Decoding nature’s hidden patterns

0:00This message comes from Whole Foods Market. Host unforgettable barbecues with savings from Whole Foods Market. Check out their ready-to-cook kebabs and made-in-house pork sausages. Stock up on bug sprays and sun care must-haves. Make your summer sizzle at Whole Foods Market. This is the TED Radio Hour. Each week, groundbreaking TED Talks. Our job now is to dream big. Delivered at TED conferences. To bring about the future we want to see. Around the world. To understand who we are. From those talks, we bring you speakers and ideas that will surprise you. You just don't know what you're going to find.

0:38Challenge you. We truly have to ask ourselves, like, why is it noteworthy? And even change you. I literally feel like I'm a different person. Yes. Do you feel that way? Ideas worth spreading. From TED and NPR, I'm Manoush Zomorodi. On the show today, translating nature. how researchers are using AI and other technology to better understand what the world around us is saying. Yeah, I think it honestly, it always shocks me how little we know about the planet. This is Sarah Beery. She's a professor of AI at MIT. When I think about the hundreds and hundreds of years that scientists have studied species, you know, all the way back from Charles Darwin until today, and all of the technological advances that have come with that.

1:34You would think that we would know every species on Earth at this point, but we really don't. Sarah says that by best estimates, humans have documented maybe 2 million species of animals, which sounds like a lot. But our best statistical estimates to try to figure out how many species on the planet estimate that we have somewhere between 10 and even 100 million species on Earth. As you might imagine, this creates a problem for conservation efforts. You can't save animals that you don't know exist. And Sarah says it's not enough to just discover a new species, give it a Latin-sounding name, and call it a day.

2:16Because that's not enough, right? If you want to protect a species, it's not enough to just know that it exists. You have to understand what it needs, what it eats, how it relates to its environment. For example, if the temperature goes up by five degrees, will that species still be able to live and survive or not, right? Without a firm understanding of these details, it's difficult to save animals and their habitats and will continue to lose species at an alarming rate. So unfortunately, I mean, if we look at our best estimates of even just the population sizes of wildlife on Earth, it's estimated we've lost over 70 percent of all wildlife since 1970.

3:02Like, imagine that, right? Within the lifetimes of people on Earth, we've lost 70 percent of all wildlife. That's shocking. but despite the dire statistics sarah is actually hopeful because over the last decade or so citizen science has exploded you might have an app like i naturalist or ebird on your phone right now where you can take a picture of an animal or plant and identify it almost instantly and then you can share and contribute your photo location information and field notes you know you don't need to be a scientist. You don't need to be formally trained as a scientist to help collect scientific data.

3:43And Sarah says researchers are mining and learning from all those millions and millions of entries. And this has already been a complete game changer for biodiversity. At this point now, community science data makes up probably 90 percent of all biodiversity data we have and have ever collected. So, you know, I've played around with iNaturalist and it's really fun. You feel like, you know, you're part of a big group of humans out in the world seeing what's out there and documenting it as you go. But you're saying that this is actually a very important source of data for researchers. Yeah. I mean, every image in a naturalist gets identified as a species by the community and then turned into this scientific record of the presence of a species in a given place and time.

4:34And then once those images have been identified to what's termed research grade, they get directly exported into the Global Biodiversity Information Facility, which is a clearinghouse of what is now over 3 billion species occurrence records. And so this has already been really transformative because it's making this information standardized and scalable. But I think one of the things that I am most excited about in all of this is the information that we're currently leaving on the table. If you look at the pixels of these images, there's a lot more ecological information hiding in those pixels than just the presence of a single species.

5:19Can you give me an example? I mean, if someone goes out into the world, they take a photo and they're like, OK, yes, that is a blue jay. They're also potentially capturing the type of tree that the blue jay is sitting on. Other species that might be in the picture. Absolutely. Not just, oh, here's a blue jay and here's the type of tree it's sitting on. You can look at the feathers of the blue jay and determine whether it's in breeding plumage or whether it's a juvenile or a male or a female. You can look and see how that blue jay is behaving in its environment. Is it calling perhaps, you know, some sort of like attractive behavior?

