The Science of Birds - Artificial Intelligence in Bird Research

Episode Date: May 28, 2021

We hear the terms ‘artificial intelligence’ and ‘AI’ all the time these days. Beyond the issue of evil robots taking over the world, AI technology is helping scientists do some pretty amazing things in the field of ornithology.In this episode, we’ll talk about what artificial intelligence is and give some interesting examples of how it’s being used to study birds.We’ll also touch on some tools that use artificial intelligence to help you in your quest to identify birds.~~ Leave me a review using Podchaser ~~Links of InterestBirdCastBirdNetMerliniNaturalistHaiku BoxDigital Guide monocular Link to this episode on the Science of Birds websiteSupport the show

It's 1850, and you're a plucky naturalist exploring the steaming jungles of Borneo. You're pushing your way through the undergrowth, looking for new bird species to document in the name of science. What tools are you carrying for your work? Besides your trusty binoculars, you've got a notebook, a rifle and ammo, and a pack to carry your gear and any specimens you collect. That's where the gun comes in, right? You aren't going to get any bird specimens unless you shoot them out of the trees. That's how it's done. Back at your camp, you'll package up the dead birds and eventually mail them back to a museum in Europe, whence you came.

You use your notebook and your modest artistic talents to sketch the birds as you see them in their natural environment. You write copious notes on their behavior in the field and on their habitats. You don't know it, but from a 21st century perspective, you're lacking some important tools. In your time, in 1850, there are no cameras or audio recorders. There are no computers or phones of any kind. No apps, no internet, no Wikipedia, no email. And as a naturalist in the mid-19th century, you're also trying to make sense of nature with no knowledge of fundamental theories we take for granted in the 2020s.

the theory of evolution by natural selection, for example, and plate tectonics, and gene theory, and population genetics. Whatever discoveries you make or conclusions you come to will be the fruits of your own brain and senses, your own natural intelligence. You don't have auxiliary devices like computers or calculators to help you. Most of the processing and number-crunching power you have for analyzing your observations, rests between your ears, inside your skull. Sure, your research might benefit from conversations with your fellow naturalists, when and if you see them again after your long journey. But in any case, it's the human brain, whether one, several, or many brains working together that does the work of science in the

1800s. Needless to say, science advances at a snail. pace in this era. Jumping forward to the present. Modern ornithologists in the first decades of the 21st century still find tools like binoculars and notebooks pretty much indispensable. I guess some things, like phone books and cursive writing, will never go out of style. But a lot of the other technology we use today would probably seem like arcane sorcery to naturalists in the 1700s and 1800s.

I mean, if I traveled back in time to show those guys my smartphone, they'd probably burn me at the stake for witchcraft. What is this Google Maps devilry of which you speak? And what, Braytel, is an e-bird app? It sounds like evil to me. He's an unholy wizard. Kill him! Cue the pitchforks and torches.

Ornithologists and other biologists these days also have unimaginable amounts of raw data at their disposal. Genetic data, photos, videos, satellite imagery, sound recordings, movement data from tracking devices, habitat data, climate data, and more. Luckily, present-day biologists also have access to ever-increasing computing power and sophisticated analytical tools for making sense of all that data. So they don't need to do all the work. on their own. Scientists use computers that can think outside the box, so to speak. Scientists these

days have more than their own natural human intelligence. They get a lot of help from artificial intelligence. Hello and welcome. This is the science of birds. I am your host, Ivan Philipson. The Science of Birds podcast is a lighthearted, guided exploration of bird biology for lifelong learners. This episode is about the use of artificial intelligence in scientific research on birds. We'll also touch on some tools that use artificial intelligence to help you in your quest to identify birds. So put on your thinking cap, my friend, because it's time to nerd out.

