Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
1:59
animal communication using AI. We
2:03
had a wide-ranging conversation that started
2:05
with the basic discovery that our world
2:08
is filled with way more animal sounds
2:10
than a lot of people initially thought. And
2:12
from there, we talked about what those sounds might
2:15
mean, whether there's some form of language
2:17
and whether scientists might someday
2:20
be able to use AI to actually
2:22
decipher what animals are saying. We've
2:25
edited this a bit for clarity and length.
2:28
Here's our conversation.
2:35
So just to start,
2:37
before we get into how researchers are
2:39
trying to decode animal sounds today, I
2:42
think it's worth talking about how
2:44
they realized they were worth decoding in the first place.
2:47
So, Karen, you have a story in your book about how European
2:50
and American scientists really started to discover
2:53
that these sounds were even worth
2:55
exploring. You talk about whale song. So
2:58
I wonder if you could just set the scene for us a little. What
3:00
was the state of research on ocean
3:02
sounds, maybe 100 years or so ago?
3:05
Yeah, so cast your mind back to about 100 years
3:07
ago. There's a widespread assumption
3:10
that only humans possess language, and
3:12
moreover, that other species
3:14
do not possess complex communication
3:18
capacity.
3:18
This
3:20
is, of course, a blind spot in Western
3:23
science that we're going to unpack throughout the course
3:25
of the conversation today. But
3:27
at the time, one of the most remarkable
3:30
aspects of Western science
3:32
was really a kind of a cinnabon mission.
3:35
There wasn't a lot of work done recording
3:37
animal sounds. There wasn't a lot of work
3:40
done trying to decode them. And
3:42
that all changed when a
3:44
small number of researchers, some of
3:47
them associated with then classified
3:50
Navy efforts to listen to underwater
3:52
ocean sounds, began attempting
3:55
to decode and categorize
3:58
the really complex underwater sounds.
3:59
hearing in the ocean. So
4:02
these scientists made early
4:04
efforts, mostly to record
4:07
the
4:07
pretty profound and amazing sounds that whales
4:09
make, which are now well known to us. But
4:12
at the time, we're pretty astounding
4:15
to Western scientists, the kind of operatic
4:17
uulations of humpback whales, the very staccato,
4:20
powerful, eardrum blasting sounds of
4:23
sperm whales. There's a
4:25
whole symphony under the ocean, of
4:28
which Western science was largely unaware.
4:30
Their work came to public attention
4:33
when a renegade scientist
4:36
named Roger Payne and his brilliant wife,
4:39
Katie Payne, a classically trained musician, took
4:42
some classified recordings that
4:45
Navy scientists had given them and published them
4:47
as an album, which
4:50
remains the best-selling nature album of all time, went
4:52
platinum, and was one of the major
4:54
things that
4:55
changed the dynamic around
4:58
the campaign to end industrial whaling, which was
5:00
arguably one of the things that saved many
5:03
whale
5:03
species from extinction. So
5:05
whales were our re-entry
5:09
as Western scientists into something indigenous
5:12
cultures had long known, that many,
5:15
many species are capable of complex communication.
5:17
And from there, many
5:19
other
5:20
species have begun essentially
5:22
to reveal themselves to Western science. But I want to emphasize
5:24
right at the start, this has long held
5:26
human knowledge that somehow we have
5:29
forgotten that we had forgotten, and
5:31
we have just begun remembering really
5:34
how to remember.
5:35
So talking about things that we've forgotten,
5:37
or maybe things that we were
5:39
not aware of until
5:42
very recently, it's
5:44
not just these sounds under the water. It's
5:46
not just deep sounds that
5:48
we haven't been able to hear because we didn't have submarines.
5:51
I wonder if you can tell us a bit about
5:54
some of these sounds that we can't hear
5:56
without technology that are around
5:58
us all the time. majority
6:00
of these sounds are inaudible to the
6:02
naked human ear. They are either above
6:04
our hearing range in the high ultrasound or below
6:07
our hearing range in the deep infrasound.
6:10
There is an evolutionary reason
6:11
for this called the acoustic niche hypothesis.
