Subscribe
Share
Share
Embed
Welcome to In-Orbit, the fortnightly podcast exploring how technology from space is empowering a better world.
Dallas Campbell - 0:08
Hello and welcome to In Orbit, the fortnightly podcast exploring how technology from space is empowering a better world. I'm your host, Dallas Campbell. This podcast is brought to you by the Satellite Applications Catapult, a UK technology and innovation company driving economic growth through the commercialisation of space. And across this new series, we're going to be in conversation with some of the most inspiring minds in the country exploring the ways that the UK is using space to make huge differences to our everyday lives, as well as gaining a better understanding of its role in shaping and sustaining our planet for the future. In today's episode, we're going to be discussing how artificial intelligence can be used to enhance Earth observation and I'm joined in the studio by Dr. Freddie Kalaitzis, Senior Research Fellow at the University of Oxford, James Parr, founder of Trillium Technologies, and Maral Bayaraa, Senior Earth Observation Consultant at the Satellite Applications Catapult. Most of the images generated by satellites will never be seen by human eyes. There simply aren't enough human eyes on Earth, let alone trained experts, to sift through the terabytes of imagery generated daily and extract valuable intelligence and insight. Big data and AI approaches can be used for innovative and cost-effective management, and the processing of data, learning to recognise patterns and finding correlations that humans would otherwise miss. New insights unlocked in this way offer the potential to deliver value to a range of users across markets such as finance, insurance, transport and agriculture. Now, if you're a little bit lost by some of these terms, fear not, we shall go slowly. Welcome to the studio. Hey, you know what? This feels a bit odd because for the last three years, I haven't really interviewed anyone in an actual studio in the flesh. It's all being remotely.
James Parr - 2:18
Yeah, you're not in a box.
Dallas Campbell - 2:20
We're actually sitting in the same room. Is that weird? Does that feel slightly odd now or is that…?
James Parr - 2:23
Oh, it’s great.
Dallas Campbell - 2:24
It's a first for you. Freddie's never sat in a room with anyone.
Freddie Kalaitzis - 2:29
Before COVID lockdown didn't exist.
Dallas Campbell - 2:31
Yeah, it feels good. Anyway, listen, hey, Freddie, welcome. James, welcome. Maral, welcome. Thank you very much indeed for coming and taking the time to be here. You're all involved in this sector but in slightly different ways I think you all sort of interconnect and yet have different specialities which we're going to come on to in a minute. This is an interesting one because we've got some terms I think we need to define first, okay so artificial intelligence for earth observation okay these are terms I know our erudite listeners will probably know all about but I think for those who are new to the subject I think we should define our terms first. So Artificial intelligence is something we hear a lot about. Who wants to have a go at explaining artificial intelligence?
Freddie Kalaitzis - 3:17
You're looking at me.
Dallas Campbell - 3:19
I'm looking at you, Freddie. The reason I'm looking at you is because before we started recording, Freddie was like, "This is going to go on for hours. How can we do artificial intelligence in less than four hours?" Have a go.
Freddie Kalaitzis - 3:29
Well, I'll start from a variant of AI that's known as weak artificial intelligence. This is the one that we use daily in our smartphones.
Dallas Campbell - 3:37
What's it called? Did you say weak?
Freddie Kalaitzis - 3:39
Weak, yeah. So this is as opposed to the strong artificial intelligence, which is like the first dream of creating intelligent machines that behave like us, that can interface like in a very human way, that can solve any task very intelligently. But weak artificial intelligence is the type of AI that we have on our phones, where they can transcribe spoken language, where they can translate from English to Chinese and vice versa, where they can automatically detect humans in photos and detect vehicles from satellite data. So that's what we mean, weak artificial intelligence. So weak in the sense that you take a lot of these menial cognitive tasks and you allow a machine to learn to perform that task, but you can be giving lots and lots of examples and yeah, like the whole point is that there's just so much data out there generated daily by satellites that we just don't have enough human eyes to plow through these images and perform certain detection tasks. So we use machines to scale that detection task.
Dallas Campbell - 4:44
So basically machines, computers, algorithms, mimicking the human brain. Because one of the things that you hear a lot about is this idea of general intelligence as opposed to specific intelligence. So you mentioned weak intelligence So where are we on this sort of curve? Is it weak intelligence because we haven't invented better intelligence yet, or does the weak AI do specific jobs, like menial tasks, like you say, and a better AI will come along in the future?
