Our final speaker is Lily Pinault, data scientist and founder of AquaAdapt. We'll close with the striking case study, Le Signail, a building constructed 200 meters from the ocean and demolished in 2023. It's okay. The time is yours. We are 20 minutes. Okay, thank you. 15 bit questions, 20. Oh, okay. Hi, everyone. Bonjour à tous. This QR code links you to a website I made for the presentation. So it is the translation of the slides in both French and English. Bonjour à tous, pour ceux qui ne parlent pas anglais, le QR code vous emmène à un site que j'ai fait pour la conférence qui reprend les slides, mais en français et en anglais. So let's start in English. This here is Je Signale. It was built 200 meters from the ocean initially and was demolished in 2023. And then 78 families lost their homes. This was predictable.This here is what a coastline looks like when it's already losing. So erosion doesn't announce itself. It's slow and then it's sudden. Here is one little number, 1 billion people. So by 2050, 1 billion people will live in low elevation coastal zones exposed to rising seas. That is nearly one in eight people on Earth. And that is not a distance that, as just in France, 371 local governments are now legally required to map their coastal risks. And that is just in France and just a small part of the coastal communities. And most don't have the tools to do it also. So, Le Signal wasn't an accident. Oops. It was a failure of foresight. It's happening again right now on coasts that haven't made the news yet.So what do they see? That's a question we ask ourselves frequently. What do developers, insurers, local governments and planners, as well as people who globally need to take decisions about coastal zones, actually see and get? They get a spreadsheet. They get a rough estimate and they get a number with no timeline, no scenario and no address, basically. So imagine you're the mayor of a coastal city and you know your coast is retreating, but you don't know which parcels, which streets, when, how and what planning decisions need to change. So yeah, that's what you're working with right now as a mayor. And nothing tells you this specific plot under this specific scenario in 30 years will be at risk.That's the gap, because decisions are being made anyway, with or without the right information. Tools were built for researchers, not for decision makers such as mayors or other people. They're slow, they require months of processing before any output reaches the people who need it. They're expensive, you need a whole team of specialists just to interpret the results. And they are static. There is no personal level resolution, no forward-looking scenarios, no risk scores to act on. AI makes it possible to do this at scale, so at speed and at the resolutions that decisions actually require.But here's something that most people don't know. A satellite has been photographing your coastline since 1972. Every year, every season, for more than 50 years, images show how the coastline is evolving, meter by meter and decade by decade. So AI is built to learn from what these satellites show, combined with climate projections and coastal physics. So the data already exists. The intelligence to read it is what we are building.So to tell you a little bit more about how it works, the output is simple. You give us a parcel, so a place, and we give you its possible futures. So across 10, 30, 50, and 100 years, across multiple climate scenarios, and with a confidence level for every number. You don't need any GIS expertise. So GIS is for mapping, basically, for those who don't know, and I guess you don't know, because it's not your field. So everyone can read this, a mayor, an insurer, or a planner, or a developer for infrastructure or real estate. So let's take an example. Let's represent ourselves as parcel in Soulac-sur-Mer. So Soulac-sur-Mer is where the signal, Le Signal, the building I showed you first, was located before it was demolished. So under a low emission scenario, in 10 years, 18 meters will have been lost of vertical, horizontal coast. And in 50 years, it's already 58 meters. But under a high emission scenario, so take the IPCC scenarios, are you all familiar with that? Not really? Well, I'll tell you about it later. It's like the...scenarios that are made by climatologists to follow the future of the climate and of people on Earth and life on Earth, basically. And so under this high emission scenario, in 50 years, we will be at 118 meters. And in 100 years, the parcel will simply be gone. So each number, we wanted to come with its own confidence interval, because at 10 years, the interval of what we know is tight, and we're pretty sure about it, but at 100 years, of course it widens, and it's not that sure anymore. And if you take the example of ChatGPP, for example, it tells you it can make mistakes, but it's not quantified. And that's what we want to do.So our model says so explicitly because that is information that changes decisions and that is information that didn't exist before also. So let's look at what these numbers mean in practice. At 10 years, at the average erosion rates, we will have lost 43 meters. At 50 years, 250 meters. But at the accelerated rate, so under a high emission scenario, and I'm going to come back to the rates in a minute, it's 400 meters, and at 100 years, it's 800 meters, which is absolutely huge. So of course it's not in eight, it's in distance from the shoreline, so where the sea stops to the land. And so basically in the winter of 2013 to 2014, a single storm happened and it removed 40 meters of land. So that exceeded what projections said would happen by 2040 at the normal rate. So oceans don't wait for planning cycles and according to the CEREMA, so it's an institution, a French institution, the rate between 1997 and 2021 was of 4.3 meters per year. So this is what happens here.If we take the current accelerated rate, which is the case right now with the current emissions on Earth, we