AI and the secret lives of whales

Summary

Podcast Summary: “AI and the Secret Lives of Whales” – Click Here by Recorded Future News

Introduction to Michelle Fournet’s Pioneering Work In the episode titled “AI and the Secret Lives of Whales,” host Dina Temple-Raston delves into the fascinating intersection of marine biology and artificial intelligence (AI) through the work of marine biologist Michelle Fournet. Michelle has dedicated over fifteen years to decoding the complex language of humpback whales, aiming to bridge the communication gap between humans and these majestic marine mammals.

Early Efforts in Whale Communication Research Michelle Fournet began her journey with relatively rudimentary methods. Initially, she and her team used simple setups involving hydrophones to record whale calls from small Zodiac boats. As Michelle recounts, “[00:02] From Recorded Future News and PRX, this is Click Here… What she heard was a cross between a truck motor turning over and an eerie siren” ([00:36]).

The process was labor-intensive and meticulous. Michelle and her interns would spend countless hours in boats, manually transcribing whale sounds. “We sat in that little boat for 90 hours and recorded data. And of those 90 hours, maybe 70 of it was, was actually usable” ([01:48]).

Evolution of Technology: From Manual Logging to AI Integration Over the years, Michelle’s methodology has evolved dramatically with advancements in AI. “Now I can sink four hydrophones to the bottom of the ocean and I can come back a year later. I can time align those files and I can find those whales” ([03:12]). This technological leap has enabled her to visualize extensive datasets simultaneously, revealing intricate patterns in whale communication that were previously undetectable.

Michelle became convinced that humpback whales engage in conversations. “All whales produce calls, and generally they're short... I feel quite confident that what I'm seeing is one whale that comes into an acoustic arena or comes into a foraging ground, announces its presence. Another whale responds” ([02:14]).

Tackling the Cocktail Party Problem with AI One of the significant challenges in deciphering whale communication is isolating their calls from the myriad of underwater noises, a problem known in AI as the “cocktail party problem.” Michelle explains, “If you have a bunch of people talking simultaneously... you need to unlayer it so you can say, okay, that voice belongs to this person” ([09:36]).

To address this, Michelle collaborates with Daniel Woodrich, a data scientist at NOAA. Together, they developed “Instinct,” an AI-driven program that digitizes, organizes, and identifies underwater sounds, effectively filtering out ambient noise to isolate whale calls. Daniel describes the initial phase: “I was sort of having to design detectors around what I expected to see... How can I come up with features and... different parameters?” ([12:10]).

Advancements Through Machine Learning Once the foundational algorithms were established, Instinct utilized machine learning to refine its accuracy. “AI allows you to explore just a scalpel, like precision, like a new signal perhaps... We really didn't have the capability to do that before” ([13:18]). This collaboration has exponentially increased the efficiency and precision of data analysis, enabling the identification of subtle patterns in whale communications.

Creating Synthetic Whale Calls: Bridging the Communication Gap Building on the ability to decode whale sounds, Michelle and collaborators are now attempting to create synthetic whale calls using AI. The goal is to generate "whoops" that are biologically meaningful and convincing to actual whales. “If we're trying to create a synthetic whale whoop call, we have to do it in such a way that it still feels biologically meaningful and relevant. Otherwise, we're just making little whale robots” ([19:15]).

Leah Buffo, a postdoctoral associate at Cornell University, highlights two primary approaches:

1. Physics-Based Modeling: Replicating the physical mechanisms whales use to produce sounds, ensuring authenticity.
2. Machine Learning Techniques: Training AI models on extensive whale call datasets to generate new, plausible sounds.

However, Michelle underscores the complexities involved: “Imagine how long it took us to get Alexa or Siri to sound like a person instead of like a robot... We're almost there” ([18:25]).

Ethical Considerations and Potential Environmental Impacts The endeavor to create synthetic whale sounds is not without its ethical dilemmas. Introducing artificial sounds into the ocean could inadvertently disrupt whale behaviors or ecological balances. Michelle warns, “What if some whoop is fired off into the deep and it sets off a cascade of unfortunate events?... It could disturb it” ([20:52]).