Freddie Kalaitzis - 5:15
It only means in a way that it's very specific to the task. Like a realistic AI nowadays is good at a specific thing, like let's say detecting agricultural fields in the UK. If you want to use the same model somewhere in Africa, it will probably fail because it hasn't seen enough data. It cannot generalise to new areas. So it's very weak in that sense. It's very specific. But general artificial intelligence, we usually think of it as a human-being level of AI that can just generalise instantaneously to another task.
Dallas Campbell - 5:48
I'm always blown away by AI. I mean, even the weak stuff. But of course, presumably AI is nothing without data. Data is the thing that it feeds on.
Freddie Kalaitzis - 5:56
Same as humans. Like, you wouldn't develop cognitively if you were stuck in a black box throughout your childhood.
Dallas Campbell - 6:03
Yes, no, exactly, exactly. Okay, so basically there's not enough brains, there's not enough eyes to look at the data. So that's kind of AI. Okay, let's move on to the next term, earth observation. Maral, tell us a little bit about what that is for someone who's never come across the idea of earth observation.
Maral Bayaraa - 6:22
Well, I like to think about earth observation as almost like these microscopes, you know. So when we invented microscopes, we can suddenly look and understand how the world works on this tiny scale. So we've solved diseases, right? But Earth observation is just these huge microscopes floating around the Earth.
Dallas Campbell -6:44
In orbit.
Maral Bayaraa - 6:46
So there, yes, exactly. Yes, in orbit microscopes.
Dallas Campbell - 6:49
When we think about that, when we think about satellites in orbit, we generally, to those people who aren't involved in the industry, in the industry, they generally think about things looking outwards, you know, the Hubble Space Telescope or, well, I suppose communication satellites, but this idea now that we've got a whole...
Maral Bayaraa - 7:03
That we are in space looking back at home, at Earth, to solve problems on Earth, right? And a really nice kind of example of this is that recently news made me really happy, which was basically a bunch of these Antarctic glaciers have been named after the satellites that have first kind of quantified climate change and how the kind of glaciers are moving. So a bunch of these Antarctic glaciers are named after the European Space Agency satellite ERS, Envisat and Sentinels, and the kind of American, you know the Landsat and even the Japanese satellites like ALOS. So already satellites have helped us understand the world at the macro level, at this kind of climatic level.
Dallas Campbell - 7:51
Interesting. So just give us our listeners an idea of when we talk about satellites. In the late 1950s there was one, Sputnik, and that didn't do very much. How many are up there and are they all looking down at Earth? What's the scope? The quantity of them, how many do we have? I mean, how many?
Maral Bayaraa - 8:11
I think that is changing all the time. And Freddie, you're looking at me like maybe you know the number. What's the number?
Dallas Campbell - 8:18
A lot. I think it's basically a lot.
Freddie Kalaitzis - 8:21
In the order of tens of thousands right now.
Dallas Campbell - 8:23
Really?
Maral Bayaraa - 8:24
But I can tell you that, you know, there's been a really big change in the way satellites have been... the kind of the way we put satellites into space. Because before, for example, we had satellites the size of big fridges, right? And to send that huge satellite into space is very expensive and it takes decades of research because it's so expensive to put it in. But now we have what is called, as you know, a small satellite, which is the size of a bread and to send that bread into space. So, I mean, we can get back into this, but it's changing.
Dallas Campbell - 8:57
You know, the launch is changing so we can send more things up there cheaper. Well, just explain to us, you touched on glaciers there and I think obviously that ties in with climate change. That's the thing that I think most people think of. Just tell us why Earth observation is so important and what are the types of things that we can learn by having these microscopes, if you like, these things, telescopes looking back at the Earth. James.
James Parr - 9:28
It's a very exciting time actually because I think the combination, as Freddie said, of AI that can actually absorb all this data and start making sense of it, plus the fact we now have the ability to see the Earth optically. We could also see it in multispectral and hyperspectral, so looking at different forms of energy so we can make more detailed assessments on things. But also radar, so technology called Synthetic Aperture Radar, which allows us to see things in 3D and also through clouds and see what's happening at night and all of these instruments and satellites are allowing us to sort of build a picture of what's happening on Earth in a way we've never been able to see before and then we use AI to actually make sense of all this data. So for the first time, we can sort of do a full body scan, so to speak, of our planet, to see the pulse of the planet, all of the variables which matter in terms of managing how it's going to change and this matters. This is important for both how we manage our systems but also planetary systems, biospheres and those sorts of things.
Dallas Campbell - 10:35
This might sound like a really stupid question, but why do you need to go into space to do that? Like what does low Earth orbit give you that the technology you couldn't just do on the ground? Like why do you need to be there?