get to, in Nouvelle-Aquitaine, so where I grew up and where this signal building was built on, it goes up to 8 meters a year, so still horizontal, not vertical. Which is crazy huge. So there's another question we ask ourselves frequently. It's what disappears first. Is it the land or the trust in the tools that are supposed to protect it? Because there's a real danger with AI, and I told you about it just a little earlier. It's not that it's wrong, it's that it looks right. So same example with ChatGPT or Cloud or whatever you like. So for us, uncertainty is a primary output. It's not just a footnote. So basically every projection we give comes with its full scenario envelope. Visible, not buried. So we say where it is safe and where it is less safe, basically. So projections are conditioned on scenarios, and they are not presented as deterministic outcomes, but rather we show you possible futures, not a single answer like ChatGPT would.So, next slide. What should AI say then? Well, this is what an overconfident AI would say. It would say, this will happen. And this is what we are building, a useful AI that says, prepare for these futures, and here's how likely each one is, and here's where our... Certainty ends. So that is the difference between a tool that informs decisions and one that replaces judgment. So Aqua at the AI is building, as I said, the useful kind, as we like to say it. The one that says prepare for these possible pushers, not the one that says this future will certainly happen, because that wouldn't be true. And, oh yeah, so where our model is less reliable, we say so explicitly and we do not qualify it only, just like saying this is not sure, we say this is how much this is sure and this is how much like this is not sure. And also, I told you about it just a little earlier, but the confidence decreases with time. So the further we look, the less sure we can be about our predictions.It's logical, but it's important to say so because for decision makers, it's not always something that they think will be the case. So, AquaLab makes one thing visible. It's which parcels will flood, when and under what conditions, before the water arrives. So coastal risks will be one of the defining challenges of the next 50 years, and also before, and also after. But the science is here, the satellite data is here, as well as other data that we use. But there's a bridge between the data and the people who need to act on it. They have nothing between this science and them. So if what you've heard today resonates with you or your experience or something like that, I would love to keep going with this conversation. You'll find me still right here after the talk. And this QR code still links you to the website that I made in case you want to follow the slides afterwards. Because it explains better than I do. Thank you. Thank you.Thank you, Lily. So my questions, I do not know much about the domain, unfortunately, it's very interesting. The question is about which solutions do we have actually? So other than that, like, so after the prediction? Yeah, oh, of course, I didn't talk about that. Because it was more about AI. So you were wondering basically what happens when you have the prediction to take a decision? Or something else? Which options do the policymakers have to counter these futures? Well, countering is not something that I would really say is possible, because you cannot counter the ocean, basically. But you can adapt to it. It's something that I rather like. And there's a few solutions. There's gray solutions first. So gray solutions. It's like a dig, a wall, a seawall. I don't know if that's the correct word in English. There's like building walls basically, or there's green solutions or blue solutions, as we call them. And green solutions are, for example, replanting a mangrove, if it's possible somewhere, not in metropolitan France.obviously, but like in Vian, for example, that would be possible, I believe. I'm not an expert in that. So yeah, there's a few solutions, but it's hard to prepare for these and it's hard to know which solution will be the best. So we would like to build a product later that will help people make these decisions as well, that's like integrated as a package. So I'm just curious, what kind of models do you use to... For now, we're still building on them. So we don't know yet. We're like, we have gathered the data. We have worked on it. And now we're trying to find the best model and the best parameters and hyperparameters for our model. So I can't tell you that in a few weeks, if you want. For now, we're just having ideas. And within the team, we do not agree on them, by the way. So yeah, this is hard. And I would love insights from other people, if you want.Oh no, no, no. No, it's like we're building maps basically and reports. So the reports are generated from the maps and the data. So we don't write them entirely, but it's not conversional. We're building a platform where you just draw on a map the zone you want to have. And then you say like, I want this scenario, like is it high emission? Is it low emission? Is it normal emissions? And when you want it. And then you have the results.I'm wondering what the team consists of. What are your backgrounds? Okay, yeah. So I'm the original founder, and I have studied earth sciences in La Rochelle, so near the ocean, and basically I... I studied oceanography and climatology a lot. Then I came to Paris to study satellites and their data, which was fun, but I stopped. And now I'm studying AI in a master's degree. And so it's funded a little by my school. So we have a team from school to help us build the AI models and everything. And I have a co-founder now who comes from legal and finance and VC stuff. VC is investment, just in case. And now we also have an urbanist with us, which is cool. And my little brother who built the web platform. So yeah, it's like a family and friends thing. Okay, thank you, Lina. Thank you very much, everyone.