Historical Context: Whale Songs and Conservation The episode also reflects on the historical impact of understanding whale songs on conservation efforts. In the 1970s, the widespread appreciation of humpback whale songs fostered a global movement to protect these creatures. Michelle reminisces, “Suddenly everyone had Save the Whales bumper stickers... The US banned whale hunting in its waters a short time later” ([22:00]). This underscores the profound influence that recognizing and valuing whale communication can have on human behavior and policy.

The Future: A Dialogue Between Humans and Whales Looking ahead, Michelle envisions a future where humans could engage in meaningful conversations with whales. “Imagine asking a whale about climate change and having it say, I’m dealing with the warm water for now. Why don’t you guys focus on all the garbage and the fishing nets and the ships that keep hitting us?” ([21:45]).

This potential dialogue could revolutionize our understanding of marine ecosystems and reinforce the imperative to protect them. Michelle poignantly expresses her hope, “I would say I’m sorry for all of the things that we have put them through... For changing their ecology, for changing their ocean” ([23:33]).

Conclusion: Bridging Two Worlds Through AI The episode underscores the transformative power of AI in unlocking the secrets of whale communication. By advancing from manual recordings to sophisticated AI-driven analyses and synthetic call generation, Michelle Fournet and her collaborators are paving the way for unprecedented interspecies communication. This not only enhances our scientific understanding but also fosters a deeper emotional and ethical connection with the natural world.

Notable Quotes:

• Michelle Fournet on whale communication: “[...] it’s a little bit like listening to outer space, sort of listening into the void and getting all these signals back with no context for them” ([02:14]).
• Daniel Woodrich on Instinct: “Original invocation of Instinct... neural net... I'm in charge of creating the detector” ([12:28]).
• Michelle on ethical implications: “We have to do it in such a way that it still feels biologically meaningful and relevant. Otherwise, we're just making little whale robots” ([19:15]).

Final Thoughts “AI and the Secret Lives of Whales” is a compelling exploration of how technology can bridge the communication gap between humans and one of the ocean’s most intelligent inhabitants. Through Michelle Fournet’s pioneering work, listeners gain an insightful look into the potential future where humans and whales might not only coexist but engage in meaningful dialogues, fostering a deeper understanding and respect for our shared planet.

Transcript

[Dina Temple Raston]

From Recorded Future News and prx, this is Click Here. Hey, it's Dina with summer here. Some of you might be heading to the beach, looking out at the horizon, listening to the lapping of waves. But just below the surface, something extraordinary is happening. Marine biologist Michelle Fournet has been listening to humpback whales for years. And now, with the help of AI, she's beginning to understand what they might be trying to tell her. It's not science fiction. It's the story of a conversation that is just beginning. Take a listen. When Michelle Fournet first started recording humpback whale calls, her technique was pretty basic. She'd gather a bunch of undergraduate interns, climb into a little Zodiac boat, motor out to where they thought the whales might be, and then throw underwater microphones over the side. Okay, hydrophones it is 14, 13 on August 5th.

[Michelle Fournet]

On August 5th, I'm going to clap three times. We were listening in real time so that we could write down every whale that we heard. And it was a very, very, very simple setup.

[Dina Temple Raston]

They'd sit in a boat, bobbing in the waves for hours, writing out the data they'd gathered longhand on paper. I hear this whale sound. It's coming from the north, they'd write. Or it was followed by this other whale sound. And then they'd write down a time code. And it was pretty tedious.

[Michelle Fournet]

We sat in that little boat for 90 hours and recorded data. And of those 90 hours, maybe 70 of it was, was actually usable.

[Dina Temple Raston]

And this is some of that usable stuff. I hadn't expected a cross between a truck motor turning over and an eerie siren.

[Michelle Fournet]

All whales produce calls, and generally they're short, and they occur in short little sequences, and they occur in exchanges. Sometimes when I'm out in the boat and I can listen and watch simultaneously. I feel quite confident that what I'm seeing is one whale that comes into an acoustic arena or comes into a foraging ground, announces its presence. Another whale responds. The two whales join. They then make these lovely little calls. As they get closer together.

