"Seed Banks for Data": Stewarding and Preserving Climate Data – San Francisco 2025

[Music] Thank you so much for your patience. This is um one of my favorite panels I probably will ever moderate, especially because of that setting, especially because of the time that we're living in. My name is David Dao. I'm the co-founder of Gainforce, which is an NGO that preserves biodiversity data around the world. We work with local communities and today's topic is very much um about data, about seat banks of data.

Seat banks, some of you might know that term is seed bankanks are used to preserve the biology, the seeds of all the plants of the world in case something goes wrong and we need to replanted agriculture and we do see that need for data specifically climate data. Mark has done a really great job from the internet archive issuing out the concern that we're seeing the times we're living in where data is not long-term but very much short-term and even archives that we believe are federal and state governed are not lasting. So how do we in terms of data that is fluctuant preserve and create seed bankanks that preserve really important issue data and the one issue data that we really want to nail down today is climate data. And I'm joined by an esteemed panelist. I couldn't be happier with um my three fellow panelists that are all worldclass in what they are doing at the intersection of climate change data and the frontier of decentralized data um provenence.

So I would like to introduce Maria Sosa from executive director of climate change AI um Alan Ransel um co-founder of Cherty and Deonian AI and Marie Claire who is leading the youth negotiator academy training many many negotiators at the climate change cup conferences. So I'll give you a word like one or two sentences to introduce yourself and then we dive straight into the topic. Okay. Thank you very much uh David. I'm Maria John Suza and as David mentioned I serve as executive director of climate change AI and I'm also a postto at uh Cornell Tech uh public interest technology initiative.

Uh climate change AI is a nonprofit that started about five years ago because we were seeing a lot of data climate data becoming available machine learning was becoming a bit more useful than it used to be. uh and so there was an opportunity to create a community at this intersection and today we really empower a global community of innovators, practitioners and decision makers to accelerate responsible climate action uh using AI uh and we address um gaps in expertise education uh research to deployment infrastructure and a lot of coordination within this community. I'm Alan. I um as David said, started a company called Deonian that's building a set of reporting tools for carbon and nature projects. And so we use web 3 for what web 3 is really good at the underlying data that backs up reports, things like feasibility studies or monitoring reports for tree planting projects or biodiversity projects.

And then we use LLMs and AI for the things that they're really good for, for using that data and filling in the actual human readable sections of these reports. And what we're trying to do with that is really speed up the rate at which people can start new climate projects, things like tree planting or tree forest preservation and biodiversity projects and lower those barriers uh so people can take more climate action. and I'm also an adviser to the Falcoin Foundation on sustainability and the uh the carbon emissions associated with the the Falcoin network itself. Thank you. And my name is Marclair as David already mentioned.

Um I'm coming more from the governance side. Um we are training and enabling young people from around the world to take part in democratic processes at the United Nations on multilateral environmental diplomacy. And to just stick with the analogies, we are basically enabling young people from around the world and enabling countries to kind of like use these seed banks of datas to then make better informed policies. And I think like in the area of not only deletion of policies but also in eraser um of democratic principles and partaking in democratic processes. I think it's more important than ever that we're building an ecosystem which is like intertwined and really rooted in its um decentralized approach.

Maria the first question goes to you. So you have been involved in the Rio conventions. These are the three conferences that the world meets on to discuss about climate certification and about biodiversity. You have been uh in that process for a very long time. I think more than seven years um as a negotiator but also training these negotiators.

Now can you give us a little bit of an insight why we talk about climate data today? Like what is the connection between data and climate change and our efforts to prevent climate change and why should we even care? Well, big questions for to start with. We could probably spend the whole day and I hope I do not have to convince the room how absolutely urgent it is that we all step up our game when it comes to environmental topic, when it comes to biodiversity loss, when it comes to climate change. Um, we all feel it, but I think there is a new layer and with our data getting more and more accurate and our tools are getting better at forecasting, um, like there's really no excuse to not act, right?

And I think we all have to like do much more to really like step up our game. Um yet I also think it's really important that we are again taking the analogy with the seed banks that when we have data like who has been creating this data what biases are there who has been not engaged in the data um kind of process um looking into how indigenous um groups and territories how we have like whole huge gaps where there is no data because only when we actually have a good data map and a much better sampling of data we can actually deviate policies um which ultimately help all of us um to get better and I think like why it is important I think we all are steered by the common aim to have a livable future for future generations and we are unfortunately fast track um to not get there um with the current um models maybe like just like one number right I mean like there is when we look into data inequality um 85% of the data biodiversity data is currently from the global north whereas most of the hotspots on biodiversity first in the global south. So obviously like when we don't have the data um very often it's kind of like a very weak element to basically erode these um biodiversity hotspots around the world. It's very hard to preserve. It's very hard to make policies out of it.

