1 00:00:02,080 --> 00:00:04,960 Speaker 1: If I were to go back, I don't know thirty 2 00:00:05,040 --> 00:00:09,200 Speaker 1: years in Kenya, what's the difference between then and now 3 00:00:09,280 --> 00:00:12,840 Speaker 1: in terms of tree cover. I'm talking to Philip, Theigo 4 00:00:13,440 --> 00:00:16,160 Speaker 1: Special Technology envoid to the Kenyan President. 5 00:00:16,760 --> 00:00:19,400 Speaker 2: Let's speak as if you think about we are now 6 00:00:19,400 --> 00:00:22,759 Speaker 2: elevent trol percent. Previously we were more than twenty percent. 7 00:00:23,160 --> 00:00:26,560 Speaker 2: So we are cutting trees more than we're planting them. 8 00:00:26,840 --> 00:00:30,560 Speaker 1: In thirty years, Kenya lost half its tree cover half 9 00:00:31,280 --> 00:00:34,760 Speaker 1: And here's why that matters. Kenya is a mountainous country. 10 00:00:35,280 --> 00:00:38,120 Speaker 1: Dotted throughout the highlands are dozens of what canyons call 11 00:00:38,280 --> 00:00:44,480 Speaker 1: water towers natural reservoirs, densely forested areas capable of absorbing 12 00:00:44,520 --> 00:00:46,760 Speaker 1: the enormous amount of water that falls on the country 13 00:00:46,840 --> 00:00:51,400 Speaker 1: during the rainy seasons. The tree roots and undergrowth secure 14 00:00:51,520 --> 00:00:55,200 Speaker 1: and capture moisture, then slowly release it into the rivers 15 00:00:55,200 --> 00:00:58,560 Speaker 1: that flow down into the country's low lying coastal areas. 16 00:00:59,160 --> 00:01:04,080 Speaker 1: But in recent years, the water towers have depleted, settlements 17 00:01:04,240 --> 00:01:08,080 Speaker 1: have encroached on them, trees have been chopped down, thousands 18 00:01:08,080 --> 00:01:11,800 Speaker 1: of acres cleared. The natural reservoirs cease to hold nearly 19 00:01:11,840 --> 00:01:15,360 Speaker 1: as much water. So now Kenya is prone to extremes. 20 00:01:15,800 --> 00:01:18,319 Speaker 1: Too much water flowing down from the highlands in the 21 00:01:18,400 --> 00:01:21,880 Speaker 1: rainy season and too little water left during the dry season. 22 00:01:22,600 --> 00:01:24,640 Speaker 2: So you have a couple of hours of water, then 23 00:01:24,680 --> 00:01:27,000 Speaker 2: you have a couple of hours with no water that 24 00:01:27,080 --> 00:01:29,840 Speaker 2: it tops off to be dry by the city authority. 25 00:01:30,040 --> 00:01:33,120 Speaker 2: So that's the significance of the water towers we have 26 00:01:33,200 --> 00:01:34,360 Speaker 2: when they cannot hold water. 27 00:01:34,800 --> 00:01:39,000 Speaker 1: Kenya desperately needed to restore its water towers by planting 28 00:01:39,319 --> 00:01:42,760 Speaker 1: as many trees as humanly possible. So in the fall 29 00:01:42,800 --> 00:01:46,679 Speaker 1: of twenty twenty three, the Kenyan government took action. It 30 00:01:46,760 --> 00:01:51,360 Speaker 1: started a national holiday, National Tree Growing Day, a day 31 00:01:51,400 --> 00:01:54,120 Speaker 1: to allow the citizens of Kenya to go out into 32 00:01:54,120 --> 00:01:57,480 Speaker 1: the forest to dominate the Kenyan countryside and plant as 33 00:01:57,520 --> 00:02:02,200 Speaker 1: many trees as they can. Government decided on a number. 34 00:02:03,080 --> 00:02:08,040 Speaker 2: The President's really focus right around how to ensure that 35 00:02:08,080 --> 00:02:10,440 Speaker 2: we do not lose more forests was in this very 36 00:02:10,440 --> 00:02:12,800 Speaker 2: ambitious campaign around fifteen billion trees. 37 00:02:13,320 --> 00:02:16,840 Speaker 1: That's right, fifteen billion with a bee. 38 00:02:17,360 --> 00:02:19,679 Speaker 2: So imagine that number will tell you the ambition. But 39 00:02:19,760 --> 00:02:22,400 Speaker 2: as he tells you the deficit, it has to be 40 00:02:22,440 --> 00:02:24,720 Speaker 2: fifteen billion in the next eight years. 41 00:02:25,080 --> 00:02:28,320 Speaker 1: Fifteen billion trees over eight years averages out to more 42 00:02:28,360 --> 00:02:32,440 Speaker 1: than five million trees per day. That's a lot of trees. 43 00:02:33,000 --> 00:02:35,560 Speaker 1: But with such a massive goal, how can you track 44 00:02:35,639 --> 00:02:38,320 Speaker 1: your progress? How do you know where to plant those 45 00:02:38,320 --> 00:02:41,480 Speaker 1: trees so they'll have the most impact. How do you 46 00:02:41,560 --> 00:02:45,480 Speaker 1: monitor where older trees are still being cut down? Well, 47 00:02:45,520 --> 00:02:49,200 Speaker 1: the answer to those questions came from IBM and a 48 00:02:49,240 --> 00:02:55,120 Speaker 1: little space agency called NASA. That's right, folks, Smart Talks 49 00:02:55,800 --> 00:03:00,359 Speaker 1: is going to space. My name is Malcolm Glabo listening 50 00:03:00,400 --> 00:03:04,040 Speaker 1: to the latest episode of Smart Talks with IBM, where 51 00:03:04,040 --> 00:03:06,760 Speaker 1: we offer our listeners a glimpse behind the curtain of 52 00:03:06,800 --> 00:03:10,800 Speaker 1: the world of technology. In this season, IBM has gone 53 00:03:10,840 --> 00:03:15,840 Speaker 1: inside elementary school classrooms, toured formulation labs at Loreel, and 54 00:03:15,919 --> 00:03:19,600 Speaker 1: spoken with the fan development team at Scuderia Ferrari HP. 55 00:03:20,639 --> 00:03:25,160 Speaker 1: In this episode, how IBM is partnering with NASA to 56 00:03:25,200 --> 00:03:29,720 Speaker 1: build geospatial models using data from satellites to better understand 57 00:03:29,760 --> 00:03:35,640 Speaker 1: our Earth and Solar system. 58 00:03:35,680 --> 00:03:43,840 Speaker 3: Five four three two one zero all engine running liptoff. 59 00:03:44,160 --> 00:03:47,240 Speaker 3: We have a liptof thirty two minutes past the hour 60 00:03:47,600 --> 00:03:48,960 Speaker 3: liftoff on Apollo eleven. 61 00:03:49,680 --> 00:03:53,320 Speaker 1: IBM has worked on space related projects since before I 62 00:03:53,400 --> 00:03:54,000 Speaker 1: was even born. 63 00:03:54,680 --> 00:03:56,440 Speaker 4: I'm all for man. 64 00:03:57,480 --> 00:04:00,760 Speaker 1: A team of four thousand IBM engineers help create the 65 00:04:00,800 --> 00:04:03,600 Speaker 1: Saturn five rocket that took Neil Armstrong to the Moon 66 00:04:05,600 --> 00:04:10,120 Speaker 1: up and when I think of NASA, I tend to 67 00:04:10,160 --> 00:04:12,760 Speaker 1: picture the Moon landing, or the team of people back 68 00:04:12,760 --> 00:04:16,640 Speaker 1: in Houston guiding the Apollo mission, or the Hubble telescope, 69 00:04:16,920 --> 00:04:21,120 Speaker 1: or astronauts aboard the International Space Station. What I didn't 70 00:04:21,120 --> 00:04:24,880 Speaker 1: think about until now are NASA's geographers. 