One of the things about dogs, for anybody who has a dog, is you know that they're very good at hiding that they are in pain.

Ah, right.

So we have had several cases where one of the buttons that get trained is ouch, and dogs being like, belly ouch, and then they puke, or ear ouch, and they have an ear infection.

And so for us, it became like, on one end, we're interested in communication, we're interested in, you know, can they combine words and so on.

But what ended up becoming more and more fascinating is what are they talking about?

Federico Rossano, welcome to Particles of Thought, my friend.

Thank you for having me.

All right, man, we gotta get into this, and that is who let the dogs talk with these buttons?

Yes.

Yeah, so, so when you saw this footage, you were like, I gotta do this and study this, right?

Not quite, not quite.

It's, it's very funny actually because I am not on social media, so I had not seen them.

Me either anymore.

I have not seen it.

So in 2019, some of my colleagues start telling me like, hey, there is this dog from San Diego that seems to be pushing some buttons.

It's been on CNN, is very popular on social media.

The dog is in San Diego.

You're in San Diego.

You should study it.

And I'm like, I send them an email with 4 papers saying, and this is why we should not be studying this.

And the papers were kind of like, Look, there's a long history of this where we tried to train chimps and other animals with sign language, and we tried to teach them symbols, and it didn't quite go very well, and it was not quite super ethical, and there were a lot of issues, and it was a big— not a waste of time, but there were issues.

And so it was like, I do not want to get involved with this.

Now fast forward a few more months.

Now it's 2020.

COVID just started.

I also had my first child that was born in February 2020.

And so I'm holding my baby and I'm clearly fascinated by the communication issue and how do you learn this?

And I get a text from an alumni of my university and he sends me clips from that dog.

He's like, what do you think of this dog?

And I'm like— So wait a minute.

At this time you've only studied humans?

I was studying non-human primates.

Oh, at the time.

I was studying chimpanzees, bonobos.

I was studying communication and I was studying children communication, language acquisition.

So I was always interested in learning and animal communication.

And I'd done some work with dogs, but it wasn't like, oh, I want to try to see how dogs communicate back.

It was more like, how do dogs comprehend human communication?

I see.

Which is how most of the research up to then was.

So I get this message.

I'm like— and it's like, what do you think of this dog?

And I send him the same email.

I was like, look, here's what I think.

I'm prepared for this.

I'm not ready for this.

I'm done.

We should never do this.

And then he starts sending me clips from another dog.

And that dog, uh, his name is Bunny.

The first dog, by the way, her name was Stella and still alive and is a great dog.

And Bunny, at that time I didn't know, was also another social media star.

Yeah.

Um, and I'm like, okay.

And my answer is like, great, now there's 2 dogs that can do this.

Still not interested.

And so he tells me, well, you know, I'm kind of interested in, in potentially selling some buttons, creating a company about this.

I think there's going to be a lot of people that would like to try because what if you could talk to the dogs?

Right.

Mainly, what if humans could have an easier time understanding when the dog needs to go to the bathroom, when they are hungry, when they are scared, when they're upset?

And then he's like, well, you know, there might be a few dozen people that might be interested in participating.

What do you think?

And I'm like, man, it's COVID time.

I can't.

I'm an assistant professor.

So I'm also like, I don't know.

This is too risky.

I start talking to my colleagues and my mentor is like, don't do it.

Have you decided never to get tenured?

Some people who had been involved in the animal language studies were like, don't do it.

This is, this is too, too risky.

You're going to destroy your career.

Wow.

And at the same time, you know, I'm like, start looking into this and I'm like, well, clearly nobody wants to do this.

Therefore, I should do it.

But not because, not because it's almost like, you know, when you see something that is like, if this is true, this would change the way we think about animal minds.

And the idea that only humans have language, right?

It's like, and if, or linguistic abilities, and if it's not true, at least people should know so that they're not, you know, it doesn't become this terrible fad of everybody spending a ton of money for something that doesn't mean anything.

So it sounds like what you're implying here is that this whole process sort of had already taken off without researchers ever verifying that there was a there there.

Absolutely.

So right now, I mean, when I started, there were like for sure, thousands of people already doing it.

I mean, some of these dogs have millions of followers on TikTok.

So you can imagine if there's millions of followers, there's going to be a lot of people being like, I want to try it.

And people are trying to do it.

And of course, we're like, this could be interesting, but could also be nothing.

Right.

And we don't want people to have sort of the illusion that they're doing something, that they are talking to the dog if the dog cannot understand any of this or if it cannot even learn it.

So the basic decision is like, OK, how are we going to do this?

And of course, imagine it's COVID time.

The advantage of COVID time is everybody's home with their pet.

They have a lot of time on their hands.

So you crowdsourced the research?

So we decided to do citizen science.

So you do like kind of community science.

You basically decide, I can't travel anywhere.

People need to start doing the training and give us access to it.

So we do start with surveys.

People need to start logging what they do.

We put cameras in a lot of people's homes.

And then when we can travel, we'll send people testing.

So we had our experimenters with masks going into people's homes.

People let us come in and be like, okay, you can do a study with my dog.

Let's check all our tests that you are allowed to do it.

And then you go and do it.

But the whole thing, and then we need to do training.

It's like, well, we tell you now what to train.

But the idea was like nobody has done this.

Train the dogs.

Train the dogs.

But it's like, nobody had ever done this at this scale.

So what ended up being is something that started basically almost like in a garage, like, you know, we have nothing, we don't know how to do it.

We now have more than 10,000 people that have been tracked for years.

Wow.

It's the largest study on animal communication ever attempted.

It's 10,000 people from 47 countries that are submitting data on, you know, the training, what their dog is doing, and how they're doing.

We have 10,000 dogs and 700 cats.

By the way.

So, and it's just like you now have this community of people who are trying to do it, and we keep getting people signing up.

All right, man, so, you know, the skeptic in me— Of course, yes, I'm a scientist, please go.

That's a lot of variables there, all these people.

And, you know, like if you look at like political polls in America, they don't match the election.

Why?

'Cause people are lying.

They're like, I'm gonna trick these guys.

Yes, yes, yes, yes.

So how do you control these uncertainties in terms— You are hitting the right chord here.

That was the biggest concern.

The first concern, the reason why most people don't like citizen science is you can't trust people.

People will tell you they can do things and it's not true.

So what you need to do is find ways to capture the liars in different ways.

And so one of the ways, for example, for us was, well, we started putting cameras in some people's homes.

And the cameras were running 24/7 so that if you tell us that the dog has pushed all these buttons and we see you pushing the buttons, then we know.

Right.

Yeah.

Or if nobody's touching those buttons.

At some point, we had automated logging.

So right now, most of the dogs, whenever a button is pressed anywhere in the world with the device that we have, we know that like, dog called Johnny pushed a button outside in Paris at 7:15 AM.

Wow.

We have that data coming in.

And so we know that something has been pressed.

And so we can now go check.

Most importantly, once you send the experimenters, you do see that some dogs are really impressive, and some dogs are not that impressive.

Right.

And some of the things we do, which we've done recently, is like, you can take the data of the button presses, and you can try to look at, once you run some random— some analysis of sort of, is it random or not?

It's like, given these buttons, given what they press, Is it going to be random or is it not going to be random?

Particles of Thought is supported by Carlisle Companies, a manufacturer of innovative building envelope systems.

With buildings responsible for over a third of total energy use, Carlisle's energy-efficient solutions are built to reduce strain on the energy grid.

For example, Carlisle's UltraTouch denim insulation, made from sustainable recycled cotton fibers, delivers energy efficiency while being safe to handle and easy to install.

Made with 80% recycled denim, UltraTouch diverts nearly 20 million pounds of textile waste from landfills each year.

Operating across North America, Carlisle is working towards a more sustainable future.

Learn more at carlisle.

com.

