Thanks very much. It's great to be here. So, today we're going to talk a bit about planetary intelligence. I'm going to start off with a little bit of an introduction to what Planet does. I'm going to be talking about this next phase of intelligence called planetary intelligence. And then I'm going to tell you why I think it's going to have a lot of practical applications to help us take care of our beautiful home, the Earth. So, let's dive in. At Planet, we set ourselves the mission of imaging the whole world every day. Myself and my co-founders were at NASA and we were thinking about how can we use space to help this planet? You see, in many ways you can say SpaceX is all about using space to go to Mars. We're founded on using space to help the Earth because it is a much better planet anyway. We love this planet and we love taking care of it. So, that's the founding mission and this is how the satellite system works. We have launched the largest satellite fleet doing Earth imaging in human history. There's about 200 satellites doing a polar orbit and the satellites stay fixed with respect to the sun. The Earth rotates underneath enabling it to line scan the entire Earth once every 24 hours. So, a new picture of every point on the Earth landmass once every 24 hours. And then we can zoom in on any particular location with higher resolution.
Now, so this is constantly updated. This is a launch we did literally yesterday from Vandenberg Air Force Base in California launching our latest satellite into space. SpaceX has helped make this look more routine. It's incredible. This used to be a huge amount of work. It's getting easier and easier to put payloads into space. This is our 42nd rocket launching our 736th satellite into space. And on this slide you can see that satellite being ejected from the rocket fairing just after about an hour after getting into space. And what is the upshot of all this space geekery stuff, the rockets and the satellites? Well, it's loads of data. We have 4 million 47 megapixel images per day imaging over 200 million square kilometers as the Earth's land mass is about 150 million square kilometers for reference. And we've done that 3,000 times. So, for each place on the Earth's land mass we have on average 3,000 layers documenting change over time. And we make that accessible over a platform that is a bit like Bloomberg terminal but for Earth data where it serves up various markets and I'll talk a little bit about some of those markets in a second.
Broadly, it fits into three major trends: digital transformation of companies and countries, sustainability of the planet, and peace and security and how we can bring transparency and accountability to peace around the planet. So, let me just whiz through a few examples. This is how we can help in agriculture. We can tell the crop type and crop yield in every farmer's field around the whole world every day, help them improve precision agriculture. So, we work with Bayer, Syngenta, BASF, John Deere helping them help farmers improve crop yields. We work with utilities like PG&E tracking power lines across California trying to stop wildfires. We help governments enforce permits. So, for example, here in Humboldt County cannabis growing is legal but you have to have a permit and a lot of the people don't have permits and we have an automated permit enforcement system that tracks for those places that haven't got their permits. We work with Swiss Re and Axa on insurance. So, instead of sending people out to check if an insurance claim is real, they just look at our satellite data, saves them loads of money. We work with disasters, disaster response. In this case in Lahaina, the before and after of fire. But, not just the pretty pictures. We actually provide building by building damage assessment. Brad Smith was on stage earlier releasing this model to the whole world. So, it's been a partnership between Microsoft and Planet so that we can do quick building damage assessment after every fire and flood and earthquake to help first responders save lives. So, and then in security realm, we do work with Ukraine, for example, protecting them. I was there just last week. We image all of Ukraine every day, all of Russia every day, tracking forces, trying to help them in their defense. And many humanitarian things besides. So, NASA harvested this work monitoring crop yields across all of Ukraine because Ukraine was serving half of the world's poorest countries with their food before the war and getting that crops out to those poor countries is really important. And our data was helping those NGOs. We helping map and track building damage across all of Ukraine to plan for reconstruction.
