This episode features audio from the Open Infrastructure panel at Open @ UNGA 81: A Conversation on Collaboration. It was recorded on September 21, 2026.
As AI grows, so does dependence on the companies and institutions that control the infrastructure on which it runs. “Public” infrastructure can mean government-built systems or open, cross-border infrastructure maintained by nonprofit, research, commercial, or volunteer communities. This panel will discuss how these models intersect with sovereignty, drawing on analogies like natural food webs to examine when openness and shared standards genuinely reduce lock-in, what dependencies persist across cloud, compute, and skills, and how public institutions can fund infrastructure that’s adaptable and accountable without every country rebuilding from scratch.
Announcer 0:01
Welcome to Engelberg Center Live, a collection of audio from events held by the Engelberg Center on Innovation, Law, and Policy at NYU Law. This episode features audio from the Open Infrastructure Panel at Open at Unga 81, a conversation on collaboration. It was recorded on September 21, 2026.
Monica Granados 0:24
Thank you so much, folks. As Anna mentioned, I'm Monica Granados. I'm going to tell you a little bit about the structure of the panel. So we'll start with just individual questions that are tailored to your specific background, and then we'll take a couple questions that I think will be really helpful to hear from each and every one of you, and then we'll leave about 1510 minutes before lunch so that the audience can ask questions. You'll see that there is a mic up here at front. So as you hear the panelists talk about the concept of open infrastructure. Think about the types of questions that you would like to ask this panel. So before we start, I want to set a little bit of context. I think there really is no coincidence that we use the word ecosystem in this space, and I particularly like the word ecosystem because I'm trained as a food web ecologist, and one of my most favorite examples of like of food webs and like how they're so interconnected is the relationship between wolves and other animals in Yellowstone. So what they found is that when wolves were removed from the ecosystem through hunting, that there were consequences on the number of elk that lived in the park. They had really high populations of elk, which ate down the grasslands, which meant that then there was less room for small mammals to hide. They it also caused erosion of the riverbanks, which had an effect on the fish. So this, you know, the wolves when they were introduced, they were able to like restabilize these connections, and I bring up this example because I think in front of us here we have the wolves of infrastructure, which you can use for your next book. And you know we're going to talk today about you know what happens when these wolves disappear, why they're so important as a stabilizing function, and like, what does like who manages the wolves? You know, the the reason why these wolves disappeared in Yellowstone was because there was no management, there was no regulation, and so I think you're going to start to see a lot of you know sort of allusions to like the ecosystem that we are in right now, and there's a reason why you know we use that that terminology. So you know, we'll we'll talk about also just like definitions. I know that was really set well for us in the first panel, and you know, like how do we create how do we create an ecosystem of infrastructure that is robust so that it doesn't collapse when one piece is removed, but also we're not creating too many similar ecosystems that are redundant? And so, the first question I'd like to give to Ellie from the GitLab Foundation. So you know definitions. I think are really important to contextualize the conversations that we're having today. So when we talk about digital infrastructure, what belongs in that category, and what's different between digital infrastructure and the physical layer, and how does that get distinguished between like software and standards and shared digital systems that are built on top of of them?
Ellie Bertani 4:06
Great, hi everyone. So nice to be with you here. Just very briefly, I'm Ellie Britani, lead GitLab Foundation. You've probably heard of GitLab, maybe less of our foundation. We focus on economic mobility. We believe economic insecurity is a driving factor in a lot of the issues we face today, including climate change, including wars and conflict globally, and including the erosion of democracy, and so that's what we focus on. But really, we're unique because we apply the principles of open source to philanthropy. So radical transparency in how we operate, and how we give grants, and what we invest in, and why, and also really intentional collaboration across the field and across the spectrum of financers. I love this question because part of what we really have started focusing on over the past three years is investing in the back end infrastructure. So critical for the nonprofit and public sector to be successful in this space, and for us, what that looks like is not just the software, the accessibility, the affordability of systems, data sets, things that tend to be overpriced and difficult for nonprofits in particular to access in order to do really important work for the public good, but also less tangible assets like the skills needed in the sector to be able to effectively work in this space, have a role to play in the broader conversation around public good, and part of what we think about is the real disparity and how much financing there is in the private sector and how that skews the market in terms of outcomes and focus on on the results that we need for the public versus the versus the nonprofit sector. So data accessibility is a critical issue, but also the platforms and the interoperability standards, as you heard in the prior discussion, are areas where we focus on investing and connecting our partners to experts in the field.