The way this system works is, first of all, it was not invented for dogs.

This system has been around for nonverbal humans for a long time.

People who might have a stroke or people who might not be able to speak, you could actually use devices that, they're called augmented communication devices.

And so you can basically use them so that others can understand what you want.

And so in this case, the way they work is you have a button, for example.

You can record your voice.

So you can record a word on it.

In this case, for example, could be something like my name.

Button: Federico.

And whenever you touch it, you hear that word.

Right.

Button: Federico.

And so the dog— and then what would happen is, for example, if you have a button, let's say— by the way, they can be different sizes and so on.

You might have a bigger dog, a smaller dog.

You can push it before you do the thing.

So for example, if we're going outside, you can put the button in front of the door and be like, now we're going outside.

Touch the button outside.

You go outside.

Or like, oh, it's— do you want water?

Water, you get water.

And so you start with this basic associative learning mechanism, which is like you associate the meaning of the use of this device and the word that you hear that maybe you already know as a dog with whatever you can get.

Right.

And what can happen then is that you place it ideally in some configuration so that the animal can also remember that the one that is here is— Ah, so it doesn't change location.

And the one that is here is your name.

So you could have, for example, you know, a space where it's like, this is people.

And now you have like— you can look at this and think, oh, this is where people are.

So it could be my name, your name, the owner, it could be the kids in the house, could be maybe other dogs or somebody else.

And then you can have something else which could be, for example— Button: Hot dog.

Food.

And now you have hot dog.

And you can have water and other things.

And so now you train that whenever you get this, you get hot dog.

And maybe you can train this.

Button: Want.

Want.

And now you can say, want— Button: Want.

Hot dog.

Button: Hot dog.

And now you can say, I'm asking for this.

And I can add now— Button: Hakeem.

Hakeem.

Button: Want hot dog.

And you could think, I'm asking you.

I'd say, you, I'm calling for you.

I want this thing.

So if there's multiple people, you could possibly start doing this.

How do you train want?

Like, do you have— The way you train want, it's fascinating because it's very similar to how children learn language.

So the way children learn language, when they learn want, for example, is they learn it as a construction.

So you learn want associated with one concept.

Let's say want milk.

Right.

And so for the longest time, the child only thinks of want as going together with milk.

So it's want milk and nothing else.

For example, my son would say, I love you so much.

So much was part of the expression because those are the things that go together.

That's right.

And then over time, you come to realize that actually milk can come by itself, and want can come combined with other things.

So maybe it could be like want toy, want outside, want food.

And the moment you start generalizing this, now you realize the want is in and of itself a tool that you can use to kind of— combined with others and get stuff.

And you can look at what are the buttons that get most pushed.

Well, you could imagine there's something like, I can train certain buttons a lot.

And now you end up pushing those buttons a lot.

More often.

More often.

And if that's the case, that's not interesting.

It's like, yeah, no wonder.

But if I train some buttons a lot and the buttons that you push the most are different, now it gets really interesting.

So let me give you an example.

So we've done this study.

We took hundreds of dogs.

We looked at the button presses for several months.

We have now hundreds of thousands of buttons.

We have replicated recently with 5 million button presses.

It's a lot of data to go through.

That's a lot of data.

For a lot of dogs.

And you know when they're pressing some.

And you can look at which buttons they press the most.

And you can see which ones are trained the most.

So you can imagine one of the buttons that is trained a lot is I love you.

The owner wants the dog to say, I love you.

Or another button that you want the dog to be okay with is all done.

Like, it's like, you know, it says finished, stop, no more.

But if you look at which button the dog pressed the most, those are things like outside, food, treat, play, water.

I love you, much further down.

So I can see the plot right now.

I can see the plot of the histogram of most trained dogs.

Exactly.

It's like, and that doesn't quite match, you know.

And it don't match.

Because of course no owner is like, oh, I want you to ask for food more often.

It's like, oh, please tell me, do you want to go outside for the 15th time today?

Well, you know, I personally would be very selfish about it because I heard about these dogs that alert their owners that they're ill some way.

I'd be like, yeah, bro, let me know.

What do you smell?

How am I today?

How am I doing?

Yes.

I mean, dogs are amazing at detecting, for example, if you have cancer or if you have other type of diseases.

And so one of the things that, for example, some owners ended up training as well was like a button that tells you when they need help so they can ask for help, which is— we've done a study now and we can see that some of these dogs are reliably capable of asking for help when they are doing something they can do.

One of the things that we're really fascinated by is them talking about being in pain.

One of the things about dogs for anybody who has a dog is you know that they're very good at hiding that they are in pain.

And, uh, but on the other hand, if you could tell that they are in pain, if they could tell you they're in pain and which body part is in pain, you could take them to the vet much earlier, right?

Right.

So we have had several cases where one of the buttons that get trained is ouch, and dogs being like, belly ouch, and then they puke, or ear ouch, and they have an ear infection.

Or paw ouch, and then you have some— something stuck in their paw that is potentially getting infected.

And so for us it became like, on one end we're interested in communication, we're interested in, you know, can they combine words and so on.

But what ended up becoming more and more fascinating is what are they talking about?

You know, it's not just food and play and so on, but sometimes it's like, I'm scared, I need help.

Yeah.

One of the things they're scared about, one of the combinations that they press a lot is stranger outside, which is like with all the deliveries that we get, you know, at the door, like delivery all the time.

No, my dog is like 100%.

She knows before they even make it into the driveway.

She's letting us know that they're on the way.

There's somebody coming, you know, and like it could be food, could be delivered.

There's anything happening in front of your door, stranger outside.

And of course that can be very distressing, you know.

And at the same time, what if you could find a way to be like, you know, It's okay.

This is like, you know now what's going on.

And if they can tell you they are in pain, yeah, or if they can tell you when— for example, one, I, I remember this fantastic clip of a cat going out and coming back, and it pushes the button, hot, did not want to go out, it was too hot.

And so you're just like, you know what, for you it might be okay, for them, they don't want to be outside.

It's like, how about that?

So, you know, we know dogs communicate.

So for me, the most common thing, I'm a great massager.

Right?

So my dogs always put their head under me.

My current dog, she'll use her head or her paw to let me know, bro, come on, you know?

And I have my 2-phrase word with her that lets her know now ain't the time, beat it.

And she's just like, okay, right?

Okay.

Um, but here's the question.

Your dog owners are from various different, um, countries, different cultures, and, you know, now you've made them a member of your research team.

And I just wonder the extent to which culture kind of inserts itself into these studies?

It's a very important question, right?

It's like, what happens when you have such a community?

And is it the case that anybody who participates is kind of driven to show that they can do something so they are ultimately inflated?

There's a bias towards trying to show you something that is not real, right?

So that's why for us, going to actually test the individual dogs is very important.

Oh, so you follow up.

So we have— experimenters that go into the house and then run the study.

Yeah.

Actually, there's a series of studies that I'm going to go do myself.

I'm going to go myself into people's homes and be like, you don't know what kind of studies I'm going to do.

But if you let me come, you tell me your dog is great at this, and I can look at what the record says they pressed.

I have all the information about your dog.

And you tell me that your dog understands this, this, and that.

I was like, great.

Now I'm going to come, and I'm going to do the test myself.

To give you an example of what the trick is, is like we did a training study where basically we just tried to teach dogs with buttons to communicate about container, let's say white container and black container.

And you would put food inside and whenever you push, let's say white container, you get whatever is inside.

Right.

And if you now, they can learn that.

That's easy.

But we want to see whether they ask about the container based on the content, so whether they want something that they like.

So now you can say, well, maybe if I know that you really like this food and you don't like this food, I'm going to put them in white and black.

And so the good food that you really like, I put it in white.

Now, you should just push white.

Right.

Because that's what you want.

And then we tried to do this.