6:04Is it looking for food? Is it maybe even actually currently eating something like an insect or a seed? You can look at the type of tree that it's sitting on. You can look at the sort of structure and the health of that tree. You can look at the amount of vegetation in the background and you can identify like the specific niche preferences of species related to their environment. Right. So blue jays prefer specific types of trees or structures of trees to be their habitat. And those preferences might be quite different for another species like a tree sparrow. And so iNaturalist has collected 200 million images.

6:48And I'm so excited about the potential of building AI systems that help scientists really quickly find new information. that's currently, it's there. It's there for the taking, but we just don't have the right systems to make it accessible, to make it possible to do at scale. Scientists have long learned about the natural world by observing it firsthand, painstakingly in the field. But new tools from the satellites orbiting Earth to remote monitors deep in forests to the smartphones in our pockets, they're all capturing more information about our ecosystems than ever before. So how can we make all that data useful?

7:34Today on the show, translating nature, making sense of the world around us to protect plants, birds, apex predators, all with a little help from technology. Which brings us back to Sarah Beery. We're sitting on an ecological gold mine, and the problem is accessing the knowledge efficiently. Here she is on the TED stage. So say you want to look through all this data. Assuming it takes you about a second to look at every image, you would need to work full time for 40 years to look through all the images in iNaturalist alone. And this is where AI is transformative. It can just help us look through all the data quickly.

8:14So an ecologist today, say they're interested in bird diets, and they want to find examples of birds eating insects in the database, what they can do is they can train an AI model to help them. So to do this, they collect hundreds or even thousands of examples to teach the model what to look for. Now, once they've trained this model, it's an incredible tool. It can very, very quickly find new examples of birds eating insects in the database. But this process of collecting hundreds or thousands of examples every time we want to look for something new, it's still too slow. So let's reframe the question.

8:53scientific discovery really begins with scientific curiosity with asking questions about the world and how it works things like how far can a grant zebra migrate what plants grow back after a forest fire do birds eat insects during the winter wouldn't it be great if instead we could just directly ask questions to our databases and get answers back. This is what my team at MIT has been working towards. And we've developed a system that we call Inquire that helps ecologists find answers in the data without collecting any examples to teach an AI model or needing to write any lines of code. Now, under the hood, what we're doing is we're developing AI models that can learn and understand similarities between images and scientific language.

9:45And this is what allows us to just ask. Can you just give us an example of how Inquire works? Yeah. So one example of a collaborative scientist that we've been working with, they're interested in how forests are growing back after forest fires and how the severity of forest fires actually impacts the growth of the forest coming back. And so what they've been doing is they've been searching within a specific area where they knew that there was several past forest fires that had burned. They're searching for evidence of burn of previous forest fires in the background of images. And then they're able to characterize how severe that burn was based on that sort of evidence of past burn in the background.

10:28And then they're looking at the distribution of what types of trees are growing back. Is it more deciduous, more coniferous trees growing back in the foreground. So it's essentially just a discovery tool, but it can be used in a really open-ended way. What that researcher found in some preliminary studies is that it appears that areas that are more severely burned have more deciduous trees growing back afterwards, and areas with less severe burns have more coniferous trees growing back afterwards. And that's important because we know that burns are actually happening faster and hotter than they were previously.

11:08So this is going to influence the way that our forests shift and change as more of the forests burn and as the severity of the burns goes up. And presumably that starts to be a data point for modeling how the world may change in the next year, five years, 20 years? Yeah, exactly. I mean, so much of science, environmental science and ecological science is just trying to understand what's currently happening and what might happen in 10 years and 20 years. Understanding the implications of the way that climate is changing, the way that, you know, larger scale human development is changing the Earth's surface helps us understand what to prioritize, where to prioritize, how to take action to protect species.