Monica Granados 6:08
So much. The next question is for you, Caitlin, and like Ellie, it'd be great to hear about investing open infrastructure. But your question I'd like to pose to you is: you know, we use the word public digital infrastructure. What should that guarantee? Should it be public ownership, open code, open standards, public interest governance? There's so many sort of layers to that. Is it should be you know a combination of that? I'd love to hear your take on that and a little bit about investing in open infrastructure.
Kaitlin Thaney 6:38
Sure, it is really wonderful to be here, not just as a frequent commoner and an old Creative Commons employee, but also leading an organization that is infrastructure in the name. So we are a nonprofit initiative that was designed really to focus on where there was both an increased reliance and need and acknowledgement for community-based open infrastructure across a number of different levels, but also a real frustration in the funding, the resourcing, and and how to get to that level of embeddedness and adoption, which I think many in this room have their own stories about working to advance. And so the when it comes to some of the language around that, and we battled with this a little bit, and I'll say battled because it felt like every first 45 minutes of every hour-long conversation, starting IOI, was around like, well, what's your definition? Is that going to put like sharp boundaries around my definition? And and I think we operate best. And the previous panelists had a great comment, Liv. I think it was credit you on this one about we need to be able to have those conversations across. We need to be able to have that coordination, and so when it comes to public and digital, and I think there's some core tenets that we have in a number of the interlocking definitions and frameworks and other structures around you know openness and accessibility and equity. What does it mean for those to have access to be able to participate? Whether you're in research, civil society, more broadly, what does it mean to be able to reuse those materials? Whether it was at Creative Commons or Mozilla or GitLab or what have you, but also the governance component. And I think that the governance and the stewardship aspect of things about who makes decisions, how are others invited in to make those decisions, really allows for there to be again that representation, which often I think is like a sub bullet of governance of, you know, are the those that are most impacted or those that are using your whether whether it's a tool, a standard, a service, what have you, content are they represented in the ability to make those decisions? And I'll say that at IOI, when we first started, it was began as a coalition of I think 22 individuals across the Open ecosystem, many who have rolled up their sleeves and worked with alongside you know, those from the open access, open science space, open infrastructure, open source, and really looking at more of an ideological, like open versus closed type of an argument of, you know, this is something that we need to make sure is preserved. We've moved to a slightly more pragmatic side of that, and I think rather than, and I and I mentioned that in terms of definitions just for a particular reason, because I think that the when we talk about infrastructure, we talk about that reliance. You know, we notice. I think all the panels today are representative of areas that you sort of notice when there's an absence of, and so it's very easy if you're not embedded in looking in those spaces to just imagine that someone's taking care of those pieces, whether it's around collaboration or data or infrastructure. The definitional piece we found impossible to extricate from what those communities' needs might be. We spend a lot of time working with researchers and librarians and scholars, but also those in civil society. And what matters is infrastructure for them. Actually, might be a system or a protocol. It might be the interoperable layer. It might be the data set. It might be the physical collection or the biological specimen. And so, I would also just kind of posit that in the conversation too, of like you know, when it comes to what you define as infrastructure, to make sure that that doesn't get extricated. And the same goes for when you mean public or digital. It is an intermix, but I think that's at least how we look at that in terms of like. Yeah, I think yeah, and
Monica Granados 10:39
it's really helpful to hear that from from you who like kind of has that bird's eye view because you have all of those organizations that are part. I want to continue on this this theme of sort of thinking about you know digital public infrastructure
Monica Granados 10:51
and you know the infrastructure that a lot of governments depend on. So I think I want to ask the next question to Jan from the Wikimedia Foundation. Love to hear a little bit more about the Wikimedia Foundation as well. Is there a distinction between digital public infrastructure and then just the broader infrastructure that governments and research institutions and civil society depend on? Many of which Caitlin mentioned earlier.