6:14
So in most ecosystems, what
6:16
you get is much like radio
6:19
stations on the radio dial. You'll have
6:21
different species, essentially broadcasting
6:23
acoustic communication at a specific set of frequencies,
6:26
a band, and also being able
6:28
to hear in that same
6:29
frequency range. We hear pretty
6:31
much at the frequencies that we are able to vocalize
6:34
at. So the ability
6:37
to record beyond human hearing
6:38
range is only about a hundred years old for
6:41
ultrasound and infrasound. But what
6:43
AI does is it allows
6:46
the parsing, the categorizing
6:48
of that data at scale. And
6:51
so we reversed a very fundamental constraint
6:54
of 20th century biology.
6:57
We used to have basically a scarcity of data.
7:00
Now we have a hyperabundance of data
7:01
because of cheap digital recording devices, many,
7:04
many such recording devices from the
7:06
Arctic to the Amazon. And now we have AI
7:08
that can do some, not all,
7:10
automated
7:13
tracking, categorizing, parsing.
7:15
That doesn't mean we can actually then make the
7:17
next step to translating, but it gets
7:19
us a lot further. So
7:22
you're talking about now using AI,
7:24
using this more data we've had, using
7:27
AI to analyze it.
7:30
Aiz, I wonder if you can talk a little bit about
7:32
the research you've done with
7:34
using AI to map things like
7:36
shapes of languages. Like how does AI
7:38
help us translate between languages
7:41
when we might not be able to understand them? Yeah.
7:44
So I'm going to tell this story in two parts. And
7:47
the reason why we started Earth Species
7:49
in 2017 was that something fundamental
7:53
changed, right? Because if you're going to try to translate
7:55
a language without a Rosetta stone, that
7:58
didn't exist in human history. up
8:00
until like 2017, then that changed. So
8:04
what changed? AI gained
8:06
the ability to build shapes that
8:08
represent languages. For those of you that are AI people in the
8:11
room, these are called latent spaces or embedding
8:13
spaces, but these shapes are really
8:15
interesting because they turn
8:18
semantic relationships into geometric
8:21
relationships. What does that mean? Think about
8:24
a language like English. Now imagine
8:26
a galaxy where every star is
8:28
a word.
8:29
And words that mean similar things are near each other, and
8:31
then words that share a semantic relationship share
8:33
a geometric relationship. So king
8:35
is to man, as woman is to queen. So in
8:37
this shape, king is the same distance direction to man,
8:40
as woman is to queen. So you just do king minus man,
8:42
that gives you a distance and direction. You add that to boy
8:45
and it'll equal prince. You add that to girl, it'll equal
8:47
princess. Like all the
8:49
internal relationships of a language are
8:51
encoded in this shape. And
8:53
if you think about it, like dog
8:56
has relationship to man and to wolf
8:58
and to fur and to yelp and to howl,
9:01
it sort of fixes it in a point in space
9:03
in this shape.
9:04
And then if you solve the massive multi-dimensional
9:07
no sudoku puzzle of every
9:09
concept to every other concept,
9:12
out pops this rigid structure representing
9:14
all the internal relationships of a language.
9:17
The computer doesn't know what anything means, it just knows how they relate.
9:19
And here was the deep insight from 2017. Could
9:23
the shape of two languages possibly
9:25
be the same?
9:26
So I'm holding a Portuguese word cloud here.
9:29
You're holding an English one. And
9:31
the mathematical relationship between
9:34
the stars and my galaxy
9:36
that represent woman,
9:39
queen, king, man, is
9:41
more or less the same mathematical, that
9:44
is spatial relationship in your word cloud. If
9:46
I were to throw up Cree or in
9:48
Nukh-to-tut, a little bit less
9:50
overlap, but more or less, a lot of
9:52
these concepts are invariant across human languages.
9:56
And that is why we are able to
9:58
really effectively translate. now using
10:00
AI between our different word clouds.