Freddie Kalaitzis - 10:47
Well it allows you to zoom out really. I think you have to understand what remote sensing really is here, right? So people don't really appreciate that remote sensing is about understanding properties of things from a distance without actually going there and physically interacting with the object and the only way to do that is by capturing the light that it emits. Think about that for a second. So when the farther out we go from Earth, the more of its light we see, and what better way to do that than being in an orbit around the Earth. So really what satellites are doing is like, you know, as James said, optical satellites capture light in particular band of frequencies that pertain to visible light and near visible light. Perhaps even, you know, sometimes you want to look into frequencies that pertain to radar, which are slightly larger frequencies, sorry, wavelengths. But the nice thing about that is that the device itself can emit that light down into the earth and then as it reflects back, it recaptures it. So by that way interacting with the object through light, it can tell certain properties about the object. So it's all about playing with light in Earth observation.
Dallas Campbell - 12:02
That's interesting. Just the way you explained it there, it's almost like you step back from a picture. If you look too closely at a painting in the National Gallery, it doesn't make sense, then you kind of take a few steps back and suddenly it's like, okay…
Maral Bayaraa - 12:16
It's like a Monet.
Dallas Campbell - 12:18
Like a Monet. Exactly.
Maral Bayaraa - 12:19
When you look so closely, the brushstrokes are so huge that you don't see what's happening.
Dallas Campbell - 12:23
If you look at an oil patch really close, you see completely different things. You see the human touch, exactly, brushstroke, but suddenly you step back and you see what was in the mind of the painter, I suppose, and in a completely different way. Okay, well, just climate change. I guess that's the obvious one. Yeah. What what else do we do these satellites tell us? When you say we get a full body scan of the earth, that's a nice way of thinking about it, what else are they looking for in their, along with their different frequencies, different wavelengths?
James Parr - 12:54
Well, I think the one that often gets talked about is shipping. So, managing fisheries, but also piracy, and of course there's lots of applications in reconnaissance and situational awareness in the military, those sorts of things.
Dallas Campbell - 13:09
I was going to ask you about military. I mean presumably having that kind of technology is pretty good If you are a military general and you want to understand the situation.
James Parr - 13:17
Well in Ukraine one of the reasons why they had suspicions that it was going to happen as they could see all of the Russian Armaments starting to build up on the Ukrainian border and this was actually from open public data. But this was something that allowed them to realise that actually something was about to happen.
Dallas Campbell - 13:33
So shipping obviously we can see ships, so I mean is it a case that you being able to see things like shipping lanes will then Inform us of how to do things better or like what what's the reason why we want to kind of look at things like shipping?
Maral Bayaraa - 13:49
I guess James you're referring to the AIS signal being also detected from… so this is very interesting because basically every ship above a certain size has to have this AIS signal and it's actually designed for ships to be able to not collide on the ocean but then someone said let's just take a satellite into space for it with an AIS detector and see if we can detect it from Space and we can and then we can even track where they're going.
Dallas Campbell - 14:16
So this is like an identification signal?
Maral Bayaraa - 14:19
Yes, exactly and then as you're tracking it, different fishing behavior show different things. So for example, like longline fishing creates these kind of loops, these hooks, right? And then you can train an AI to detect these hooks or, you know, like trawling looks different or fishing looks different. So there's this whole world of proxies we didn't think it was possible. There's a case study. This is kind of why we're doing this is an example of why we're doing this. So, for example, the Thai government, to trade fishes with the European Union, you have to have this kind of green card system, right? So if you have the green cards, then you can sell your fishes. Whereas if you have the amber card or red card, you can't. The Thai government was on the amber card, but then they've used this AI for EO within this particular use case to get their green card and so if that's not impacted, I don’t know what is.
Dallas Campbell - 15:10
Basically all human activity, you know, as well as the health of the planet can now be monitored using satellites to some degree.
James Parr - 15:19
Even large animal, yeah.
Dallas Campbell - 15:21
Okay, well let's, okay, we might come back to climate in a moment. Give us some other ones that people might not think satellites are being used for. Earth observation.
James Parr - 15:30
Penguin dung.
Dallas Campbell - 15:31
Penguin dung?
James Parr - 15:32
Yeah.
Dallas Campbell - 15:32
Why do we…
James Parr - 15:33
So they can see the colonies of penguins from looking at their dung on the ice, all those sorts of things. Yeah, and elephants, you can, although elephant is just about a pixel, you can see their shadows.
Dallas Campbell - 15:47
Oh my God.
James Parr - 15:48
Yeah, and then we're…
Dallas Campbell - 15:48
Actually, when you say pixel, What kind of resolution are we talking about, particularly in visible light?