Um and I think there we can all do better and with the technologies we are all engaged in and building and with the focus on um on public goods I think we really can can steer up and support um so many more communities to get the data sets to then get the policies. So we can all steer in the right direction together. A quick follow up on this. Do you think politicians act on top of data data driven or do you think data serves a certain kind of pressure tool for politicians to act or would they act without the data itself? Well, yes and no.

There are these politicians and these politicians, right? Um I do think that um having very solid data sets is helping a lot in the advocacy especially also for um for example for NGO groups for investigative journalism and we have seen over and over again how the pressure has been translating into policies. We also know that there are a lot of um environmental ministers, head of climate change office who deeply deeply care about these topics um and they also need the data that they can prove to the financial department that we need to implement this we need to protect this area right and it's not only like kind of like I feel or I heard which is also a very important component yet we also have the underlying data and I think all of this really helps um to either unmask what's being done or to support what should be done and again we all can do much together in kind of being informed citizens voting for people who actually take care um and have data informed and evidence-based policies. Nice. So just for the other panelists in case you want to respond or answer to any of your discussions feel free to do so.

This is meant to be a conversation but I want to move a little bit now from the human and the data to the machine and the data. So Marad, you're leading climate change AI and the two terms are already giving up my question spoilering it. So how important is data in terms of decision-m from machines? How important is that climate data when it comes to um artificial intelligence research? What do we see how those two soundly very distant disciplines are connected?

Yeah, that's a a great question, David. And as David spoiled it, I lead climate change AI, the two things that are more uh imminently going to kill us. Um uh so maybe bundling them together will help. Uh we're trying to do so. So we'll see.

Um definitely um data like I think uh one of the first things that uh computer science would learn when they would start look at machine learning would be you know garbage in garbage out. Um you know now we see um uh scenarios where uh a bit less data curation can can yield uh better results than maybe 10 years ago but it's still very much uh the quality of data uh uh a great concern in this space and we at climate change AI have worked a lot on both trying to build uh uh uh incentives for people within our community to bridge bridge some of these gaps. So, um four years ago, we started a program that was trying to uh bridge those research to deployment in infrastructure gaps that uh we created an innovation grants program that was funding particularly the creation of data sets, tools or simulators that then would enable the community to work on these problems because in many cases the data wasn't there at all. It's not on the internet in many cases. uh and a lot of the machine learning develops have been uh built uh on uh building models that are very good at parsing information that is structured in a particular way.

And if you start to look at addressing climate change from mitigation, adaptation uh and even climate science problems all within all the sectors that intersect with this be it power sector, be the biodiversity, be it transportation, uh be um uh disaster management and relief. All of these communities sort of uh deal with data, right? But they all deal with data with different ways with different taxonomies. And so it is a very hard uh uh work to bring that to to to structures that then machine learning can be applied to. Uh and that program sort of it started to create incentives for for the community rally around building those data sets.

And more recently we have started to look more at what are the data gaps. And so we develop uh uh or started to develop a a repository of what are meaningful data gaps that we need to address. uh and the taxonomy as it evolved as uh particularly focus on things that are critical that we see and are crosscutting across domains right sometimes the gap is usability sometimes is reliability of the data sometimes it's we need to measure things that are not observable or measurable directly and we need proxies for it so it is definitely a complex pro problem and it's not just like for researchers to address right like It's the utilities is the the do the practitioners and and and domain uh experts that are dealing with this data every day that uh play a role in bringing that this data to the four from a technical perspective. Um oh Ellen, you wanted to address yeah could I just uh jump in really quickly about this this question between the this question about the link between sort of data and decision-m and politics, right? Like as as someone who spent a lot of time in academia, I have to believe that you can use data to change people's minds because I don't know where we are without that.

And I I want to speculate a little bit that in the past, one of the reasons that it was difficult to use data on climate to change people's minds is that the world around them wasn't obviously changing. And so in the 90s when you held up these these curves showing the the concentration of CO2 going up or the you know these these data sets that were pretty clear if you're a scientist even at that point that the world is getting warmer. um it didn't like it didn't feel right to people, right? Like it was it was way too easy for people in the 90s to to say, "Well, you know, it's it's kind of fine." Like, "All right, we have some warmer days, but like it doesn't it doesn't like feel like the world is changing that much."