10:02
That's exactly right. You literally just rotate
10:05
one shape on top of the other, and even
10:07
though there are words in one language that don't exist in the other,
10:09
the point which is dog ends up in the same spot in both,
10:11
and you sort of blur your eyes in the same shape, and that works
10:14
for English and Spanish, and you're like, cool,
10:16
those are related languages, obviously, but
10:18
also works for Finnish, which is a weird language, and
10:20
Aramaic and Urdu, Esperanto.
10:23
Every human language sort of roughly fits in this
10:25
universal human meaning shape, and that
10:28
was the moment that we're like,
10:29
there's a path through.
10:31
Do you think if we build this shape for animal
10:33
communication, it fits somewhere
10:36
in the universal human meaning shape, and the parts that
10:38
overlap, we should be able to directly translate into
10:40
the human experience, and the parts that don't overlap, we
10:42
should be able to see complexity, and so
10:44
this gives us the ability to start getting like blurry
10:47
polaroid images of
10:50
things that are beyond the human imagination.
10:52
Let's go back one step to this notion
10:55
that we can translate between our different
10:57
word clouds, you know, English and Inuktutut,
11:02
and then extrapolate that to non-human communication
11:04
systems. So, first of all, you have to imagine
11:06
that scientists now have the ability to create these kind
11:08
of latent
11:09
spaces or word clouds with non-human
11:11
communication regimes. For example,
11:14
there is now an elephant dictionary with
11:17
thousands of sounds, and field
11:19
biologists have painstakingly documented
11:21
what each of those signals mean. And
11:24
elephants, for example, have a specific signal
11:26
for honey bee. They're terrified
11:28
of honey bees. They can get
11:30
into their trunks and their ears and sting them,
11:33
and so there's a very specific behavior
11:35
elephants display when they hear the sound.
11:37
So tested through playback experiments,
11:40
we have this elephant dictionary, but
11:42
there are many ifs here because
11:44
the assumption that the underlying
11:47
worldview, the felt-lived embodied sense
11:51
of an elephant, the umm-veld,
11:53
as researchers call it, is anything
11:55
like a human is one we haven't
11:58
yet proven. And so my word cloud means...
11:59
not contain any concepts that
12:02
actually overlap with ASAs, with the exception
12:05
of very few. So it may
12:07
be that AI translation is either
12:09
a dead end, or it would only
12:11
allow us to develop a small subset
12:14
of translatable concepts, and
12:17
that there are
12:18
other rules governing these non-human
12:20
communication systems that we've yet to figure out. I'll
12:22
just give a couple examples. It
12:25
may be that animal species have different languages
12:27
for different times of year. A little
12:29
bit like if you're familiar with Indian classical music,
12:32
ragas.
12:32
There's the morning raga, the evening
12:34
raga. So you have to throw out all
12:36
of your assumptions about language. Maybe
12:38
they have different languages for different parts of the
12:40
world if they're migratory. The
12:43
Bering Strait may have a different language. Then
12:45
the warm waters of Hawaii,
12:46
where you give birth. So
12:49
we are just really, at the beginning, of
12:52
trying to figure out what ASAs is trying to do is achievable.
12:55
My personal bet, and I'd love to hear your view, is
12:58
that there will be an incomplete translation.
13:01
We will be able to detect names. We'll
13:03
be able to detect alarm calls. And
13:06
we'll be able to detect the labels
13:08
that are given to
13:09
features of the environment that are linguistically
13:11
invariant. But there are many, many
13:13
more complex concepts that we're going to have
13:15
to invent entirely new types of science
13:18
to begin understanding. And those are going to combine field
13:21
observations with AI.
13:22
Absolutely. I just want to say
13:25
that all of our work is built off
13:27
of decades of painstaking
13:29
research done by biologists out in the field. And
13:31
everything we do is hence in collaboration. But
13:34
then just to add a couple of thoughts
13:36
to what you're saying is, one, the way we're describing
13:38
doing this kind of translation with
13:40
rotation, that's 2017 AI tech. It's
13:43
now sort of Stone Age. There are many
13:46
other techniques which I can talk about that that
13:48
becomes just one tool among many.
13:51
But why should we expect, I agree, the unvelte
13:53
of a sperm whale might be so completely different. It
13:55
spends 80% of its life in complete
13:57
darkness, a kilometer deep in the ocean.