71 00:04:25,680 --> 00:04:27,599 Speaker 5: In order to go places, you need a map things. 72 00:04:28,040 --> 00:04:32,120 Speaker 1: This is Kevin Murphy, chief Science Data Officer at NASA's 73 00:04:32,240 --> 00:04:33,560 Speaker 1: Science Mission Directorate. 74 00:04:34,320 --> 00:04:37,320 Speaker 5: But I think that there's an assumption that NASAs all 75 00:04:37,360 --> 00:04:41,159 Speaker 5: about rockets and astronauts, and certainly that's a really large 76 00:04:41,160 --> 00:04:42,480 Speaker 5: part and important part of NASA. 77 00:04:43,240 --> 00:04:46,640 Speaker 1: NASA sends people to space and looks out of the stars, 78 00:04:47,120 --> 00:04:50,680 Speaker 1: but NASA also looks down at the Earth. The agency 79 00:04:50,720 --> 00:04:57,599 Speaker 1: has about one hundred and fifty satellites that use radar, lightar, landset, Aquaterra, cloudset, AURA, 80 00:04:58,000 --> 00:05:02,960 Speaker 1: low Earth orbit, Medium Earth orbit, geostationary orbit, on and on. 81 00:05:03,600 --> 00:05:07,760 Speaker 1: In one sense, NASA makes hardware to build rockets and 82 00:05:07,800 --> 00:05:11,280 Speaker 1: spacecraft and all those satellites that circle the Earth. But 83 00:05:11,400 --> 00:05:17,599 Speaker 1: fundamentally NASA also collects data. It's scientists and engineers people 84 00:05:17,640 --> 00:05:21,160 Speaker 1: like Kevin want to make the best use possible of 85 00:05:21,200 --> 00:05:25,240 Speaker 1: all the information gathered by all those many dozens of instruments. 86 00:05:25,920 --> 00:05:29,599 Speaker 5: Right now, we gather around twenty five petabytes of new 87 00:05:29,760 --> 00:05:32,680 Speaker 5: observational data per year. In the next couple of months, 88 00:05:32,760 --> 00:05:38,279 Speaker 5: we're about to launch a high resolution global radar. When 89 00:05:38,279 --> 00:05:41,840 Speaker 5: that launches, will double how much we collect every year 90 00:05:41,960 --> 00:05:44,000 Speaker 5: to about fifty petabytes of information. 91 00:05:44,680 --> 00:05:49,080 Speaker 1: Actually, since we recorded this conversation, NASA launched that global radar, 92 00:05:49,279 --> 00:05:53,640 Speaker 1: what they call NYSAR. So NASA is already generating new 93 00:05:53,680 --> 00:05:57,160 Speaker 1: data at the rate of fifty petabytes each year. To 94 00:05:57,160 --> 00:06:00,679 Speaker 1: put that in perspective, a single petabyte could hold about 95 00:06:00,720 --> 00:06:05,280 Speaker 1: five hundred billion pages of standard printed text. Now can 96 00:06:05,480 --> 00:06:07,960 Speaker 1: anyone sort of apply to use this data is. 97 00:06:08,279 --> 00:06:11,039 Speaker 5: They don't even have to apply. It's free and open data. 98 00:06:11,160 --> 00:06:14,960 Speaker 5: It advances how we understand what we do on Earth 99 00:06:15,000 --> 00:06:18,840 Speaker 5: and how we see ourselves within the universe. People can 100 00:06:18,880 --> 00:06:21,680 Speaker 5: take it for so many different downstream applications. So you 101 00:06:21,720 --> 00:06:24,560 Speaker 5: can go to our websites today, you can search through 102 00:06:24,600 --> 00:06:29,000 Speaker 5: our tools, and you can download information from the Mars rovers. 103 00:06:29,040 --> 00:06:32,720 Speaker 5: You can download information from the lunar reconnaissance orbiter or 104 00:06:32,760 --> 00:06:34,920 Speaker 5: any of the science data satellites, and. 105 00:06:34,880 --> 00:06:38,919 Speaker 1: Give me an example of a really cool application, a 106 00:06:38,960 --> 00:06:41,640 Speaker 1: really cool use that someone I don't know in academic 107 00:06:41,760 --> 00:06:44,080 Speaker 1: whatever has used your data for. It is there? 108 00:06:44,120 --> 00:06:47,360 Speaker 5: It okay. So one of the really kind of cool 109 00:06:47,440 --> 00:06:51,320 Speaker 5: but unexpected observations that we had is that we launched 110 00:06:51,360 --> 00:06:55,720 Speaker 5: a pair of satellites in their early two thousands called Grace, 111 00:06:56,120 --> 00:06:59,000 Speaker 5: and these satellites orbit the Earth and they can measure 112 00:06:59,080 --> 00:07:01,800 Speaker 5: very precisely the distance that they're away from each other 113 00:07:01,839 --> 00:07:04,520 Speaker 5: as they orbit the Earth, and as you go into 114 00:07:04,560 --> 00:07:07,960 Speaker 5: gravity wells, you can actually see a satellite accelerate and 115 00:07:08,000 --> 00:07:11,720 Speaker 5: the other one accelerate after it. Right, And using that information, 116 00:07:12,160 --> 00:07:14,840 Speaker 5: we were trying to map kind of the gravity fields 117 00:07:15,120 --> 00:07:17,880 Speaker 5: of Earth. What they found is that they can actually 118 00:07:17,960 --> 00:07:21,600 Speaker 5: map below kind of the mass of Earth to where 119 00:07:21,680 --> 00:07:25,240 Speaker 5: water storage is. For instance, so aquifers, right, so you 120 00:07:25,240 --> 00:07:30,400 Speaker 5: can monitor through gravity how much water is being depleted 121 00:07:30,520 --> 00:07:33,840 Speaker 5: or added to an aquifer or the density of glaciers. 122 00:07:34,440 --> 00:07:37,760 Speaker 1: So, just to back up for a moment, the presence 123 00:07:37,960 --> 00:07:42,280 Speaker 1: and density of water deposits below the Earth's surface have 124 00:07:42,360 --> 00:07:47,280 Speaker 1: an effect on gravitational fields that are being measured in 125 00:07:47,320 --> 00:07:51,480 Speaker 1: space correct. Yeah, And so does that tell you presume 126 00:07:51,720 --> 00:07:53,760 Speaker 1: you learn things like where there's an aquifer where you 127 00:07:53,760 --> 00:07:55,600 Speaker 1: didn't think there was an aquifer. 128 00:07:55,640 --> 00:07:58,840 Speaker 5: Or if it's being depleted faster. Yeah. Yeah. 129 00:07:59,240 --> 00:08:01,320 Speaker 1: So who's using that kind of data? 130 00:08:01,800 --> 00:08:05,320 Speaker 5: All sorts of different organizations, whether they're you know, NGOs 131 00:08:05,600 --> 00:08:08,920 Speaker 5: or government agencies or people that are planning a large 132 00:08:08,960 --> 00:08:09,960 Speaker 5: agricultural product. 133 00:08:10,040 --> 00:08:12,160 Speaker 1: How did you Was that an intentional decision? 134 00:08:12,200 --> 00:08:13,520 Speaker 5: It wasn't. It was accidental. 