Jan Gerlach 11:15
Yeah, great question. Thinking about this from a perspective of a wolf of infrastructure, I guess, which is still a new concept to me. So, yeah, hi everyone, great to see you all. My name is Jan. I'm a public policy director at the Wikimedia Foundation, which is a nonprofit that hosts and supports Wikipedia and other open knowledge projects. Wikipedia and Wikidata are actually digital public goods recognized by Liv's organization. The Wikimedia Foundation is also a member of the Digital Public Goods Association, which I think is important to mention here. The distinction, I think, is is is an interesting one to your question. We think of Wikipedia as as in a way both physical infrastructure because we host the servers, right, which are being crawled and scraped right now. Like there's no tomorrow, and at the same time, I think Wikipedia is really also the intellectual infrastructure in a way that a lot of people depend on, be it like people like us using it. People in the global south who have limited access to educational content, but it's also verifiable content that, as we see now, a lot of AI companies that have been mentioned a lot today are really interested in, because it's it's facts, it's verified, it's it's maintained and curated in content and knowledge about our evolving world, right? And so we think about those, or I think about those two dimensions of critical digital infrastructure. Really, there's the there's the physical, the axis, but there's also the the content that is quite important because it's also part of every large language model, to our knowledge. It's really hard to quantify that, but we know that every company that builds large language models crawls Wikipedia and Wikidata and includes that in their rag in all their systems, and so we're really interested in finding ways to sustain all of this, and I guess we can talk about that later. What that looks like, I think there's a there's a really a funding component that is well represented here in the room as well and on the panel. But there's also really a component that's regulatory. How do we support these distributed governance systems where people can work together to build this content together instead of companies or individuals or nonprofits making top-down decisions about what content looks like, what governance of these systems look like, I think there's there's two really important aspects to this question.
Monica Granados 14:05
Yeah, we'll definitely return back to to sustainability, but you know I want to surface what are elements that are hard to see until something fails. You know, with with the wolves, that was it was very obvious. The wolves are really big. It's a lot harder to see when something, you know, like a smaller invertebrate, you know, leaves the ecosystem. So I'm going to turn over to the question to Yaël from Open Athena. We'd love to hear more about Open Athena and then ask you the question, you know, what dependencies are hardest to see until they fail. Cloud hosting, hardware maintainers, financing, technical skills. There's so many different pieces, in your opinion and your experience, which which are sort of hardest to see until they're no longer there.
Yael Elmatad 14:58
Thank you so much, Monica. Thank you for the organizers for having this lovely event and for hosting me. I'm Yaël. I lead the engineering team at Open Athena. Open Athena is a very young nonprofit. We've been in existence for only two years. Initially, we were founded to help build science foundation models, so to work with academics and to bring sort of industry level technology experience in building models into those academic spaces. In the course of that project, we acquired the Marin project, which is a attempt to build a truly open, truly transparent large language model at frontier levels, and in doing so, we have found that it is there are so many different aspects, as you can imagine, that go into building these kinds of models, and so many places where the infrastructure can be a little bit fussy. From anything like we heard earlier about vendor lock-in, we know that one chip manufacturer controls something like 80% of the sort of HPC chips out in the in the ecosystem. And one thing that we've been doing at Open Athena has been really trying to take the software that we run to build the LLMs and make it portable and make it so that it can run on other kinds of hardware. Because that's such an important thing to remember: is that if only there's only one sort of proprietary player in the market, it's very challenging to kind of work around that. We've also found that even if you build these sort of open source projects, I think like like we heard in the lightning talk, hardware involves money, and you have to pay it, obviously. And I think what we know is, I was asking one of my colleagues the other day. Okay, great, we built a model. We've released a model. We've released the open weight model. Everyone is super happy. What next? And I and then they said, well, realistically, it's very challenging for the kind of players that we want to have access to this model to actually deploy that and to do the inference for the model or to do the post training on that model, and so I think about we all think about the model, but we don't think about the serving of the model and and how critical it is to create infrastructure at that layer, and it's such an important part of model of model and AI sovereignty is that ability to control where your model is run, even if you didn't control every aspect of the development of that model. We also have obviously a huge skill issue in the infrastructure: is how do we get engineers sort of away from some of these very closed model factories into convincing them to join our efforts and to work in the open, we've been very fortunate that we have a lot of folks who are just very much believers in open software and open development, and therefore they choose to spend their time with us, or they have perhaps this is sort of the second act in their career, and they've sort of already done the the big tech thing, and now they want to do something nice for humanity. But but it is challenging. I think like we we really have found that the only way to do it is to be competitive, and I think it's very challenging when I go to an engineer and I and they say, hey, we work at a nonprofit, and they'll give me kind of a look, and then I say, no, no, no, I swear we don't pay that little, and I think so. We kind of have to be competitive, if we want to see this this market kind of evolve, and it's, I think we assume that we can get this sort of cheap labor to work in this space, and I think the reality is, you know, you get what you pay for, and you have to be willing to kind of get those those really high level talents and cutting edge engineers. For sure, I think this is a great transition into talking about sustainability. I'm going to keep drawing on my wolf analogy. I study fish. I didn't even study this ecosystem, but I really like it because you know again, there's a lot of parallels. So you know,
Monica Granados 19:00
the wolves were able to be driven out of the park because there was no like regulation on hunting. They, folks who had you know ranches near the park just hunted the wolves to to like local extinction because there was again no regulation around, or no protection for the wolves by the government, but it was the government that reintroduced the wolves back into Yellowstone. So they, you know, having that having the government be a part of that introduction really facilitated their success. So thinking about the role of government in this ecosystem and this sustainability, and making sure that, like you know, the we don't remove the infrastructure and important infrastructure from the system. I'd like to hear from each and every one of you. You know, is there a way for government to use procurement and funding or standards? Legislation to be able to facilitate the sustainability, the competitiveness of the infrastructure, and particularly like the pieces or the layer of open infrastructure that each of you are representing or stewarding.