What exactly is artificial intelligence? I don't know about you, but I see this term pop up all over the place, artificial intelligence, AI. If we can believe what we hear, it seems pretty clear that artificial intelligence is going to be the downfall of human civilization. Scientists are going to make all these super-smart robots that will one day become sentient and suddenly turn on us. Humans will be hiding out in the smoldering ruins of buildings, cowering in fear as the world gets overrun by gun-toating robots that look an awful lot like

former governor of California, Arnold Schwarzenegger. And don't forget about Hal 9,000 from the movie 2001, or The Matrix, Battlestar Galactica, Westworld, and all the other evil robot stories. We can even trace this sort of tale back to Frankenstein and his monster. We might have a good laugh and sleep easy at night, thinking that such science fiction scenarios will never come to pass. But there's genuine concern about AI among scientists and other big thinkers. Professor Stephen Hawking said, quote, the development of full artificial intelligence could spell the end of the human race. It would take off on its own and redesign itself at an ever-increasing rate.

Humans, who are limited by slow biological evolution, couldn't compete and would be superseded. End quote. Ah, comforting words from one of the greatest minds of our time. But artificial intelligence isn't just about evil robots eventually taking over the world, world. It's also about evil robots helping us perform mundane tasks in the present. Virtual assistants like Siri, Alexa, and Google use artificial intelligence to understand human language and to give us the information we're asking for. I have an Android phone, which means I use Google. Hey, Google, what is the airspeed velocity of an unladen swallow?

It really depends if you're talking about an African or European swallow. See, even the robot that lives in my phone gets my Monty Python references. Other examples of AI that are all around us include face detection, autocorrect when we're typing, Google and Apple Maps, search engine algorithms, Amazon product recommendations, chatbots, and self-driving cars. But I still haven't answered the question, have I? What is artificial intelligence?

The dictionary puts it this way. Artificial intelligence is the capability of a machine to imitate intelligent human behavior. That's probably what most of us imagine when we hear the term artificial intelligence. Another definition I came across is a little more specific. AI is a system's ability to correctly interpret external data, to learn from such data, and to use that learning to achieve specific goals and tasks through flexible adaptation. This is the sort of artificial intelligence we're talking about today regarding bird research. A computer program interpreting data, learning from that data, and adapting itself, if necessary, to reach a goal.

One of the main reasons computer scientists developed AI is because the more standard computational approaches of the late 20th century just can't keep up anymore. They can't keep up with the ginormous heaps of data that humans generate every set. second of every day. We are now in the era of big data. That's big data with a capital B and capital D. So much of what we humans do now gets translated into digital information, digital data, lots and lots and lots of it. Traditional data processing methods can't properly deal with these great volumes of complex data. AI is based on algorithms that can parse and sort and churn through all that data to arrive at some answer, to achieve some goal. Algorithms are sets of rules that AI uses to solve a problem. Some algorithms are better than

others for certain tasks. What's really cool about artificial intelligence algorithms is that many of them can learn from data. The algorithms fine-tune and improve themselves. They get better through experience. This is what we call machine learning. In my experience, which is admittedly limited, the term machine learning seems to be used more often than artificial intelligence in scientific papers about birds. I could be wrong, but I wonder if scientists sometimes shy away from saying artificial intelligence because of all the evil robot stuff in pop culture. Machine learning doesn't have quite the same connotation, does it? Generally, for machine learning to work,

we need to first train the machine. We use a training dataset to teach the machine what to look for. So if you're training an AI algorithm on how to identify bird species from photos, you feed the machine a bunch of photos and then you tell it the correct species name for each photo. And maybe you take it one step further and label the various parts of the bird in each photo,

the eye, the tip of the bill, and so on. This training dataset ideally includes thousands of photos, or even millions. This is called supervised machine learning. You're giving the AI some useful info that will help it improve the algorithm. Then it will perform better when it's confronted with the real task, which is identifying birds in unlabeled photos, photos the machine has never seen before. Okay, at this point, you might be ready to yell at me.

But what about birds, Ivan? I'm here to learn about birds, for Pete's sake, not bloody computers. Fair enough. Let's move on to talk about birds and how artificial intelligence is being used to study them. I'm going to cover three major types of data, image data, radar data, and acoustic data. There are certainly other types. like genetic data, movement tracking data, etc. But today I'm just going to focus on those first three.

Starting with image data. Everybody and his grandma is a photographer these days, right? We all have smartphones that have great cameras built in. And there are more amateur and hobby photographers than ever, running around with expensive DSLRs and mirrorless cameras that can take excellent photos of birds. So photos and videos are the image data we have for birds. And with so many people out there capturing images of birds, the world is positively overflowing with this kind of data.