135 00:08:14,200 --> 00:08:20,800 Speaker 1: It was accidental. NASA has assembled a historically unprecedented mountain 136 00:08:20,800 --> 00:08:24,640 Speaker 1: of data about the physical world, free and open to anyone, 137 00:08:25,040 --> 00:08:27,800 Speaker 1: and the possibilities for how that information can be used 138 00:08:27,840 --> 00:08:33,840 Speaker 1: are so vast that even NASA is still uncovering them. 139 00:08:34,040 --> 00:08:36,760 Speaker 1: When I was a kid, I loved legos. I had 140 00:08:36,760 --> 00:08:39,880 Speaker 1: a huge bin full of them. At the time, legos 141 00:08:39,960 --> 00:08:43,480 Speaker 1: were really just colored bricks of various sizes. They weren't 142 00:08:43,520 --> 00:08:46,440 Speaker 1: as complicated as they are today. And what I realized 143 00:08:46,480 --> 00:08:49,040 Speaker 1: even then was that there were more possibilities in a 144 00:08:49,040 --> 00:08:51,960 Speaker 1: box of legos than I could ever imagine on my own. 145 00:08:52,760 --> 00:08:54,440 Speaker 1: I played with my brother and he would show me 146 00:08:54,480 --> 00:08:56,720 Speaker 1: something that hadn't occurred to me, and I go to 147 00:08:56,760 --> 00:08:59,080 Speaker 1: my friend Bruce's and see that he was off on 148 00:08:59,120 --> 00:09:01,920 Speaker 1: some legos. Tan engined that I'd never even thought of, 149 00:09:02,480 --> 00:09:04,520 Speaker 1: like a cool bridge, or a castle or a truck. 150 00:09:05,280 --> 00:09:08,960 Speaker 1: I use legos one way. Bruce used his legos in 151 00:09:09,000 --> 00:09:14,520 Speaker 1: a completely different way. NASA's data treasure trove is like 152 00:09:14,559 --> 00:09:18,040 Speaker 1: a very very big box of legos. And here's the question. 153 00:09:18,640 --> 00:09:23,760 Speaker 1: With so much data, containing so many possible connections, could IBM, 154 00:09:24,080 --> 00:09:29,960 Speaker 1: and specifically IBM's artificial intelligence help NASA scientists uncover patterns 155 00:09:29,960 --> 00:09:33,360 Speaker 1: and connect systems in a way they've never done before. 156 00:09:35,840 --> 00:09:38,280 Speaker 4: Everything started with a question, right. 157 00:09:38,400 --> 00:09:42,520 Speaker 1: I'm talking to one Bernabe Moreno, director of IBM Research 158 00:09:42,559 --> 00:09:43,120 Speaker 1: in Europe. 159 00:09:43,840 --> 00:09:47,720 Speaker 4: As we advance AI, we have new tools to understand 160 00:09:48,520 --> 00:09:52,120 Speaker 4: the surroundings, understand the world, understand the language, and understand 161 00:09:52,120 --> 00:09:55,199 Speaker 4: our planets. And the question that we were asking ourselves 162 00:09:55,400 --> 00:09:57,880 Speaker 4: was all these new advances that we see in language. 163 00:09:57,960 --> 00:10:01,280 Speaker 4: It was a post GPT moment. Could we apply the 164 00:10:01,320 --> 00:10:05,000 Speaker 4: same idea on the same architecture and technology to add 165 00:10:05,000 --> 00:10:06,200 Speaker 4: a double do our planets? 166 00:10:06,760 --> 00:10:10,120 Speaker 1: The advent of AI created a new opportunity. What if 167 00:10:10,120 --> 00:10:13,680 Speaker 1: all of NASA's mountain of data could be organized, analyzed, 168 00:10:14,000 --> 00:10:19,280 Speaker 1: understood by artificial intelligence. The original idea was to create 169 00:10:19,360 --> 00:10:23,000 Speaker 1: a geospatial foundation model for the Earth and from there 170 00:10:23,440 --> 00:10:28,160 Speaker 1: create additional specialized models for other scientific priorities of NASA, 171 00:10:28,920 --> 00:10:32,640 Speaker 1: and finally create an AI system that can understand all 172 00:10:32,720 --> 00:10:36,480 Speaker 1: the data across those specialized models in order to uncover 173 00:10:36,559 --> 00:10:41,720 Speaker 1: hidden insights and relationships. Together, these models could unlock an 174 00:10:41,920 --> 00:10:46,480 Speaker 1: infinite number of potential applications. I asked Kevin Murphy at 175 00:10:46,559 --> 00:10:49,840 Speaker 1: NASA about the beginning of these Earth models. 176 00:10:50,440 --> 00:10:53,240 Speaker 5: Has some colleagues, and we were investigating a number of 177 00:10:53,320 --> 00:10:58,480 Speaker 5: different avenues of using AI with our data, but also 178 00:10:58,760 --> 00:11:01,360 Speaker 5: kind of the management and stewardship of the data, so 179 00:11:01,480 --> 00:11:03,600 Speaker 5: not only like the observations, but how we make it 180 00:11:03,640 --> 00:11:07,320 Speaker 5: available to people, make it discoverable. And they said, hey, 181 00:11:08,000 --> 00:11:10,520 Speaker 5: we see these transform architectures. We think that they can 182 00:11:10,600 --> 00:11:14,760 Speaker 5: be applicable to some of the sequential observations that we make. 183 00:11:15,280 --> 00:11:17,400 Speaker 5: We'd really like to work with IBM on that. And 184 00:11:17,440 --> 00:11:21,360 Speaker 5: I was like, I'm really skeptical. But while because I 185 00:11:21,400 --> 00:11:27,480 Speaker 5: hadn't seen those types of tools really produce results that 186 00:11:27,520 --> 00:11:31,520 Speaker 5: were commensurate with the amount of effort you put into them, right, 187 00:11:31,600 --> 00:11:33,920 Speaker 5: So we were getting some really good results and deep 188 00:11:34,000 --> 00:11:37,840 Speaker 5: learning approaches, but they took a lot of effort. 189 00:11:37,520 --> 00:11:39,120 Speaker 1: But Kevin came around quickly. 190 00:11:40,000 --> 00:11:44,680 Speaker 5: When we typically develop a new data product or an algorithm, 191 00:11:45,120 --> 00:11:48,800 Speaker 5: it takes anywhere from you know, twelve months, eighteen months, 192 00:11:48,880 --> 00:11:54,280 Speaker 5: twenty four months to go from data and hypothesis to 193 00:11:54,480 --> 00:11:58,440 Speaker 5: results which is validated. We were able to get approximately 194 00:11:58,480 --> 00:12:03,560 Speaker 5: the same precision for some well known types of benchmarks 195 00:12:03,920 --> 00:12:06,000 Speaker 5: with and I think it was about four months, oh, 196 00:12:06,040 --> 00:12:07,080 Speaker 5: instead of starting the work. 197 00:12:07,200 --> 00:12:11,360 Speaker 1: Yeah, yeah, so it happened faster than you thought, much faster. 