13:59
Why should we expect there to be any overlap whatsoever?
14:03
And I'll give two examples for why there might be
14:05
overlap. And then I'll sort of talk
14:07
about the parts that, like why I think this goes even
14:09
beyond language. The
14:11
first example is the mirror
14:14
test. It's like, how do you know
14:16
whether another being has self-awareness? One
14:18
way you might discover that is you would paint
14:21
a dot on them
14:22
where they are unaware of that dot. Then you
14:24
put a mirror in front of them. And
14:27
if they see
14:28
the dot and they start to try to get it off, that
14:30
shows that they're connecting the image in the mirror
14:33
with themselves, that they have a self-image.
14:36
Now, if they don't respond, that doesn't actually
14:38
tell you anything. Researchers thought
14:40
for the longest time that elephants couldn't pass the mirror
14:42
test, but it turns out they're just using small mirrors.
14:45
Right? But
14:49
a number of species do pass this kind
14:51
of mirror test. And that means if they're
14:53
communicating, they may well be communicating
14:55
about a rich interiority,
14:58
like a self-awareness, one of the most profound things
15:00
that we have. Another example, as
15:04
I have this in my presentations, an incredible
15:06
video of lemurs biting
15:09
centipedes to get high. So they're like
15:11
taking hits off of centipedes. They get super
15:13
cuddly. They enter these trance-like states. It's
15:16
sort of like a proto-burning man. And
15:20
like dolphins do the same thing. They will intentionally
15:22
inflate puffer fish also
15:24
to get high off of their venom and pass them around
15:26
in the original puff puff pass. Right?
15:30
So gorillas and chimpanzees will
15:32
spin. They'll hang on a vine and spin to get really dizzy.
15:35
Transcendent states of consciousness appears
15:38
to be a thing that we share and desire
15:40
across many species. So that, too,
15:42
is like a very profound thing that if we communicate or if they
15:45
communicate, they may well communicate about.
15:47
So there are some, I think, really interesting
15:49
areas of overlap.
15:51
This conversation we're having, on one
15:54
sense, it can feel like we're just speeding
15:56
ahead. And there's this one other hurdle we
15:58
have to pass, which is like. like figuring out this translation,
16:01
the AI is almost there. We're
16:03
also using a lot of anthropomorphic
16:06
language, and I feel like that is definitely,
16:10
I understand why we would do that. We don't have
16:12
better words for those concepts. I
16:15
mean, I want to clarify that
16:16
scientists do not use that language.
16:18
They use very technical terms. So
16:20
for example, scientists would not use
16:23
the term name. They
16:25
would say individual vocal label
16:28
or vocal signature. Equally,
16:31
most of the scientists studying the communicative regimes
16:33
of non-human species would
16:35
use the term communication, not
16:38
language. Because language is sufficiently
16:41
anthropocentrically defined in terms
16:43
of complex
16:45
combinatorial capacity, symbolic content,
16:48
syntax, so on and so forth, that
16:51
it is as yet to be proven that other
16:53
species have, quote unquote, language.
16:56
So
16:57
I just want to clarify that
16:59
although Aysa and I, in
17:02
a sort of a public communication of science way, are
17:04
using these terms, the scientific community is
17:07
pretty rigorous, perhaps
17:10
incorrectly so, but nonetheless
17:12
pretty rigorous about setting a
17:15
boundary between humans and
17:18
non-human species. But one of the things that this research
17:20
may eventually do is create
17:22
a sufficient weight of evidence that
17:25
we do indeed say, ah yes,
17:27
other species have language. We
17:29
may need to change our definition of language in order
17:31
to do so. Or, ah yes, other species
17:33
do convey symbolic
17:36
meaning through language. And here's how. We're
17:39
not there yet. So progressively,
17:42
I think this science is going to lead us somewhere very,
17:44
very interesting in terms of asking
17:46
these fundamental questions on the basis
17:49
of a huge amount of
17:51
empirical evidence.
17:53
I wonder, you know, we briefly
17:56
mentioned the Umm Velt question about what
17:58
does it mean to be a bat or a whale. or
18:00
any of these things that perceive the world completely differently.