It's safe to say that this qualifies as big data. Those images are easily uploaded to the internet and shared with other people, including scientists. Online databases are now chock-full of bird photos and videos, ripe for the harvesting by ornithologists. And of course, ornithologists themselves can generate great gobs of bird photos or videos when they need to for specific research projects. Artificial intelligence is being used, with great success, to identify bird species from photos. We'll talk some more about species recognition a little later. But let me tell you about some recent research using artificial intelligence to identify individual birds from photos.

I'm not talking about species now, but individuals. As in, Boy, that bird looks like old redneck Joe, the brush turkey from Brisbane. That bludger owes me 20 bucks. It's one thing for a computer with AI to distinguish between two species, like the Australian brush turkey and the yellow, legged brush turkey, for instance. But this new research aimed to identify individual birds,

like old redneck Joe, a one-of-a-kind Australian brush turkey that I just made up. Of course, technology like this is already being used by face recognition algorithms for humans. But until recently, it hadn't been used with small birds. So this study was published in 2020 in the journal Methods in Ecology and Evolution. The researchers are from institutions in Europe mostly. They wanted to develop AI algorithms that could identify an individual bird from a photo. They worked with three bird species,

the sociable weaver of Southern Africa, the Great Tit of Europe, and the zebra finch of Australia. The weavers and tits in this study were from wild populations while the zebra finches were captive. The AI method in this case involved deep learning, which uses algorithms inspired by the structure of the brain. We call these algorithms

artificial neural networks. In order for this deep learning approach to work, the researchers needed to build large training datasets to quote-unquote feed into the machine. Each individual bird needs to have hundreds of correctly labeled photos in the training dataset. Building this data set manually would be incredibly tedious and time-consuming, or maybe impossible if you don't actually know which bird is which in your training photos. So the researchers came up with a cool automated process to generate the training data. Individual birds were marked with tiny pit tags. These are little pill-shaped transponders that emit a unique code when activated by a pit tag reader. pit tags are the so-called microchips that you can have a veterinarian inject into your dog or cat or small child.

Wild sociable weavers and great tits were captured and fitted with these tags. Then when a bird with a tag landed at a special feeder in the study area, an automated pit tag reader got the ID for the bird and simultaneously triggered a camera to take the bird's photo. Each individual weaver or tit eventually had hundreds of photos. These were then used to train the artificial neural network algorithm. The zebra finches were a little easier to work with since they were captive birds. So how did it all work out? The AI could correctly identify individual birds of all three species about 90% of the time or better.

The system isn't perfect, and they, There are some challenges, but still, 90% accuracy is pretty good. This sort of approach for automatically identifying individual birds from images opens up a lot of exciting research possibilities for the future. This is super cool stuff. Artificial neural networks and deep learning show up in all the example studies I'm giving you for image data, and in some of the other studies we're looking at today as well. One study used video footage of purple martins at their nests.

It was published in the Journal of Avian Biology in 2020. The researchers wanted to determine what factors determine the rate at which the parent Martins brought food to their chicks. In other words, the provisioning rate. How many trips back and forth to the nest do these birds make and what affects the number of trips? There was continuous video footage recorded for each of 20 Purple Martin nests in Iroquois National Wildlife Refuge in New York State. The researchers wanted to know how many adult birds were at the nest at any given moment. The old way of answering this question would be to have some poor graduate student gather the data by spending their entire summer watching these videos frame by frame in a dark basement. But in this study, the researchers trained a robot to do the grunt work instead

because a computer with artificial intelligence isn't going to get bored and mess up the data by spacing off or looking at its phone every 30 seconds. Of course, the algorithms in this case still need to learn, right? This is machine learning after all. So some poor grad student did have to watch hours and hours of video initially to count birds, to generate the required training data. But once it was sufficiently trained, the AI algorithm got to work and was adept at counting the number of martins at the nest in each frame of video.