198 00:12:12,440 --> 00:12:16,120 Speaker 1: In twenty twenty three, IBM and NASA launched a foundation 199 00:12:16,320 --> 00:12:21,120 Speaker 1: model trained on NASA's harmonized landset sentinel to satellite data 200 00:12:21,480 --> 00:12:25,520 Speaker 1: across the continental United States. They named the model Prithvi, 201 00:12:26,040 --> 00:12:30,120 Speaker 1: the Sanskrit word for Earth. The first version of Prithvy 202 00:12:30,480 --> 00:12:34,480 Speaker 1: used only Earth observation images and just that was enough 203 00:12:34,520 --> 00:12:38,560 Speaker 1: to totally change Kevin's idea of what foundation models could do. 204 00:12:39,559 --> 00:12:43,480 Speaker 1: But they didn't stop there. IBM and NASA were encouraged 205 00:12:43,640 --> 00:12:47,960 Speaker 1: at how well Prithvy worked for Earth observation tasks, so 206 00:12:48,240 --> 00:12:51,079 Speaker 1: they decided to create a more complex version of Prithvy 207 00:12:51,480 --> 00:12:55,559 Speaker 1: that could understand whether and climate data. They hope this 208 00:12:55,640 --> 00:12:58,760 Speaker 1: new version of PRITHV would allow researchers to answer new 209 00:12:58,840 --> 00:13:02,440 Speaker 1: questions about the Earth, from short term weather forecasting to 210 00:13:02,559 --> 00:13:06,160 Speaker 1: longer term climate effects. Imagine you have a map of 211 00:13:06,360 --> 00:13:11,319 Speaker 1: all the different temperatures, pressures, clouds, rainfall, and more from 212 00:13:11,400 --> 00:13:15,640 Speaker 1: around the globe. With this map, IBM and NASA could 213 00:13:15,720 --> 00:13:19,679 Speaker 1: implement advanced tasks. They could track the formation of El 214 00:13:19,760 --> 00:13:22,400 Speaker 1: Nino or predict how the path of a hurricane would 215 00:13:22,440 --> 00:13:26,640 Speaker 1: change if the ocean temperature went up by half a degree. 216 00:13:26,960 --> 00:13:30,040 Speaker 4: I will always remember this moment was when we created 217 00:13:30,080 --> 00:13:35,160 Speaker 4: the Weather and Climate Foundational Molive. The senior methodologist of NASA, 218 00:13:35,640 --> 00:13:38,480 Speaker 4: it was like, I cannot believe that it has changed 219 00:13:38,679 --> 00:13:40,800 Speaker 4: the way I think about the AI and ever since 220 00:13:40,840 --> 00:13:43,120 Speaker 4: he's been kind of preaching with this A sample. 221 00:13:43,960 --> 00:13:46,440 Speaker 1: One and his team then took the model and decided 222 00:13:46,480 --> 00:13:50,559 Speaker 1: to test it, really tested it. Took away ninety nine 223 00:13:50,600 --> 00:13:53,839 Speaker 1: percent of the data points and ran the experiment again. 224 00:13:54,520 --> 00:13:56,560 Speaker 1: What they were trying to figure out is if the 225 00:13:56,640 --> 00:13:59,680 Speaker 1: model had learned enough about the basic principles of the Earth, 226 00:14:00,120 --> 00:14:03,040 Speaker 1: the underlying physics of the way the planet works, to 227 00:14:03,160 --> 00:14:06,480 Speaker 1: fill in the blanks on its own with just one 228 00:14:06,520 --> 00:14:10,120 Speaker 1: percent of the original data, would it still be accurate 229 00:14:10,200 --> 00:14:16,760 Speaker 1: in its predictions. What happened the model crushed it, so 230 00:14:17,000 --> 00:14:19,440 Speaker 1: it was able to extrapolate on the basis of one 231 00:14:19,480 --> 00:14:23,160 Speaker 1: percent of the data what the entire picture looked like. Yes, 232 00:14:24,600 --> 00:14:28,280 Speaker 1: because pre learned everything right, Yeah, it learned the kind 233 00:14:28,280 --> 00:14:32,880 Speaker 1: of principles of exactly. Yeah. Oh wow, that's very very impressive. 234 00:14:33,000 --> 00:14:35,560 Speaker 1: So at that moment when when you realize you could 235 00:14:35,640 --> 00:14:40,360 Speaker 1: do that, and just curious about your emotional I mean, 236 00:14:40,360 --> 00:14:41,800 Speaker 1: did you jump up and down? What did you do? 237 00:14:42,200 --> 00:14:45,400 Speaker 4: That's I was like, wow, it was a very emotional 238 00:14:45,440 --> 00:14:49,360 Speaker 4: meeting because you know, having this person say now I'm 239 00:14:49,400 --> 00:14:52,920 Speaker 4: convinced right, Yeah, it was kind of a quite a 240 00:14:52,920 --> 00:14:55,840 Speaker 4: special moment. These moments make your life as a researcher. 241 00:14:57,280 --> 00:15:00,880 Speaker 1: IBM and that's a launch prithe for Weather and Climate 242 00:15:01,040 --> 00:15:04,040 Speaker 1: in twenty twenty four. And while IBM and as a 243 00:15:04,080 --> 00:15:08,040 Speaker 1: scientist could use Prithvy to run interesting experiments, they were 244 00:15:08,080 --> 00:15:11,640 Speaker 1: even more excited about how prithy could help people in 245 00:15:11,640 --> 00:15:19,920 Speaker 1: the real world. So let's go back to Kenya Ambassador 246 00:15:20,040 --> 00:15:23,600 Speaker 1: Philip Digo and the country's Great tree planting project. 247 00:15:24,480 --> 00:15:27,440 Speaker 2: So on those initial months, there was a massive effort, 248 00:15:27,480 --> 00:15:30,240 Speaker 2: including a couple of national holidays. 249 00:15:30,400 --> 00:15:31,440 Speaker 1: For tree planting. 250 00:15:32,240 --> 00:15:35,760 Speaker 2: Yes, where the entire cabinet was sent. 251 00:15:36,040 --> 00:15:38,640 Speaker 1: Ah, did you plant trees as I did? 252 00:15:38,720 --> 00:15:40,600 Speaker 2: Oh my god, I said, the entire cabinet plus someone 253 00:15:40,800 --> 00:15:41,600 Speaker 2: we have to be seen. 254 00:15:42,040 --> 00:15:43,840 Speaker 1: Are you good at that plantet two weeks ago? 255 00:15:44,200 --> 00:15:46,080 Speaker 2: Well it's very easy to go hole put a tree 256 00:15:46,160 --> 00:15:46,760 Speaker 2: in the ground. 257 00:15:47,400 --> 00:15:49,120 Speaker 5: The show Well wow, what. 258 00:15:49,520 --> 00:15:52,480 Speaker 1: Planting a tree is easy? But remember it has to 259 00:15:52,520 --> 00:15:57,720 Speaker 1: happen fifteen billion times. IBM Research has been operating in 260 00:15:57,800 --> 00:16:01,920 Speaker 1: Nairobi since twenty thirteen, and what Kenya wanted, at least 261 00:16:01,960 --> 00:16:05,880 Speaker 1: in the beginning was straightforward. The prith Fee model that 262 00:16:05,960 --> 00:16:09,200 Speaker 1: IBM and NASA built could be used to essentially make 263 00:16:09,240 --> 00:16:13,560 Speaker 1: the world's greatest map, and Kenya, with IBM's help, could 264 00:16:13,680 --> 00:16:16,720 Speaker 1: use that model to make the world's greatest map of Kenya. 265 00:16:17,760 --> 00:16:20,000 Speaker 1: The first step was to lay a grid across the 266 00:16:20,080 --> 00:16:24,200 Speaker 1: topography of the country, break the forest into manageable bite 267 00:16:24,240 --> 00:16:27,560 Speaker 1: sized pieces, each of which could be analyzed separately. 