And you're like, well, out of quite a few dogs that we had, it was like 60 dogs participating.

there were only 7 that reliably could use the buttons to ask for things that they were clearly preferring over the other ones.

The other ones, they would do it, but it wasn't like as reliable as you'd expect.

It's important because when people think of buttons, you know, they have this idea like everybody is equally good or every dog is going to start doing amazing things.

The reality is that there's huge individual differences just like in humans.

Some dogs are very good on this.

Some dogs can learn new things easily, right?

And, and some others don't.

So, you know, if you look at the distribution of how many buttons most of these dogs have, you know, you have a few that have 180 buttons, and then the median is 8.

Oh wow.

So you're like, you know, it's a big difference, right?

There's some of them are like, yeah, we tried with a couple of them and they didn't really go well.

So is there an association with particular breeds?

That's a great question.

So traditionally, Border Collies are supposed to be the better ones with language, mainly because the test that was used with Border Collies was go fetch the object I'm telling you about.

So if I say cup, you go get the cup and bring it back.

Right.

And of course, they are herders, so they go and bring back the sheep.

That's what they do.

That's what they like to do.

Right.

So as a task, they're very good at this.

Yeah.

The other dogs, take a poodle, they don't like that game.

They don't like to play that game.

What we find is that in terms of animals that breed that try to communicate, there is a huge variety.

We have a lot of poodles, a lot of terriers that seem to be good with this.

But there's a lot of mixed breed.

And I think that's the other reminder, which is like, it doesn't need to be a pure breed for it to be a smart dog.

Actually, mixed breed are often— So has anyone done this with wolves since wolves— It's funny.

They've asked me to do it with dogs— so with wolves.

And we haven't— We haven't started the training, but we could.

We could try to do it.

I think, you know, again, one of the questions is how far can you go and what can you actually reveal about their intelligence?

Well, here's another element of it.

What if you took 2 dogs that were great at it and bred them so that they produce puppies?

Can you get them to a point where they're like writing novels?

Writing novels, probably not.

But I can tell you what we know.

One of the famous stories about one of the bonobos that was very good at using one of those kind of touchscreens, his name was Kanzi.

He was not trained directly.

They were training his mom, but he happened to be there watching all the time.

And at some point they realized that this baby bonobo was actually learning through observation by watching his mom getting trained.

Mom never learned, but him and his sister was very good at it.

Oh, interesting So one of the things we're now looking at is a lot of these owners that started training one dog bring in a second dog, a new puppy.

And now what we're interested in is like, is a new puppy sort of seeing the first dog using it, are they learning faster?

Just like a second child might start speaking faster.

Just like Steph Curry.

Exactly.

Just because you have another one that can model it for you.

And then of course you can imagine now, if they have offspring, is there some genetic predisposition?

Is it the training?

Is it they just happen to be— obviously— Well, could you distinguish between them?

Because if there's something in your mind that makes you more susceptible to the training or pick up the training better, right?

So that's the thing is we obviously— Rather than being better at language, you're better at being trained.

Right, exactly.

And one other thing we know is that dogs that are particularly motivated by than to be better at these things.

If you are mainly motivated by food, you can do it, but that's mainly what you want.

But if you enjoy the process, if you like engaging with others, you think it's kind of a way of doing things with others, that actually gets you to potentially engage more with this.

And we know that the dogs that are better at learning names of toys and so on, they are much more motivated by play than by anything else.

So one thing I learned about being an educator, right, of humans, is that, you know, there's no substitution for someone actually being interested in the topic, right?

You know, if they're not interested, it doesn't matter how smart they are.

It's, you know, they're not gonna do it.

And so with dogs, I guess the interest moment is play.

So is there, how do you identify this is a play-interested dog?

Is it, again, is it a breed thing or is it an individual thing?

It's an individual thing.

I see.

So it's really, you know, there's different, different little tests that you could do.

But basically, I don't want to send the message that if your dog is mainly motivated by food, it cannot learn it.

It definitely will learn it.

It mainly will ask you for food and other things.

Right.

But if you are interested in dogs that might kind of keep going and be curious and try stuff— so one of the things that we notice is some dogs learn because I show you 10 times that when I push this, you get food, or when I push this, you go outside.

So that's how you associate the concept.

But other dogs, they start going there, they touch it, and then they see what happens, right?

It's like, and I touch it and I see what happens.

Now those dogs have a different strategy for learning.

It's like, I'm just assuming that this is doing something, and I want to look at what happens around me and figure out what goes with it.

Yeah.

Now those kind of dogs probably are going to learn in a different way and might be even more curious about this device rather than you need to tell me this 100 times before I'm actually gonna get it.

So I wonder if you were to put a little dot of smell on each button, if that would change the outcome since they're so smell-focused.

So some people have done that and that's how they sort of got some of the buttons to be used.

So for example, if you have, you know, something associated with the meaning, so let's say a specific treat or the scent of a person or a scent of a toy or water or something, then it might get them them to sort of use it a little faster.

What is interesting for me is like, what is the smell of help?

What is the smell of ouch?

You know, it's like, is it like, how do you find that scent?

Right, right.

Well, what if you do it in a mix, right?

So there's like the look of it, the sound of it, the smell of it, the taste of it.

Right, right.

You could imagine that it needs to be— Different sensory inputs.

Exactly.

It needs to be multimodal.

Multimodal.

Multimodal, yes.

So given this, what is the most surprising thing that you've been aware that a dog has communicated?

I really am fascinated by how much communication there is about the humans and the other animals living in the household.

Here is why.

When you think about dogs, as I said, what we knew about dog understanding of language, it was always like, do they learn names of objects?

And we think that that's what we care because that's why we care as humans.

Working on language acquisition, I realized that when you actually look at the first words that a child learns, the first 2 are mom and dad.

But after that, it's like hi, bye, more, yes, no, things that allow us to do things together.

Social interaction.

Social interaction.

And mom and dad are the meaningful ones in your house.

They might say dog because there's a dog in the house, right?

And so now you have dog as a word for a child.

Dogs are social animals.

You know, they're pack animals.

They live, you know, they live in the group.

They're not solo animals that live in isolation.

And so, you know, we might think that, like, oh, they probably just care about objects and food and water and going outside and pee.

But the reality is, when you're missing, when you're gone, very often they push the button of, that refers to you, you know.

So like— Button: Federico.

This would be me— Button: Hakeem.

This would be you.

So you are now gone, right?

And I go to the other person in the house and I keep pushing— Button: Hakeem.

Hakeem.

And I look at you.

Yeah.

And you're like, well, clearly I know that this is associated with you, and I'm basically asking you, where is this?

Where is this person?

And when— and we have— we had instances where another pet would die, and then they would keep asking, calling the name of the pet that was gone.

Wow.

And I'm sort of pushing that button and looking at the owner and be like— and again, it's like, we need to do more work on this.

But it is completely reasonable that they represent who they live with, because we know other animals do it.

Wolves howl when somebody is missing from the pack, right?

It's like, you know, baboons have the lost call, where like when they're lost and isolated, They call for others.

They say, I'm here, where are you?

And these vocalizations are common in the animal kingdom.

There's a sense of like, somebody is missing, and we know, and we are calling for you, or you are calling for us.

And so I think that, to me, kind of changed my perspective of what it means to be an animal, is like, there's the social nature of their representation network.

And emotional.

The emotion, they care about you.

I've had instances where they would— comment on what happens to others.

And so there would be some animal being sick or some animal that has some problem.

And one of the buttons that some people have introduced is concerned.

And so the dog would just push the button, like, concerned, and then the name of the animal.

How the heck do you train concerned?

So they're very good at reading your facial expression, right?

So they can see when you're like— frustrated or concerned or like a little like, I don't know what's going on.

Right.

And so you can just like, you can say, let's say, Federico concerned, and you just describe what you are going through.