Anyway, there you go, an example of AI being used to automate the collection of data from video. I'll link to this Purple Martin study in the show notes on the website if you'd like to actually read about the results. The next study is pretty amazing. Seriously, we're getting into the realm of what feels like science fiction here. A group of researchers, based mostly in Finland, conducted a study on northern lapwing nests in agricultural areas. The northern lapwing, Vanellis Vanellis, is a gorgeous shorebird in the plover family, Karadriadi. Both males and females are a metallic sort of iridescent green on the back and wings.

They have a white belly and bold black and white markings on their heads. To top it all off, literally, these birds have a graceful, curving black crest on their heads. This study was published in the journal Scientific Reports in 2020. The researchers were experimenting with a new, automated way of locating northern lapwing nests. This species nests on the ground. making a simple scrape in an area with short grassy vegetation. Unfortunately, northern lapwings and some other species in Europe like to place their nests in agricultural fields.

Heavy farm machinery routinely comes along and destroys the nests. This is the primary threat to northern lapwings in Europe, where the species is endangered. Locating the nests of these birds manually is difficult and time-consuming. There are groups of volunteers that hunt for lapwings, wing nests, but wouldn't it be nice if a robot could do that instead? Artificial intelligence to the rescue! The researchers in this study outfitted aerial drones with thermal imaging cameras

and flew them around over agricultural fields in Finland. The drones followed pre-programmed flight patterns, snapping thermal images along the way. So these photos could pick up the heat signatures of nests on the ground. many such nests would be hidden in the vegetation and hard to find for mere humans. The thermal images were then transferred to a computer equipped with AI. You know the drill now. Before AI could tackle this problem effectively, the algorithm had to learn from a training dataset.

But eventually, the AI algorithm in this study could spot lapwing nests correctly about 90% of the time, not too shabby. In their paper, the researchers propose a system where drones fly around over fields in the nesting season. When the drones locate a nest using artificial intelligence, they alert any farmers who are operating machinery nearby. The machinery can then avoid the nest in one way or another, either manually or in an automated fashion. This makes me envision a future where an ornithologist could be sitting on her couch, watching some Gilligan's Island reruns, when she comes up with a research question.

She wonders, what is the average time that tufted puffins spend foraging underwater during the nesting season? Specifically, off the coast of British Columbia, Canada. The ornithologist

opens up an app on her phone and asks that question out loud. The app records and interprets her question. Then she taps a green button labeled Execute and several squadrons of solar-powered drones launch into the sky from a hangar in the backyard.

The ornithologist is like, Fly, my pretties, fly! You know, I thought that was one of the famous quotes from the Wizard of Oz, but apparently the Wicked Witch never actually says that line in the movie. She doesn't. Not quite like that anyway. In any case, there are all these drones

zooming off towards the horizon, like the flying monkeys of Oz. The drones communicate with, each other and with an AI-powered computer back at the ornithologist's lab, using several orbital satellites as intermediaries. The drones fan out along the coast of British Columbia and start recording video data on tufted puffins. The AI in the lab crunches the data to figure out how much time the birds spend swimming around underwater. Meanwhile, our ornithologist just lounges on the couch, eats pop-tarts, and mostly forgets about the drones for the next few months.

Then, one day in July, our ornithologists goes to her office and finds an email in her inbox. It's from her AI computer. The Puffin study is now complete, the email reads. The data was gathered and fully analyzed. The AI also took the liberty of writing the first draft of the manuscript, which is attached to the email. Bada Bing, Bada boom. Okay, so this is mostly fantasy, but it is within the realm of possibility in the coming decades.

There are some major hurdles, of course, not the least of which is that birds get spooked by drones pretty easily. So using drones to study natural bird behavior isn't going to be easy. And I'm not actually sure I want to see this become a reality. I mean, who wants to see hordes of drones zipping around among wild birds? Not me. Let's look at a different kind of data now, that which comes from Radar. As you probably already know, Radar uses radio waves reflected off of objects to determine certain qualities of those objects, qualities like location, size, and velocity.