And here's us here this last week working with Halo Trust that is doing reduction of mines and removal of mines from fields after the aftermath of war regions in Ukraine. And they use our satellite data to hone in on where, looking back through the archive to determine where did the Russians put the mines so that they can go and take them out. In sustainability, we help track coral reefs with Allen Coral Atlas. We help track deforestation across 64 countries with Nick Ve and the food and agricultural organization here. And I just show you a little example here. This is what we do in Brazil where we don't just look at deforestation, we find illegal logging roads and they're the early signs of deforestation. And we do that across the entire Brazilian Amazon and we give daily alerts to the federal police and they have about 100,000 people using our data to try to use these alerts to go and stop illegal mining, illegal deforestation and illegal narcotic activities. And what's most important about that is not just that they did lots of interventions and seized lots of assets, but they've helped in the last couple of years reduce deforestation rates by 60% in the last couple of years with these combined tools. That is AI powering great stuff in the real world today. But there's a long way to go and I want to get to the... here's the one more project we do with monitoring carbon of every tree on the planet every 3 months to underpin carbon markets. We do a lot of work in research and one more example of impact work. We've launched the first satellite that is doing hyperspectral imaging that can enable us to detect methane and CO2 gas leaks around the world. The idea is that if we track those gas leaks, then the operators of gas leaks don't want them, the environment doesn't want them, we can shut them off. And we've detected thousands of leaks and you can access that on a public website. It's just carbonmapper.org and you can find where there are gas leaks and the operators can go and stop them. It's a digital public good. And we believe that satellite data can really help up to 14 of the 17 sustainable development goals because you can't manage what you don't measure.
However, there's a big gap. And the big gap is that satellite data is really cool in principle, but in practice you need teams of people with satellite imagery processing to get value out of it. The Federal Police in Brazil have a huge team that we work together with to process all this data, but most NGOs, most small businesses, most small countries can't get value out of satellite data. So, I'm here to tell you about Planetary Intelligence because I think that will bridge that gap finally and for all. So, about 8 years ago I gave a talk at TED called Queryable Earth. And I imagined that if you could use machine learning, you could index everything on the Earth. The buildings, the roads, the ships, the planes, the trains, everything. And if you could index everything on the Earth every day, then you could search the planet. You can search the planet for objects. So, a bit like Google indexed the internet, we're trying to index the Earth to make it searchable. And if we did that through time, you could search the Earth through space and time. And that is essentially what we mean by planetary intelligence. Planetary intelligence is about combining Earth data and LLMs to build a new kind of machine intelligence. You can call it large Earth models and they could power planetary intelligence. And this is both useful for extracting out the value that's latent in all this satellite data, but it is also about upgrading AI. You see, AI models that all the AI leading companies are trying to build real-world models, but of course AI is only as good as its training data. And for real-world models, you need real-world data. And that's where satellites come in. So, let me step back 1 second and tell you what are LLMs doing? LLMs have sifted through all the text of the internet, the images on the internet and put them into a model and you can chat with it and it's incredibly versatile and incredibly powerful. It can obviously write code, write your talks, write your speeches. It has most of human knowledge encapsulated in a model. I ask it about hard questions in theoretical physics and it knows the answer. It's incredibly deep in all the human fields of knowledge. But yet, for all that power, it's essentially blind. So if you are a farmer trying to understand how to improve your crop yields, the LLM is going to tell you lots of theories about agriculture, but it isn't going to know anything about your farm field today and what you can do to improve the crop yield or stop that blight or what have you need to do to improve the yield of your farm. If you're a journalist investigating a flood, you don't want a theory of floods that you can get from LLMs. You want actual information about this flood, which town is it affecting, how to send first responders in and so on. I.E. you need to add real-time information to our LLMs to be able to answer questions about the physical world. And may I add that as we build this planetary intelligence, I believe that it will help lock these AI systems down to the truth of the planet. You see, they spur off a lot of hallucinations, but the more we tie it to physical reality, the more it is tied to a truth. And so I think it helps with that vector, too. And this might sound far out, satellites and AI, but it actually can help us upgrade AI to have more practical applications for that farmer, for those civil government agencies doing emergency response, those first responders. For journalists trying to investigate what's going on in the world. We really hope that planetary intelligence actually makes AI much more practical to real-world applications, real-world people on the ground.