268 00:16:28,400 --> 00:16:30,680 Speaker 2: So, because our forest is massive when you look at 269 00:16:30,680 --> 00:16:33,360 Speaker 2: it in terms of green rite, but only lay it, 270 00:16:33,600 --> 00:16:36,240 Speaker 2: you're able to break it into pieces like into boxes. 271 00:16:36,480 --> 00:16:40,120 Speaker 2: And for us that was important because then it's easy 272 00:16:40,160 --> 00:16:42,920 Speaker 2: to tackle it. When it's in a greed system than 273 00:16:43,120 --> 00:16:45,600 Speaker 2: just as a massive forest. So that was also what 274 00:16:46,280 --> 00:16:47,680 Speaker 2: the model was able to do. 275 00:16:48,120 --> 00:16:51,680 Speaker 1: Then the model painstakingly sorted through each of those boxes 276 00:16:52,040 --> 00:16:54,360 Speaker 1: and look for what Philip calls hotspots. 277 00:16:55,040 --> 00:16:57,440 Speaker 2: So you can see, for example, very quickly, which other 278 00:16:57,520 --> 00:17:01,160 Speaker 2: areas are being eroded very fast, and that you need. 279 00:17:01,040 --> 00:17:01,840 Speaker 5: To quickly protect. 280 00:17:02,320 --> 00:17:05,120 Speaker 2: Yeah, because you sometimes and that's where you want to target, right, 281 00:17:05,200 --> 00:17:07,320 Speaker 2: I mean, it's not possible to do everything at the 282 00:17:07,320 --> 00:17:07,760 Speaker 2: same time. 283 00:17:07,960 --> 00:17:09,760 Speaker 1: Do you have a definition of a hotspot and how 284 00:17:09,760 --> 00:17:12,480 Speaker 1: many hotspots are there according to that definition. 285 00:17:12,840 --> 00:17:13,480 Speaker 5: Oh, there are a lot. 286 00:17:13,560 --> 00:17:16,840 Speaker 2: So we have more than forty water towers, and I'll 287 00:17:16,880 --> 00:17:19,639 Speaker 2: tell you all of them have hot spots. And the 288 00:17:19,680 --> 00:17:23,200 Speaker 2: hot spots in my definition areas that are being degraded 289 00:17:23,640 --> 00:17:26,600 Speaker 2: fast and in a very unusual way. Right, you can 290 00:17:26,720 --> 00:17:30,320 Speaker 2: literally see how human activity is seriously degrading that particular 291 00:17:30,320 --> 00:17:32,800 Speaker 2: area that if you do not have a direct intervention, 292 00:17:32,960 --> 00:17:36,119 Speaker 2: we'll lose the entire forest. So that's the hot spot 293 00:17:36,200 --> 00:17:38,760 Speaker 2: for us, because you think about cutting a hundred trees 294 00:17:38,800 --> 00:17:40,800 Speaker 2: a day and cutting a million trees a day, So 295 00:17:40,880 --> 00:17:43,800 Speaker 2: that's a hot spot. You want to look at places 296 00:17:43,800 --> 00:17:48,399 Speaker 2: where there's just unusually high activity of deforestation. 297 00:17:48,480 --> 00:17:50,600 Speaker 1: In a hot spot. The size of each box in 298 00:17:50,640 --> 00:17:53,520 Speaker 1: the grid was ten by ten meters, about half a 299 00:17:53,560 --> 00:17:57,439 Speaker 1: tennis court. That's how closely they were examining the forest 300 00:17:58,560 --> 00:18:02,760 Speaker 1: so very crudely. The model ingests all of this satellite 301 00:18:02,840 --> 00:18:06,879 Speaker 1: data and it helps you answer some very specific questions 302 00:18:06,920 --> 00:18:11,560 Speaker 1: like where should we prioritize our tree planning efforts, which 303 00:18:12,000 --> 00:18:16,600 Speaker 1: areas down to an extraordinary levels of specificity are eroding 304 00:18:16,720 --> 00:18:20,560 Speaker 1: most quickly. You know, all those kinds of practical questions 305 00:18:20,600 --> 00:18:22,159 Speaker 1: about how to direct your strategy. 306 00:18:22,480 --> 00:18:24,560 Speaker 2: So if you think about a smart forest, right, and 307 00:18:24,600 --> 00:18:27,200 Speaker 2: that's really for us according with smart fencing, smart forests, 308 00:18:27,200 --> 00:18:30,960 Speaker 2: everything that's smart because of AI. If you think about 309 00:18:31,680 --> 00:18:34,359 Speaker 2: your usual what you can see with your eyes and 310 00:18:34,400 --> 00:18:37,399 Speaker 2: then the satellite layer which just zooms in and you 311 00:18:37,440 --> 00:18:40,320 Speaker 2: see green. So what the model has been able to 312 00:18:40,320 --> 00:18:42,680 Speaker 2: do is to create a smart layer, right, and then 313 00:18:42,720 --> 00:18:46,320 Speaker 2: that smart layer you can actually see many things from 314 00:18:46,320 --> 00:18:49,400 Speaker 2: analytics to the greeds to a dashboard on a lot. 315 00:18:49,760 --> 00:18:53,480 Speaker 2: So able to layer to those blocks, you can quantify 316 00:18:53,600 --> 00:18:57,919 Speaker 2: degradation by blocks. You can match intevations, you can match reforestation. 317 00:18:58,440 --> 00:19:00,679 Speaker 1: I asked Philip to imagine what would have been liked 318 00:19:00,680 --> 00:19:04,720 Speaker 1: to attempt the tree planting project in an era before AI. 319 00:19:05,480 --> 00:19:11,880 Speaker 1: His answer was plant fifteen billion trees, restore the water towers. Impossible. 320 00:19:12,920 --> 00:19:17,080 Speaker 1: With Prithvi on Kenya's side, though it's really happening. What 321 00:19:17,160 --> 00:19:20,000 Speaker 1: should be clear by now is how versatile Prithvy can be. 322 00:19:20,600 --> 00:19:24,159 Speaker 1: Want to know how to combat deforestation, Prithvy can model that. 323 00:19:24,600 --> 00:19:26,280 Speaker 1: Want to know when the best time in the year 324 00:19:26,320 --> 00:19:30,400 Speaker 1: to plant your crops is, Prithvy can help predict that too. 325 00:19:30,800 --> 00:19:35,160 Speaker 1: Last year, six months after IBM started helping Kenya with reforestation, 326 00:19:35,680 --> 00:19:38,840 Speaker 1: Kenya needed Prithvy's help on something else and it was 327 00:19:38,880 --> 00:19:39,680 Speaker 1: an emergency. 328 00:19:40,560 --> 00:19:43,280 Speaker 2: So something was happening in the world that we sort 329 00:19:43,320 --> 00:19:46,040 Speaker 2: of had these floods that we didn't expect. 330 00:19:46,440 --> 00:19:49,040 Speaker 1: In the spring of twenty twenty four, Kenya was hit 331 00:19:49,080 --> 00:19:53,240 Speaker 1: with thunderstorms and torrential rain, days and days of. 332 00:19:53,200 --> 00:19:56,440 Speaker 2: It, and so I got a call from the Red Cross, 333 00:19:56,600 --> 00:20:00,080 Speaker 2: the one of my friends, and they're like, Ambassador, we 334 00:20:00,119 --> 00:20:03,160 Speaker 2: need a little bit of help on how we deal 335 00:20:03,240 --> 00:20:06,240 Speaker 2: with response because what we're seeing is unusual, right, because 336 00:20:06,240 --> 00:20:09,119 Speaker 2: normally you would only have one area, all of a sudden, 337 00:20:09,240 --> 00:20:12,439 Speaker 2: we had an entire country flooding. In April we had 338 00:20:12,480 --> 00:20:16,840 Speaker 2: about three thousand, eight hundred kilometers square kind of total 339 00:20:16,960 --> 00:20:21,280 Speaker 2: land flooded, which is unusual for Kenyon. And so when 340 00:20:21,320 --> 00:20:23,600 Speaker 2: I got this call, we were like, okay, there's someone 341 00:20:23,640 --> 00:20:26,320 Speaker 2: could do with IBM. We only did one function for 342 00:20:26,359 --> 00:20:29,160 Speaker 2: the trees. It was actually a climate model, and we said, 343 00:20:29,200 --> 00:20:34,880 Speaker 2: can we use this to help us better respond to floods? 