Yael Elmatad 20:17
I have a very small one that I so a lot of what we do at Moran is we built on what's called preemptable compute, and it's sort of I like to think of it as nose to tail open model development because most of the big players they sit on what's called reserved compute, but every so often those little reserved instances they go a little bit idle and they're sort of leftover cycles. And so what we have done is we've built a bunch of software that allows us to do experiments on this kind of ephemeral compute that may or may not get pulled out of the rug for people that are higher up the food chain than us. Maybe we're not wolves. Maybe we're small rodents. But I think one thing that is sort of really bizarre to me when I first came on and said, "Well, why do these companies give it to us? Is it because they get a tax break? And I learned that is that you you actually cannot get a tax break on donating compute in your data center to a nonprofit, which I thought was very bizarre. It turns out if you give them hardware, that is okay. That is the gift. There are ways around this. You can give them money to purchase time on supercomputers, but just that kind of little policy switch to encourage these big players in the infrastructure place to donate some of their resources, even in that shared context, would make it a lot easier to go procure more of that kind of resource.
Kaitlin Thaney 21:34
Maybe to Jimena on that. So, so IOI came about, and also previously, Jan and I were colleagues at Wikimedia, where I was helping to build an endowment to think about. I mean, it's kind of the dream use case of trying to sustain open in a big way, right? And also, there's some really interesting things which I'll let Jan talk about about Wikimedia and its relationship to government and support and things there too, because there's some areas where I don't think government is actually the right group to be able to support. But what we do at IOI, coming into looking at how we can better bring funding and resources and sustainable models and actually help with the growth and success of embedding these infrastructure principles and technologies, we did a deep and we continue to do a deep examination of where funding is flowing, and also who are the groups that should be at the table? Can we open the aperture and get more groups at the table? Moving beyond grant funding, moving beyond just institutional funding, moving beyond certain pockets of resourcing from different government agencies. And I think we're in a really fascinating time. You know we're here in the U.S. IOI does work across four continents at a deeper level right now, but I think we're also at a time in the last 18 months to 24 months of a fundamental reshaping of how we think about government and supporting public goods. A good portion of our work right now is how to restabilize infrastructure that is no longer being seen as something that the government can steward alone, and should not be sustaining alone. Whether it's around publicly funded research, data access to things that allow wildfire prevention, disaster preparedness, public health emergencies, etc. And so I think that there's. I like to think of it as not that the government should not have a role; it should not be the single point of reliance, especially as we're not only looking at the geopolitical issues that I know we're focused on in the previous and mentioned in the previous panel, but also just the fact that a single point of reliance is a brittle system, right? And we've had even in early days back when there were nicer times, which I'll talk about. You know, we we brought different government agency funders, different private funders, different public funders together to say, you all have a vested stake in thinking about the long-term sustainability and investment maintenance of open infrastructure in various ways, right? So, what are you doing to dedicate? And these are private conversations. But what are you doing to, you know, dedicate more of your portfolio to these calls? You also came to the table and said IOI needs to exist to be able to help with this. And every single one of them, not and starting almost in like your usual, like, well, we're here to just fund the innovative stuff. We're not here to sustain it. That's his job. That's her job. And then they got to the government representatives, and they said, "We have to do that too. Like we can't just sustain everything behind the scenes, right? And especially now that we're seeing different personalities and governments, and it's not just the U.S. We are a bit of a special snowflake, but we're also seeing it across South America. We're also seeing across Central America. We're seeing it across Europe. We're seeing it like where there's very narrow tilts into spaces where fundamental investment at the at the public infrastructure level could teeter systems over, and the level of investment that's needed if we don't think about diversifying that is significant. You can't replace that overnight, and we've even tried. And I will give just a very tangible example that I was telling, talking to Michel about earlier. We were brought in to help look at what does an open network solution look like for the data infrastructure across the United States. As we were looking at data sets being deleted, or what happens if NOAA goes down or a CDC goes down, etc.