[Dina Temple Raston]

Michelle became convinced that what she was listening to was a conversation. Fast forward 15 years, and now Michelle's ability to listen in on whales has leveled up. Whale.

[Michelle Fournet]

Now I can sink four hydrophones to the bottom of the ocean and I can come get him back a year later. I can time align those files and I can find those whales. Using software that helps me to visualize data over long time periods, I can look at a year of data all.

[Dina Temple Raston]

At once, which has allowed her to see patterns she might not have otherwise noticed. Before and to ponder some remarkable things. Like what are the whales actually saying? Kind of like two people who might meet each other. Hey, I'm over here. Hey, how's it going?

[Michelle Fournet]

It's a little bit like listening to outer space, sort of listening into the void and getting all these signals back with no context for them. And I got really interested in figuring out what that context was.

[Dina Temple Raston]

In other words, Michelle didn't want to continue to passively listen to whales. She was keen to figure out what they might be trying to say. I'm Dina Tap and this is Click Here. A podcast about all things cyber and intelligence. We tell true stories about the people making and breaking our digital world. And today on the show, look at how AI might allow us to not only talk to the animals, but help us stop just long enough to let them talk to us too. Think of the amazing repartee. If I could walk with the animals, talk with the animals, grunt and squeak and squawk with the animals, and they could talk to me. Make a hole. Dr. Dolittle. AI is coming through. Stay with us.

[Daniel Woodrich]

This show is supported by Odoo. When you buy business software from lots of vendors, the costs add up and it gets complicated and confusing. Odoo solves this. It's a single company that sells a suite of enterprise apps that handles everything from accounting to inventory to sales. Odoo is all connected on a single platform in a simple and affordable way. You can save money without missing out on the features you need. Check out odoo@o-o o.com. that's O D O O.com you're listening to.

[Dina Temple Raston]

Click here. I'm Dina Temple Rastin for Michelle Fournet. The thing about humpback whales that got her hooked was the idea that they might have their own language.

[Michelle Fournet]

There are certain natural laws of communication that most vertebrates follow and humans follow them, and birds and, you know, dogs for example. And so because we have a shared evolutionary lineage, we have the ability, at least in, in some form, to listen to something and know is it hurt or is it aggressive? And humpback whales, you can listen to them and you can understand something about how they're interacting with one another.

[Dina Temple Raston]

Michelle was living in Alaska in her early 20s and she picked up a job on a whale watching boat, which is kind of funny because before that she'd never seen a whale or ever been on a boat.

[Michelle Fournet]

One of the things that we would do when I was working on this boat that I loved and our passengers would love, and I think the whales probably also appreciated is we would turn the boats off and drop a hydrophone in the water, and we'd sit out in the sort of the pristine Alaskan ocean in perfect quiet, and we would listen to the whales.

[Dina Temple Raston]

The ocean went from being the sound of motors and bouncing on waves to.

[Michelle Fournet]

Something otherworldly, a very vibrant and living place. And so the ability to listen to the whales and sometimes even just to sort of sit with my toes in the water and my eyes closed and listen to the whales gave me so much insight into how these animals were interacting with each other, because you can hear them countercalling back and forth.

[Dina Temple Raston]

And she came to find out there were all kinds of really intelligent humpback whale behaviors, like this thing called bubble net feeding.

[Michelle Fournet]

It's where you get large groups or sometimes small groups of humpback whales. They blow bubbles around a school of fish, and then they. One of them will begin to produce this coal, and it sounds like. And they'll do this for anywhere from 10 to 45 seconds.

[Dina Temple Raston]

And apparently, one of the humpback's favorite fish, the Pacific herring, hate that sound, so they swim away from it.

[Michelle Fournet]

So the whales use this sound to push the fish towards the surface of the water. And then when the fish come to the surface of the water, they get concentrated there because they get stuck. At which point the wheel will stop calling and the group will all come to the surface of the water, open their mouths, and devour the fish.

