I'll feed you some of those. But here's where I want to come specifically back to the procurement question. If the future is what we shape of it right now, very concretely, these deals are getting negotiated, contracts are getting signed. In some cases, they're not getting signed, as you point out, and as is being discussed in the chat, AI is just showing up in places on contracts that are already signed. But suddenly something is AI enabled. You know, and it's correcting your grammar, which thank God. And sometimes it's something we're signing, the governor's breathing down your neck going, you need all our employees to have access to AI, because we want them to start building product with it. And by the way, they are and they're building some great stuff. Very concretely, bring us back to the procurement question, or the other levers where you can use right now.

[37:27](https://youtu.be/4I2Kt9XPHxQ?t=2247) Again, anything that we haven't said yet that you want to add here? And what would you do differently? What would be the pushback? What's the thing you want to sign? I just want to make sure to be really concrete here. Yeah. So a few things I want to say first is I don't want to be seen as like anti-innovation, 100% in favor of Citi's experimenting. But I just want us to be really constrained and thoughtful about that experimentation, and not be racing forward to actually then go to full production right now on things which we have very like, we do not know what the practice looks like. So what should we say no to? Or what should we say yes to? I'm like $30 million, $30 million deals with Anthropic to like bake them into, you know, every service in the city. I'm like, do we know what we want yet? Do we even know what we're trying to solve? Like, I'm like, what I find interesting are the experiments that I know, Beth, you've been involved in, like with Santee in Boston, where they've been using AI to like, look at like all of the different work requests that are going on to figure out what what are the work requests to get bundled so they can then package those as a UX for users.

[38:35](https://youtu.be/4I2Kt9XPHxQ?t=2315) That's super interesting. But that's like, that's like, that's basically like, I suddenly have a data analyst that's a lot cheaper and is allowing me to scale all sorts of work. That's not fundamentally rebuilding the architecture of how the city operates around AI. Like, that is in the future for us. So I'm like really interested in those types of experiments. The other thing I would say is I would be in favor of Citi's pursuing strategies that give them maximum autonomy. And so notions of like, we should be pushing things towards MCPs or MD files as much as possible. Like as Joshua said, like API access to things as opposed to baking them in. And Citi's could be co-designing or at least publishing like, here's the MD file for how to access these services. We'd like to see if others have a similar approach. Can we begin to standardize around this so that companies can start building against that? And then we can use that to then shop for vendors who are all willing to work against that MD file. That's the kind of optionality and standards work that will give you choice in the future that you have the agency to do right now. Assuming that agentic approaches become the dominant model, which nobody knows, but at least we're building some muscle. So sorry, go for it, Josh.

**Joshua Tan**

[39:52](https://youtu.be/4I2Kt9XPHxQ?t=2392) So just to be clear, the question is kind of largely about like procurement practices, right? How can we... To the extent to which you see that as being the most powerful lever we have to wield now. And you may say procurement isn't the most powerful lever. The lever we should wield is training or hiring a particular role or firing somebody or, you know. So I think like, you know, my emphasis tends to be like getting governments to deliver these services themselves. I think procurement 100% is one of the most powerful tools you can make. I think there's a level of, you know, I don't want to put the entire onus on the folks here, on government, generally speaking. I mean, you know, public sector, public servants have a lot of things on their plate. I think, honestly, the open source movement in the open source community that builds, you know, these AI alternatives, that tries to emphasize interoperability and public goods or public benefit, needs to do more to actually productize its work. It needs to make it actually accessible to kind of function in the way that some of these like larger closed source projects function. So it is easy to adopt. So it is easy for, you know, public agencies to come in and actually say, oh, this is a live auction for procurement. I don't have to spin up like a, you know, like a three-person team just to be able to integrate the software or the solution into my, you know, my existing flow, right? I think that stuff is emerging or is coming into play. And hopefully, if we do our jobs right, more and more of those options will become live. So once again, you can still go with, you know, the chat GPT, but now is the point is you have an option.