18:03
And I guess
18:05
I just wonder if we
18:07
could be communicating something to the animal,
18:10
that the animal would be potentially understanding
18:12
in a different way and acting in a way that
18:14
looks like what we expect
18:17
the animal to act, because that's how we are understanding
18:19
things. But will we ever... Is
18:21
it even possible to imagine that we could actually
18:24
communicate where we both know that we
18:26
are understanding each other?
18:28
I mean, the same problem exists between any
18:31
two humans. Yeah. The myth of communication
18:34
is that it ever happened in the first place. Yeah.
18:37
So I think the
18:39
practical, pragmatic scientific responses
18:42
play back experiments. And that
18:44
is how these are tested. We assume this acoustic
18:47
signal means this. We can play
18:49
it back in the field or in lab-controlled conditions.
18:52
We see the response as what we predict. Elephant
18:55
honeybee alarm call leads to very specific
18:57
physical behaviors, the group of elephants coming together,
19:00
dusting themselves.
19:01
Now, beyond that,
19:03
I mean, the act of communication is
19:06
a profound mystery. The ability
19:08
of any two beings on Earth to
19:10
believe they could actually understand one another,
19:13
it is actually quite magical. So science
19:15
is one of the ways
19:16
of approaching some of the great mysteries,
19:19
and communication is one of them.
19:22
But the reason I think this captures so much public
19:24
imagination, Ays and I have talked about this in the past,
19:26
is because this is also a great mystery,
19:29
which has been the subject of much reflection
19:32
in various mythological
19:34
and spiritual
19:34
traditions. And
19:36
so that is some of the richness of this
19:38
work. It's also some of the controversy that it inspires
19:41
in the scientific community.
19:47
We're going to take a quick break from the conversation
19:49
here. When we come back, we're going
19:52
to ask whether we should be trying to translate
19:54
animal communication at all.
20:00
This episode is brought to you by
20:02
Shopify. That's
20:03
the sound of switching your business to Shopify, the
20:07
global commerce platform that supercharges your selling. Harness
20:10
the best converting checkout and same intuitive
20:12
features, trusted apps, and powerful
20:15
analytics used by the world's leading brands. Stop leaving
20:17
sales on the table. Discover
20:20
why millions trust Shopify to build,
20:22
grow, and run their business. Sign
20:24
up today for your $1 per month trial period
20:27
at shopify.com slash tech23.
20:30
Chinese fashion behemoth Shein is
20:32
one of the fastest growing companies in
20:34
the world, which means the
20:36
people of planet Earth like it. But
20:40
it has so many critics. Designers
20:42
say Shein steals from them. Shein
20:44
straight up just stole my pirouette
20:47
skort design. Influencers say
20:49
they've been misled. I make mistakes.
20:51
I'm imperfect. I'm forever
20:53
growing and evolving, and that's like probably
20:56
one of my favorite things about myself. I'm sorry
20:58
about that trip to China. Congress
21:00
said Shein's being deceptive about who
21:02
makes the clothes. Consumers don't, all
21:04
things being equal, want to purchase goods
21:07
that are contaminated by slave labor or forced
21:09
labor. And a Vogue editor says the
21:11
clothes are just not that good. These are not very
21:14
well made clothing. They are not made to last.
21:17
As Shein flirts with going public, why
21:20
is this hated company so
21:22
beloved? And why is this beloved
21:24
company so hated? Today Explained,
21:27
it's in your feeds every weekday at 2
21:29
p.m.
21:38
One issue we haven't touched on is sort
21:41
of a should issue. Yeah. And
21:44
assuming we can use
21:46
AI to analyze these language shapes,
21:49
to figure out in some basic sense
21:52
what animals may be saying to each other, maybe
21:54
we can communicate with them, maybe we can listen
21:56
to them, knowing
21:58
humans... I can imagine there
22:01
would be some maybe not friendly ways
22:03
to approach that situation. And I'm wondering what you
22:05
think about the danger
22:08
of being able to understand animals
22:10
and being able to communicate with them.