Right.

Which, by the way, is the same thing that we do with our kids.

So if you go to preschool, you'll see how do they learn happy, sad, angry.

They see faces with little facial expression, and it's like happy face, sad face.

And so they come to you and it's like, Daddy, are you happy?

Are you sad?

Right.

And I'm like, why are you asking me?

He's like, oh, because you're trying to match.

Right.

This is what it looks like to be in that state.

So, you know, it's, it's, I think it's, you know, we often have an idea of how, how we as humans learn this, that it's like this magical, like, emotions, we know what they are.

It's like, no, we're socialized into it.

And then we learn to communicate about it.

And often, you know, this thing works, but we also know that sometimes, you know, what I think means to be happy is very different from what it might be for you to be happy.

Absolutely, yeah, yeah, it's personal.

So let me ask you this question that is related to something you said earlier.

You mentioned how the Border Collie likes to do this particular thing.

So different breeds of dogs are bred for different reasons.

So I, like my little doggy, I'm like, she must have been a hunting dog because she literally goes crazy.

She sees a squirrel, she's actually taking one out.

And so I always, before I let her out, I go out back and warn the squirrels, "hey, get outta here!"

And so there's hunting dogs, there are dogs that are just meant to be carried in a little purse and petted, right?

So do you see differences based on their domestication purposes?

We do.

We have not kind of tried to— We've not quite systematically looked at that specifically for that.

Right, right.

But we definitely noticed that, you know, certain breeds that are supposed to be, you know, the dog that is outside to protect the house does not like to chat with you.

You know, it's just like it's supposed to just watch out.

While, you know, the ones that are— The strong, silent type.

The poodles, they're like the animals that were, you know, kind of bred to be social companion dogs.

You know, the ones who are like, supposed to be in the house, they might be more keen on communicating with you, which is like, you know, as you say, motivation is everything.

So it's like, you need to have a dog that, at the end of the day, A, trusts you, that if I ask you for things, you might give them to me.

B, that cares about you, so that I'm like, I actually pay attention to your communication and how you're feeling, because that might help me bond with you.

Yeah.

And if you care about the human, you trust that the human cares about you, then you might be more likely to try to use it.

I mean, at the end of the day, you don't need a PhD to have a conversation, you know.

Right.

When you let these dogs just put together words, right, do they start making jokes?

Do they start saying naughty things?

Like, how far does this go?

Well, I mean, we haven't studied it systematically, but yes, there are very funny things happening.

I like to say that some of these dogs go through their poop phase, which is kind of what toddlers go through with everything.

You know, you say something and they look at you and it's like, poop.

Well, I will tell you this.

My daughter Chloe, who's 3 turning 4 this summer, she always calls me Poopy Daddy.

Yes.

Yeah, yeah.

Poopy Dad.

Yeah, it's like, hi, Poopy Face.

I'm like, excuse me?

So some of these dogs, Some of these dogs do that.

So they get into constantly talking about that and pushing that button a lot, which is kind of fascinating because again— Do they have an equivalent to laughter?

But go ahead.

So one other thing that I've seen a couple of times, and again, we need to sort of study more systematically, is things that I think would only make sense as a joke from a dog.

And so there was one dog, Bunny, that as— there was a new dog in the house, and Bunny doesn't like this dog.

Mm-hmm.

And so Bunny is like telling the owner something like, you know, I want to— I have something— I want to tell something to this other dog.

So you push the button, the name of the other dog, and then it says, I don't remember the name, but let's say the other dog is called, uh, whatever, Jenny.

And it's like, Jenny, cat.

And then looks at it.

And it's just like, it's like, I mean, clearly, you know what a cat is.

Right.

And calling a dog a cat, you're just like, whoa, is this real mean?

And so this is the kind of stuff that I'm like, Do they have a sense of humor?

Maybe.

Maybe.

So we're trying to look into this.

Wow.

Yes, yes, yes.

So how can you be sure that they're not just random pressing things?

How do you know that it's truly meaningful?

So the one thing we're trying to look at is if you know which buttons they have, the first thing you can look at is do they match the training that they get?

And so you're like, you know, traditionally, if I train this 100 times and that 5 times, then you should be more likely to press the button that has been trained 100 times because that's the one that you got reinforced more, got rewarded more often.

And we don't see that.

So, you know, when you think about the buttons that get trained the most, things like, I love you, which is what humans like to hear and would like their pet to tell them.

But the buttons that actually a dog likes to push the most are things like food, treat, outside, play, water.

Now, fun fact, we look at this into cats as well.

We have 700 cats in the study.

Do you want to know what is one of the top 3 buttons that the cat pushes?

Leave me the heck alone.

No.

Oh, no.

It's just like, no.

And their favorite combination, because they do some combination for a cat, is food now.

Now.

While for dog is play outside.

Again, isn't it amazing?

Doesn't it tell you something about being a dog versus being a cat?

So the first thing is like the kind of use of the buttons from the dogs and from the cats does not seem to be like human-like, seem to be more like what a dog would probably ask.

The owner is not going to want to go play outside all the time.

Right.

It's probably what the dog wants.

Now of course we have cameras in the house for some houses so we can check on this and we do experiments.

Well, the dogs already communicate that.

They bring you the ball.

They bring you the Frisbee.

Right, right, right.

So clearly we know that those are things that they care about.

Then the other thing you can look at is once you have the buttons available, you can kind of see if when they combine them, is the likelihood, for example, of pushing these 2 buttons just higher than me pushing this button and this button.

So it could be that because they're close to each other, I just happen to spam them.

Or spatial.

Or spatially, some spatial relationship is like it's much easier.

And also, if it was completely random, it should be that the likelihood of me pushing this and this and this and that is the same.

It doesn't matter.

And what we find is actually that certain combinations are more likely to happen.

As I say, play outside much more than others.

Or food now for a cat, while others are not.

And so you can kind of see what would be a random distribution versus what these dogs are doing.

Now, a clear point to make is some dogs are random.

So some of them are not really learning much of it.

They're using it, but they don't really know what they're doing.

Yeah.

But we can tell from the data, having hundreds of thousands, actually now millions of button presses, that some of these dogs are definitely not random.

Now, not random does not mean interesting, because not random could be you learn a sequence, right?

You learn that 1, 2, 3 gives me the Kit Kat, and, and 3, 4, 5 gives me chips, right?

And it's like— You don't care what the meaning is.

Exactly.

It doesn't matter.

I have no— like, it's just like, this is what it— what leads to an effect.

So the next thing is to sort of try to see, you know, do we think that they understand the meaning of each button, and can we see them sort of using it without cueing.

And so one of the studies that we've done was basically dealing with a phenomenon that is called the Clever Hans effect.

Clever Hans was this very famous horse from the beginning of the 20th century that people claimed was able to do mathematical operations.

Right, yeah.

Kind of additions and subtractions.

And what they really realized was that what Hans was very good at was reading cues from the audience.

And so it would be something like if you tell Hans 2+2, Hans would do 1, 2, 3, would look at your face, 4, you smile, and so it would just stop because he realized, oh, you got it.

I look at your facial expression to know when to stop.

And so Hans knew nothing about numbers and math, but Hans was very good at reading humans.

Right.

And in the same way, one of the fears is that maybe all that is happening is that these dogs are just learning cues from humans.

So the example is when you— you might think that your dog understands the word outside, but really what they pay attention to is when you put your shoes on.

And when you put your shoes on, you're going outside.

Right.

So they have no understanding of the words, no understanding of what you're trying to do, but they associate shoes with getting out.

On the other hand, it could be that they actually understand the meaning of the word independently of how you're dressed and what you're doing and so on.

So what we wanted to do was, if you try to tell, for example, a dog outside, what does your dog do?

Does the dog run to the door or does the dog just look around and doesn't know?

