(water splashing) - You are seeing something almost no human has ever witnessed.

A whale's eye view of life in the ocean.

Footage like this is only possible, thanks to cutting-edge technology that's changing how we study these species.

But to truly understand what it's like to be a whale, what you see isn't the important thing.

It's what you hear.

(whale groaning) Whales exist in a world of sound.

Many have wondered if their songs and clicks might even be a form of language.

This video is about how new technology, like artificial intelligence is being used to decode this sonic world.

Scientists are now using whale songs to train AI language models, like ChatGPT for nature.

Their hope is that these tools will help us better understand what it's like to be a whale.

Join me on one of the most fascinating and perspective-changing explorations we've ever researched into a whale's world of sound.

As we meet whale scientists, animal communication experts, and artificial intelligence innovators on a quest to answer one question, "Can we use AI to talk to whales?"

(upbeat music) (whale groaning) Hey smart people, Joe here.

You are listening to a record called, "Songs of the Humpback Whale."

In the 1950s a Navy engineer named Frank Watlington was using top secret underwater microphones to listen for Soviet submarines when he recorded these eerie, alien-like sounds.

(whale groaning) In the 1960s, these recordings made their way to biologist Roger Payne.

When Roger and his wife Katy analyzed the sounds, they found complex, rhythmically repeating collections of notes.

These whale sounds were songs and those songs were in danger of disappearing.

See, in the second half of the 20th century, whales were on the brink of extinction.

Thanks to unrestricted slaughtering for use in soap, lubricants, pet food, margarine.

We killed around three million whales in the 20th century alone.

Some populations shrunk by more than 90%.

When you look at the sheer biomass that humans removed from ocean ecosystems, scientists think this was the biggest annihilation of any animal in human history.

Well, Roger Payne had a crazy idea to try and save whales from extinction.

- You know, how do you describe somebody like Roger Payne?

- [Joe] Iain is the CEO of Ocean Alliance, a whale conservation group Roger Payne founded in the 1970s.

- He was probably the father of modern whale research.

But what's more important than that, I think, is Roger understood scientific papers or science alone, won't save the wild world.

- In 1970, Roger produced an album of these humpback songs, and it changed the world.

"Songs of the Humpback Whale" went multi-platinum.

Those whale songs even made it onto the Voyager golden records that are now flying off somewhere outside our solar system.

This version here?

Was included in an issue of "National Geographic Magazine."

It is the largest record pressing of all time.

Michael Jackson?

Beyonce?

Whitney Houston?

No.

It's whales.

It is honestly hard to imagine today how big a deal this album was.

Not just for people, but for whales too.

"Songs of the Humpback Whale," helped spark a revolution in ocean conservation.

The Save the Whales Movement was born.

In 1972, the United States, passed the Marine Mammal Protection Act.

There were protests and actions against whaling ships.

By the 1980s the International Whaling Commission had banned commercial whaling worldwide.

Fast forward to today, most of the species we used to hunt are recovering.

No joke, this album might be the reason we still have whales today.

- You know, I think it really did change how people thought about wild animals.

Wait a minute.

These aren't cat food and dog food and lubricants.

These are singer-poets.

Musicians filling our oceans with their haunting voices.

- Sadly, Roger Payne passed away in 2023.

But all of this happened, because humans realized whale songs mean something.

In the decades since, scientists have learned whale songs have dialects.

They carry family connections, name-like identifying calls.

Unique songs are composed and passed across whole oceans, like culture.

- We nearly all love music.

Who can't appreciate a great composer, a great singer.

And these are clearly nature's greatest composers.

The songs that whales compose are rich and complex.

But, if a song doesn't have words, how can you begin to understand what it means?

Especially if the composer can't talk back.

(suspenseful music) Well, this is where technology might be able to help.

Some scientists think AI's ability to comb through enormous volumes of data in search of meaning might make it possible to figure out what whales are saying when they sing.

When I first heard about this, I was really confused, honestly.