[41:46](https://youtu.be/4I2Kt9XPHxQ?t=2506) And ideally, there'll be additional infrastructure that keeps improving, so that you have more and more of an option. So you can, you know, liberate. Nowadays, it's not even like you make a procurement decision, you're like using anthropic now, can technically like, you know, I have an API that's set up like I can swap between anthropic and chat GPT and everything else and a purchase and Omo and Meta and Lama, every like, pretty much every second, right, every other query, I can send to a different different version, the API or the service that works best for it, right. And this is, I think, an emerging practice that I will, I think people as people are thinking about procurement, that is, that is a possibility. I will certainly make a plug, though, that I think it's important, if nothing else for the learning sort of opportunity experience, or, you know, the public sector to come forward and try to offer some of these services, it doesn't take that much effort, it doesn't cost that much. And you can, you know, build a functional AI service that people can sort of call on, not just something that you use internally, or, you know, you put a little chatbot on the on the corner of your web page or something.

[43:04](https://youtu.be/4I2Kt9XPHxQ?t=2584) Okay, have you had a chance to take a look at the questions? And is there one you want to grab? Or shall I facilitate more of them? Some of them we have asked and answered already. But let me know if there's something you want to grab on to. And if you're still reading, then let me, Josh, go ahead. I can answer Jorge's question quickly. I think we talked a little bit about what the first steps were, you know, we spent a lot of time on the first steps. But the specific question is, like, you know, you know, is it publicly trained models, for example? Remember, people can't see the q&a. So just the question is, yeah. I love the discussion about public infrastructure, what public AI infrastructure looks like? How do you see this branching away from this market, our current market driven approach? What do the first steps look like? We talked about that. And one example is, is it publicly trained, available models, for example? I think this is a really interesting question, because obviously, I care, and I really try to emphasize open source models.

[44:10](https://youtu.be/4I2Kt9XPHxQ?t=2650) And some applications like, you know, that open source models, maybe like they're actually very, very close to the quality of the closed source models, but they're like, not quite there yet. On certain use cases, and even in those cases, I think like you don't have just to reemphasize, you don't you can use those models without like buying that whole kit and caboodle. The generally speaking, I would not advise individual governments, unless you are a very rich government, like the Swiss, for example, to try to train your own models, I wouldn't, I would even advise them not to do their own fine tuning or post training, unless there's a very, very specific application of this AI that you like really, really want to emphasize. I would figure out other ways of, you know, wrapping those, those services, supplying tools. And this is where like orchestration actually As a really value-add here, if you are able to control your own orchestration, it means you are able to add your own tooling on top, your own customizations, your own context, managing context, integrating your data in the way that you want, so that ultimately, your patients, your users, your citizens, you know, get the experience and use the AI in the way that you want them, you know, you want that experience to look like.

[45:25](https://youtu.be/4I2Kt9XPHxQ?t=2725) Is there... I know, David, there's probably something you want to latch on to here, but I want to ask either of you, where does collaboration come into this? And there's a question in there about smaller cities. Is there... So, if you could also just answer from the perspective of, like, you're talking a lot about what one jurisdiction can do for itself. Is there a version, and is public AI a forum for this, for multiple cities or states or countries or otherwise to sit down together and to try to have the leverage that we didn't have and we didn't do with digital? And Charles points out in the chat, you know, this is a conversation, there's a version of this we've been having since 95, and we all went it alone, and we didn't engage in that kind of collaboration. Like, is there a forum for this? Are we making one as part of what these conversations are hopefully going to lead to or spur more of?

**David Eaves**

[46:24](https://youtu.be/4I2Kt9XPHxQ?t=2784) So, my first career is as a negotiation consultant, and I'm of two minds of this. On the one hand, I've always been confused by the fact that more small cities don't band together to share more services or to share more code or to kind of, like, work in collaborative ways. And the other side of me is, like, collaboration carries enormous costs, like, enormous costs to work. So, like, I'm not interested, actually, in a public version of, like, a purely public version of AI in the future. Like, we are going to need private capital to scale and be part of this process in various ways. Like, the internet is not a public good. In fact, there's a great debate to be had about what actually is public, if anything, about the internet. Almost all the infrastructure on the internet is privately owned. All the bits, like, all the cabling that you're using is privately owned.