My dog runs to the door.

Exactly.

You say play, would he look for a toy and bring it back to you?

And of course they do.

Then the next thing was, what if you now have an experimenter coming into your house and there's a button that say outside and a button that say play and there's a bunch of other buttons.

What you do is the experimenter does not know which button is which because we put stickers on it so that you have no idea which button you're playing.

The experimenter has headphones on and has glasses on so that you can't look at the face of the experimenter.

The experimenter doesn't know which button they're pressing, but they have been told that they need to— imagine there's a red sticker here and a green sticker here, and the experimenter has been told, press the red sticker, the button with the red sticker, and then stand still.

So you push it and imagine that that button says outside.

Now you see what the dog does.

The dog has never seen you before.

Right.

You're not the owner.

You're not giving any cue because you can't hear it because you don't know what you just pressed.

Does the dog run to the door or not?

And the dog does exactly the same thing that they were doing when the owner was saying that.

So what you find is that at the very least, they seem to react to hearing the word produced by the button the same way that they would react— push buttons.

What if you change context?

What if there's a second dog or another member of the household and you say, is Amy outside?

Right?

So you ask it, is a person outside who is outside, and you ask it, is Amy— or is John outside?

And John is actually inside.

And they just have yes/no buttons.

That's exactly the study we want to do next.

So we want to see— because some of these dogs have yes or no buttons.

And so you can see, it's like, when I describe things, is this the case or is this not the case?

And now you can see, can you actually understand whether this is true or false?

And if you can do this, sort of understand truth conditions— Right.

It's basically what Wittgenstein was talking about.

It's what philosophers have been talking about for hundreds of years.

It's like this is the level of understanding of reality that humans have.

It's like is this the case or this is not the case and I can have a way to communicate about states of the world to others.

So what if you have a button for lick and a button for paw or front paw?

Yeah, you should, that is exactly what we want to do because the next question is do they have what we call compositionality?

So the idea that you can combine words to describe, for example, actions on an object.

So lick paw could be like— Versus smell paw.

Run outside versus walk outside.

It's like a different kind of actions.

So could you sort of elicit that kind of response by describing what you want them to do?

So do you see that now they have— From words they know in other contexts.

Exactly.

Once they know the context, they can actually distinguish those things.

Are we actually going to get to a point where we can literally converse with our pets?

That is a very open question.

I don't really know.

What I know is the kind of challenges that I've been faced with and the kind of question that I like to ask, which is like, if some people tell me, why don't you just stop?

Hey, you've already done 2 or 3 studies.

You found that they can do some of this.

Isn't that enough?

I always like to say, imagine you are trying to learn a human-used language and you have one child exposed to something for 1 year or 2 years and then you're like, we got everything we needed.

It's like, well, you never know that when they are 20, they can do things that at 2 they could not do.

And the other thing that people forget is like a human child hears around 20,000 words in one day.

Dogs, our dogs, see a human pressing 20,000 buttons in more than 1 year.

The amount of training they get is so much less than a human child gets just in 1 day, just of hearing what those things are.

So what if you did more training?

Right.

What if you could actually do a more targeted training?

What if you could— there's many different ways in which you could think, is there a way to enhance the learning so that now you could actually expand it.

Or do it with an octopus.

Or do it with other species.

So for example, there are right now, uh, you know, people who are training pigs, horses, goats, a lot of domesticated animals that have been trying to get trained with this animal, with these buttons.

Um, the other thing is, is like, if they could communicate about pain, if they could communicate about how they are experiencing what they are, you can only imagine the implication for animal welfare, right?

It's like, if I can tell you when I'm sick, then maybe you can take care of me.

If I can tell you when I'm scared, when things happening around me are particularly stressful.

I like to think, when you bring a shelter dog home, one of the big issues— Oh, boy.

Yeah.

Like, they destroy the house.

They pee everywhere.

They rip things.

What if they could tell you, I pee.

Like, I need to go out.

Or like, I need— I'm hungry.

Or I am scared.

And you had a way to just react to that and just be like, Oh, let me help you with this.

It's like now, if this could be done before it comes into your home, now you're giving them a little way to communicate to let you know just the basics of like, you want a little scratch, you want a little cuddle, you want to go outside, you want food or anything.

So I see this as like there's the scientific side of like what does it tell us about their mind?

Right.

And I think that there's a lot of work that we could do to kind of see how far this can go.

I don't think this is the end of how we're going to learn about this.

Obviously, we want to study their spontaneous communication and many other things.

But what these tools could help us see is that there's a mind behind these animals that is more complex than we thought.

Absolutely.

And ultimately, what we know is whenever we have given— whenever we have acknowledged intelligence, that has changed the welfare of those species.

You know, once, once the octopus was shown to be able to know much more than we thought, Yeah.

now there are things that we don't want to do to octopuses that we were okay with before.

Right.

And when you can show that the lobster experiences pain, maybe you do things differently, and this, and so on and so forth.

So it's like, you know, the more we know about their intelligence and their awareness of what's happening to them, the more we can sort of tailor how we interact with them and improve sort of our way of being with them.

And to be honest, again, it's like a lot of people during COVID were like, oh, I'm alone, I'm lonely.

And I'm like, you're not alone.

You have your dog in the house with you.

There's 2 of you.

Now, it might not be your best friend.

It might not be like your roommate.

But to be honest, it's like very often— It's better than a roommate.

It's better, yeah.

It can be better than your roommate.

There's real affection and there's really a real care and real connection that you could try to sort of understand and sort of foster.

Right, right.

Man, this is so interesting.

So there's this $10 million prize out there for anyone who can actually communicate with animals, which is, I have to see the details because I feel like we already are doing that, right?

Before this research, we were definitely communicating with our dogs, right?

We are definitely communicating with our dogs.

Interspecies communication is happening every day.

Just think about working dogs, right?

Think about when you're doing detection, you're trying to find somebody who is in the rubble after an earthquake.

You tell your dog, go find it, and the dog will tell you it's here.

So you're clearly communicating.

We have hunting dogs that will tell you.

So we clearly have been doing this as a species for a long time.

I think what is the question is sort of like, is it always just a minimum domain that we can only say something is here?

Right, yeah.

Or can we communicate about more than that?

Right, right.

And also, again, it's like, you know, I think we tend to underestimate the idea of two-way communication.

So I think a lot of, What we can do is trying to translate what's going on, right?

So you can be like, I know you're upset, kind of like what we're doing with babies, right?

It's like, I know you're upset, you must be hungry, this thing is what's going on.

But can we do it back and forth?

So here's another way I look at this.

When we look at animal intelligence generally, right, and we compare human intelligence, there are certain places where we feel like we're superior, and language is one of them.

So I wonder if, you know, it really goes back to something super fundamental, because even bacteria can sense and respond to their environment, right?

So do they, you know, like, how far off are we in our human-centric, you know, thinking?

You know, is it— they're not doing what we're doing, that's obvious.

Yes.

But it doesn't seem to be that clean break.

It's not a clean break.

And I think even worse than this, I think, is we are constantly overestimating our abilities and underestimating their abilities.

I mean, just think about, you know, we discovered only in the '30s about echolocation, sonar abilities, right?

Oh, really?

That was the 1930s?

Yeah.

We didn't know that they could do it.

And now, of course, we use sonars.

Holy cow.

The bees are communicating through a little dance, a little— A little dance, yeah.

We learned it in 1927.

That was when they talked about it for the first time.

It's like a lot of these things we really did not know until somebody took the time to really document what was possible.

And we often talk about these things as like, oh, maybe, you know, think about we have airplanes because somebody spent time watching birds, right?

It's like, and so just think about a dog.

A dog can hear way better than we can.

So they can hear a mouse vocalization.

They can hear your heart beating.

They can hear things that you cannot hear.

What?

They can smell.