Kaitlin Thaney 25:22
and as we were having those conversations, and that work has been incredibly rich, but also to a point that when we started to model out the numbers and others started to model out some plans, it sort of had this elastic rubber band effect that we also saw during COVID of, oh, the government's the only group that can sustain this, and it's like, but we're literally threat modeling the government. But when we talk to other funders that had the capital to be able to move forward, it's like I don't want to be on the hook for hundreds of millions of dollars to try to replicate the whole system writ large. But like, so I think trying to also think about you know it's not just getting different groups at the table, thinking about public-private partnerships, thinking about how we fund public utilities-all of these various components. But it's also trying to think about where are there areas that we might be able to pilot a different way, rather than just expecting the system will rebalance, because we are at a place where I think expecting that the government will be supporting this writ large in its totality, or in a larger percentage, is actually keeping us in a space which puts these infrastructures more. Yeah,
Monica Granados 26:30
I will say the most stable food webs are those that have the most number of connections. Yeah. So yeah, because you have you get you have redundancy. And I'll just
Kaitlin Thaney 26:38
say one little bit too, because and then I'll yield the floor. This is what I spend my days in, so my apologies. But I will also note that we have colleagues across Europe and other, and even our Canadian colleagues say, "Well, we've been doing it, and it has worked. I will tell you that there are little cracks in all of those different systems, whether it's around not enough capital being diverted to those resources, and I'm not saying that there aren't incredible examples of where starting small can allow for some of this work to move forward, but I don't think that we-it's-it's a longer memory span. We have not proven that out yet. I think in terms of what that looks like, and it still is going to need its own sort of diversification. Thank you for coming to my touch. I will stop.
Monica Granados 27:19
Ellie, Jan.
Ellie Bertani 27:20
Yeah, just a couple thoughts here. Two main thoughts. One, I think your point on the cost of talent is a really critical issue we see in our in our space, and not enough funders are focused on how do you actually pay for the for the talent you need in these organizations. I saw data recently that 41% of nonprofits rely on one single person to run all of their data and systems, and that's incredible weakness at a time when this is becoming more and more critical. So we think about that in terms of our funding. We try to encourage other funders across the table from us to do the same. But I think it is an underappreciated risk as we think about the potential for public good and the risk if if we don't achieve that, also just from a practical standpoint, I like the question. And you know, as I sit here in the U.S. and think about our government, I'm not seeing any at the federal level. I'm not highly optimistic about regulation in the short and medium term. We are much more focused at the state and local level these days, and how do we influence the 50 odd governors? We're partnered with organizations like the Center for Civic Futures, which is really focused on bringing together all of the chief technology officers of each of the states, in understanding how collectively they can affect their procurement standards and things like that. So we heard a lot about that in the panel. There's a lot one can do with procurement in the United States. Changing procurement practices is incredibly challenging, but is I think a really important potential to tie funding to to some of these goals that we're talking about today. So just a couple thoughts from this side.