Freddie Kalaitzis - 15:53
Well, the best resolution commercially available right now is something like 30 centimetres per pixel. So this is enough to resolve the shape of a car or a very small hut in the desert. Yeah, and this is just commercially available. I think if you go into more spy kind of level of intelligence.
Dallas Campbell - 16:18
That's what I want.
Freddie Kalaitzis - 16:19
You could probably look into it.
Dallas Campbell - 16:20
That's what I want.
Freddie Kalaitzis - 16:21
Well, appropriately it's 10 centimetres or even lower than that.
Dallas Campbell - 16:24
Is it a case, Maral, that you know we have this technology and people see it being used in particular areas that suddenly people are like "I know we can use it for this" and come up with new ideas that can be used for?
Maral Bayaraa - 16:34
Absolutely. So for example, mining, you might think, so for example, you know, we are in the midst of an energy transition and metals will be the key to energy transition and so where are all those metals that are going to come? So satellites are actually used for example, mineral exploration. So at the Catapult, for example, we've done some research on, you know, looking for lithium in the UK using satellite data, or my particular research is very much concerned with making sure that mine waste doesn't collapse and destroy the environment. So we need mining because it's at the core of our energy transition. But then this metal hungry future means that there's going to be so much mine waste and all these other things that nobody's thinking or people we need to think more about and that needs to be solved and satellites, from my perspective, have a huge role to play in this.
Dallas Campbell - 17:27
So really, it's like we're thinking about the problems that we face, things like energy transition and then being able to apply what we learn from our eyes in the sky as it as it were.
Maral Bayaraa - 17:36
Exactly.
Dallas Campbell - 17:37
And well that's that's my next question you know because obviously you can generate lots and lots of data and then you can have things like artificial intelligence that can process some of this data and make sense of it but ultimately it's up to us in terms of what we then do with it and how we and how we behave.
Freddie Kalaitzis - 17:53
That actually touches on kind of the second main use case of AI. So the first use case that I mentioned was about scaling menial cognitive tasks. The second one is about defined complexity. So there are certain patterns in nature that we humans are just not wired to perceive in the right, let's say temporal scale or special scale, like because we're not tall enough or because our lives are too short or we just, we don't, cannot perceive the right wavelengths, right? So what AI is really good at doing is by using statistics and correlations of data, it can perceive captured cues that were just not wired.
Dallas Campbell - 18:33
So pattern-seeking.
Freddie Kalaitzis - 18:34
Pattern recognition, pattern-seeking.
Dallas Campbell - 18:36
I mean, humans are pretty good at pattern recognition generally.
Freddie Kalaitzis - 18:41
Yeah, within our senses, like at the right temporal scales, you know, in the order of ours. But if something lasts years, we're probably going to forget about it. So we need something to capture this data and then something to analyse it.
Dallas Campbell - 18:65
Yeah. Do you have any examples of that where AIs use kind of pattern recognition to…?
James Parr - 19:01
Well, there's a good one, which is the relationship between aerosols and the persistence of clouds. So clouds form because of droplets in the air, aerosols. We can't see them, but of course, satellites can. But an AI plus that satellite data can then compute how long those clouds are going to stay in the sky and that matters because clouds are reflective and so we can then use that prediction to figure out how much energy is going back into space.
Dallas Campbell - 19:29
It's really interesting. So both of these technologies, EO, Earth Observation and Artificial Intelligence, they kind of need each other, don't they? They do. EO doesn't work on its own without having the grunt work that artificial intelligence sort of offers in terms of painting a picture that we can understand.
James Parr - 19:45
Well similarly with methane, this is another climate use case, methane requires human beings to see the plumes, just because of the characteristics of methane. The spectral signature also interferes and you can see rooftops and requires a human to go well actually that's a plume and not a rooftop. But we can now use AI to do that task and so the huge task of mapping the world's methane output can now be done by AI.
Dallas Campbell - 20:12
Yeah, Freddie?
Freddie Kalaitzis - 20:13
An interesting one I think also is you can, and this relates to food security and you know having enough food on the planet, you need AI to look at the temporal patterns of how a crop field looks like across time. So a healthy crop field looks a certain way and in fact that crop field looks a very different way and a human cannot just be there all the time to observe it and measure all the time. This is really easy to do using Earth observation data and having a very crude model to understand healthy versus not healthy.