They have 300 million receptors for smell.

Humans have 6 million.

Just imagine how much better they are at smelling things in the environment.

And with vision, yes, they don't see red and green fine, but they have better detection of motion, movement, because they have more rods, so they can actually detect movement better.

They have better vision in darkness compared to us.

So we look at them as like, oh, you know, it's only smell.

It's like, no, they're actually better in many other modalities.

And once you know this, well, why would you think that they're lesser than us when they can see better, they can hear better, they can smell more?

It's like, so what else?

And it's like, I think once you appreciate this, you might ask yourself, what is intelligence about?

And intelligence ultimately is our ability to adapt to the environment and survive.

The first rule of evolution— Or problem solving?

Yeah, problem solving.

The first rule of evolution is survive, which means you need to figure out how to get your food, how to protect yourself from predators, and how to get some shelter so that you don't die.

And that's the first rule.

And it's like, clearly other animals have their way of living in the world that allows them to survive, and they're perfectly adapted to the conditions they're in.

We have found a way to colonize the planet.

Yeah.

Fun fact, the other group of animals that has done the same— ants.

And they're very good at cooperation.

So what is common about it is they're very social.

Yeah.

So, you know, sociality has allowed us to kind of overcome the challenges of dealing with an ever-changing environment.

Because once you get out of Africa and you are exposed to a new environment, you're like, what's this?

You know, how are we dealing with this?

Yeah, so now this brings us to the current research topic of the moment, which is AI.

So, you know, with AI, you know, when you design this kind of study, you typically have a training dataset where you have your observables and the truth that they are related to, this known truth.

So can we use AI and AI models to decipher?

'Cause I would imagine that, you know, if there's an animal that makes complex vocalizations like a whale, right?

Right.

Then you need to see it in different contexts and capture the vocalizations and somehow understand the pattern.

So how do you get to that truth dataset?

And if you can, where is AI, how is AI gonna play a role in uncovering animal communication more deeply?

So this is a great question because there's a general belief that AI will solve this problem.

And I think— Among the researchers?

Among, I would say among the people who are investing into this.

And I think part of the idea is because we've seen a success with large language models.

So it's like nobody believed that you could just, by giving text, that you would be able to have some technology that would allow you to answer questions and provide information that seems structured and has some meaning and is not completely random.

And one of the ideas was, oh, they have no world knowledge, so it's impossible.

Now, what happens is that they are getting knowledge from how information is organized.

So from syntax and context that you get around it, right?

So it's kind of like when I say the cat chased— the dog chased the cat.

Now you're learning something about relationship and agents and so on.

And now you can put some other animal in front of chase and then you can learn more about like, oh, this dog and lion and whatever might have similar properties because they can be used in combination with Chase.

And so now there's something that they can do that make them similar to some respect.

And so what happens is that when you think about sentences, you learn a lot about the world by, by just looking at their position in a sentence using grammar, and also by looking at the context in which they occur.

Now the problem is that animal communication doesn't come in sentences, right?

You might have like a scream for like eagle, like you might have maybe 2 vocalizations combined, which would be kind of the example of water outside or want food.

It's okay.

It's helpful.

But you can't get very far with that.

It just is very minimal.

And so then you need to get context.

You need to get what this means in another way.

And the only way to do it is to look at when the signal is produced.

And what happened next?

In what context did it happen?

And what was the response of the other individuals?

Right now, we're not quite there yet with the data that we have.

A lot of the data that people are using actually doesn't even know who's producing the vocalization.

They just know somebody talked, somebody vocalized.

So what would be ideal?

You'd have a group of animals in a zoo, and you keep video cameras on them at all times?

So that is one of the ways in which some people have been trying to tackle this.

So we have been doing some of this, for example, with computer vision.

We have a model that we call AlphaChimp where we actually tracked chimpanzees in a zoo for many months.

And then we tried to look at who is interacting with whom, when they produce a signal, how does the other one react, so that you can start to label those activities.

Mm-hmm.

And like, this is the kind of action they're doing.

This is the kind of signal they're trying to produce.

And you try to see, OK, it looks like that's how the other ones interpret it.

Because it's not enough to say, that's what I think they're doing.

It needs to be, that's what the other chimp thinks they're doing.

That's right.

Otherwise, it's like, I can imagine anything and that doesn't mean anything.

So what you need is ground truth.

So one issue is you need to know the context and you need to know what they're trying to do.

For that, you need to look at the response.

And to see the response, you need to have a tool like transformer models that allow you to sort of track sequences.

You can't interpret things just with a frame, with one shot.

You need to know what happens next and what happened before that.

So you need to have some way of sort of connecting this.

But you need to know who is vocalizing, who is producing the signal, who is it addressed to, and what is the response.

So what would you say is the take-home story on the study of dogs so far?

Like, how far have you gotten?

What is the take-home?

I think that the main take-home is, until a few years ago, people were quite convinced that dogs had the intelligence of an infant, maybe a 9-month-old or maybe a 12-month-old.

And I think more and more work is showing that their communicative abilities and the kind of things that they seem to be able to represent and imagine are more similar to what a toddler could do, more like a 2-and-a-half-year-old, maybe even a 3-year-old.

Which might not seem like a big difference, But if you have a child, what a 9-month-old does and what a 3-year-old does is very different.

Very different.

And again, the next question is like, well, some people say, yeah, but that's not what humans do.

That's not language, or that's not what grown-up adults do.

And I'm like, OK, but do you think that your 3-year-old is speaking?

And do you think that your 3-year-old has a mind?

And if you think so, then why would you think the dog doesn't?

Or why would you think it's OK to treat the dog that way when you would never do it to a 3-year-old?

Right.

So part of understanding where they stand in terms of their cognitive abilities is helping us situate what we think is acceptable in terms of how we interact with them.

Yeah.

Not to mention understanding how we can better care for them in terms of their needs and wants.

And I think ultimately one of the big things is that if you know what the dog wants and what the dog understands, Then, you know, you can provide better for them and they will be nicer to you.

And that's the job of science, right?

It's like collecting the data and showing them.

And method is ultimately all it takes.

So you can see a future where they're like, yes, back in the 21st century, they didn't know how to communicate with the animals, so here's how they treated them.

But nowadays we have our— every animal has their own smartphone.

Well, it'll be something different, right?

Can I give you one example of something I've seen seen recently.

So I'm working with working dogs as well.

So the dogs that do detection— narcotics, explosives, and so on.

So I did this event just a few days ago.

And the way you do a search, there's like 3 cars.

You hide, let's say, the narcotics in 2 vehicles.

And many dozens of participants.

The agent takes the dog and starts the search left to right because that's what we do, left to right, 1, 2, 3.

One of them talks to the dog.

When the dog is waiting to do the search, it's already smelling stuff.

It's already doing the search.

Oh, really?

So while they're waiting to start, the dog is already sensing the world.

It's not waiting for you to start the search.

And so the agent talks to the dog.

And it's like, OK, where is it?

Of course, that's what they're trained to do is find the stuff and lets it go.

Instead of guiding them left, the dog goes straight to the right, 7 seconds, find the narcotics.

7 seconds.

The other one would go around potentially for minutes without finding anything.

Interesting.

And so the difference is most people think that you need to guide the others, and then you know better, and you know how to do it.

And then they'll show you.

Others, if you find a way to communicate, understand that the dog smells things better than you do.

So if you actually trust the dog, the dog might tell you.

And it might be better for everybody if the dog finds it in 7 seconds rather than in 2 minutes, right?

Right.

And so that, to me, is the attitude shift that I think we need to get at, is the idea that as of now, we have this idea that we are completely superior to any other creature on Earth.

And that they can only— we can document, we can teach, we can, we can look at them, but they have nothing to teach us, nothing to show us.

And sometimes if you let them go first, maybe you can actually see and discover things that you didn't know were possible.