11:54So triage, essentially. Yeah, triage, but proactively, right? So there's a great example of this that I think is very intuitive, which is we know that a lot of birds die during migration by being sort of distracted and moving into cities because of lights. And then they'll run into windows in the cities. And we have estimates of, you know, hundreds of millions of birds that are dying each year during the migration season just by hitting windows in cities. One of the adaptive management campaigns that people are looking at is lights off campaigns. So when we actually understand, when we've detected that the migration is going through, now there are campaigns for cities that are in the path of that migration to just turn off all their city lights at night.

12:44Because that migration happens just a couple days a year, right? The vast majority of birds are migrating through the U.S. just a couple days a year. So just turn off your lights a few days a year and you're already having a huge beneficial impact. So easy. So easy. But we need to know when and where it's happening to be able to take that step. When we come back, Sarah Beery takes us into the field and explains how she can train her AI without using massive amounts of energy. On the show today, translating nature. I'm Manoush Zomorodi and you're listening to the TED Radio Hour from NPR. Stay with us.

13:30Support for NPR comes from IBM. On Smart Talks with IBM, host Malcolm Gladwell speaks with leaders who are pushing the boundaries of AI and technology in partnership with IBM. Hello, hello. I'm Malcolm Gladwell, host of Smart Talks with IBM. I sat down with Alon Cohen, who leads research and development at UFC, to discuss the complexity of using technology to analyze fight data. With kick to the head, it makes contact with the outset of my arm, which I brought up. In our world, that's a blocked strike. Yeah. But teaching a computer what exactly that means and when and how. Like, when my arm is up, that's a block.

14:12When my arm is down and hits my shoulder, that's not. It's those nuances that proved incredibly difficult for machines to be able to handle for a very, very long time. That is until IBM entered the octagon. Listen to Smart Talks with IBM wherever you get your podcasts.

14:34It's the TED Radio Hour from NPR. I'm Manoush Zomorodi. On the show today, translating nature. And we were just talking to Sarah Beery. She's an MIT professor who specializes in AI and ecological conservation. Sarah and her team have built a new AI tool they call Inquire, which lets researchers search huge treasure troves of data just by asking normal open-ended questions. Kind of like how ChatGPT works. It's helping scientists do their work faster and easier. But I asked Sarah, at what cost? Whenever I talk to people about AI and the environment, people are like, well, AI is part of the problem.

15:17It is everybody, all these techies say, oh, AI is going to solve our climate problems. But actually, all of the energy that these data centers require are just making it worse. How do you see it? Yeah, so AI is actually, I mean, it's a term that is incredibly broad. And there are many different ways that AI can be used. Things like ChatGPT, some of these other large generative language models and vision language models that you're interacting with if you're, you know, playing with some of these AI chatbots. This is really only one type of AI, and it is probably the type of AI that is most energy intensive, right?

16:02Both in the training of the models and the models themselves are very, very large in the use of the models. It does take a lot of energy. But much of the AI that we're building for environmental science is AI that is designed to be not computationally intensive from the beginning. Because if you need to use AI in a remote rainforest or out in the middle of the savanna where you don't have a large-scale data center, you don't have access to GPUs, you don't have the bandwidth to move the data to the cloud, then by design, you have to work within the resource constraints. So often we're building AI that is powerful but small and that is actually far less computationally intensive than maybe some of the traditional statistical methods that are used in ecology.

16:53So we've talked a lot about the data and using computers to make sense of it all. But what about fieldwork? Because I know that you do get out of the lab quite a bit and get out there yourself, right? What have you been looking at recently? Yeah. So one of the most important things as we're starting to see AI grow and be deployed in the real world is direct integrated co-design and development with the stakeholders who are going to be using the models. And to me, that means going to the field and seeing the models in use, in deployment, understanding what gaps remain, and then prioritizing our research to be innovation that's driven by these applications of the models.