I mean, like a lot of you, I'm pretty impressed by what AI can do.

But I'm also a little bit skeptical?

Like, it's one thing to ask ChatGPT to help me come up with a title for this video or something.

But how do you ask a thing that's not alive, what a totally different species is trying to communicate?

This is my quest to answer that.

We're about to talk to whale scientists, animal communication experts, AI innovators, to get some answers.

And lemme tell you, what I learned in the end, completely changed how I looked at this entire question.

Okay, so I'm about to interview an animal communication expert, so we can get an idea of how you figure out this question of what information is an animal trying to convey when it makes a sound?

And that question centers around this word umwelt.

(cow mooing) - It's my favorite word.

(Caroline laughing) - [Joe] This is Caroline Casey, she researches animal behavior and communication.

- Oh my gosh, yeah.

So, umwelt refers, well, it was developed by this researcher, named Jakob Von Uexkull.

I'm probably butchering that, but it basically refers to the self-perceived world of an animal and how it interacts with the world around them.

And so if you think about our sensory lenses, they're very biased towards vision and hearing, what we are most attuned to.

But other animals, have completely different sensory lenses, through which they perceive the world around them.

So, sensory systems that they are reliant on, we cannot assume that they're similar to our own.

- Okay.

So, you're saying I have to imagine what it's like to be a whale.

To imagine what it's like to be a whale, first, close your eyes.

Because underwater, a mammal's vision might reach 10, maybe 15 meters.

But sound travels faster and farther underwater than it does in air.

Down here, your primary sense is going to be sound.

I mean, take blue whales.

They aren't just the largest thing that's ever lived.

They make the loudest call of any animal ever.

They can make sounds that can be heard over a thousand kilometers away.

That would be like me having a conversation, across the entire State of Texas.

The umwelt of a whale is about sound, on scales of distance and time that we can't even fathom.

Fathom is an underwater joke.

- [Audience] Boo.

- But just because whales are all about sound, doesn't mean we can't turn it into something visual so we can understand it better.

This is a spectrogram.

Basically, a spectrogram is a way to represent sound on a graph.

And when it comes to whales, knowing what vocalizations look like can teach us a ton.

For instance, this is a blue whale vocalization.

Baleen whales like blues make sound with a larynx, kinda like we do.

Toothed whales, on the other hand or other fin.

(whales squealing) Like orcas, they make sound through their noses, basically these balloon-like lips that can be opened and closed to make sounds like clicks and whistles.

(ballon squealing) I'm a dolphin.

Toothed whales also have this thing called a melon, a chunk of special fat on their forehead, which they can use to aim and focus sound.

Sperm whales string their sounds together in vocalizations called codas, they're almost like Morse code.

And sperm whale codas are incredibly complex.

Probably because they have the biggest brain of any animal.

And that's the thing.

If we're looking for meaning in whale sounds, it's not just about making sure that whales have the right hardware.

They need the right software too.

How do you figure out what makes a whale brain tick?

It's not like we can send whales through an MRI machine.

But scientists have looked at brains from beached whales and learned a lot about they work.

For example, the part of a whale's brain associated with intelligence is bigger than a human's relative to body mass.

Whales also have special neurons in their brains that we only find in animals that are highly social, intelligent, and chatty.

So, that means whales have the right hardware and software to communicate in really complex ways.

And now we're at the point where we can't avoid it any longer.

We've gotta talk about the L word, language.

What would it take for whale sounds to count as language?

This is a kinda controversial question.

Language researchers have this loose set of rules for something to be a language, it has to have individual words with specific meanings.

The ability to create new words when you need them.

And the capacity to describe things that have happened in the past and may happen in the future.

So baby talk?

(baby babbling) Not language.

Klingon?

(speaker speaking in a foreign language) Language.

Language is this really complex thing, so we cant look at one whale song and slap the L word on it.

We've got to look for patterns in tons of songs over a long period of time.

It's sounds like we're gonna need a bigger boat.