All right.

So we've talked a lot about our pets, our dogs and our cats.

Let's get a little closer to our species and talk about some other primates.

That's where you started, right, with your research?

That's right.

I started working with chimpanzees and bonobos and other great apes, and I got to work with dogs afterwards.

Yeah.

But that was my first love and still is in many ways.

A lot of people, of course, think of domesticated animals like dogs and cats and horses.

It's like, yeah, of course we domesticated them, they care about humans.

But if you're interested in language evolution, you want to know what about our closest living relatives?

What about chimpanzees?

Or what about bonobos?

So what about other animals that have, you know, they're so closely related to us that genetically they are so similar, and yet, you know, their communicative abilities are not quite identical to us, but they seem to be remarkably smart.

So the work has kind of developed in 2 different directions.

On one end, they've tried to document their cognitive abilities.

Right.

And one of the things that happened about 20 years ago was that when they tried to run studies comparing their physical cognition versus their social cognition and compared them to humans, they realized in terms of physical cognition, they're pretty smart.

They're very good.

And I'll show you and tell you an example.

But in terms of social cognition, a human child is already better than they are.

Oh, interesting.

Let me give you an example of physical cognition.

So one of the studies, one of my favorites, is called the floating peanut study.

Okay.

Imagine you have a long tube here that is empty, and you drop a peanut at the bottom of it.

You have no stick, you have no tool.

How do you get the peanut out?

Every single great ape knows that if you spit water into it, the peanut will float, and then they can take it out.

Right.

And if you don't have water, they can pee in it, and the peanut will come out, and then they can take it out.

So they can all do it.

They understand floating.

Now they've tried to do the same thing with kids, and they've tried to put kids in front of the same situation with a peanut at the bottom, and they even put water, a bucket of water underneath the thing.

Do you know at what age 50% of kids could figure out that they should pour water into it to get the peanut up?

15?

50%.

So, so what age do you have that like at least half of the kids can figure this out?

Yeah.

8.

8.

Before 8, they have no idea how to do it.

You just look at these things and you're like, they know about floating.

They know that you can use water as a tool.

They understand it.

We can understand it once you think about it, but it's like, but that's sort of like part of their biological— So I wonder how biased this study is, man, because, you know, growing up in the country versus growing up in the city, you are made to solve problems more often.

Right?

There's less reliance on others and experts, so your mind is already in that problem-solving mode.

So do we know that they chose the right group of kids?

We don't, and that's why we definitely need to replicate this in other locations, because of course the possible issue is that those kids could only do it by age 8 because those were urban kids.

So maybe rural kids would be much better at this.

But the message is like it's more about these animals are not stupid.

That's right, they are not stupid.

You know, they're capable in terms of like understanding.

You know, one other thing that I love is, is the idea that like imagine you see this, this cup and you see some shade here, uh, sorry, some shadow.

Yeah.

They know that if they see a shadow that doesn't match this cup, there must be food over here, there must be something hidden behind, and so they will come and check what is behind this.

And if you have a different shadow, they will not look for it.

So they can look at the shadow to figure out something must be behind it that I cannot see.

Now again—That is pretty advanced.

To me, it's like it kind of shows that they have somehow the ability to process information in the environment and figure out this fits or this doesn't fit, or how do I solve this.

Yeah, yeah.

But social cognition, that is the interesting thing.

So what we are amazing at is figure out how we can teach others and learn from others, how we can communicate with others, and how we can represent others' minds.

So what we call theory of other minds.

Yeah.

So I can think, what are you thinking now?

I should say this so that you will think that I have done this.

So once you get to that, that's where sort of things become a little different.

And so one of the general claims is been for a long time that a child approaches humans as a cooperator.

Like, what your experience as a child is, everybody's trying to help me.

When I have a problem, they were trying to help me.

And actually, growing up is learning that not everybody's going to help you, but most people are going to help you.

I learned that really young.

Yeah, exactly.

You go to preschool, you're like, okay, wait a second.

No, I was home with my family.

Oh, with your family.

Oh, wow.

I like to think of preschool as like prison yard.

It's like you realize, oh my gosh, it's like all these 4-year-olds do not like me so much.

Why are you not like mom?

I want this.

Why can't I take it?

And so you start fighting.

But the general idea is when we are looking at each other, we kind of trust each other.

And we assume that the other one, when they are communicating, they're communicating to our benefit.

They're trying to help me.

Yeah.

And one claim is that when you look at non-human primates, most of their communication are more in a competitive mode when I'm like, I either force you to do things like give me this, or, or I'm trying to trick you into things so that like now you're gonna give me this even though you didn't quite realize that that's what I was trying to go for.

So if you think of others as competitors, now you're not really constantly trying to communicate.

Indeed, they engage in much less back and forth than we do.

Like what we're doing right now would be inconceivable as a chimp.

They're not gonna hang out talking about what they think about the, you know, science or the weather or what are the dog— what do dogs do?

They have no, like, exchange about this.

It's more like, give me this, food there, danger over here, I'm hot, you know?

Right, yeah.

And so those kind of interactions become more, like, aimed at obtaining something right away.

People talk about it as being more manipulative, rather than cooperative.

And one example in which you see this is the fact that every single child around age 1 in the world understands pointing, right?

So if you point to this, they know that you're looking at that.

They might even start doing it around that age.

And chimps don't point to each other.

Bonobos don't point to each other.

They just don't do it.

Wow.

Not only that, the claim up to now was that they also don't quite understand it.

They can understand They're like, you know, if I point to this, then maybe, you know, you're gonna try to do something.

But most importantly is like, if I point to this, you're gonna look at my— and you're a chimp, you're gonna look at my finger and you're like, what is this?

Right.

And we thought for a long time it's like, oh, it must be because they don't understand that you're trying to help.

So they tried to do the same thing with children, with 2-year-olds.

And now you have— imagine you have 3 cups.

Yeah.

Okay, and I show you the— sorry, there's 3 cups.

You're a 2-year-old and I'm an adult, and I point to this cup.

And then the question is, which cup are you gonna go for if you think that I'm trying to help you?

So you open this.

Now I'm a 2-year-old, you're a 2-year-old, I point to this cup.

Now you're completely at chance.

A 2-year-old does not trust a 2-year-old to know anything.

They don't assume that you're trying to help them.

They just assume you either know nothing or you are competing with me.

So the same 2-year-old that would trust an adult would distrust a 2-year-old.

Correct.

Got it.

And so the idea now is like probably the issue is not that they don't understand pointing.

The issue is what do they think of the one who's doing the pointing?

Like, are you trying— are you the kind of person who's trying to help me?

Or are you the kind of person who either knows nothing or you probably are messing with me?

Right.

So I shouldn't trust your communications.

And this is one of the interesting things about how we can get to figure out how animals represent others' minds is like maybe you can do it if I realize, for example, that if you give me— if you're a chimp and I am a human and I give you a signal, I've established that I try to help you find food, and I now produce a vocalization that says food and I point to it, now you can do it.

But if I have not established this before, you're like, why would I be helping you finding anything?

Usually in nature, I— you don't have humans helping you find stuff, right?

So now you don't trust me.

Yeah.

So now what you're getting at is, it's not that they don't have the cognitive ability to process the signal.

The pointing.

It's that what makes it work is whether you trust the other.

And what do you think is the other one trying to— the motivation.

Why is the other one pointing for me?

Are they trying to help me, or are they trying to mess with me?

And if you assume cooperation, now a lot of communication becomes possible.

And if you assume competition, then it's much harder.

And so what this reminds me of is how impactful our early experiences are on how we interpret life for the rest of our lives.

So is it the case that if you're If you intervene with the chimp or primate early in life with this communication process, that you get a different result than a later intervention?

Yes.