[Dina Temple Raston]

It's like an acoustic net.

[Michelle Fournet]

Yes. Yes.

[Dina Temple Raston]

There are lots of amazing things about the humpback whale. For example, they live in oceans all around the world. They have one of the longest migrations of any mammal on the planet. Some whale pods will actually swim three thousands of miles to breed in the tropics and then hightail it to colder places like Alaska, where the feeding is better. They grow up to 60ft long. They can weigh 80,000 pounds. Think fully loaded tractor trailer. And if all goes well, they have a life expectancy that comes close to matching ours 80 to 90 years. To eavesdrop on whale conversations, you have to start by isolating their calls from all the other competing noises under the sea. This is known in AI circles as the cocktail party problem.

[Michelle Fournet]

If you have a bunch of people talking simultaneously, their voices are all overlapping, how do you tease out whose voice belongs to. To who?

[Dina Temple Raston]

That's the cocktail party effect. Humans are really good at focusing on one person and then isolating their voice, and machines have to be trained to do that.

[Michelle Fournet]

If you think of sound as layering that, you need to unlayer it so you can say, okay, that voice belongs to this person.

[Leah Buffo]

This is some party. Who are all these people anyhow?

[Dina Temple Raston]

Who knows?

[Michelle Fournet]

The word gets out, this voice belongs to that person. And if I can get all of these different voices separated out now, I can begin to look for some of those patterns that are going to help me to understand speech or language or communication.

[Dina Temple Raston]

And one of the tools she and other acoustic biologists are using to do that is something called a spectrogram.

[Leah Buffo]

So it's in a waveform. It's like the spectrogram. It uses a mathematical technique called a Fourier transformation, which kind of reprojects the data into two dimensional space.

[Dina Temple Raston]

This is Daniel Woodrich. He's a data scientist and researcher at the national oceanic and Atmospheric Administration, better known as noaa.

[Leah Buffo]

For whatever reason, it's a lot more efficient, I would say, to make a spectrogram that contains like three minutes worth of acoustic data. And analysts can just look at it, tell exactly what's going on there, say, oh, there's that call at 56 seconds. Oh, there's that call at a minute 30. And so the, the rate that brains can process imagery just happens to be quite a bit faster.

[Dina Temple Raston]

So if you're recording the ocean like Michelle does, it would sound something like this. You can use some algorithms on the spectrogram to strip out the ice noise and seals barking and boats rumbling until you isolate what it is you really want. Whales. Daniel helped create a program that essentially took care of the undersea cocktail party problem. He and his team digitized and organized and identified all those underwater sounds so a computer could pick them out automatically. The program is called Instinct, like that.

[Leah Buffo]

Original invocation of Instinct. I was sort of having to design detectors around what I expected to see. You know, I would go in and I would look at a signal and I was, I would say, how can I come up with features and how can I come up with, you know, different parameters?

[Dina Temple Raston]

Parameters or algorithms that would tell the program that this is what a waveform of ice cracking looks like. And this is what a waveform of a seal barking looks like.

[Leah Buffo]

It's got to be a frequency upsweep. It's kind of in this certain frequency range.

[Dina Temple Raston]

And then once the machine got the hang of it and incorporated all of Daniel's understanding into the program, he took a step back and let Instinct's large language bottle just take over.

[Leah Buffo]

I've offloaded a lot of thinking and just said, you know, neural net, kind of do your thing. I'm going to make sure that the statistics look good on the back end, but you're in charge of creating the detector.

[Dina Temple Raston]

An instinct used machine learning or neural networks to keep refining, making lots of complex calculations that Daniel and a room full of graduate students would never be able to do as accurately or as quickly.

[Leah Buffo]

AI allows you to explore just a scalpel, like precision, like a new signal perhaps, maybe a man made signal, maybe a new animal that is in these waters for the first time because of a force like climate change. We really didn't have the capability to do that before.

