David Gruber Can we learn to talk to sperm whales TED
You are about to hear the sounds
of the largest-toothed predator
on the planet:
an animal bigger than a school bus
with perhaps the most sophisticated
form of communication
that has ever existed.
(Video: whale clicking)
These are the sounds
of the mighty sperm whale,
a fellow mammal
that can dive almost a mile,
hold its breath for more than an hour
and lives in these amazingly complex,
matriarchal societies.
These clicks you heard,
called codas,
are just a facet of what we know
of their communication.
We know these animals are communicating,
we just don’t yet know
what they’re saying.
Project CETI aims to find out.
Over the next five years,
our team of AI specialists,
roboticists, linguists
and marine biologists
aim to use the most
cutting-edge technologies
to make contact with another species,
and hopefully communicate back.
We believe that by listening
deeply to nature,
we can change our perspective of ourselves
and reshape our relationship
with all life on this planet.
This of course seems
like an impossible goal.
People have been trying
to make contact with other animals
for hundreds of years.
How could we do what others could not,
especially given that I’m sitting here
on my couch in New York City
in the middle of a pandemic and protests?
I’ve spent the last 20 years
as a marine biologist and oceanographer,
studying the ocean
from all different perspectives,
from microbes to sharks.
I’ve assembled interdisciplinary teams
that have built the first shark-eye camera
to see the world
from a shark’s perspective,
and have collaborated with engineers
to design robots so gentle
that they don’t even stress a jellyfish.
But it wasn’t until 2018
when I was on fellowship
at the Radcliffe Institute
for Advanced Study
that I realized that perhaps the best way
to understand the ocean
and its inhabitants
wasn’t just by seeing the world
through their eyes,
but by listening –
by really, deeply listening.
I became interested in sperm whales
when I heard their sounds.
They sounded like they were
coming from another universe;
a siren song being broadcast
from the darkest reaches of the sea.
These weren’t the typical
harmonious whale songs
that I had been accustomed to.
These sounded more
like digital data transfer.
We assembled the future Project CETI team
and began discussing how to use
the most advanced technologies
to communicate with whales.
One of the principal conclusions
was that machine learning
had a really good chance
of understanding the patterns
of sperm whale communication.
And the time to apply
these technologies was now.
Cracking the interspecies
communication code
didn’t just seem possible,
it almost seemed inevitable.
But how can analyzing patterns
help us converse with whales
and other animals?
Well, step one is to understand
the elements of sperm whale communication.
These codas you heard don’t appear
to be sentences as we know them,
but there’s clear structure
in how these animals communicate.
Sperm whales send codas
back and forth to each other
in sequences,
and there are regional dialects
like British and Australian accents.
This is exactly why machine learning
is such a powerful tool.
These approaches analyze patterns
in relationship and map meaning to them.
Just a few years ago,
scientists used machine learning
to translate between two
totally unknown human languages.
Not by using a Rosetta Stone
or a dictionary,
but by mapping them on patterns
in higher-dimensional space.
But for machine learning
to work effectively,
it needs data –
it needs lots and lots of data.
In the past half-century,
marine researchers
have painstakingly collected
and hand annotated just a few thousand
sperm whale vocalizations,
but in order to learn
sperm whale communication,
we’ll need to collect millions,
if not tens of millions
of carefully annotated
sperm whale vocalizations
correlated with behaviors.
We’ll do it with noninvasive,
autonomous, free-swimming robots,
aerial-aquatic drones,
bottom-mounted hydrophone arrays
and more.
We’ll work with our close partners
at the Dominica Sperm Whale Project
to cover a 20-square-kilometer area
that is frequented by over 25
well-known families of sperm whales.
We’re going to put specific focus
on the interactions of mothers and calfs,
training our machine learning algorithms
to learn whale language
from the bottom up.
All this data we’ll have sent
through a pipeline
and analyzed by the Project CETI
translation team.
The raw audio and context data will
go through our machine learning engine
where it’s going to be first
sorted by structure.
The linguistics team will then search
for things like syntax
and time displacement.
For example,
if we find an event where a whale
was talking about something yesterday,
that alone would be a major finding,
something that has thus far
only been shown in humans.
And once we’ve really mastered listening,
we’re going to try
really carefully to talk back
even on the most simplistic level.
Finally, Project CETI will build
an open-source platform
where we will make our data sets
available to the public,
encouraging the global community
to come along on this journey
for understanding.
These animals could be the most
intelligent beings on this planet.
They have a neocortex and spindle cells –
structure that in humans
control our higher thoughts,
emotions, memory, language and love.
And all the platforms that we develop
can be cross-applied to other animals:
to elephants, birds,
primates, dolphins –
essentially any animal.
In the late 1960s,
our team member, Roger Payne,
discovered that whales sing.
(Recording: whale singing)
A finding that sparked
the Save the Whales movement
led to the end of large-scale whaling
and prevented several
whale species from extinction
just by showing that whales sing.
Imagine if we could understand
what they’re saying.
Now is the time
to open this larger dialogue.
Now is the time to listen deeply
and show these magical animals,
and each other,
newfound respect.
Thank you.