So the evidence is that animal chimpanzees, for example, that were raised by humans, that were kind of enculturated, which means they spent more time with humans, maybe humans fed them when they were babies and so on.

Tend to be much better at understanding human communication than ones who were not human-raised.

I've done a study, for example, about— I see.

Vocalizations.

And so one of the things that children can do is they can tell— 1-year-old children can tell which direction you're looking at just by listening to your voice.

And so if you close your eyes and you have— let's say I'm looking this way and I say, hey, come and see.

Or I'm looking this way, hey, come and see, you can tell that I'm looking this way or I'm looking that way.

Wow.

And so we've done— so it's not just where the voice is coming from, right, but where my head is oriented to and what I'm looking at.

So if I'm saying, you know, hey, look this way, and there's a box here and a box here, the child, a 1-year-old, would crawl to that box if I'm looking that way just by hearing my voice.

Wow.

You now do this with the chimps.

Chimps could not do it.

But one could do it.

The one chimp, he actually was raised by humans the first 2 years of his life.

Oh, wow.

Same study we now do with dogs.

Most dogs are great at it.

Some dogs are bad at it.

The dogs that were bad at it were not living with humans.

They were living kind of by themselves and were just, like, kind of given food.

The ones who were living with the humans, they could all do it.

So now you have a different relationship with the humans, you trust the humans, the humans is a cooperator, is a provider.

Now you pay attention.

And now you actually process that communication as being helpful rather than competitive with you.

So to me, it's just like, it's not just about the signals.

It's about who produced the signals and why we do it to each other, which is why AI needs to care who is doing those signals to whom.

And it's also your individual history.

Yes, exactly.

Who are we for each other?

Who's signaling to whom?

What is our relationship?

Without that information, we don't know why is this signal working here and not working there, right?

If you could decode animal communication, what is that going to do for us?

So for example, one of the things that comes up in my topic of study, astrophysics, is what happens when we find life somewhere else, right?

And what I often say is that, like, look, if you only have one example, you don't know what is specific, you don't know what's general, you don't, you know, you don't know the, the full range of possibility.

So once you're able to decode this animal communication, I, I imagine that leads directly to animal cognition and their brain, which goes— which tells you something about brains and intelligence generally, because a bird brain is not a human brain, right?

Absolutely.

In their construction.

But if they're able to converge to the same abilities, then that means that there's multiple pathways to get there, right?

Which means it could be quasi-infinite.

Right.

So in your mind of the future of this study, first, if you could communicate, what's the first thing you'd be asking these animals, right, in order to, you know, do better research and get more information more quickly?

And then what do you see the future of that impact?

You know, it's like you cross a threshold and human history is like coming out with AI.

Everything's different from this point forward.

Right.

Well, what would you ask if you could ask?

What is your one question?

Yeah, but you're different because you study this, right?

So you're looking for those keystone questions.

Right.

For me, it's like, you know, what we're interested in when you study animal minds and animal communication is the limits.

What is the boundary, right?

It's like, for humans, you're told you can learn an infinite number of words.

and you can combine it in infinite ways, and this infinite potential is— not that people actually do it, but in principle, you could do it.

In principle, yeah.

And the question is, is there a limit?

So is it a memory limit?

Is it a computational limit?

You cannot put together more than X number of signals.

You cannot remember.

Is it a memory limit?

What kind of issue is— what would prevent you from being able to go one more step?

Where is the boundary?

And that to me is what we're always looking for.

It's like, what is the upper limit?

Where is the ceiling?

What is the ceiling for you compared to what the ceiling is for us?

And there's one— So that's the question then.

Does that inform our ceiling?

It should.

It should.

Because I think a lot of the times we kind of have assumptions about what humans can do.

And it's like, because in principle, we could do it, but realistically, we never do it.

I think if what you're trying to do is predict what— as a scientist, what you're trying to predict is what people would do in the natural environment.

You're not only trying to predict what in principle they could do if one day something happened.

These are 2 different questions.

I think being able to understand what the average individual can do is equally important as understanding what you know, the Einstein can do.

Right.

I think a lot of the times people, when they work with animals, they're looking for the Einstein because you want to figure out what is the upper limit for them.

But what is also the kind of average member of that species capable of doing?

Because understanding that gives you a sense of like what you should expect of them when you interact with them and what would be surprising, right?

It's like, it's almost like everybody wants to know, is my kid smart or is my kid just like totally average or even below average?

You don't know.

Right.

And so it's just like, you know, And you're like, well, what would you look for?

It's like, what would you look for?

And I think finding those boundaries, finding those upper limits is particularly helpful because it then allows us to see— if you only find that only a few individuals can learn 1,000 words, then we should never expect that the dog that I bring home should be able to learn this.

It's like maybe it's possible, but really, why are we even expecting it?

Well, there is that other word.

Engineering.

We could potentially engineer brains.

That is true.

We could try to find a way.

So, and that would be the next thing, which is like, once you know what works, you're trying to figure out, can you develop it so that more individuals could get there, right?

So what is it that you need to get there?

And as humans, we are very good at pedagogy, right?

At teaching others.

We constantly try to figure out how can we teach better, faster— how can we prolong the learning phase so that then we can sort of apply our knowledge to other domains?

And at the same time, that's what intelligence is.

Intelligence is not that I can learn A and B. It's that once I'm confronted with C, even if I've never seen it before, I can use what I've learned with A and B to solve C. Huh.

And that is what makes us so special as a species, is our flexible cognitive ability where when we get in a new environment confronted with a new challenge, even if we have never seen it before, we can try to retrieve from whatever we know if there's a solution, probably works like this, and we can get there.

And that ability to kind of extrapolate from knowledge that we have into new domains is a thing that people say many, current large language models do not have.

If it's in the training set, that's great.

If it's a completely new task, we don't really know if they can get there, if they can figure out how to solve this problem because it was never part of their training set.

Yeah.

But I guess we can program them to reprogram themselves in order to— And see how far it goes.

It's a little scary, but yeah, we can think about it.

But what you just mentioned reminds me of the process of physics growth that I had.

And here's what I mean by that.

You know, you go to the university, you go to colloquia, you know, you get scientific papers, and I can confidently say that I never understood anything ever.

Like, I would go to colloquia and fall asleep as a first-year grad student, right?

First, second-year grad student.

Then I look up 4 years later, and it's like, no matter what talk I go to, if I've never heard of it, because I have this toolbox now of understanding the universe and how it works, I can figure out everything.

Yes.

Never have seen it before, right?

Yes, yes, yes.

And so I, I, I, it makes me wonder if, you know, I took my brain, which was, you know, ignorant, untrained, and trained it, and it seemed like it improved its ability to interpret the world because I gave it a new set of knowledge.

So I guess the question becomes is if you could communicate, could you up their knowledge, right?

Is there a way to increase chimpanzee knowledge?

Because maybe one species has the ability to do things, you know, maybe some species are great at computation in their heads, right?

Maybe some species are great at making connections, right?

So where does the— So exactly, so the way I think about it is kind of those superhero movies where you're like, or like, you know, the A-Team, like how do you put together a team?

Where you have the one who's really good at this, the one who's really good at that.

And if you could imagine a world in which you can communicate with other animals, and you can tell that this animal is really good at smell, this is very good at vision, this is very good at hearing things.

Yeah.

This is very good at thermal, detecting thermal variations and so on.

This one can tell you when earthquakes are coming or when hurricanes are on the way.

And you can just communicate with them so that you can figure this out and figure out some kind of solution, that would be amazing.

Bro, I think we have the beginnings of a script here.

We can, uh— The new A-Team.

A-Team where A is like the animal team.

Animal team.

Frederico, sir, you are amazing.

Thank you for sharing the world of animal communication with us.

This is far beyond my expectation.

I had no idea.

So now you got me stuck.

Now I've got to go down another rabbit hole of learning.