344 00:20:35,440 --> 00:20:38,520 Speaker 2: And so that was how we started having this discussion 345 00:20:38,560 --> 00:20:42,480 Speaker 2: with IBM in terms of repurposing the model to help 346 00:20:42,720 --> 00:20:46,320 Speaker 2: us deal with this new challenge around floods. 347 00:20:46,880 --> 00:20:52,840 Speaker 1: Again, prithvy is versatile. Prithvie could use everything it new 348 00:20:52,880 --> 00:20:57,240 Speaker 1: about the land, the forests and infrastructure to analyze how 349 00:20:57,320 --> 00:21:01,560 Speaker 1: and where and when floods would occur. The Kenyan government 350 00:21:01,600 --> 00:21:04,000 Speaker 1: could then use the model to help the Red Cross 351 00:21:04,119 --> 00:21:08,440 Speaker 1: organize its response, show areas that needed to be evacuated 352 00:21:08,760 --> 00:21:11,840 Speaker 1: or safe places where the Red Cross could set up camps. 353 00:21:12,520 --> 00:21:15,119 Speaker 1: That information was invaluable. 354 00:21:16,080 --> 00:21:18,640 Speaker 2: Historically, what has happened is that they would set up 355 00:21:18,720 --> 00:21:23,879 Speaker 2: camp based on population congregation right where people assembly is 356 00:21:23,920 --> 00:21:26,240 Speaker 2: where they set up a camp, not based on any data, 357 00:21:26,440 --> 00:21:29,960 Speaker 2: right simply because people are there, they will come there 358 00:21:30,000 --> 00:21:33,600 Speaker 2: to provide services and emergency response. What we realize is 359 00:21:33,600 --> 00:21:36,080 Speaker 2: that that model doesn't work. So what we've been able 360 00:21:36,080 --> 00:21:38,560 Speaker 2: to do with IBM is be able to sort of 361 00:21:38,640 --> 00:21:42,199 Speaker 2: give red cause very specific locations or options where to 362 00:21:42,240 --> 00:21:44,680 Speaker 2: set up camps. So if people come here, just tell 363 00:21:44,720 --> 00:21:48,480 Speaker 2: them no, move here, that's the safe place you really 364 00:21:48,480 --> 00:21:50,040 Speaker 2: want to go. So I think for me that was 365 00:21:50,080 --> 00:21:52,240 Speaker 2: really amazing. So we are calling them a very funny 366 00:21:52,240 --> 00:21:54,480 Speaker 2: word for it, flood assembly points. We always have fire 367 00:21:54,960 --> 00:21:56,879 Speaker 2: for assembly points, but now we can say we have 368 00:21:57,520 --> 00:22:02,280 Speaker 2: literally flat assembly points that are safe or citizens. 369 00:22:02,680 --> 00:22:07,240 Speaker 1: That's fascinating. So the model has ingested this incredibly granular 370 00:22:08,080 --> 00:22:14,679 Speaker 1: picture later of the topography and weather patterns of Kenya. 371 00:22:14,800 --> 00:22:18,399 Speaker 1: It's just giving you a set of useful predictions about 372 00:22:18,400 --> 00:22:20,320 Speaker 1: how you should shape your response. 373 00:22:21,280 --> 00:22:23,840 Speaker 2: Yes, and what we did remember is that, as I said, 374 00:22:23,840 --> 00:22:27,760 Speaker 2: it was a full multistate called capability. What IBM gave 375 00:22:27,840 --> 00:22:30,520 Speaker 2: us was a base map. We didn't have that before, 376 00:22:30,920 --> 00:22:33,480 Speaker 2: and a base model. So you cannot have these layers 377 00:22:33,520 --> 00:22:34,879 Speaker 2: up on layers up on layers to be able to 378 00:22:34,880 --> 00:22:37,000 Speaker 2: make intelligent decisions. 379 00:22:40,720 --> 00:22:43,640 Speaker 1: Throughout my reporting on this episode, I've been really impressed 380 00:22:43,680 --> 00:22:46,720 Speaker 1: by what Prithvie can do. But it doesn't stop at 381 00:22:46,720 --> 00:22:50,520 Speaker 1: floods and reforestation. Prithvie has also been used to look 382 00:22:50,560 --> 00:22:54,800 Speaker 1: at wildfires and floods in the UK, and Kevin told 383 00:22:54,800 --> 00:22:58,040 Speaker 1: me that researchers in Africa have even used prithvy to 384 00:22:58,200 --> 00:23:02,120 Speaker 1: identify locust breeding grounds, which could help them prevent swarms 385 00:23:02,160 --> 00:23:07,479 Speaker 1: that destroy crops. But all these are issues on land. 386 00:23:07,920 --> 00:23:10,119 Speaker 6: I mean, I always say to people, seventy percent of 387 00:23:10,119 --> 00:23:12,000 Speaker 6: our land mass is ocean. 388 00:23:12,520 --> 00:23:15,240 Speaker 1: Kate Rice is the director of the heart Tree Center, 389 00:23:15,600 --> 00:23:20,440 Speaker 1: which focuses on adopting AI into UK's public and private sectors, 390 00:23:20,840 --> 00:23:26,320 Speaker 1: and one of those sectors is the blue economy. Oceans, fish, shellfish. 391 00:23:26,960 --> 00:23:31,720 Speaker 1: But oceans are huge, and getting data from motions is difficult. 392 00:23:31,600 --> 00:23:34,240 Speaker 6: So you're dealing with something where there's not a lot 393 00:23:34,240 --> 00:23:40,000 Speaker 6: of people walking around collecting data. So the real difficulty 394 00:23:40,160 --> 00:23:44,719 Speaker 6: is understanding that collecting enough data to make anything makes sense. 395 00:23:45,320 --> 00:23:51,199 Speaker 6: And oceans are very complex in terms of their interaction 396 00:23:51,760 --> 00:23:55,000 Speaker 6: with our climate and how they interact with the climate, 397 00:23:55,480 --> 00:23:59,520 Speaker 6: so understanding the physics space models is pretty challenging too. 398 00:24:00,200 --> 00:24:05,639 Speaker 1: Once again, enter IBM. IBM created a new geospatial model 399 00:24:05,880 --> 00:24:09,840 Speaker 1: to help us better understand our oceans. Heart Tree and IBM, 400 00:24:10,080 --> 00:24:13,359 Speaker 1: along with the Plymouth Marine Laboratory, the UK Science and 401 00:24:13,440 --> 00:24:17,239 Speaker 1: Technology Facilities Council and the University of Exeter have all 402 00:24:17,280 --> 00:24:21,040 Speaker 1: partnered to focus the model's power on the waters around 403 00:24:21,080 --> 00:24:26,399 Speaker 1: the United Kingdom, which ultimately will help the UK's blue economy. 