Radar was invented for military applications around the time of World War II. In those early days, radar technicians were confused when they detected mysterious blobs or wisps in the sky that clearly weren't man-made aircraft. People called these enigmatic signals angels or ghosts. Eventually, we figured out that at least some of these ghosts are actually flocks of birds. Over the decades since then, radar has become a valuable tool for studying the movements of birds, especially bird migration. We call this radar ornithology, and it's come a long way since World War II. There are several types of radar used to detect or track flying birds. The most common is weather surveillance radar. That's what you see on your evening

newscast or in your weather app. Besides rain, storms, and snow, data from weather radar can also be analyzed to detect mass movements of birds through the atmosphere. And there are even ways to discriminate between birds, flying insects, and bats. But now let's get to the artificial intelligence stuff. There are no doubt multiple ways AI is helping to answer questions in the world of radar ornithology. Before AI came on the scene in recent years, and before we had such powerful computers, it used to take a ridiculous amount of manual effort to make sense of all the radar data with respect to birds. I'll give you an example of one fairly recent study. It was published in 2016 in the International Journal of Avian Science. In this study, the researchers used data

from marine radar systems in Portugal. As I understand, marine radar works at a fairly small spatial scale compared to weather radar. The data in this study was processed by a handful of machine learning algorithms. The researchers were comparing the algorithms in terms of their usefulness in radar ornithology. What I found especially interesting in this study is that the ornithologists were testing not only the various algorithms' abilities to separate birds from other objects, like storms or background clutter, to separate the wheat from the chaff, so to speak,

they were also testing how artificial intelligence performed at discriminating among several classes of birds, classes like herons, storks, gulls, and swallows. It turned out that there was one AI algorithm that performed pretty well for most of the classification tasks. Not surprisingly, it performed best when simply separating birds from everything else, from the clutter. It was more challenging to sort out the various size classes of birds.

But from what I understood in this paper, the winning algorithm still correctly classified groups of birds as, say, gulls or swallows or herons, roughly 80% of the time. These are size classes. This is not species recognition. In any case, 80% seems pretty reasonable to me. If it's possible for radar data to include enough fine-scale details, maybe future algorithms will get really good at classifying birds this way. The authors of this study in Portugal offered some real-world applications for using AI like this in radar ornithology. One example was using AI on wind farms to shut turbines down if incoming birds are detected. In case you haven't heard, wind turbines can be bad news for birds. The fast-moving

blades are really dangerous to them. There are already some systems that do this or something like it. For example, there's a product called Identiflight that uses image data in real time to monitor bird species on wind farms. Turbines can be automatically shut down to avoid killing eagles and other birds that approach the wind farm. Identiflight uses images, not radar, but the basic idea is the same. Another application suggested in the paper was that AI-powered systems could be set up to protect agricultural areas where crops like rice are grown, protect them from birds, that is, because it turns out that some birds really, really like to eat rice seeds. So, for example, imagine you're in Louisiana, near the Gulf of Mexico. Before you are acres and acres of rice fields.

It's a beautiful sunny day and everything is peaceful. But suddenly, the air raid sirens go off. Red alert. Red because the local AR radar system has detected an inbound flock of red-winged blackbirds. Farmers are scrambling and running in all directions. In his panic, one guy trips over a shovel. There are little kids sprawling in the dirt, crying. for their moms, it's pandemonium. The AI defense system automatically deploys an army of mechanized scarecrows and airhorns that spring up in the fields. With any luck, these deterrence will keep the

blackbirds from devastating the rice crop. The farmers huddle together in their storm cellars, praying that the ravenous cloud of birds will just pass over. At least the artificial intelligence radar system gave them some warning. All kidding and exaggeration aside, this sort of thing could work. That paper really did suggest a system like this. Real-time AI systems may someday be able to help protect crops from swarms of hungry birds. And there are already AI systems on the market that use radar to protect birds from collisions with things like buildings and aircraft. Now briefly, I want to mention Birdcast, the very excellent service provided by the Cornell Lab of Ornithology.

Birdcast is an online resource that uses radar data to provide real-time bird migration updates and forecasts of mass bird movements. This info is freely available to anyone. The scientists working on Birdcast used machine learning combined with decades of radar and climate data to generate accurate predictions of when and where birds are going to move during their spring and autumn migrations. This is pretty amazing stuff, people. Unfortunately, birdcast only covers the United States for now.