[Dina Temple Raston]

Which led researchers to think about what else AI might be able to do if it could identify all these sounds. How hard would it be to teach AI to create them too? When we come back, the quest for the perfect computer generated humpback whale. Stay with us.

[Recorded Future]

In cybersecurity, your greatest fear isn't the threats you see, it's, it's the critical signals lost in the noise. Every day, security teams sort through millions of potential threats. That's why Recorded Future exists, to give you precision intelligence tuned to your needs. Our advanced AI detects patterns humans might miss, while our threat intelligence experts, veterans of military and intelligence services, provide crucial context. With Recorded Future, you gain the confidence to identify critical threats and the precision to act before they become attacks. Learn why 1900 customers, including 45 plus sovereign governments, trust us to detect threats faster and achieve 350% plus ROI within a year.

If you're looking for a daily guide to cybersecurity news and policy, sign up for the Cyber Daily from Recorded Future News. It serves up today's most interesting and important cyber stories from our sister publication the Record, and then aggregates all of the big cyber stories you might have missed from news outlets around the world. Just go to the Record Media and click on Cyber Daily to get all you need to know about the world of cybersecurity right in your inbox.

[Dina Temple Raston]

So the humpback whales have a certain call. It's known as a whoop. Well, can you do a whoop?

[Michelle Fournet]

Oh yeah, absolutely. So a humpback whale whip call sounds like wow, yeah, that's, that's, that's, that's what they sound like, only they're all a little bit different.

[Dina Temple Raston]

Do they have like a, a different note or range? Like a, is a bigger male have a deeper whoop and a young one maybe have more like a childlike whoop?

[Michelle Fournet]

That's a great question. We're not quite sure what it is that gives humpback whales a unique sense of voice. And we are just now in the process of really investigating that sense of voice and whether or not this particular call is unique to the individual.

[Dina Temple Raston]

Acoustic ecologists like Michel Fournet, the whale researcher we met at the beginning of the episode, typically study whale behavior with an experiment known as a playback. They use recordings of real whale calls and then play them into the ocean through an underwater speaker. The idea is to link the playback to some specific observable behavior and then tease out what the call means by a whale's reaction to it, which is a little fraud. How do you know you're not swearing at the whale?

[Michelle Fournet]

How do we know we're not swearing at the whale? Yeah. Well, the idea is that we picked a call that was really, really common, that occurs in contexts that don't appear to be agonistic. There are probably the sort of the ecological equivalent of a whale swear word, like a whale fu probably exists. There is probably a sound that a whale is making that means, screw you. I don't think it's this one. And mostly we picked this one because we've observed it in so many contexts that appears to be social.

[Dina Temple Raston]

This kind of speak and spell level of communication, though, only gets you so far. What Michelle and whale researchers like her really wanted was to come up with a meaningful whale sound themselves, to essentially create a synthetic whoop. Michelle is working with an AI group called the Earth's Species Project, and they're trying to write an algorithm that would create a whoop that a whale would find convincing, One that makes the whale think they're actually talking to another whale. And the hope is that would be the first step to toward creating enough words to piece together what a whale might be trying to communicate. But it's a lot harder than it sounds.

[Michelle Fournet]

Imagine how long it took us to. To get Alexa or Siri to sound like a person instead of like a robot. And we're like, we're almost there. You know, Alexa almost sounds like somebody who's hanging out in my living room playing music for me.

[Dina Temple Raston]

We got to that slightly less robotic but still not real Alexa by training AI on millions and millions of examples of human language in various voices. And we actually understand the rules that govern human communication, the syntax, the grammar. And because we're human, we know that even after all that training, Alexa is still a little off. Why did the whale cross the ocean? To get to the other tide. And that's one of the problems that Michelle and the Earth Species Project are solving for.

[Michelle Fournet]

So when we're trying to create a synthetic whale whoop coal we have to do it in such a way that it still feels biologically meaningful and relevant. Otherwise, we're just making little whale robots.

[Dina Temple Raston]

In other words, how do you avoid a mechanical whale sound if you aren't a whale and can't hear the difference? Can you explain to me how you make a whoop? Is it a whoop or a whoop?