22:12
So I want to give a really specific example of
22:18
the new responsibility that we
22:20
as a species are going to have to show up
22:22
to in the very, very near future.
22:25
So you guys, I'm sure, have encountered Chat
22:27
GPT. You can
22:29
build chat bots
22:31
like Chat GPT in Chinese, even
22:33
if you do not speak Chinese. And
22:36
you've probably also seen all of the deep fake stuff.
22:40
Now, it is possible with just three seconds of anyone's
22:42
voice to continue to speak in their voice and
22:44
say what they were saying. And what
22:46
this means is that within months
22:50
to short number of years,
22:52
we will be able to build, essentially,
22:55
synthetic chat bots, synthetic whales, synthetic
22:57
belugas, synthetic tool
23:00
using crows that can speak in
23:03
a way that
23:04
they don't understand. They're not speaking to
23:06
one of their own. Imagine you had the superpower,
23:08
and your superpower was able to walk up to
23:10
somebody whose language you don't understand.
23:13
You sort of cock your ear. And you listen, and you're like, OK,
23:15
I see this pattern after this pattern. You start to babble
23:18
with those patterns. And you don't know what you're saying, but
23:20
the other person's like, yeah,
23:21
wow. It's Douglas Adams' Babelfish. It
23:23
is, except here's the plot twist. You'll
23:26
be able to communicate before
23:29
you understand. And so this is actually
23:31
the case. We're actually starting our first
23:33
experiments with Zebrafinch likely later this
23:36
year. We're doing real time, two
23:38
way communication with captive
23:40
population to see can
23:42
we do, start to
23:44
cross this communications barrier
23:47
by being able to speak before we
23:49
understand. And this is fascinating.
23:51
It obviously lets us start to get to decode much
23:54
faster. But Humpback
23:56
whale song goes viral, right? Like songs
23:58
sung off the coast of Australia. Australia can
24:01
go 1,000 kilometers. The
24:04
Humpback whale song will be picked up by the world population
24:07
within a season or two sometimes. And
24:09
so if we're not careful, if we just create a
24:11
whale that starts to sing, especially before
24:13
we understand what it means, we
24:17
could be messing up with wisdom tradition, creating
24:19
a whale QAnon. We don't know. And
24:22
that means before that
24:25
happens, because that means it's
24:27
a very crazy thing to think
24:29
of. I didn't think we were going to get here this quickly. In
24:32
the next 12 months, five years, certainly before 2030,
24:35
we will have the capacity to
24:37
do real time, two way communication,
24:40
animal to AI, not necessarily
24:42
animal to human.
24:45
And
24:46
we need to have a kind of Geneva
24:48
convention for cross species communication, a
24:51
prime directive, sets of norms,
24:53
ways that IRBs review. There are a whole
24:56
bunch of things we
24:58
need to set up. And I think you can talk about them too. Yeah, and
25:00
I think there are even more nefarious uses. Precision
25:02
hunting, precision
25:04
fishing, of course. Yeah, poaching. Yeah,
25:06
poaching. This will enable the
25:10
acceleration of the kind of cat and mouse
25:12
game between poachers and game keepers,
25:15
no doubt. There
25:17
also is the specter of being able
25:19
to domesticate species that
25:22
were formerly not domesticatable by
25:24
humans. So we may
25:26
be able to use this in certain contexts,
25:29
and this is what my next book is about, for
25:31
biodiversity conservation goals.
25:34
At the same time, it could allow
25:36
bad actors, and
25:39
keep in mind how big
25:41
the multi-billion dollar global
25:43
illegal wildlife trade is, right, to
25:45
further capitalize on their ability to ensnare
25:48
animals
25:48
that have so far been out of reach. So
25:51
the Geneva Convention long term for multi-species
25:54
dialogue, great. Prior
25:56
to that, I think we've got a
25:59
more immediate problem.