17:37So one example, we have a project where we've been working with a bunch of different state agencies, NOAA and Cal Trout, to try to improve our estimates of salmon escapement to build more sustainable fisheries. So escapement is basically as the salmon are coming upstream, you want to count the number that are going upstream to spawn so that you have a good estimate of how many will actually successfully breed and then kind of go back out to the sea and then come back in three years. So to keep a sustainable population of salmon, you need to ensure that enough have escaped upstream. Now, one of the things that is very exciting is across the state of California, Washington and Oregon, we're seeing a lot of dam removal projects going in.

18:22Because we've realized that dams have been a pretty significant barrier for salmon populations. So the Klamath Dam, there was a series of four dams on the Klamath River in Northern California that were all taken down in the last year. It's the largest dam removal project in the history of the world. And it's opened up 800 square miles of habitat for salmon. And so what we've been doing is we've been working with Cal Trout. They have sonar cameras in the river, and we're using that to automatically count the number of salmon that are escaping upstream. So we were just there at the dam removal site, actually working with the team that's managing the sonar, that's kind of using and evaluating our models and still trying to understand what the gaps are.

19:08You know, what do they still need? What does the technology need to do to improve? And I think this type of kind of interactive partnership, this interdisciplinary understanding, it's vital if we want to make sure that AI is used in a way that's truly beneficial. It's not a hammer looking for a nail. We really want to design the right tools for the right problems. And sorry, forgive me for sounding obtuse, but what did the salmon get out of it? So there's actually the estimate is that there's potentially been close to 10 ,000 salmon that have gone up into this previously inaccessible habitat in the very first year.

19:48And salmon are what's called a keystone species. Oh, they are? Yeah, yeah. So keystone species, they're essentially species that, another very famous example of a keystone species is wolves. Like when the wolves were reintroduced in Yellowstone, we saw this kind of expansive effect where the entire food chain, the entire ecosystem started to really flourish in response to having the presence of this keystone species. So for salmon, not only do they provide food for a huge number of different types of organisms, right, both like large predators like bears, wolves, coyotes, but also as they decompose after they've spawned, they're introducing a huge amount of nutrients in these rivers.

20:35and their presence as a keystone species, we're hoping we'll actually really be able to, now using again a lot of these technologies, we're hoping to actually measure how that trickles outwards. So to be able to, at this type of comprehensive scale, monitor the regeneration or the kind of bouncing back of an entire 800 square mile river system, that just wouldn't have been possible 10 years ago. And we're still figuring out how to do it today, But I'm very optimistic that we're going to be able to really understand things at scales that were just not possible in the past. I mean, Sarah, it feels so good to hear optimism and, you know, progress.

21:22And you don't hear a ton of that when it comes to the environment and certainly not when you layer on AI usually. But I know as much as you are an optimist and see the amazing applications of this technology, you have also been very pragmatic and critical of those who claim like, oh, yeah, no problem. AI is going to fix so many things. I know earlier this year, Google came out with a tool called AI Co-Scientist, and you sort of were like, yeah, let's be careful what kind of promises we're making big tech. Can you explain what your sort of concerns are with this tool and maybe others? Yeah, I mean, not to even pick on that tool in particular, but I do think that in general, computer scientists have had a bit of a problem with what some people call tech saviorism, where they'll sort of say, oh, like AI could do that.

22:24And it turns out that hard problems are hard for a reason. And AI isn't a magic wand. And so I think a lot of my work, my research, and also some of the ways that I kind of act as a leader in our field is really thinking about how do we do AI robustly and rigorously? And how do we build systems that ensure that we're evaluating AI in ways that validate the claims we're making? I think AI can be a really amazing tool to help an expert do something more efficiently, to help our experts spend less time like looking through millions of images and instead spend time analyzing them, understanding the trends that they kind of help us establish, and then working to in the policy space to sort of shape and take the actions that are going to have an impact.