[Unknown]

I would say whoop, but not a native English speaker, so that might not be the best.

[Dina Temple Raston]

Leah Buffo is a French postdoctoral associate at Cornell University, and she's studying conservation acoustics. And she says acoustic biologists are trying a couple of different ways to make the perfect whoop. First, you start with physics.

[Unknown]

We know actually a few things about how whales are making sounds. And so we can look into the physics and the acoustic side of it and basically recreate the sound. We can model it so you could.

[Dina Temple Raston]

Build a bottle that takes the whale itself into account, how its vocal cords work, how sound resonates in that big whale head of theirs.

[Unknown]

Another way of going about it, which might be even more fun, would be to use machine learning.

[Dina Temple Raston]

Or someone could build an AI model that would learn from all these whale recordings we have, and it would learn.

[Unknown]

The statistics of the signal, and from that learning, it would be able to recreate a new signal.

[Dina Temple Raston]

The problem is, though, it might sound great to us or to the AI algorithm that created it, but it would be completely lacking if you were actually a whale. Which raises the broader question about all of this. We don't know what. We don't know. What if some whoop is fired off into the deep and it sets off a cascade of unfortunate events? Maybe it would commit some cultural whale faux pas or shift the balance of the ocean environment, because the world under the sea was doing just fine, thank you very much, before we came along. And now all this could disturb it. Michelle says all this AI research is sweeping her backward through rooms she didn't even know existed when she was on those whale watching boats in her 20s. But if all this works, the results could be profound. Imagine asking a whale about climate change and having it say, I'm dealing with the warm water for now. Why don't you guys focus on all the garbage and the fishing nets and the ships that keep hitting us? Imagine what it would mean for the way we view the world and our place in it. Back in the 1970s, early acoustic biologists recorded humpback whale songs, and it turned into a record album that went multi platinum. I remember pulling a small plastic version of the record out of a National Geographic magazine that came in the mail and then playing it at 33 RPMs on a record player. Judy Collins included those very humpback recordings in one of her songs. In hunts to find riches in hunting the whale. And just hearing the humpback songs created this odd kinship. Suddenly everyone had Save the whales bumper stickers. The US banned whale hunting in its waters a short time later. And all this happened because the humpback whale songs beg the question, how could you kill something that sings as beautifully as that? Our ship is well rigged and she's ready to sail. The crew, they are anxious to follow. Michelle says if we could decode what whales or any other animal is saying, it might focus minds more on what we're doing to the planet. Now, if you could say something to a whale, what would you say?

[Michelle Fournet]

That's. I would say I'm sorry for all of the things that. For all that we have put them through for. For the mass removal of whales from the ocean, for the noise that we inundate them with, for our deep misunderstanding for so long of. Of their culture and their intelligence and their social structure, for. For changing their ecology, for changing their ocean.

[Dina Temple Raston]

I think you probably also say I love you.

[Michelle Fournet]

Oh, without a doubt. But I would say that with no expectation that the whale loves me back.

[Dina Temple Raston]

This is Click Here. Click Here is a production of Recorded Future News and prx. I'm Dena Temple Raston. Our producers are Megan Dietre, Sean Powers, Erica Guida, and Zach Hirsch. Click Here is edited by Karen Duffin and fact checked by Darren Ancrum. It contains the original music by Ben Livingston with some other music from Blue Dot sessions. Our staff writer is Lucas Riley and our illustrator is Megan Gough. Martin Peralta and Jesse Niswonger are our sound designers and engineers. Tune in on Friday for Mic Drop, which features our favorite interview of the week. We'll see you then.

[Recorded Future]

If you're looking for a daily guide to cybersecurity news and policy, sign up for the Cyber Daily from Recorded Future News. It serves up today's most interesting and important cyber stories from our sister publication the Record, and then aggregates all of the big cyber stories you might have missed from news outlets around the world. Just go to the Record Media and and click on Cyber Daily to get all you need to know about the world of cybersecurity right in your inbox.