25:59
on our hands given the biodiversity crisis
26:02
with respect to nefarious
26:04
uses of these technologies. The
26:07
only saving grace is that the
26:09
AI may not be really as good as we
26:12
think. So you know, first
26:14
of all, we're being very self-centered
26:17
here as usual, we're humans. We're assuming
26:19
other species actually want to talk to us. They
26:22
may be like boring, you know,
26:24
or they may just assume that
26:26
these sounds which are gibberish, you know,
26:29
are to be avoided, rightfully so. Or they
26:31
may simply be able to detect it's not being made
26:33
by another living member of their
26:35
species and avoid. So
26:38
my hope is that they're
26:39
going to reveal, we're going to reveal ourselves to be slightly
26:42
stupider than we think. They're going to reveal
26:44
themselves to be smarter than we believe. And
26:46
maybe that'll create a bit more breathing room. But no
26:48
doubt, longer term,
26:50
deep fake AI technology creates a whole
26:52
bunch of risks. And do you think we should, given
26:54
these risks, you think this is something that
26:57
scientists should push forward? Yeah.
26:58
I mean, I believe we should have a moratorium. Ays
27:00
and I don't agree on this point, I think.
27:03
I think there are certainly thresholds that if we cross,
27:05
we should have a moratorium. We should stop. Absolutely.
27:09
Here's the thing. We are mutilating
27:11
the tree of life. And at some point,
27:14
we are going to cut the branch upon which humanity
27:16
depends. Right?
27:18
So we are in the land of Hail
27:21
Mary passes.
27:23
The hope for,
27:25
I think, working on
27:28
showing and really, the
27:30
point is not really to talk to animals. The
27:32
point is really to understand and listen. And along
27:34
the path to that, we are creating the
27:36
technology that solves the fundamental
27:38
problems we see across all of conservation,
27:42
biology and ethology research. Every
27:45
biologist we talk to needs to do classification,
27:48
denoising, detection of signals
27:50
to understand biodiversity, to understand their
27:53
behaviors. And so the tools we build
27:55
as we head towards decoding
27:58
are the fundamental tools we use. that
28:00
are accelerating conservation biology,
28:02
which to the extent that conservation science
28:05
accelerates conservation, like
28:08
we're trying to broad scale do that. But
28:10
then there are these moments when we get shifts
28:12
in perspective and that changes everything. We talked about
28:14
songs of the humpback whale, but also when human
28:17
beings went to the moon and
28:19
when human beings were dosed with that overview effect
28:22
and seeing us as a pale blue dot suspended
28:24
in space, planet, spaceship earth,
28:26
right? That's when the EPA came into existence. NOAA
28:30
was born, modern environmental movement started, clean
28:32
air acts was passed, and that was in the Nixon
28:35
era, right? And so the goal here
28:37
is
28:38
like there are moments in
28:40
history which superpower
28:42
movements. There are no silver bullets, but maybe
28:44
there's silver buckshot. And maybe
28:47
if
28:47
we know this is coming, we
28:49
can arm every other conservation
28:52
org out there, rights for nature, personhood
28:55
for non-humans, E.O.
28:57
Wilson's half-earth, much bigger marine
29:00
protected zones. When this becomes the thing
29:02
that the entire world sees
29:05
and becomes the top of politicians'
29:08
priority list, suddenly I think
29:11
we can accelerate and be a force multiplier
29:13
for every other conservation
29:17
and climate action out there. And
29:19
that I think is the reason why it's worth pursuing.
29:21
I wonder if, yeah, I mean, just before we finish, I wonder,
29:24
could you just say a bit about why you think we should have a moratorium?
29:28
I will, but I do also want to build on Asia's
29:30
point. So
29:31
the climate change and biodiversity
29:34
crisis are intimately interrelated. And
29:37
the fundamental
29:40
challenge of the next 20 or 30 years,
29:42
as we add a couple billion more humans to
29:45
the planet, is a sort of arc, Noah's
29:47
arc-like challenge. How many species
29:51
will be around at the end of our
29:53
lifetimes? And
29:56
acoustics, regardless of whether
29:59
we actually achieve it, we can do it. to achieve interspecies communication
30:02
is a powerful tool in the conservationists
30:05
toolkit. Because simply
30:07
through the use of digital bioacoustics, you
30:09
have a very low cost, very effective
30:12
monitoring regime that
30:14
is much less invasive than human
30:16
monitoring. And this
30:18
is now something that is being set up around
30:20
the world.