Jan Gerlach 29:03
Yeah, I think it's important to to understand or or examine what what the hunting here looks like. What's the actual problem? What's the sustainability crisis? What's the ecosystem breakdown that we're looking at? And and I assume I'm here on this panel because it's really easy to understand is that the example of thinking about the example of Wikipedia. So Wikipedia is view is has about 13 billion human page views or requests per month. This number has dropped by roughly 8% I think that's a conservative estimate right now, year over year, just over the last 12 months. My assumption is that this percentage is actually much larger, and it's going to grow. At the same time, Wikipedia's bot requests have skyrocketed. We have tightened our API policies a little bit because the new bots also are a very different breed, different kind of hunters, I guess. Sorry, trying to stay within within the metaphor here. And so, so I think historically, when when the Google bot comes by. The etiquette says, "Hi, I'm the Google bot. Just gonna take a few bits here and there, and as soon as the server slows down a little bit, it backs off, comes back two hours later. I'm not a technical person, but that's how I can try to describe it. What this new sort of kind of AI bot does is it comes, it masks its identity, it pretends to be somebody entirely different, it boofs IPs, and when it notices that the server goes slow, it actually doubles down and tries to extract as much content as possible in as little time as possible. This is a huge problem, obviously, for networks, for servers, etc. And we currently, the foundation currently blocks 1.5 billion bot requests per day, and those are just the ones that are out of our policy, that do not comply with our AI policy, API policy. And to be clear, we want you to take the content from Wikipedia and do great things with it, right? What we do not want is systems to be abusive, to be extractive in ways that actually hurt the integrity of Wikipedia, and to really like bring up our costs, because those costs really are much higher when there's a a bot involved that systematically just goes through Wikipedia or erratically actually because that's content that we cannot cache in our systems. Automatically, sort of like extracted content is much higher, much more expensive for us to deliver. People go from like they. You all have probably done this. You you go from one blue link to a next on Wikipedia. They're all kind of connected. People look at the same things because they see a news cycle, etc. Bots don't do this. They just go from wherever to different places, and this is not content that is cash. So this is much more expensive. It's about 65 percent of our actual cost when we when we serve content. At the same time, we are now in a position where we actually have to add new infrastructure at a time when prices are most of you know this are really just going up at at at a pace that's that's insane, right? And so this is what sustainability crisis looks like for open systems right now. The prices for hardware are going up. The prices for delivering content are going up, and people are actually not the main users of our systems anymore. And so when systems, rack systems, any AI chatbots deliver content to or responses to questions. They also don't do a great job at bringing those users back to to Wikipedia either, right? And this brings us to the, I guess, the sustainability of content, the curation, the maintenance problems, because if you look up something on ChatGPT or in Gemini, you do not become a Wikipedian, right? There's around 250,000 active Wikipedians out in the world. Those are the folks who maintain all of the content, right? But they have become Wikipedians because they, at some point, came to Wikipedia and saw something that was off, right? There's maybe a grammar error, or maybe there's a number that's not right, or something, right?
Jan Gerlach 33:48
We've all heard these stories, but when you when you just chat with a bot, you don't do not become a Wikipedian, and that's so that's the crisis of attribution, and and I think from a democratic perspective, there's also a crisis of provenance, of authenticity, etc. But I think it's really important in this context to think about what is the actual problem, right? And in the example of Wikipedia, it's it's really visible.
Monica Granados 34:14
Yeah. So I want to ask one more question to the panel before we turn it over to the audience, trying to wrap up all the couple of things that we've talked about today about you know sustainability, maybe regulation at different levels, sustainability and like recruitment of talent. So if there was one policy decision that you could make tomorrow, like maybe it's you have to attribute, and that has to be, you know, when if you take data by law, you have to attribute. Maybe it's that you know you have to contribute to a fund for for talent. What would be the one policy decision, if you could snap your fingers today, that I think would would advance the. Particular sort of gap that you see from your perspective and where you sit in like the infrastructure and OpenStack.
Yael Elmatad 35:12
Oh, I should start. Okay, I I actually think that for me, that ability to sort of serve models for folks, so even once we develop open source models for them to actually use them, we see lots of people developing open source models. Not so many of them actually have the capability to use them. Investing in infrastructure for use of the models, to me, would probably have the biggest unlock for sort of a sort of take up of our model.
Ellie Bertani 35:46
I came in with a different answer than I think I'm going to give, which is why I'm sitting here pondering. I really, you know, I I think there is value in what I alluded to earlier, which is at different levels of government thinking about procurement and tying funding to interoperability standards, I think that's really, really powerful at the scale that we need. I also continue to think about the issue we've been discussing. It's just the costs continue to rise for the public good, and it's crowding out the ability for the nonprofit sector, in particular, to to have a voice and and make impact in the space. I was struck recently. I saw another statistic that the the largest 10 venture rounds in 2025 invested 84 billion dollars in the private sector to advance AI, and it was about $2 billion in contrast to the public could, and that 40x difference is the difference between value capture and value delivered to the public, value captured from versus delivered to, and so there's something there where if I were a you know if we're thinking about policy decisions, I'd really try to rebalance this market somehow, force some investment towards the public good away from just value capture, is is part of what I'm thinking about today as well.