And that boat is called AI.

- There are now tools that allow pattern recognition to be much more quantitative.

- Ari Friedlaender is one of the leading whale behavior researchers on Earth.

- The D call of a blue whale looks like this and has this shape.

Once you know sort of what that is, you can look through volumes of data and pull out all of those cases of finding that individual call.

The volume is so high at this point.

If you've got 10,000 or 100 thousands of images, having something that can go through and process and look for that similar pattern in real time almost is going to change, everything about how that type of data is collected.

Well, it turns out AI has already changed the way we study whales.

I'm talking whale tails.

Whale tails, or flukes, are like human fingerprints, no two are alike.

And we can use them to identify and track individual whales.

And luckily, researchers, and whale-watchers on vacation, have taken hundreds of thousands of whale tail photos.

It used to be, if you wanted to identify the whale you just saw, you had to flip through literal binders of pictures and match photos by hand.

This worked, but it took thousands of hours.

So, scientists decided to take a bunch of these fluke photos, feed them into a computer and let it learn.

They created a neural network that studied whale tails over and over, until it could correctly ID them.

Now, AI can ID a whale 90% of the time.

That's better than humans can do.

And it can even ID whales from fuzzy or bad photos.

What this proved is that AI isn't just some gimmick.

It can help people do real science.

- This is an opportunity, if there are people that are interested in the animals that I work on, to help us process the data that we have or facilitate more data collection that'll help us learn about these animals in a new way, I'm all for that.

- I guess I need to talk to somebody who actually works on this AI stuff.

- Hi, I'm Katie Zacarian.

I am a co-founder and CEO of the Earth Species Project.

There's this thing called artificial intelligence that you may have heard of.

And there's been some breakthroughs in the last decade which give rise to a whole new way of approaching this problem.

- The biggest breakthrough came in 2017, when researchers from Google published this paper, called "Attention is all you need."

Before this, computers had trouble understanding sentences, because they had to read them one word after another, like how we normally read.

But this new idea was like having a magic spotlight that shines on all the words in the sentence at the same time, to figure out how they fit together like puzzle pieces.

- For the first time you didn't really need to label the data and it can detect meaning.

- This made computers much faster at understanding language.

And because they can understand language better, they can do all sorts of cool things, like translating languages, summarizing articles and answering questions.

Like, this is what made ChatGPT possible.

- But also what was really cool is when applied to things like audio, sound, music, human speech, it also worked.

Well, why not for animal communication?

There's some things that human beings are really great at.

Like if I were to play you the sound of a bark and play the sound of a meow, I'd be like, "What animal is the bark?"

And you be like.

- It's a dog, okay.

- And the meow is a cat.

We human beings are really good at being able to do that from very few examples.

But what if I were to play you 10 million barks?

How much time with that take you to go through hours and hours and hours, tens of thousands, maybe hundreds of thousands of hours of audio and be able to accurately assess a bark from a meow.

- AI is really good at these kinda things.

Taking scales of data that exceed our tiny insignificant finite human lifetimes and finding meaning in it.

Sometimes meaning that we can't even see or hear.

- There may be things that are just kind of escaping us because of our humanness, our anthropo-ness.

If we could accelerate that with machine learning and AI, that would be incredible.

- Earth Species Project and other groups are out there right now, trying to train computers to comb through immense datasets of animal sounds, turn them into math that computers can understand and look for patterns.

But just because we can find patterns in these sounds isn't the same as knowing what they mean.

These sound patterns sorta don't mean anything, unless we know what the animal is doing.

- It's a really complicated question.

I think far more complicated than most people might think.

My name is Michelle Fournet.

I am a marine acoustic ecologist.

Do I think that AI can help?

Sure.

If it's used the right way.

Do I think that AI is ready?

No, it requires an enormous amount of data.

Thousands, millions of data points in order to paint an accurate picture of what's occurring in nature and we don't have that yet.

- So, what would it take to sorta paint that picture?

- So, step one is what is the whale doing when it produces the sound?