And I feel like we're at a point right now, which is exactly where the science said we were going to be right now. We're sort of at this tipping point where it's really obvious to people all of the sudden, right? It's it's obvious because hurricanes are getting worse. It's obvious because heat waves are getting worse because LA is burning down. There are flash floods.

You know, David, you grew up in Germany. There there have been like insane floods in Germany over the past few years that no one would have expected. Um, and I feel like there's that that like gut sense that the world actually has changed, that is growing in people and is going to get more intense, right, and more urgent. And I feel like that's where data can be really powerful, right? When you're telling people a story that feels correct to them and then you also have the data to back it up, I think that's how you can really change people's minds.

Um, which I think is why we're at such a crucial moment right now where it's really important that we preserve these these data sets, right? That these data sets are disappearing off of US government websites rapidly like you just heard about from the internet archive. And we need to we need to keep that data so we can as people's gut feeling that like something is changing fundamentally about the earth's ability to support our type of society. We need to preserve these data sets sort of now more urgently than ever. Maybe I have a quick followup to that.

I think there's a moment for everyone that uh it becomes obvious that this is happening but that moment is not for everyone at the same time. So the one of the challenges is this haha moment about climate happens to people around the world at different points. It's not like for instance co that it became uh obviously and and sort of there was like this collective effort simultaneous and and climate sort of it takes the years sometimes decades for for some of these effects to to to play and and th those effects also um are very sparse geographically distributed uh sparsely distributed geographically and so it's a bit more hard for for that uh collective effort to to emerge um naturally. Yeah. So we already heard the importance of data in human decision making changing people's mind and also for artificial intelligence algorithms to make informed decisions.

Now as we seeing data being deleted in an unprecedented way not just in this term but also in the previous term of the current administration. We are seeing dangerous consequences of due to the fact that our society and our actions depend on that data. What happens if we just remove we rockpool b society from that data right it's really important to now address the concerns and maybe hopes and ideas of building a seat bank for this data I think the internet archive has been an amazing place perfect place to host this panel to discuss how we can build a better library how we can build a better seatbank how we can preserve this really essential so Alan you've been building on decentralized data infrastructure for a while already at Falcoin and now with your company could Can you give us a little bit like a nerdy nerds can you nerd snipe us a little bit about what technology is out there to help us preserve this data in a more decentralized way? Yeah, 100%. And I think that I I may be preaching to the choir in a church which uh you know maybe is a good place to be.

Um but when we use web 2 tech the the problem with that the like the deep and fundamental problem is that we are addressing data based on where it is not what it is and then the Trump administration can give you a 404 error when you try to access this this data set that is really important to you. The data set probably exists still somewhere in the world, but you can't find it because the link that you have goes to a place on a server that that data is supposed to be. Which means that if someone moves that data, if someone makes that server unavailable, if someone corrupts that data, right, you could worry about actors going in. I'm not sure that the Trump administration is sophisticated enough to do this yet, but you could you could imagine them going in and just like not deleting data sets, but changing them uh to make their point. Actually, they use Sharpies, right?

That's how they do it. They just draw they redraw the past of hurricanes, right? Yeah. So, um the point is if you address data based on where it is, then the person who controls the server where it's supposed to be has all the power and you have absolutely no recourse if they change that data set or move that data set or make it unavailable. The like core realization of data management in web 3 that underlies IPFS and underlies sort of a lot of the philosophy around how we work with data is that if you address data based on what it is, then you suddenly are able to take advantage of this universal coordinate system for data provided to us by hash functions that gives us the ability to access that data no matter where it is in the world as long as it's, you know, connected to a decentralized network and prove that that data set is the data set that we requested.

And so this is how content addressing works. You don't you know that that random looking string of letters and numbers that uh that you use to to address data on IPFS for example is the output of a hash function run on that data set that lets you search for that data no matter where it is, no matter what server it's on on the IPFS network and then verify that it is indeed the data set that you wanted. And so you know I think that's that's like that's a fundamental set of primitives that is totally necessary if we are going to take data and preserve it beyond the lifetime of the like goodwill of a particular set of stewards right you actually can verify that this data set is being preserved over you know over the world over time um because you're you're like tying it to to the digital fingerprint of that data not just that application. So building the seatbank requires of course on one hand technology but it requires on the other hand agency humans that want to build this and also access to the data sources. Marla, you've been working with the UN for for a while now.