404 00:24:26,760 --> 00:24:29,959 Speaker 6: You get these major blooms in algae, so the ocean 405 00:24:30,000 --> 00:24:33,480 Speaker 6: goes green and you might see it in lakes as well. Now, 406 00:24:33,720 --> 00:24:37,960 Speaker 6: if you are shell fishing and that's what you're harvesting, 407 00:24:39,119 --> 00:24:45,880 Speaker 6: you can't harvest cockles muscles to be very colloquial, when 408 00:24:45,920 --> 00:24:49,479 Speaker 6: you have algae blooms because they're poisonous. So there are 409 00:24:49,520 --> 00:24:51,320 Speaker 6: certain times of the year where you can harvest. In 410 00:24:51,359 --> 00:24:54,040 Speaker 6: a certain times of year you can't if you keep 411 00:24:54,119 --> 00:24:56,960 Speaker 6: having the algal blooms. Just to put it on an 412 00:24:57,000 --> 00:25:01,359 Speaker 6: economic terms, that's a problem. So if we look at 413 00:25:01,400 --> 00:25:05,680 Speaker 6: it that way, that's an issue. So we really do 414 00:25:05,760 --> 00:25:09,840 Speaker 6: need to try and understand where these algal blooms will happen, 415 00:25:10,240 --> 00:25:13,880 Speaker 6: when they will happen, and how to limit them because obviously, 416 00:25:13,960 --> 00:25:17,520 Speaker 6: if you're shell fishing as your livelihood, that's going to 417 00:25:17,560 --> 00:25:18,320 Speaker 6: really impact you. 418 00:25:19,200 --> 00:25:23,440 Speaker 1: Kate told me that understanding these algal blooms, how they form, 419 00:25:23,720 --> 00:25:27,040 Speaker 1: why they form, and how they move would allow people 420 00:25:27,160 --> 00:25:28,240 Speaker 1: to better manage them. 421 00:25:29,359 --> 00:25:31,679 Speaker 6: What is it you're putting in the water. Are you 422 00:25:31,800 --> 00:25:35,960 Speaker 6: putting fertilizers in the water in the near shore environment 423 00:25:36,040 --> 00:25:39,639 Speaker 6: that is causing those algal blooms? Is it because we 424 00:25:39,680 --> 00:25:43,840 Speaker 6: are heating up the oceans and particularly our near shore 425 00:25:43,920 --> 00:25:45,639 Speaker 6: environments that is causing that. 426 00:25:45,840 --> 00:25:46,479 Speaker 1: I don't know. 427 00:25:46,760 --> 00:25:50,439 Speaker 6: I'm not a specialist, but that's what you're trying to 428 00:25:50,440 --> 00:25:53,840 Speaker 6: figure out. Is there something we are doing that is 429 00:25:53,880 --> 00:25:58,920 Speaker 6: creating those environments that is causing those algal blooms or 430 00:25:59,119 --> 00:26:02,560 Speaker 6: is it natural? And natural was always a difficult one 431 00:26:02,640 --> 00:26:05,280 Speaker 6: because I would say we live in a very managed environment, 432 00:26:05,800 --> 00:26:10,480 Speaker 6: particularly in the UK. Very few landscapes are natural. Most 433 00:26:10,480 --> 00:26:14,520 Speaker 6: of it is managed in some way. Are we managing 434 00:26:14,560 --> 00:26:16,960 Speaker 6: it in an appropriate way? Is there changes in how 435 00:26:17,000 --> 00:26:18,920 Speaker 6: we behave that could make things better? 436 00:26:19,920 --> 00:26:22,000 Speaker 1: Not that I needed more examples to sell me on 437 00:26:22,040 --> 00:26:24,919 Speaker 1: how useful the Perythvia models are, but Kate gave me 438 00:26:25,160 --> 00:26:28,840 Speaker 1: a few more use cases that reinforced just how exciting 439 00:26:28,880 --> 00:26:31,840 Speaker 1: foundation models are for our oceans. 440 00:26:32,640 --> 00:26:37,240 Speaker 6: These big brown seaweeds can really help with carbon sequestration. 441 00:26:37,880 --> 00:26:41,520 Speaker 6: Imagine if we could improve the environment enough so that 442 00:26:41,560 --> 00:26:43,840 Speaker 6: we could have more of that so that we could 443 00:26:43,880 --> 00:26:47,520 Speaker 6: sequentch more carbon. The other thing is wind power. In 444 00:26:47,560 --> 00:26:50,240 Speaker 6: the UK, we have a lot of offshore wind farms 445 00:26:50,640 --> 00:26:53,840 Speaker 6: and we're doing really well with our renewable energy resources. 446 00:26:54,080 --> 00:26:56,000 Speaker 6: So where do we put that and how does that 447 00:26:56,119 --> 00:27:02,200 Speaker 6: impact sand movements? So these sandbars things aren't static, they move, 448 00:27:02,760 --> 00:27:06,160 Speaker 6: so understanding that is really important for where you're going 449 00:27:06,160 --> 00:27:10,680 Speaker 6: to put your suboceanic infrastructure. So you've got cables going 450 00:27:10,720 --> 00:27:14,640 Speaker 6: across the oceans. If we're going to use our oceans more, 451 00:27:14,720 --> 00:27:18,880 Speaker 6: we need to understand what that environmental impact is going 452 00:27:18,880 --> 00:27:20,160 Speaker 6: to be long term. 453 00:27:20,920 --> 00:27:23,639 Speaker 1: The Ocean Model launched at the end of September twenty 454 00:27:23,680 --> 00:27:33,800 Speaker 1: twenty five. The research is only beginning. When I sat 455 00:27:33,840 --> 00:27:36,800 Speaker 1: down with Kevin Murphy at NASA, I wanted to understand 456 00:27:37,040 --> 00:27:40,520 Speaker 1: where all of this impressive work was going. And one 457 00:27:40,520 --> 00:27:43,280 Speaker 1: of the signature aspects of this work is that it's 458 00:27:43,320 --> 00:27:47,360 Speaker 1: not just for IBM and NASA researchers. Anyone can use 459 00:27:47,400 --> 00:27:48,320 Speaker 1: these models. 460 00:27:49,080 --> 00:27:51,760 Speaker 5: So before, if you were a researcher, or let's say 461 00:27:51,920 --> 00:27:56,200 Speaker 5: you were a farmer or maybe a technology informed person 462 00:27:56,280 --> 00:27:58,720 Speaker 5: that was interested in something like this, you would have 463 00:27:58,760 --> 00:28:02,119 Speaker 5: to learn about how to do remote sensing, how to 464 00:28:02,280 --> 00:28:06,040 Speaker 5: calibrate the imagery, how to stitch it together, because you 465 00:28:06,080 --> 00:28:07,920 Speaker 5: know they come in kind of postage stamps that you 466 00:28:08,040 --> 00:28:11,360 Speaker 5: have to squash together. And then you'd have to learn 467 00:28:11,440 --> 00:28:14,400 Speaker 5: about the algorithms necessary to do all the processing right, 468 00:28:14,440 --> 00:28:17,360 Speaker 5: So a lot of work, and then you could actually 469 00:28:17,840 --> 00:28:21,000 Speaker 5: do the mapping that you were interested in. Today. What 470 00:28:21,040 --> 00:28:23,119 Speaker 5: you can do is you can go to hugging face, 471 00:28:23,400 --> 00:28:27,320 Speaker 5: which is where this model exists in the open using 472 00:28:27,359 --> 00:28:30,040 Speaker 5: kind of our open science principles, and you can apply 473 00:28:30,160 --> 00:28:35,040 Speaker 5: it to future or historical observations. With how having all 474 00:28:35,080 --> 00:28:36,920 Speaker 5: of that background information. 