So, this is no longer theory. I'm actually going to show you what this looks like today. Because Planet has built an application, an AI application. This is it. It looks like a map, obviously, an image of the whole Earth. But importantly, you will see along the bottom there's a time axis. It has every day of imagery going back 10 years. So, it's your Google Maps satellite layer, but with a time axis. Not several years old from today or from 10 years ago and every day in between. And we've added a chat-like interface to it. And this is how it works. So, basically, you type your natural language query, and it can spit out answers about that. It can describe the imagery it sees, in this case zooming into this region in California. It can pick out imagery through space and through time, and help you do actual analysis of that imagery. So, it can describe the imagery. It's doing a sort of visual analysis. It can write code, plot graphs of what's going on on the world. And let me show you just one other example. A lot of people are interested in this meta world of everyone being excited about AI is, 'Well, the data centers, are they on track? What's going on?' And so, this is a little search I did on how are the data centers in this particular region being built up? How is that changed in the last few months? And it can quantify the answers of that. And we've just added one more feature to this application. Instead of just doing, you know, able to describe imagery and do charts, we've added a deep research mode such that it can spit out reports. So, this is a little example where it looked at data center construction across all of China. Literally over the space of a few minutes, it went off and did all this research and found all the data centers, modeled their construction, saw how they were going. And the whole point of all of this is to empower people with all of this satellite data. So, a few things about this app because I'm going to give you all early access to in a second. What it can do, it can browse the imagery, describe it, chart things. And I didn't sort of mention it can search for things. So, you can actually specify an object. I want to see blue boats near green bridges on a lake, and it will go and find all in the world those type of objects. Or you can find an object on the Earth and search the whole world for similar objects. And it can do that in a matter of a couple of seconds using an embedding technology we use in collaboration with Google. And what it can't do though, it can't give you driving directions. A lot of people go into it when they first use it and ask for driving directions. It can't do that. It can't see small objects. Somebody literally the other day asked for it for golf balls and looking for golf balls on golf sites and it can't see that because it's satellite imagery. It's pixelated as a couple of meters per pixel. And it's got a delay of imagery to protect our business because this is a free app at the minute. And it, like any LLM, is not very good at counting things. It's sort of just looking at the imagery and doing a visual analysis. And it can, of course, hallucinate and tell you incorrect things. You have to really probe it. You have to ask it the right kind of questions. But with those caveats in mind, this application, I think, can serve a number of user types. In conservation, for example, we had some really early users look at it and already detect encroachment in parks in Africa, which the rangers hadn't found, and sent the rangers to those locations so that they can stop that encroachment. We had people farmers looking at how they can improve their crop yields. We've had researchers do really interesting research using the data. And so, we really believe there's an opportunity to scale up the accessibility of satellite data. And so, I'm very, very pleased to say here to say today that you can all access this app. It's in beta. If you go to this, we're giving early access. We're announcing this here today. Early access for anyone, but especially those people in conservation NGOs, people doing humanitarian assistance, researchers, and things like that. Go on, try it. It's just in beta right now, but we hope that it can empower organizations around the world and experience yourselves the power of this earth observation. And it's free. Ultimately, it'll be a freemium model where there'll be some paid tiers, but for now, we just want your feedback so that we can test and improve this capability to make it better at serving the needs of everyone around the planet. And as I said, the ultimate vision of this is to democratize all this earth data and make it the foundation of people's response. Having toured around all the NGOs and startups here, so many of them based their stuff on satellite data as the foundation of their planning. This is a way to make that more useful, easier for everyone.
Now, that to me is what I would call phase one of planetary intelligence. I want to speculate a little bit about phase two. You see, we're working on a project with Google called Suncatcher. It's a moonshot, but it's putting compute into space. And you might think, why are you putting computers into space? It sounds like science fiction, but I'm here to tell you it's just going to be practical and cheaper than putting them on the ground. We did a study with Google about 8 or 9 years ago that looked at all the costs of terrestrial data centers and all the costs of putting them into space. And we calculated that when the launch cost come to about 2 to 300 dollars a kilogram, it's simply going to be cheaper to put the compute in space than on the ground. We're about 1,000 dollars a kilogram now, but the launch costs are rapidly coming down and within a few years, it'll be just cheaper to put them in space. And I think that's important for a number of reasons. I think this is the natural evolution of planetary intelligence. By the way, the insight behind why it's going to be cheaper is simply that data centers are primarily a power problem. And you would naturally use solar panels to power your data centers. Only then you get intermittent power and everyone wants their data centers working 24/7. But the cheapest way would be solar power. A solar panel in space gets five times more energy than on the ground because you don't have the day-night cycle, you don't have atmospheric absorption, you don't have angle issues. You can put it in an orbit where it has 24/7 access to sun. You don't have to have batteries, you don't have to have other generators and stuff. You would literally just have the solar panels, radiators, and the chips and ways of communicating that back and forth. Sci-fi authors have been dreaming about this for decades. In fact, Isaac Asimov talked about this. Arthur C. Clarke talked about some of these things. But the idea of putting energy-intensive infrastructure into orbit makes a lot of sense long run. This is the first thing to put in space. Putting energy centers in space to beam that to the Earth is actually very hard. But for data centers, all you have to do is put the data center up and you beam up your questions and beam back the answers and we know how to do that. So, it's just going to be cheaper in the next few years to put data centers in space. So, we're doing the first one where we're putting TPUs in space later this year, the first time those will be put in space. It's the early days of that project. But, the reason that relates to planetary intelligence is the following. You see, planetary sensing is naturally done from space. We've been doing that for many decades. NASA started it with Landsat and we're continuing that. But, the compute should sit next to the sensor data. Biology learned this. You will notice that your brain in mammals evolved next to all the sensors, the eyes and the ears. And that's because you want your compute right next to all the data so that you can process it fast and get real-time answers. What inspired that biological analogy, sensors already in space, the compute is going up there, they'll be together because the sensors and the compute belong together, and then we'll be sort of phase two of planetary intelligence.