Maybe someday this sort of rich information will also be available for other parts of the world. I'll put a link to the Birdcast website in the show notes. You should definitely check it out. Go out into nature on a nice spring day and look for some birds. I pretty much guarantee you're going to see some. But if you stop to listen, you'll probably realize that you can hear many more birds than you can see. You hear more individual birds and more species. This is why if you're a birder, it really pays to learn some bird songs and calls. The skill of birding by ear opens up a whole new dimension to your

experience. Anyway, birds are noisy critters, and that means they unwittingly generate lots of audio information for humans to record and analyze. This brings us to the third data type we're talking about today, acoustic data. As with image and radar data, there's a bunch of cool ways that AI can help scientists analyze acoustic data. One of the most obvious applications is species recognition using songs or calls. Photos and video can work well for automated species ID, but if you think about it, it's often much easier to record a bird's song than it is to get a decent photo of the bird. So there are some advantages to using acoustic data for this purpose. Researchers in Argentina, for example, published a study where they used algorithms developed

for human speech recognition to automatically distinguish 25 different bird species from their songs. Specifically, these were birds in the diverse family, Fernariati, the oven birds and wood creepers

of South America. From what I can tell, there are many approaches and different AI algorithms being tested and applied to solve this kind of problem, the problem of automatic bird species recognition

from acoustic data. There may not be a clear winner yet among all the algorithms, but I want to talk a little about Birdnet, a tool that I find incredibly impressive. Birdnet is a research collaboration between the Cornell Lab of Ornithology and the Chemnitz University of Technology in Germany. Birdnet, the product, is an artificial intelligence platform based on an artificial neural network algorithm. At present, the algorithm has been trained to identify 984 common bird species in North America and Europe. Birdnet is free and you can try it out in a web browser or by using a

mobile app. With the app, you record a stretch of audio with a bird singing or calling. Then you manually select a short segment that you want Birdnet to analyze. The segment is sent to remote servers where the AI lives. The app also uses your phone's GPS to share location data with the algorithm. That's important because it helps narrow down the list of possible species that the algorithm needs to search through. The analysis is lightning fast, in my experience. In a few seconds, the AI sends its results to your phone and you're presented with the most likely species. The app may also list a couple other possible species, if there's some uncertainty. uncertainty. There are now a bunch of competing bird song apps, but BirdNet seems to be the

best, by far. My understanding is that Birdnet is being built first and foremost as a very serious tool for bird research, research in the fields of bioacoustics, conservation, and ornithology. Unlike some other apps, Birdnet isn't a for-profit venture, and the app is designed to promote citizen science. The Birdnet team and other researchers are working on AI systems that work passively. Such a system would consist of microphones, recorders, and computers that can continuously monitor and analyze natural soundscapes. So an ornithologist might hike off into a remote jungle in Southeast Asia or Central Africa, then set up one of these passive AI recorders. They leave the contraption there for a week and then return to pick it up.

Hopefully it wasn't trampled by an elephant or torn to pieces by a troop of rambunctious monkeys. The ornithologist goes back to camp, downloads the data to a laptop, and voila. The AI system ended up identifying many of the bird species in the week-long recording. And it tagged any unique songs that it couldn't positively ID. This sort of passive system could be in a single. incredibly powerful and unbiased way of surveying the diversity of bird species in remote or little explored regions. Passive AI recording systems have also been proposed as a way to monitor bird migrations. Remember how we were talking about using radar to classify types of birds?

Well, with acoustic data, it's easier to identify the actual species of migrating birds, at least in theory, because many birds call while they're migrating. Songbirds, for example, migrate at night, and many of them make species-specific nocturnal flight calls. So if you have a passive AI system that continuously listens to the sky during migration periods, it might be able to detect not only the general mass of birds passing over, but also their species composition. Wouldn't that be amazing? there are many challenges to overcome before we have technology like this. But the foundations are being laid and we may see this sort of thing become reality within a decade or so.

Maybe you're thinking, that's all fine, Ivan, but what has artificial intelligence done for me lately? Well, besides the things I listed earlier, Google Maps, Autocorrect, and all that, AI is providing you with some nifty tools to enhance your birding and your understanding of the natural world. Much of what I've talked about so far has been Ivory Tower academic stuff, or applications of AI in bird conservation and all of that. But let's talk a little about some free apps that you can use right now. These all help you identify bird species. Let's say you're out cruising around in nature and you see an unfamiliar bird.