30:22
I can't go into the technical details for lack of
30:24
time, but very simply, bio
30:26
and ecoacoustic
30:27
and disease allow us to tell
30:30
simply by listening the
30:32
extent to which climate change is disrupting species,
30:35
species migrations and movements, and
30:39
species abundance or disappearance, et cetera. So
30:41
we may never achieve interspecies communication,
30:45
but what we can do and we all should be doing if we're
30:47
interested in environmental work is supporting
30:49
the inclusion of digital acoustics,
30:52
bio and ecoacoustics, into conservation
30:54
work as a low cost, minimally invasive,
30:56
very powerful tool. So
30:59
I hope that's a take home message for all of
31:01
you. The question
31:03
of the moratorium, I think, is really
31:06
one about something
31:08
humans are not very good at. That
31:11
is
31:11
having
31:12
the ability
31:13
to
31:16
create a space in
31:18
between what we think we're capable of doing
31:21
and reaching for that thing. So
31:24
there is a
31:25
in the tech community, in the scientific community,
31:28
there is a can do ethos. I
31:32
can do that. And so I want to do
31:34
it. But there is a should
31:36
do question here that I think
31:38
requires very, very careful consideration.
31:41
And I see no, and
31:43
I hear I disagree with you, compelling reason
31:45
right now to continue the
31:48
work that could be so damaging to other species
31:51
that could
31:52
lead to precision hunting and poaching. Without
31:55
getting a lot of the ethical frameworks
31:58
in place, it would mean updating the
31:59
Convention on International Trade in Endangered Species,
32:02
updating a lot of international environmental regulatory
32:05
frameworks. AI governance poses a more
32:07
general
32:07
problem, right? So
32:09
my view is we need to get our house in order on that.
32:11
And just like from time to time human
32:14
genomics research has hit pause
32:16
and there's certain no-go areas, like cloning
32:19
humans, I think we can come
32:21
up with a set of no-go areas for
32:24
AI science in this regard that
32:26
would allow technical progress to still be made,
32:29
but not be invoking the kind of risks
32:31
that we barely have even begun to understand.
32:33
And on that we actually I think agree,
32:36
which is that as we show up to
32:41
the new responsibility of the power, you have
32:43
to ask how is that power bound to
32:45
wisdom and how knowing
32:48
that this technology is coming, because
32:50
regardless the ability to emulate
32:52
any signal,
32:54
that's being pushed by
32:56
the market forces on the human domain.
32:59
So we need to accelerate as fast
33:01
as possible
33:02
all of the ethics and legal
33:04
updates.
33:05
It's Pandora's box, right?
33:08
Every new technology creates a new responsibility.
33:11
I think there's a small enough set of researchers doing
33:13
this that we could actually do a better job this time
33:16
at sorting out the responsibility before we unleash
33:18
the technology. I completely agree.
33:21
Thanks so much to Aza Raskin and Karen
33:23
Bacher and to the Aspen Institute. If you
33:26
wanna read more
33:26
about the history here, I recommend Karen's
33:28
book, The Sounds of Life. It's got
33:31
the whale story we mentioned, but also some fascinating
33:33
stories about everything from
33:35
chatty turtles to bee communication. You
33:38
can also watch a video of the full discussion we had at
33:40
aspenideas.org. This
33:43
episode was produced by the Aspen Institute and
33:46
the Aspen Institute Foundation. This
33:49
episode was produced by Bird Pinkerton. It
33:51
was edited by Brian Resnick and Meredith Hodnut,
33:54
who also manages our team. We
33:56
had mixing
33:56
from Christian Ayala, music from me, Serena
33:59
Solon, check the facts. and we're so happy to
34:01
have her on the team. And Manding Nguyen
34:03
is not afraid of spiders. If
34:05
you enjoyed the show, it would bring us a lot of joy
34:08
if you'd leave a review, or just send
34:10
us thoughts directly. You can email
34:12
us at unexplainable at vox.com. We
34:15
read every email. Unexplainable
34:17
is part of the Vox Media Podcast network, and
34:19
we'll be back next week.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More