And I want to end a little bit with some speculation and some musings about where we're going. You see, we're at an incredibly important juncture. And AI is the most important invention, arguably, that humanity has ever had. The ability to be smarter than if we invent a technology that's smarter than ourselves, is incredibly important. It's not just like another technology. It's the technology that can invent all future technologies. But, it also brings with itself existential risks. And as a physicist, I look at the cosmos and I often think about the Fermi paradox. And why is it that we haven't seen life out in the universe? I know some people think they've been abducted, but they're wrong. Roughly speaking, we haven't seen life in the universe yet. And we should expect it given the vast numbers of galaxies and planets, but we haven't yet found it. And one of the potential explanations for the Fermi paradox is that we blow get up to a certain point of technology and then blow ourselves up. And all you have to postulate for that, that there's some sort of great filter. And AI could be that great filter. Nukes, we just about got through the nuclear age, although it was touch and go. Will we get through the AI age? A lot depends on the answers to how we steer AI, which is why it's so important that we having this discussion about the future of AI and AI for good and steering AI for the positive benefit of humanity. But I want to say just the one thing about how planetary intelligence may be able to help that. You see, valuing stuff and learning about stuff are related. So, AI is basically blind to the physical world. With all its power. But if we tell it about the physical world, maybe it'll value it more. Maybe it'll value life more. Humans certainly do that. I've got into birding recently because my partner, she's into birding. And the more I learn about birds, the more I value them. It's not just some birds over here, there's all these different kinds and they tweet and all this stuff and different aspects of them. It's so exciting when you get into it. The more I learn, the more I value them. Perhaps, just perhaps, the more we teach AI about the planet, the more it will value the planet. The more we tell it about the rivers and the forests and the coral reef systems, the more it'll value those things. I don't think we should rely on that for AI alignment, but I think it could help AI alignment. Because in many ways we have an alignment triad. We want to align AI with humans, we care about humans and AI. We want AI to care about humans, obviously, and align with our interests. We also want AI to care about the environment and sustainability and nature and not just suck all the power and resources from it. We want it to positively help us and it has lots of potential for that. And we want, of course, humans to be better aligned with nature. We're still destroying a lot of the precious ecosystems of our planet. I think planetary intelligence can help us. It can help us be a smarter species taking care of this planet as the de facto stewards. It can help us be better aligned with AI because AI will value nature the more it knows about it and it will value humans the more it knows about them, too. That's speculative, but I think it's important. So, let me bring back and wrap up with planetary intelligence. We founded Planet on the simple notion that the more we understand the planet, the more we can take care of it and be smarter stewards of our precious spaceship Earth. I think planetary intelligence is that idea matured such that we can get to a smarter governance on our planet. Hopefully, planetary intelligence will first help us with insights, with those specific tactical things we can help take care of planet like deforestation and peace and security. Then it will upgrade to an intelligence, a way in which we can help steer the planet in a collective way more smartly. And then ultimately, hopefully, that will lead us to some sort of planetary wisdom, an ability to steer this planet in a smart way that makes sense. So, with that, I want to thank you all for listening and just leave you with a couple of links in case you want to learn more, including about AI and governance of AI and planetary intelligence and the link to that Planet app. Thank you very, very much for the time.