Step two is who is the whale with?

And then step three is how does the whale respond when it hears the sound?

And if you could put all of those things together, then you could begin to ascertain what the what the sound is for.

So, we are just barely scraping the surface of the amount of data that we would need for AI to build a model that would help us to decode something.

When you think about how long it is we've been studying medicine or chemistry or natural science.

We have several hundreds of years of thinking about natural science.

We've only been studying whale song, really since the '60s.

So, it's only been 60, 65 years that anyone's been listening to these vocalizations.

- And that is when it hit me.

All these amazing things that humans are training AI to do?

They're so easy in comparison to decoding whale language.

Because, well, remember that word umwelt?

- It's like the things that we would even say to a whale, you have to remember that is within our umwelt, right?

Even the greeting, "Hi, how are you?"

Asking another animal how they feel internally, that is a very human way of communicating and we don't know and we can't assume that whales communicate in that same way.

- I mean, think about that.

We might share the planet with whales, but they live in a completely different world.

- Whales in general have been on Earth for 150 million years and they've been communicating for 150 million years.

And we have been here for 100,000 years in our current form.

When we think about language, it's arrogant to think that the language of an animal that evolved in the ocean, over millions and millions of years is going to be analogous to our own.

And so until we can come up with a version of AI that is capable of ingesting 150 million years worth of data, I don't think it's going to answer the questions.

- But I just can't let go of this idea.

Let's face it.

What you and I really want to know is, what if AI does get there?

I've seen headlines saying we're gonna have Google Translate for humpbacks.

I've watched YouTube videos saying we're gonna have ChatGPT for whales.

Serious people are genuinely trying to make these things right now.

How do we know that AI won't get there soon?

The power of AI just continues to grow and grow.

Every few months we see it do some new thing that we thought was impossible.

What if WhaleGPT becomes real one day?

- I mean, the desire to learn how to talk to a whale and have it talk back to you, that's very enticing.

I think there are times where the ability to communicate with a whale is really important.

If it is in danger, like I want to go to the ocean and drop a speaker in the water and I want to have a conversation with the whale.

And AI might help me do that.

But does that benefit the whale?

And if it doesn't, why do it?

At the beginning of all this, I wanted to figure out how technology, might let us understand whales and talk back to them.

I mean, that would be so cool.

What I realized along the way is, technology absolutely is changing the way we study and protect these and other species.

And AI will absolutely be a part of that.

I mean it allows scientists to do things they could never do before, study more than they ever could in a lifetime.

But when we think about how far we take that technology, what I've realized is, the question isn't can we talk to whales, it's should we.

Like, so much of this time we spend, trying to figure out nature, we make it about us.

When more of the time, we should be making it about them.

Okay, I've been asking everybody this question.

Let's say, hypothetically, technology and AI does allow us to decode and communicate back to whales.

What would you say?

- If I could talk to a whale, I'd probably say what pisses you off.

- I don't have a need to talk to a whale.

I would prefer to leave it alone.

It's a wild animal.

Let it live in peace.

- Everything that I would think to say is a very human construct.

Like, I'm sorry.

- To be able to understand would be profound.

I don't know about communicate it's a very big question that deserves a lot of thought from many, many people and maybe also animals need to be part of that deliberation.

- As someone who's had the privilege of sitting in a small boat with a hydrophone in the water listening to whales, it's one of the most amazing experiences I've ever had.

I do not necessarily think that that experience is made better by making a sound myself.

My hope is that we can, as a species, that we can learn to be better listeners.

And I think I have great hope that being better listeners, will bring us enormous amount of joy.

- When you think about it, listening is what started all of this.

And maybe being better listeners, whether it's with our ears or AI, is what we need to keep fixing our ocean ecosystems, from the smallest plankton to the largest species that have ever lived.

As for me, what would I say to a whale?

That's obvious, stay curious.

That's why they call it the aquarium of the world.

Delicious.

Whale poop does not smell good.