[47:26](https://youtu.be/4I2Kt9XPHxQ?t=2846) All of the infrastructure that's, like, the servers is privately owned. The genius of the internet was actually the alignment around standards and the ability to align people's business models with the notion that actually adhering to those standards was going to keep making the pie bigger and bigger and bigger. And so, it kept plowing more and more public, I'm sorry, private money in and kept pulling more and more people and increasing access. And there was a private version of the internet, like, Minitel in France was what a private version of the internet looked like. And it didn't scale beyond France because they weren't willing to share those. And so, what I don't want to have is a purely, like, single version of AI that's only public. I think we're going to need to have a variety of approaches. What we need to be doing is figuring out, how do we drive interoperability? How do we drive the business model? How do we drive the incentives to enable portability and sharing and competition in that space? And if it turns out there's only going to be one AI that's dominant over all those, then we have a much bigger problem in our hands.

[48:27](https://youtu.be/4I2Kt9XPHxQ?t=2907) And there we have to be thinking about kind of monopoly and competition.

**Joshua Tan**

[48:33](https://youtu.be/4I2Kt9XPHxQ?t=2913) Okay, just really quickly to Beth. Go ahead, please. I think you're creating this forum for banding together. And it's absolutely amazing work. And just want to celebrate that for itself. Second, I think there's... Yeah, I'll share. You know, we're doing this stuff with Maine and we're hoping to expand this to different states, a more scalable model. This is kind of on the mode of banding together, but it's a little bit more political as opposed to policy or governmental. You know, actually, I don't want to take too much time. There's a lot of questions. I think some really good ones. But I'll share some more stuff around, I think what we're trying to do and to shape the politics of the moment so that we can more effectively work together as a coalition to build more of this public interest technology. I think the kicker to this is like, how do we convert all that data center backlash to something good?

[49:34](https://youtu.be/4I2Kt9XPHxQ?t=2974) Okay, we have not enough time today, but we do have, I might add, four more of these. And then we can always continue. So let me just see, is there anything that's jumped out at you from the questions that you want to just grab onto? In these last few minutes, if not, I will ask you. David, this is over to you. Is there something that you want to jump in on?

**David Eaves**

[50:04](https://youtu.be/4I2Kt9XPHxQ?t=3004) Again, I would be wanting to try to find the single cheapest way for us to be sharing best practices. So I would come back to this notion of like standardized MCPs and standardized MD files is like very, very, very low cost. Like anyone can publish them. Anyone can grab what someone else has done and take it and just use it or build off of it. And the more, like particularly cities start to align around those, the more likely you're going to create a market where multiple vendors are serving you and you will have optionality. So I think you want to be looking for those leverage points, particularly as like small and mid-sized cities. The big cities are going to do whatever they want. They're like states. But like, if you're a million people less, you will not have market power on your own. So don't, and my big fear is everybody's going to want to have like, oh, it's got to be my values. Don't be a creator, be a re-user. Go and support other standards that are out there.

[51:05](https://youtu.be/4I2Kt9XPHxQ?t=3065) Josh, let me give you a last word here.

**Joshua Tan**

[51:12](https://youtu.be/4I2Kt9XPHxQ?t=3072) I think what makes public AI really important or differentiates from all the, a lot of the other conversations about like, wouldn't it be nice if AI looked like this, if it was more open source, was more democratic? Is that it's not just, it is absolutely policy. It's about government, but what are governments, right? They're fundamentally political institutions. They are power because of the buy-in and the fact that they represent and act on behalf of the people. And AI for better or worse is already, and it's only going to become more and more political, right? That is the flip side of this conversation around public infrastructure, which we often don't think of as political. I think, you know, back in the day when electricity was new, for example, it was fundamentally, it was like a very, very deep conversation about what we wanted our societies to be, what we wanted to make sure people got access to these tools and what like fairness looked like. I think, you know, if we want public, if we want AI to be public, you know, we need to lean into that, you know, how do we think about the politics of this moment that we work together to build out a kind of a shared vision and how do we collectively organize?