Do you think UN will use IPFS to store their climate data? Well, they have to and there is actually like now like a lot of effort to um digitize the the old data, the old decisions um because basically like they have been around since the the 80s and they still have like a huge archive of paper paper print documents um which we cannot currently access right and as I mentioned we are training young negotiators to partake in these processes and it's pretty challenging for them to go into the history because basically everything stops um around like the 2,000 years and they cannot like dig into the history which happened before. Um so I think like the UN really needs to um needs to digitize and archive their data to make it more accessible and additionally I think we also have to build on these data sets on how they can be like more easily be accessible how they can be like used better for not only decision-m but also for building your own opinion right because so the latest um there on climate change there is the so-called IPCC the intergovernmental panel on climate change you might have heard of it the US has also been withdrawing their participants, their scientists who have been um working in this council. So basically it's a global council of science climate scientists who are bringing all the scient all the science and make kind of like a state-of-the-art report um an assessment report and they just published the last one and it's like it's thousands and thousands of of of pages right this is the summary of what they have been doing. I mean I have read it but I mean I'm not going to ask anyone else to read this um and I think also like it's not really like smart to read it.

Um but I think like we can really use um kind of technology and agentic AI to to build on top of these tools to make the data more accessible and to also make it more user friendly. So it actually can inform us that we can actually take um evidence-based and scientific based um decisions because we we always talk about it but it's really really hard because it's exceeding our our human capacity to read through all of these different data and then they have like the lens of oh well there is also indigenous data which we haven't even grasped. So how do we go there? So um I know that and David you have been working on an on an AI assistant for indigenous data. We have been working on an AI assistant um working it's called poly um to support negotiators to read through documents to read through the science to better understand quite quickly what it is about and if it is identified it uh the right documents to then really dig into um and all of this is I think like really like a compass and navigation through um what is there because I think like in the vastness of all the data we have we need some compass otherwise we are kind of like being flooded and it's exceeding our capacity to actually take the good decisions.

So we are achie we are not achieving our goal right. Um and I think with the with the technology and especially also the the AI agentic um move which is coming up we can um we have to make use of it. Um so we are on the forefront of the good development of what these agents can do. That's really interesting. How the Trump administration right now is currently preventing us from accessing data is actually in a three-fold strategy.

Number one, it of course deletes the data set. Number two, potentially keeps the data set, but it restricts legal access to the pages. And number three is actually to make it just super hard to find by digging it deep into a website. So those three different kind of like barriers that they are building up on the data sets. How do you see potentially technologies such as AI or other um fields can change what Marla said making it more accessible making us filter through the information explosion but also making sure that we find the data that we want to archive.

Yes. Uh thank you for the question. Um definitely I think uh within the AI for climate community now with this change in administration that have been several efforts are trying to to to protect that that data and uh it would be great that these kind of uh adversarial sort of uh environments didn't have to emerge for people to rally about around some of these issues. But definitely it's um I think it sort of propels people to to to join forces in many ways that uh to to to to preserve that data. I think also we have seen or at least I I beyond the AI for climate community, I think the civic tech community uh is also looking a lot at these issues and trying to uh preserve that data and also build interfaces for people to to have access to that data.

You know, we have this great example of the internet archive uh and the way back machine that is a great interface that anyone with any level of literacy can use and in dealing with climate it's I I I work closely with the AI community but often uh technologies alone won't solve anything right uh the policy makers alone won't solve anything uh uh developers alone won't solve uh everything right and so it's creating interfaces that people can access the data in ways that uh they need like policy makers will need the data uh in in uh particular ways. Uh you know AI developers will want them in machine readable ways. Uh and often machine learning can also be a tool that can be used to to to distill that raw data and uh uh embed that that those solutions in uh those kind of interfaces. One one great example that I always give is climate trace uh which is the platform that is using a lot of uh AI to uh map uh global greenhouse gas emissions and is using a lot of AI to to to measure that information and and make it available for you know any kind of uh uh application be it in machine learning but also in policy making as well. Yeah.