It's singing on a tree branch about 30 feet away. You get a good look in your binoculars, but you still don't know what species it is. You pull your phone out of your pocket, moving slowly so as not to spook the bird. You snap a couple photos before the little bugger flaps off into the bushes. Merlin and I Naturalist are two free apps on your phone that can analyze your photos with AI. Merlin is specific to birds, whereas I Naturalist can handle birds, mammals, bugs, plants, and pretty much any living thing. Each of these apps has been trained with data from huge citizen science databases. Merlin connects to eBird observation data and has been trained with images from the Cornell Labs

Macaulay Library. I-Naturalist has its own vast database. A strength of I-naturalist is that species identifications can be crowdsourced by users. You upload a photo, let the AI give you its best guess, then let other I-naturalist users weigh in as well. Some of those users are bona fide experts that will either verify your conclusion or correct you if you and the AI got it wrong. I love Merlin and I-Naturalist. Check them out if you haven't already. Even if your photos are kind of lousy, the AI algorithms in these apps

will often do an impressive job of at least narrowing down the possible species you saw. Now, let's come back to Birdnet, which we talked about already. The Birdnet app is much newer than either Merlin or I-Naturalist. And the Birdnet team makes it clear

that their app is still in beta testing. It's not yet a fully fleshed out tool. Anyway, remember that bird you saw earlier? It was singing, so you could have used Birdnet to get a few seconds of acoustic data. Assuming the bird is among the 984 species that the AI was trained with, there's a good chance Birdnet will help you ID the bird. Here's a true story.

Just the other day, my wife and I were out working in the yard, gardening, plucking weeds, and all that. She stops and says, Hey, what bird is that? I don't recognize that song. We listen and we wonder if it's a new bird for our yard. We keep a list of birds we've seen or heard on our property and we're up to about 75 species.

I'm pretty sure the birds singing in the bushes is a new world warbler, but I don't immediately recognize the species. I take off my work gloves and bust out my phone. I open up Birdnet and start recording. One thing I really love about Birdnet is that it gives you a real-time display of the spectrogram. A spectrogram of audio data is a visual representation of the frequencies and signal intensity. So you can literally see the shape, the pattern of the bird's song as it's recorded in real time. I wait until I get a few decent seconds of the Warbler's song, then I select a shape,

short segment on the app and tap Analyze. Here is the actual acoustic sample I submitted to BirdNet. That's a pretty low-quality recording, right? Here it is again. Two seconds after I tapped Analyze, Birdnet told me that the bird is almost certainly a McGillivray's warbler, geothelipis Tomi. The result was actually labeled

almost certain. I was a little shocked because we've never documented that bird in our yard before. I recorded the song a few more times, ran more samples through Birdnet, and it kept coming back as McGillivray's warbler. I've heard this warbler singing before, in other places, so it wasn't a completely new sound to me. We've just never heard it in our yard. So, like a good little scientist, I was being skeptical and conservative. I didn't completely trust. the results from Birdnet. Robots can make mistakes too, you know. I didn't have my binoculars, but I stood still and waited patiently for a bit. And then, sure enough, here comes a male McGillivray's warbler hopping towards me down among the tangle of twigs in the shrubbery. Wow,

there he is! He's got a yellow body with a hood of slate gray, a broken white eye ring, and circles his dark eyes. Unmistakably, McGillivray's Warbler. So even with that crappy short recording on my phone, Birdnet was able to quickly and accurately identify the species. Hooray for Birdnet. Hooray for my wife. Hooray for artificial intelligence. And of course, hooray for McGillivray's Warbler. There is a theme I see running through a lot of the research papers I looked at for this episode. Most of these studies are from the last few years, and they present tests of AI algorithms in the early stages of development, early-ish anyway. Like the radar study in Portugal that proposed a way to protect rice crops, these studies often pointed to some near future where the technology has greatly advanced.