[52:26](https://youtu.be/4I2Kt9XPHxQ?t=3146) Speaking to sort of David's questions about like, you know, coordination has costs. There's a lot of institutional development and infrastructure. I say like, in the sense of like people coming together, you know, having bodies like this where people are talking and sharing ideas and coming to consensus, more and more of that needs to be built. And I think it's so important, so urgent. So thanks again, Beth, for doing that. So I'm gonna squeeze every last bit of knowledge out of you with three minutes to go. If you love talking about electricity, come next time, because Ganesh Sitaraman has written the book on public option, not about AI specifically, though he writes about that more recently, but it's all about electricity and public schools and public parks and how we think about publicness and the techniques we have used to achieve publicness with private market competition at the same time. So that will be coming next time. And I think Luca, who is here, is joining us for that to make sure we have a global perspective on these questions of what public means.

[53:28](https://youtu.be/4I2Kt9XPHxQ?t=3208) But let me just ask you, because in addition to being lots of government people in the room, we're also all citizens. Is there one thing that as individuals we should be doing differently or better to move the world towards more public interest technology with our consumer choices or our vote?

**David Eaves**

[53:54](https://youtu.be/4I2Kt9XPHxQ?t=3234) Oh man, I don't have a good answer to that because the consumer choices feel relatively limited. But what I would say is twofold. One is I would be, if I was a manager, I'd be pushing my employees to be trying to use these tools. I joked last week in Ottawa that if I had a whole group of people doing, if I had a whole group of analysts, I'd take away their data analysts and try to make them use cloud code to be a data analyst for them, because I think those are the skills that they're gonna need for the next 10 years. And I wanna be scaling my people up for that future. So I do think you have a response. Like, again, I think I worry people will come away thinking I'm anti-innovation. Quite the opposite. I think right now is the moment for innovation at the individual level. And you as a manager should be pushing your people to be experimenting and playing with these tools because the innovations right now are all happening at individual level, not institutional level, individual level, and you could be equipping your team up. And building those muscles with your individuals is one of the best ways to prepare yourself for the institutional shifts that are gonna come in the future.

[54:55](https://youtu.be/4I2Kt9XPHxQ?t=3295) Okay, one minute, Josh.

**Joshua Tan**

[54:57](https://youtu.be/4I2Kt9XPHxQ?t=3297) David is on fire, so I'm just gonna add a bit more fuel here. I think ultimately you as individuals, what you can do right now, starting this second, is just stop believing the narrative that the big companies tell you. Do not believe that they are the only option, that only these few dudes in San Francisco or these big tech companies could possibly build or provide AI. That is just not true. Challenge that narrative. Build it yourself. Test it. Try out some of the things that are online, and just through one additional Google search away, and prove it to yourself, and tell the other people that you know that this is how we change that narrative and start saying these are not options. These are options that really exist and that we can really use. Thank you to both of you for sharing of your time, your passion, your fire, and enthusiasm. Thanks to the 200 of you who showed up for this and the many more who will watch this after on video.

[56:01](https://youtu.be/4I2Kt9XPHxQ?t=3361) Again, join us next week for more on public AI. Earlier time, no, so same time, sorry. Next week, same time, same channel. And we'll talk to you all soon. And again, all of the resources will be openly shared afterwards. If you missed anything, videos, et cetera, will all be online. See you next week. And thank you to everybody. Here's to a more public future. Take care. Bye-bye. Bye, folks.

* * *

Original page: https://innovate-us.org/what-is-public-ai-and-why-it-matters · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org