Can I just tag on there really quick? I think that yeah like like you're saying like like building these interfaces and tools to allow people to easily access those data sets and know what to do with them and as as you were saying very clear too like building we have this this sort of exploding new set of of tools um that can can really improve those interfaces and like make a huge difference. I think the the other thing I'd add a fourth layer to this strategy that you're talking about, David, which is just like firing all of the people that are updating those data sets and using those data sets, right? If you are interested in public goods funding and sort of the history of public goods funding, I'd really point you back to this short book written by Vanver Bush after World War II called Science: The Endless Frontier. And we've gotten a huge amount of benefit out of this idea that that Vanderver Bush had, which was that we had in World War II, we had built up the capacity to do a lot of science and engineering very quickly.

And we had sort of changed the paradigm from the system, the world that had come before which was where private companies and individuals had to basically fund research to this world where because we had to fight World War II and we had to build radar and you know atomic physics and and all this stuff, we had built up this massive capacity to do science and fund that um using the the federal government, right? And it's been the most successful mechanism for funding public goods in the history of the world. And many of the the you know advances in medicine and in um in tech and computers like all the stuff that we rely on batteries certainly um I come from from energy. Um that's all being the consequence of this like great vision that Vanver Bush had. And the direction that I would really like to go in, right, is one in which public goods funding is such that if you are interested in working on something that could push the boundaries of human knowledge, you're just going to be able to do that as your job.

And I think that we we should, you know, go further in that direction, right? And it seems like we're going in the opposite direction right now. Yeah, seed bankanks are also a symbol of hope, right, for humanity because in case of it connects so well to climate change. Um, in times of disaster, we always have hope that long-term the next generation can use the knowledge that we created. So in in that way I would like to open up questions from the audience to the panelists.

Um if there are any questions open up the discussion to from basically the panel to the church back any other questions if not I have one more question to all panelists um so data oh yeah please thank you I'm curious how open data And this idea of data seed bankanks can be wired together to create funding mechanisms particularly for grassroots communities and indigenous communities who are already doing this work. The last one you guys want me to address the specialist? Yes, I think you're the perfect person. So gainf um the NGO that I started seven years ago. We pay people for conservation data.

Um we kickstart it by paying them for a Starling access because that is a human right we believe to have access to broadband and then based on that we teach them essential basic skills of initial small sensor like a bio acoustic sensor to collect bio acoustic data and every time they do this in a certain part of the world they're the very first person who ever collected this data set that's very valuable for science for decision making um for empowerment of the community for feeling proud they did a part to support uh the scientific consensus and knowledge and we pay them currently it's governed through a small endowment fund that we we were able to raise um by winning the XP price um and that endowment fund pays them 1 cent per minute of bio acoustic recording and 5 cent per megabyte of drone data imagery from the forest that pricing is determined through the committees themselves know exactly how much money we can give out through the interest. It's a predictable outcome. Um it's a little bit like a small octant for the people who know octant here in this community. Um the pace it out. So call out great project.

Um and yeah, this really helps on board the communities and make connect them to the digital world. We believe there's so much opportunity with the data economy um to trickle down the wealth from the global north to the global majority. Can I do a really quick plug too which is that if you're interested in this topic uh Falcoin is about to do a workshop on preserving climate data. Uh Sarah's going to lead that downstairs. We're out of time.

Yes, there's one more question here please. Then I think we are almost out of time. This is more of a an observation or I guess a question for Alan. Um hi Alan. How's it going?

uh there's a assumption among all the panelists that data is somehow generic that data can data sets can be mixed and I worry a lot when I hear people talk in the blockchain space and certainly web 3 about aggregating data that strips out the identity of how the data was originally gathered and who gathered it and what their confirmation biases were. And it's interesting when you talked about the value of data, you talked about fires in LA and you suddenly switched over from a climate data set mentality to a qualitative measurement analogy talking about qualitatively the value of expressing the impacts of something. And I think that's the challenge with all of this. How do you reconcile qualitative with quantitative data? especially since what you're doing is essentially stripping out the core identity of that data when it's put onto these chains.

Um, yeah, I'd love to kind of talk with you after maybe and understand what you mean a little bit more. Um, I think that when we are addressing data based on this digital fingerprint, the goal of that is to preserve the integrity of that data. And that's sort of a a hook that you can hang provenence information on that I think, you know, addresses some of what you're talking about, but I'd love to have a longer conversation about that. Well, wonderful. We hope that we gave you a little bit of hope um and a direction of building a seatbank for data.

Um hope is very rare nowadays in these times but I do believe data technology and human are the right way to steer forward that. Thank you so much panelists for this wonderful discussion and they are available here today as well to talk to. Thank you, David. [Music]