Maral Bayaraa - 20:46
At this point, I guess maybe to build up on what Freddie is saying, to introduce the concept of… I like to think about materials on the ground kind of have their own spectral signatures like the way humans have fingerprints, right? So to build up on what Freddie is saying, the kind of healthy vegetation and the unhealthy vegetation have different spectral signatures in the infrared part of the electromagnetic spectrum. So then you can train an AI algorithm to kind of understand which fields on a huge scale are showing this healthy pattern and which fields are not showing this unhealthy pattern and if you even go further, then you can say, this time last year, this many fields showed healthy signature at this time compared to this year. So this year, do we expect the same amount of food grains as last year? And what is the global implications of that?
Dallas Campbell - 21:45
It's unbelievable. Actually, a few years ago, I did a documentary series where we used exactly that technology looking near infrared for archaeology, actually looking at the signatures that the ground gave us in Egypt. So we found all kinds of things that nobody had and actually, in archaeology, it's quite interesting. Back in the day, people would do exactly that. They would in early airplanes have a look to see, just with our eyes at the ground, to see how the Earth had changed. But actually being able to do it from space now, and being able to do it not just with visible light, but through infrared and other signatures.
James Parr - 22:21
And AI, to see methyl-like structure.
Dallas Campbell - 22:24
Plus AI, exactly. I think maybe AI wasn't invented by digital. It was just us looking at things going, "Wait, what's that? "It's a pyramid."
Freddie Kalaitzis - 22:30
There's an interesting type of instrument called LiDAR.
Dallas Campbell - 22:34
I like LiDAR. I’ve got it on my phone, my blinking iPhone has them on it. I'm like, crikey.
Freddie Kalaitzis - 22:38
Yeah. So it's done from an aeroplane, and it shoots hundreds of thousands of laser beams on the ground per second and in the Amazon, they've done it and they found all the infrastructure from civilizations that we didn't think were possible, you know, dated back in the day. So now it's creating a new revolution in archaeology.
Dallas Campbell - 23:01
I think actually new revolution, in every form of human activity, we seem to be new revolutions in terms of what AI and what EO can do. Just the clarity by which it presents the world and the new ways of seeing and then being able to interpret that is just phenomenal.
James Parr - 23:20
I think it's bigger than people think.
Dallas Campbell - 23:22
Well, here's the thing, how come, like we're sitting here in Oxford and you'll work in this area and we know what it is, but generally I don't think people are aware. Like for example, we have conversations, like we shouldn't be sending things into space because it's bad for the environment. We wouldn't understand climate change if it wasn't for putting things into space presumably. Presumably what we know about climate change is because we have things like satellites and…
Freddie Kalaitzis - 23:50
Yeah, I think that kind of touches on the other main use case of Earth observation, which is like, we talked a lot about the sustainable development goals and how it can help nonprofits. But I think earth observation is also being used to answer fundamental questions on climate, ecology, society, and sustainability. Whereas SDGs is more about mitigation and protection, like scientific questions is more about acquiring a fundamental understanding of the processes in the earth and every good way to do that is looking at it at the micro scale. In large spatial scales, but also in very large temporal scales.
James Parr - 24:33
There's a big one we haven't mentioned, and that's weather.
Dallas Campbell - 24:36
Yes.
James Parr - 24:36
And so this is this incredible feat of science and engineering that we don't think about, but the ability to have like a seven-day weather forecast, which we're now used to, this is incredible.
Maral Bayaraa - 24:47
Especially in the UK.
James Parr - 24:48
Yeah. But it's amazing for everything from shipping to airlines to our supply chains. We rely absolutely on space data for weather prediction. But to your point, people have taken that for granted and I think it's about a trillion dollar a year value to the human race, just the ability to predict the weather. But if you add the ability to join climate predictions to weather predictions and close that gap, which is what AI is starting to do, then you have the ability to start looking years into the future and start making plans and knowing how that world's going to change and this is the world of… sorry the study of some earth systems.
Dallas Campbell - 25:28
Yeah, and it's interesting having all this technology satellites, earth observation that we get this clarity and we get an interpretation of it through AI It doesn't necessarily though make the world a better place presumably is still up to us about what we do with it we still have to make decisions and you know is having all that information. Is it all good or is it can it be a hindrance as well? I wonder.
James Parr - 25:52
Well no because I think that's where it actually matters is decisions don't happen when there's fogginess about the outcome and In a way one of the reasons why we've been sitting on our hands in terms of preparing the future which is now a policy, you know sort of changing climate AI and the sort of the best tools we have to actually understanding and predicting how things are going to change and this is the world of resilience and so how do we design our cities and design our coastlines to really understand how things are going to change we can now predict for example storm surge using AI. So then we can say well listen, this is what resilience should look like for the decades ahead.