The theme is that we are at the cusp, the beginnings of a new era, where AI will be ubiquitous in the future. in the field of ornithology and biology in general. We're already well on our way. I've already talked about slash joked about some future possibilities. What else can birders expect in the near future? Well, for one, get ready for smart binoculars. Imagine that you're out in the field

and you get focused on a bird in your smart binoculars. The image is then transmitted via Bluetooth to your phone. which uses AI to identify the species, then the name of the species pops up in your binoculars while you're still looking at the bird. So this would be a form of so-called augmented reality. It's like having a sci-fi heads-up display, like what Iron Man sees in his helmet. For the low, low price of $2,300 American dollars,

you can already buy yourself a device that does this. Swarovsky Optic makes a monomone, that does this. The product is called the digital guide. The monocular connects to your phone and works with Merlin. I haven't tried it out yet, but it sounds interesting. I say it that way because I'm a flesh and blood guide. I lead birding tours. Among other things, my tour participants rely on me to help them find and identify birds. So where is all this heading? Will this be one of those cases where AI robots put humans out of work? Should I be worried? Hey Google, should I be worried that robots will replace me? I try to be a helpful and fun AI. Birds are so interesting.

I can help you learn about different species like the hummingbird, tell you how fast they can fly, or explain how they build nests. In the future, I might be scrubbing floors in a hotel bathroom somewhere while you're out on a fun tour in Costa Rica with your robot guide. The robot's name is probably something like Birdotron 3,000. You'll be like, Hey, robot, birdotron or whatever. Get over here and show me all the birds in this forest. I want you to identify every single species and list them in alphabetical order. Wait, no, in taxonomic order. Then make me a sandwich and then drive me back to the hotel. I'm sorry, Dave. I'm afraid I can't do that. Can't do what? Drive? Well, you're

you can identify the birds, at least, can't you? Then stop calling me, Dave. My name's not Dave. Yeah, I'm not actually worried that AI is going to replace me as a birding guide. A robot or smart binoculars won't be able to replace my charming demeanor or my crowd-pleasing sense of humor. At least, I hope not. I think I still have some job security. Anyways, another cool device in the works is called the haiku box. It looks like a birdhouse. You set it up in your backyard and it continually

records acoustic data from the birds singing out there. It connects to your home's Wi-Fi network and, I imagine, links up with Birdnet to analyze the data. You could get some amazing information about all the species singing in your yard. The Haiku box is being developed as a collaboration between an engineering company called Loggerhead Instruments and, surprise, surprise, the Cornell Lab of Ornithology. Haiku boxes are being beta tested

and are expected to cost about $500 once they're widely available. So this would be a consumer version of those passive recording devices we talked about earlier. I know that I would totally get one of these haiku boxes. Bring it on. So let's

rewind and think again about our intrepid naturalist from the intro to this episode, the one bushwhacking through the jungle in Borneo. That person may not have had all the gizmos that we have today, but they still could make important discoveries about birds. We owe a lot to those old-school naturalists and ornithologists. Fancy technology isn't always required in the practice of good science. Even today, often the best tool we have is that gelatinous mass of gray matter quivering in our heads. Good old natural intelligence. As artificial intelligence and digital technology become ever more pervasive in birding and ornithology, we might want to, at least occasionally, take a break from it all. Maybe put our phones and smart binoculars

down for a bit. Take a deep breath and just observe the natural world with our unobstructed senses of sight, hearing, smell, and touch. Augmented reality is cool and all, but let's not lose sight of how wonderful actual reality can be. Thanks so much for listening to this episode and for learning about birds and robots with me today. I want to thank all of my patrons who support this podcast. You guys are really helping to make this possible. And here's a special shout out to my newest patrons, Beverly, Terry, Mark, and John. Your support is super appreciated.

Thank you guys. If you are interested in becoming a patron of the science of birds, you'll find a link to my Patreon page in the show notes. Or you can simply hop on over to to patreon.com forward slash science of birds. If you have something you'd like to share with me, maybe some thoughts about the podcast or your own solution for dealing with swarms

of marauding redwing blackbirds, well, just shoot me an email. The address is Ivan at scienceofbirds.com. As always, you can check out the show notes for this episode, which is number 29, on the science of birds website, science of birds.com. This is Ivan Philipson. Thanks again for being here with me today. If you're ever concerned that I'm taking too long to publish the next episode or you're wondering where I am, don't worry because I'll be back.