Jan Gerlach 37:09
I think there's a there's a sort of a a short to medium term answer, and then there's a longer term answer. I think the short to medium there is a huge problem with what is called residential proxies being used to scrape the internet. These are, as the term says, and I don't think it sort of like hides the problem with those. These are networks of basically residential routers that are being hijacked sometimes legally because somebody has clicked a consent button somewhere to to really scrape websites through residential networks of people's internet access from their homes, and this is becoming a huge problem. We see this as a huge problem to our soft to our stack to our network and systems, and I think just addressing that through regulatory measures is a not easy fix, but it should be a focus for regulators and lawmakers. I think in the long term, really just recognizing the value of what we call a critical digital infrastructure of the physical and intellectual kind will be very important for governments to just acknowledge how important these systems are for the public interest, for us not to be locked in, and for us to have really like sustainable democracies that benefit from access to information. I think that's a long-term, big-picture thing, from which a lot of, of course, regulatory measures and support measures will follow. Oh,
Kaitlin Thaney 38:50
we spend a lot of time trying to figure out what's a way to relate to different stakeholder groups and different groups with capital, how to move that forward. So knowing this will not solve all of the problem. But I think if I had a magic wand, folks are usually very aware of overhead costs. Whether you're a program officer running an R and D department in a for profit, etc. say 10 to 15% of every dollar that goes to R and D discovery or a grant out the door that relies on open content or data goes into reinvesting into those systems. I think the amount of money that is going in-you gave some great numbers-the amount of money that's going into private investment for whether we look at AI, whether we look at moonshots, whether we look at focused research organizations, governments even pushing on all these various components, it is not getting reinvested into the system, whether it's teaching critical thinking skills, Wikipedians, or the underlying systems. There, 10 to 15 percent. I think one of the recent examples that we gave was I think is the Simons Foundation that gave to CUNY, the City University of New York, 75 million. Dollars for seven years to build an AI institute. 10% of that would support one of the data infrastructures that they heavily rely on for, I believe, five and a half years in its entirety. That was at risk of collapse from about seven institutions that had to trickle down effect pull their funding this past calendar year, and they lost a third of their operating budget, so I think some of those components it is a rounding error, yeah, and also putting that to task for those that are also utilizing this language where there needs to be a little bit more reciprocity or innovation.
Monica Granados 40:34
Well, I can only hope that some of the folks down at the UNGA meetings can hear these really great suggestions. I'd like to turn it over to the audience. Do we have some questions for the panel? Ideas too about you know if you had a magic wand, what would you change in order to help with sustainability And sovereignty. If anybody, yes, you mind coming up to the mic just so that. Thank you so much, and feel free to introduce yourself. Sure.
Speaker 1 41:09
Hi. Thank you for the panel for this one, the morning panel. I'm Neha. I recently graduated with a master's in philosophy, and I work part time at a nonprofit in fundraising, on tech policy initiatives. One of the things I've been thinking about is so much of the public discourse and narrative is set by Silicon Valley. There are so many conversations right now about the inevitability of an AI apocalypse and things like that. But the tone in this room is very different. We're thinking of and imagining and working towards a different future, and it's not so great that that's not a big part of the public discourse. And because public discourse has a lot of power in telling governments and media what people think is important, what can we do to shift public discourse to know more and dream more about open infrastructures. Thank you.
Yael Elmatad 42:10
So I think that was to anyone here on the panel who'd like to respond. Okay, I'll try. I think one of the things that you said you mentioned that the big frontier labs have been the loudest voices, and we don't know their motives for its press, but or other things. But I do have concerns as a person who builds open source AI, open weight, open development. I am I do live a little bit worried about regulation that would make what I do seem to be sort of outside of what the public thinks is the right approach. Right, that they see the fact that models are closed, or that software is closed, or that infrastructure is closed as a as a way to keep them safe, which I think many of you in this room would agree is probably not the best way to keep us all safe, and so I would love for regulators to hear that message more-that there is actually safety and openness, and there's value in showing people, showing the world how these things are made beyond closed doors, where there's a single group controlling the alignment of these models.