Dallas Campbell -26:31
Slightly off topic… not really off topic, but we've become or we're becoming incredibly reliant on this and just how good this technology is How is it a danger we become over reliant on it? Like how safe is the system? Like what happens if it all stops suddenly?
James Parr - 26:48
Does anyone want to mention GPT?
Dallas Campbell - 26:51
What's GPT? Well, you just mentioned it.
James Parr - 26:56
Well, it's the ability of language models to make extremely convincing-sounding responses and so we're witnessing this massive change in AI, where instead of putting a question to a search engine, the AI will provide a written response, sounds like a human being. The problem is it does it with such authority that people trust it automatically.
Dallas Campbell - 27:18
Crikey. So it's like kind of deep fake technology.
James Parr - 27:21
It is. It's very similar to that, but it's…
Dallas Campbell - 27:25
Everyone thinks it's Tom Cruise. It's always Tom Cruise for some reason. And then it's not Tom Cruise. It's like some kind of weird.
James Parr - 27:30
Yeah. But imagine a response which is, which, you know, it's, it sounds as authoritative as Oxford PhD, but it's all hallucinated.
Dallas Campbell - 27:38
Okay. Well, let's think about the future. We've got this revolutionary game changing technology, and we've all given some really interesting examples of that. Where are we going with all this? You know, you talked about weak AI. Let's just project ourselves into the future a decade or so. Like what's coming online? What are we going to be seeing? What's exciting? What's terrifying?
James Parr - 27:57
So chat GPT, which is what we just mentioned, can write code.
Dallas Campbell - 28:00
This is the AI can write it saying, oh, crikey, are we going to get into some terrible singularity and it's all going to get paperclip thingies.
James Parr - 28:10
So what it can do, which is really incredible, is it can also write code that goes into a 3D program. So for example, we could say, hey, write me some code that designs a chair and it'll produce that chair in 3D and then you can 3D print that. So you get a chair out of thin air, a little bit like the Star Trek replicator, if you remember that. But if you could also say, all right, design me a sea wall and build it in the 3D and the AI will know, as Freddie pointed out, it'll know things that no human being could know. Like how that, what the rocks like, what the storm surge is gonna be in 20 years, like all those variables. It can build into the design of that sea wall and then build that in 3D that then goes to the engineers to construct.
Dallas Campbell - 28:53
That's amazing.
James Parr - 28:54
That is five years away.
Freddie Kalaitzis - 28:56
Well, I think that a lot of what chat, GTP or GPT, I was confused. It's predicated on existing data that humans have already generated. So if no one's done the exercise, as a counter argument to your point, if no one's done the exercise of actually creating that stone wall, done that case study, correlating with all the different kinds of physical phenomena, then the best it can do, it can probably extrapolate by using existing knowledge base and we don't know what level of accuracy it can do that.
James Parr - 29:37
Have you used MidJourney? Have you used Dall-E?
Freddie Kalaitzis - 29:39
Yes.
James Parr - 29:39
And so these are generative tools based on the same foundational models that drive GPT, but they can give you options and so you say, well listen, can I have a dog wearing a spacesuit? And five minutes later you have 40 of them, and then you choose the one you like. So imagine that for seawall design.
Freddie Kalaitzis - 29:59
That's very nice. I think these nice things about these generative models like Midjourney is that they all appeal to our human senses. We like pictures. We're very visual beings, right? We're less better at sound, less better at smelling things like dogs, but we're very good at like pictures. So Midjourney works really well for that. For something like megaprojects, we have no idea how to assess that.
James Parr - 30:27
That's a good question though.
Dallas Campbell - 30:28
Maral, tell me about your thoughts and hopes and fears for the future with this technology.
Maral Bayaraa - 30:34
Thank you. So for me, I think the hopes and fears is that I think the EO for AI will really help kind of connect us in a sense. So for example, you know, I don't want to go too much into this, but just to mention like carbon credits on the table, right? So a lot of industries who are going to have to create some kind of emissions as part of their industrial working, they are trying to reach net zero by then maybe, you know, growing some more trees or helping fund somebody who's growing some trees in some other part of the world. So using this technology, maybe, you know, a nomad in the Mongolian Gobi Desert can be funded to help restore the nature that it needs or the kind of the grasslands it needs. So that's the case. Or a farmer growing a bunch of crops, maybe they can get the information they need on the ground from all this global satellite data set to say that their crops might be in danger because of all this data that we're analysing with the EO for AI or the miner, the person who has this can then be told, Your mine waste in these areas is looking very dangerous from our satellite data. So please send more engineers on the ground to help that. So I guess my one is more kind of like the infrastructure of society, all these different industries. How can you for AI combine to become so common that everyday problems are solved by that?