23:22But I really think it's vital that we have these experts very tightly connected in the loop, right? We design systems where AI is a beneficial tool, but not a solution in and of itself. We stand at a unique point in history. We have both an unprecedented biodiversity crisis, but we also have unprecedented tools to address it. We have millions of people around the world eager to contribute to nature conservation and scientific discovery. And we have AI tools that enable scientists to find patterns in all of that data at scales impossible for humans alone. The future of conservation doesn't just lie in remote rainforests or deep ocean trenches.

24:10The future of conservation is hiding in our ecological databases, both the ones we have now, but also the ones we have yet to collect. And that is where all of you come in, because everyone can contribute. Everyone can collect data and upload it to platforms like iNaturalist. Every photo uploaded, every sound recorded, every observation shared is a piece of the puzzle. We know that we need to act now to save nature under threat. And together, with scientific AI tools in our toolbox, we can help by building the complete picture of life on Earth. That was Sarah Beery. She's an assistant professor of AI at MIT's Computer Science and AI Lab, and she runs the Beery Lab.

24:59You can see her full talk at TED.com.

25:06Today on the show, translating nature. And now we want to go deeper into decoding one particular voice in the wild. A wolf's howl, a primal sound that's been heard and mimicked by humans for thousands of years. People have been howling forever. I'm sure the Clovis people that came here 13 ,000 years ago were doing it. The Mountain Shoshone, who used to live in Yellowstone, they were described by trappers as being the best imitators of animals around. They could howl like wolves, they could neigh like horses, and imitate any bird. Hunters have used it to draw in wolves and kill them. Some people did it during COVID.

25:55They'd go outside their back door on the porch, and they'd howl to prove to the world that they were still alive.

26:08Across cultures, wolves are symbolic. For some indigenous nations, they stand for loyalty and family. In European stories like Little Red Riding Hood, they're a reminder that danger lurks everywhere. And, of course, wolves were domesticated, becoming man's best friend and pet. But a century ago in the U.S., wolves were widely seen as predatory enemies, hunted to near extinction. In places like Yellowstone, they were wiped out entirely. So Jeff Reed, who grew up on the outskirts of the National Park, never heard their howls as a child. I was born in the late 60s in southwest Montana, which we call the Absorkee Mountains here, just north of Yellowstone National Park, one of the premier ecosystems in the world.

27:01And we lived outdoors all the time. Without wolves, Yellowstone lost its apex predator. The elk population surged and plant life was devastated. It took several decades until 1995 when conservationists reintroduced wolves in the hopes of restoring the park's delicate ecosystem. By the time Jeff Reed moved back to the area after spending his adult life in Silicon Valley, wolf howls were everywhere. And he wanted to know what the wolves were saying. So when I moved back, wolves were back. And I always knew that there's nothing more iconic than a wolf howl to tell you about nature. you know I was going out and studying them just as a casual observer and it really dawned on me that wolves are hard to study like you can't walk up with a mic and say you know what did you mean by that and I knew a lot of the biologists who were working in the park and they're amazing researchers and I just started talking to a few of them and said hey what if we did a large-scale bioacoustics project where we put recorders out on the landscape and then really started studying what the wolves were saying, so to speak.

28:17Jeff's background in tech and linguistics made him the perfect person to build on the work already being done to understand wolves. Wolves have been researched for their sounds for a long time, and there are captive facilities, often sanctuaries, for wolves. And so you can hear them in captive centers. But we were really interested in wild wolves. Like, what are your kids saying when they leave the home and they're with their friends? That's really what we wanted to get after is how are they communicating in their natural soundscapes. What were the tools that they were using before you got there?

28:52So wolves in Yellowstone are the most studied wolves in the world. So they already have this huge database, genetics, you name it, observations. They fly planes over, they put GPS collars on. And so we just started to amplify that with an acoustics project. To do that, Jeff needed to upgrade the technology. There was certain acoustic technology that you could record, but it really hadn't been innovated upon from a biologist's perspective on the landscape. And so I was like, of course we can do this, right? We do this all the time in Silicon Valley. It's just we're only doing it to make money. And so I was like, OK, let's start reinventing these devices, these camera traps and these acoustic recorders, and have it record for a year.