Ellie Bertani 43:22
Ah, for. some thoughts on this. First of all, proud of you, philosophy masters, who would be very interesting to an OpenAI or anthropic, has chosen to work in the nonprofit sectors. Kudos! Very pleased to hear that. You know, I think a couple of thoughts here on the discourse and the narrative in general. First, we as a foundation, just to give an example of how we think about this, really try to present a balanced message around the risks and the opportunities of AI. Right? There's a lot of focus these days on existential risk in a variety of different ways, we try to present a lot of use cases of how AI is really helping drive the public good. So, as an example, three weeks ago we held a demo day. We GitLab Foundation have funded over 50 different nonprofits for sophisticated use cases of implementing AI around economic mobility issues. We brought together nearly 200 funders in the room to hear these 15 different organizations present a really interesting, diverse array of projects from MIT, which is used and which has a project called Project Iceberg, which is thinking about labor market data again, open source access to labor market data, trying to project out the impacts on the workforce across the world due to the due to the effects of AI. We have U.S. digital response, which is building access to benefits platforms in the United States, SNAP benefits, unemployment insurance, but have built. Using publicly verified language access, really ways to translate these systems into 15 different languages, so that people can access benefits. So we're really trying to shine a spotlight on and provide additional follow-on funding for the projects that are really going to inspire people. What I see in the funding community, at least on the philanthropic side, is a paralysis for many funders these days. The level of uncertainty and the lack of technical proficiency, I think, is really constraining capital from getting to the market quickly. So, part of the role we're trying to play is investing in talent, articulate. the vision and the upside potential, because I think when people are in a very negative frame, it really constricts the ability for good to be done in this space.
Jan Gerlach 45:55
Yeah, how do you change that narrative? I think part part of it is actually us being here and then going out tomorrow or later today to meet those people who don't understand this space so well, right? I think this is to me that that's my job, right? Going out there and meeting folks who are not the likely allies, right? Who who may not think about Wikipedia as a living and breathing organism that actually needs support, but it's just like an encyclopedia that will be there forever, right? Or who don't understand that it's also a platform that that is regulated by the same laws or affected by the same laws that Facebook and TikTok are affected by, right? Thinking of intermediary liability and those things. That's I think it rely really depends on all of us to to make sure that narrative slowly changes. I don't think there's one silver bullet solution for this, unfortunately, but it's it's the conversations that we need to have.
Kaitlin Thaney 46:57
I mean, just plus one to everything that was said, and also kind of bringing together Ellie and Jan's comments there too. I think there's so much nuance, and there's I think that Ellie, your comment about bringing some of that positivity of like you know where are there opportunities, but also where are there risks, and being really open about it. And this may be I'm going to channel like an inner Wikipedia, and in terms of like also seek out areas that challenge that narrative. If you're seeing it be concentrated in Silicon Valley, look for you know where you can be curious about other sources. And I think that that critical thinking component is one of the reasons that you know, especially with how much work we do in higher education and in the research ecosystem, it worries me the most with how much funding is being moved outside of that ecosystem. That if we're not actually teaching people to be curious in that sort of way, and kudos again on the philosophy degree in that side, then you know we have an entire next generation that might not know to look outside of some of those conventional narratives that might have louder voices.
Monica Granados 48:05
Yeah, I think that's a great question to to end on. This you know this has been alluded to before in other panels, but it's about having these types of gatherings and then going to like the gathering tonight and the gathering on Tuesday because the voices together in this room and in similar rooms can be louder than that dominant narrative because you know they have a stage, but there are more of us than them. And so I think it's about you know fomenting that collaboration and putting our voices together and being loud about what what we are doing, what's you know the value add of like public open infrastructure and the need to sustain it, and what the consequences are if we don't. So with that, I really want to thank our panel. Thank you, Yael. Thank you, Ellie. Thank you, Jan, and Caitlin. And I'm going to turn it back over to to Anna, who has great announcements about lunch. The
Announcer 49:12
Engelberg Center Live Podcast is a production of the Engelberg Center on Innovation, Law, and Policy at NYU Law, and is released under a Creative Commons Attribution 4.0 International license, our theme music is by Jessica Batke and is licensed under a Creative Commons Attribution 4.0 International license.