475 00:28:37,040 --> 00:28:40,560 Speaker 1: And with the partnership between NASA and IBM, these foundation 476 00:28:40,680 --> 00:28:44,400 Speaker 1: models are multiplying. The new version of prithviy I mentioned 477 00:28:44,480 --> 00:28:47,880 Speaker 1: launched in September twenty twenty four. Then in August twenty 478 00:28:48,000 --> 00:28:53,120 Speaker 1: twenty five, NASA and IBM launched another foundation model called Syria, 479 00:28:53,160 --> 00:28:56,320 Speaker 1: based on data from the Sun. Soria can help predict 480 00:28:56,480 --> 00:29:01,320 Speaker 1: solar flares which can disrupt communications and increase radiation for 481 00:29:01,480 --> 00:29:04,800 Speaker 1: high altitude flights. And then there's the Ocean model I 482 00:29:04,840 --> 00:29:08,360 Speaker 1: talked about with Kate Royce. So what does the future 483 00:29:08,400 --> 00:29:12,120 Speaker 1: look like for all these foundation models built from NASA data? 484 00:29:12,680 --> 00:29:15,000 Speaker 1: If I wanted to look five or ten years out 485 00:29:15,040 --> 00:29:17,920 Speaker 1: to understand erosion patterns in a coastal town. 486 00:29:18,080 --> 00:29:20,960 Speaker 5: Yeah, you could give me Eventually, I think we'll get there. Yeah, 487 00:29:21,240 --> 00:29:24,520 Speaker 5: you know, we've really only been doing this for the 488 00:29:24,560 --> 00:29:28,840 Speaker 5: past few years. There is a lot of I think 489 00:29:29,040 --> 00:29:33,800 Speaker 5: capabilities to still discover and uncover with how we use 490 00:29:34,240 --> 00:29:38,120 Speaker 5: these models for like especially long term predictions. Like you're talking. 491 00:29:37,880 --> 00:29:41,440 Speaker 1: About what do you think you can't do and that 492 00:29:41,560 --> 00:29:43,920 Speaker 1: you really love to do. What's the kind of like 493 00:29:44,200 --> 00:29:45,440 Speaker 1: great white whale problem. 494 00:29:45,720 --> 00:29:47,480 Speaker 5: We can't do this today, but I'd like to be 495 00:29:47,480 --> 00:29:49,320 Speaker 5: able to do it in the future, which is really 496 00:29:49,400 --> 00:29:52,640 Speaker 5: the linking of the models together. Right. So right now 497 00:29:52,680 --> 00:29:57,160 Speaker 5: we have these isolated areas where you know, we have 498 00:29:57,240 --> 00:30:03,080 Speaker 5: the harmonized lansat Sentinel or GSPA. We have the weather model, 499 00:30:03,800 --> 00:30:06,240 Speaker 5: which can look at short term predictions. We're building out 500 00:30:07,240 --> 00:30:10,560 Speaker 5: the heliophysics model to look at the Sun dynamics. But 501 00:30:10,640 --> 00:30:13,640 Speaker 5: they're probably going to have to be additional models built 502 00:30:13,640 --> 00:30:17,040 Speaker 5: so that we can understand how they interact with one another, right, 503 00:30:17,840 --> 00:30:22,560 Speaker 5: And that is you know, kind of towards a digital 504 00:30:22,600 --> 00:30:26,400 Speaker 5: twin of kind of the Solar system or Earth systems, 505 00:30:26,880 --> 00:30:29,680 Speaker 5: which I think is a big Harry problem. But if 506 00:30:29,680 --> 00:30:32,080 Speaker 5: we understand it, we might be able to address some 507 00:30:32,080 --> 00:30:34,000 Speaker 5: of the questions that you just asked about prediction. 508 00:30:34,560 --> 00:30:37,720 Speaker 1: So if you linked all of those models together, Basically 509 00:30:37,800 --> 00:30:39,960 Speaker 1: what you're saying is, can I you say a digital 510 00:30:39,960 --> 00:30:46,880 Speaker 1: twin you're essentially replicating holistically how our world works. 511 00:30:47,360 --> 00:30:47,560 Speaker 5: Yep. 512 00:30:47,760 --> 00:30:49,680 Speaker 1: And do you think that is achievable? 513 00:30:50,600 --> 00:30:53,440 Speaker 5: I don't think it's immediately achievable, yeah, but based on 514 00:30:53,560 --> 00:30:55,520 Speaker 5: kind of the progress that we've seen in the last 515 00:30:55,520 --> 00:30:59,360 Speaker 5: three or four years, I think it's more achievable today 516 00:30:59,400 --> 00:30:59,880 Speaker 5: than it was. 517 00:31:00,720 --> 00:31:04,600 Speaker 1: Do you think you'll see it in your Yeah? 518 00:31:04,880 --> 00:31:06,840 Speaker 5: Sure, I'm hopeful, and I've got a couple of years last. 519 00:31:21,680 --> 00:31:25,480 Speaker 1: Smart Talks with IBM is produced by Matt Ramano, Amy Gains, McQuaid, 520 00:31:25,960 --> 00:31:30,240 Speaker 1: Trina Menino, and Jay Harper. Were edited by Lacy Roberts. 521 00:31:30,680 --> 00:31:34,840 Speaker 1: Engineering by Nina Bird Lawrence, mastering by Sarah Bruguer, music 522 00:31:34,920 --> 00:31:40,640 Speaker 1: by Gramoscope, Strategy by Tatiana Lieberman, Cassidy Meyer and Sophia Derlin. 523 00:31:41,480 --> 00:31:46,120 Speaker 1: Special thanks to the team at NASA's Science Mission Directorate. 524 00:31:47,240 --> 00:31:50,440 Speaker 1: Smart Talks with IBM is a production of Pushkin Industries 525 00:31:50,680 --> 00:31:55,240 Speaker 1: and Ruby Studio at iHeartMedia to find more Pushkin podcasts. 526 00:31:55,520 --> 00:31:59,480 Speaker 1: Listen on the iHeartRadio app, Apple Podcasts, or wherever you 527 00:31:59,560 --> 00:32:03,680 Speaker 1: listen to podcasts. I'm Malcolm Glawell. This is a paid 528 00:32:03,680 --> 00:32:08,760 Speaker 1: advertisement from IBM. The conversations on this podcast don't necessarily 529 00:32:08,800 --> 00:32:29,720 Speaker 1: represent IBM's positions, strategies, or opinions. Since we recorded this episode, 530 00:32:30,000 --> 00:32:34,800 Speaker 1: IBM and NASA released Syria their solar weather model. In 531 00:32:34,840 --> 00:32:38,680 Speaker 1: early testing, it showed us sixteen percent improvement in solar 532 00:32:38,720 --> 00:32:42,680 Speaker 1: flare prediction accuracy. This is the kind of improvement that 533 00:32:42,800 --> 00:32:46,840 Speaker 1: helps protect our satellites, our power grids, and our GPS 534 00:32:46,840 --> 00:32:51,240 Speaker 1: systems from the Sun's unpredictable nature. And the next step 535 00:32:51,320 --> 00:32:55,080 Speaker 1: in this partnership another model coming in twenty twenty six. 536 00:32:55,400 --> 00:32:58,840 Speaker 1: Looking beyond the Earth and the Sun, the universe of 537 00:32:58,920 --> 00:33:01,400 Speaker 1: possibilities just keeps expanding.