Dallas Campbell - 32:06
That's interesting. So really we're looking at the health of the planet itself the various earth systems, climate, weather, the earth itself, geology, but also human elements, human geography, if you like. Are you sort of suggesting a sort of more equitable world, do you think?
Maral Bayaraa - 32:23
Absolutely and I think that is my hope. I mean, if it's not serving that, why should it exist? Sorry, that's a big statement.
Dallas Campbell - 32:34
No, it is. But my question is, does this technology, which is awe-inspiring, does it then lead lead to change human behavior, I suppose is my question. Or, I mean, the hope it would be, but what we do know is about things like facts do not necessarily change minds. People make decisions based on values, based on trust, and I wonder, especially with the speed of progress of this technology, I wonder if it's how people are going to react to it.
James Parr - 33:04
I think we are emotional beings, But in the end, data is extremely powerful for decision-making. We have a problem that the Earth's are heating up really rapidly. In fact, the latest reports are looking at between three and four degrees. The IPC reports from last year. That is the end for the human race. Let's not beat around the bush. This is a sterilisation event for life as we know it. Imagine every pixel on Earth is currently suboptimal in terms of its carbon sequestration. Could we use AI and EO to look at every pixel and say listen, if we just crank that up by 5% or 10%, we could start to use the land and use the sea to absorb more CO2 and these technologies allow us to think like that and start to optimise for that future. We have to do it. We have no choice. We've got about 20 years to pull it off.
Freddie Kalaitzis - 33:59
Remember when I said that it helps to be high enough to see the large picture? not just in space but also in time. So there's no better way to demonstrate James's point by other than looking at the polar caps and how they shrink and expand throughout the year and as you look year after year, the shrinking and expansion becomes smaller and smaller and smaller to the point where it will flatline at some point.
Dallas Campbell - 34:26
That's it. It is the technology that hopefully will get us out of the problems we're in.
James Parr - 34:32
It’s like Gandalf
Dallas Campbell - 34:33
Because I always think it's not just a question of just turning the taps off. We actually have to be much more fundamental in terms of how we use technology to do things. Well, carbon sequestration is a really good example and actually, yes, being able to step back from the canvas and look at it gives us that tool.
Maral Bayaraa - 34:48
And I think earlier, Freddie mentioned about sustainable development goals and how the technology can feed for that. You know, in some ways, the power of EO for AI is also in monitoring how well we are going towards the sustainable development goals and kind of keeping governments and companies accountable for that.
Dallas Campbell - 35:10
It's funny actually looking at the when you look at the UN sustainable goals, EO and AI are across all of them pretty much. This is a technology that can be really used across the whole suite of goals that they have. It's really interesting. Actually, you did a white paper on this. I just want you to explain.
Freddie Kalaitzis - 35:28
The state of AI for Earth observation.
Dallas Campbell - 35:30
Exactly. So this particular white paper, just explain to us where that came from, like why you did it and what were the conclusions.
Freddie Kalaitzis - 35:41
So let me kind of touch on the vision of Earth observation because what we do now is towards those end goals. The question I'm trying to answer with Earth observation and something that AI is only helping in us doing is the question, how can we tell what happens on Earth based on observations from space? How can we let the data tell a story of a natural or an anthropogenic phenomenon? How can we meaningfully combine all the sensors, maybe even coming from very fundamentally different mechanics like radar versus optical cameras. You know, we're getting all the sensors, how can we place them accurately on the earth in a harmonious and continuous way? And lastly, you know, all these sensors are really noisy. How can we deal with those noisy sources? And how do we know what we don't know?
Dallas Campbell - 36:38
Yes, the unknown unknowns. Yeah, that's really interesting. How can we get the general population to understand this? How do we bring it? Because I think it's too important for it to be stuck in laboratories in Oxford or wherever it is. It's too phenomenal a technology, I think.
Freddie Kalaitzis - 36:56
So this report was funded and paid for by the Satellite Applications Catapult and it's aimed towards the general public and any jargon that you might encounter, there is a really nice…
Dallas Campbell - 37:10
You have a very nice, I read it. You have a very nice…
James Parr - 37:14
A jarg-inary?
Dallas Campbell - 37:15
Yeah. Exactly.
Freddie Kalaitzis - 37:15
So I think it's reports like these that allow these new types of technologies to bring them into the public awareness.
Dallas Campbell - 37:27
Yeah, public awareness. Exactly. Exactly. I mean, when you when you were writing that white paper, did you have that in mind? Was that of... You know, I want everyone to be able to...
Freddie Kalaitzis - 37:35
When on day one, when I was hired by SAC, Satellite Applications Catapult, kind of like what we did, it was like a little workshop to say, okay, what do we want to do? And like the first thing we we wanted to achieve was make these technologies approachable, understandable by the public.
Dallas Campbell - 37:51
When I first really became aware of it, when I did that archaeology documentary, actually making things like television programs and things that people can conceive of, you know, discovery in Egypt, that kind of thing is a fun way of doing it. But I guess climate is the one that's really gonna...
James Parr - 38:06
Well, yeah, I mean, I think the thing that will bring it to life is AI is now poised to really be a huge tool for disaster response. So flooding, fire, tornadoes, hurricanes, you know, this is all stuff which is now possible, and this includes putting AI on spacecraft themselves, so they're able to report directly.
Dallas Campbell - 38:28
Yeah, they did that in 2001, it was called HAL, it ended really, really badly. There is that thing, you know, you talk to people about things like AI, and of course, they come up with.
James Parr - 38:38
He was running chat GPT.
Dallas Campbell - 38:40
Exactly, killer robots and how 9,000 computers. There's always that, there is something quite sci-fi about the whole thing, isn't there? Quite matrixy and quite…
Maral Bayaraa - 38:48
So AI for EO, maybe we need like an identity. So maybe the microscopes in orbit, we should create a cartoon to create…
Dallas Campbell - 38:57
Yes, kind of brass microscopes in orbit. Listen, we're out of time, we're running out of time. Thank you so much for taking the time to come and actually sit in a room as human beings and not do this virtually via AI algorithms.
Freddie Kalaitzis - 39:11
I appreciate it.
Dallas Campbell - 39:12
Something nice about just talking to humans, you know what I mean?
James Parr - 39:14
Especially about this subject.
Dallas Campbell - 39:16
Yeah, you're all excited about it, I think.
James Parr - 39:19
Oh yeah, I mean we have this analogy, a bit like in the movie, "The Lord of the Rings" and all is lost.
Dallas Campbell - 39:22
Heard of it.
James Parr - 39:23
Yeah, in Helms Deep, episode two, and then Gandalf appears on the horizon with his staff. That's like AI right now, I think.
Dallas Campbell - 39:36
In what way? I don't understand.
James Parr - 39:37
All of the multitude of problems they call them, what's the phrase they use? Not a polyglot, but a poly-problems, poly anyway, that's the jargon. A lot of them, I won't say all of them, but a lot of them, I think AI is going to be able to help us to ameliorate.
Dallas Campbell - 39:53
Maral, what's the thing you're most excited about when you turn up to work? What keeps you?
Maral Bayaraa - 39:58
I feel like I've kind of spoken about this already, but I guess the thing that excites me about is solving kind of almost mundane-y looking problems, but they really matter. Solving the problems of farmers, solving the problems of miners, solving the problems of the everyday people using this technology because I think that's what really matters for me.
James Parr - 40:20
One other thought is, you know, we used to be really optimistic about the future and I think over the last few decades that optimism has sort of turned into a sort of a begrudging pragmatism especially with you know amount of problems we have but I think that the thing about EO and AI is it does give us a window of possibility that we can actually turn this around, we can the human race can solve this problem.
Dallas Campbell - 40:47
I think that's a really good point I think Neils Bohr, the famous Danish physicist, I remember he said prediction is very difficult especially if it's about the future and actually the problem with that quotation is that actually you don't want to predict the future, you want to make the future. If you have the information to be able to make decisions to make the future yourself in the way that you want it, then suddenly the prediction becomes very easy because you just do it how you want to do it.
Freddie Kalaitzis - 41:12
My philosophy towards AI is that it's an amplifier of human intent, like any technology and what I'm excited about is enabling the right kind of people and, you know, creating common good and this is why in my job I help domain scientists better understand AI, better understand the data that they're using so they can get better at their jobs. It's not about publishing papers. It's not about becoming an esteemed scientist. It's about the end goal, whether I do it through academia or through the industry.
Dallas Campbell - 41:34
Listen, thank you so much, Maral, Freddie, James. An absolute pleasure to have you here in the flesh. It's been great and a fascinating discussion. Thank you very much.
James Parr - 41:51
Thank you.
Freddie Kalaitzis - 41:51
Thank you.
Maral Bayaraa - 41:52
Thank you.
Dallas Campbell - 41:53
That's it for this episode. Thank you very much for your company. To hear future episodes of In Orbit, be sure to subscribe on your favourite podcast app and to find out more about how space is empowering industries between episodes, why not visit the Catapult website, or you can join them on Twitter, LinkedIn or Facebook. See you next time.