# Ep23 "How can we learn to speak alien?"

An episode of [Inner Cosmos with David Eagleman](https://invisiblenetwork.ai/podcast/1677842672), 28 Aug 2023, 1 h 01 min.

Audio: https://podtrac.com/pts/redirect.mp3/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/534dc725-2971-42a1-b1bf-afaa0107416e/d1a4684d-a82b-4ff5-8440-b06b0065d69e/audio.mp3?utm_source=Podcast&in_playlist=bee00526-7528-4581-b6b6-afaa01083a3d

If we meet extraterrestrials someday, how will we figure out what they're saying? We currently face this problem right here at home: we have 2 million species of animals on our planet... and we have no Google Translate for any of them. We’re not having conversations with (or listening to podcasts by) anyone but ourselves. Join Eagleman and his guest Aza Raskin to see the glimmer of a pathway that might get us to animal translation, and relatively soon.

## Transcript

[0:04] Speaker 1: Imagine that one of these centuries we make contact with
[0:08] Speaker 1: an alien civilization.
[0:10] Speaker 2: It's a big.
[0:11] Speaker 1: Cosmos with quintillions of planets, so it's bound to happen
[0:15] Speaker 1: at some point. But how the heck are we going
[0:18] Speaker 1: to understand what they're saying? How are we going to
[0:22] Speaker 1: decode their language? After all, they might not communicate with
[0:27] Speaker 1: air compression waves. Maybe they do something visual, but in
[0:30] Speaker 1: ranges of light we can't even pick up with our eyes.
[0:33] Speaker 1: We won't have a Rosetta stone, So how are we
[0:37] Speaker 1: going to decipher what they are trying to say to us?
[0:40] Speaker 1: And this might seem speculative, but what I want to
[0:43] Speaker 1: draw our attention to is that we currently are in
[0:46] Speaker 1: the same position right now, right here at home, which
[0:50] Speaker 1: is that we can't tell what a single one of
[0:54] Speaker 1: the two million species on our planet are saying, not
[0:58] Speaker 1: even the six six hundred species of mammals, which are
[1:03] Speaker 1: presumably kind of like us. We're not having conversations with
[1:07] Speaker 1: anyone but ourselves. With all these species, I'm willing to
[1:12] Speaker 1: bet you don't listen to a single podcast not.
[1:15] Speaker 2: Made by a human.
[1:17] Speaker 1: But today we're going to see some hope, some pathways
[1:21] Speaker 1: for how we might get to animal translation and relatively soon.
[1:31] Speaker 1: Welcome to Inner Cosmos with me David Eagleman. I'm a
[1:34] Speaker 1: neuroscientist and an author at Stanford and in these episodes
[1:39] Speaker 1: I examined the intersection of science and our lives, and
[1:44] Speaker 1: today we're going to talk about understanding animals. When I
[1:54] Speaker 1: was a kid, I saw some episodes of Star Trek,
[1:57] Speaker 1: the original one with Kirk and Spock, and the thing
[2:00] Speaker 1: that always struck me was how every week on schedule
[2:04] Speaker 1: they discovered new alien civilizations, which is not so crazy
[2:10] Speaker 1: given that the universe is presumably teeming with life. There
[2:14] Speaker 1: are about one hundred billion galaxies, and each of these
[2:18] Speaker 1: has about one hundred billion stars, and most stars have
[2:23] Speaker 1: some planets rolling around them, so it's extraordinarily unlikely that
[2:28] Speaker 1: we are the only planet with life on it. But
[2:31] Speaker 1: before I get into alien communication, let's quickly address something first.
[2:35] Speaker 1: You've probably heard of the Fermi paradox, and if you haven't,
[2:39] Speaker 1: it's a very important question. It's the question of why,
[2:43] Speaker 1: if there's all this life in the cosmos, why have
[2:46] Speaker 1: we not heard a peep from anyone. This paradox is
[2:50] Speaker 1: named after the physicist Enrico Fermi, who raised this question
[2:54] Speaker 1: if there are so many potential alien civilizations, why haven't
[2:59] Speaker 1: we detected any signals or encountered any of them. Yet
[3:03] Speaker 1: we're living in a moment of history where there seems
[3:06] Speaker 1: to be this very strange contradiction between the high probability
[3:11] Speaker 1: of extraterrestrial civilizations and the lack of any shred of
[3:16] Speaker 1: evidence for them. So over the decades, people have proposed
[3:19] Speaker 1: all kinds of possible explanations for the Fermi paradox. The
[3:25] Speaker 1: first is that maybe aliens don't exist, which is the
[3:28] Speaker 1: simplest explanation but presumably not terribly likely given the size
[3:34] Speaker 1: of the cosmos. So some people point out that maybe
[3:37] Speaker 1: the problem just has to do with the enormous distances,
[3:41] Speaker 1: the vastness of space, and the limitations of our current
[3:46] Speaker 1: technology that might make it hard to detect other civilizations
[3:50] Speaker 1: even if they exist, because the distances between stars are
[3:54] Speaker 1: enormous and signals may take thousands or millions of years
[3:59] Speaker 1: to reach us. Okay, so that's a possibility. Or A
[4:02] Speaker 1: related idea is what's called the rare Earth hypothesis, which
[4:06] Speaker 1: is that Earth like planets, which are capable of supporting
[4:11] Speaker 1: complex life, are exceedingly uncommon in the universe, and that
[4:16] Speaker 1: makes the emergence of intelligent civilizations a rare event. But again,
[4:23] Speaker 1: given that there are something like seventy quintillion planets, that's
[4:27] Speaker 1: a seven followed by nineteen zeros Earth like planets can't
[4:33] Speaker 1: be too rare. So another idea is the technological singularity idea.
[4:39] Speaker 1: Some thinkers have proposed that advanced civilizations might always end
[4:44] Speaker 1: up reaching a technological singularity when their technology suddenly accelerates rapidly,
[4:52] Speaker 1: and one possible outcome of this is that these civilizations
[4:56] Speaker 1: tend to self destruct, and a related hypothesis is that
[5:00] Speaker 1: when other civilizations hit this singularity, this leads them to
[5:05] Speaker 1: a post biological existence. They're no longer products of nature,
[5:10] Speaker 1: but instead they build themselves into other sorts of devices,
[5:14] Speaker 1: which would be hard for us to detect given the
[5:17] Speaker 1: ways that we're searching. And other people suggest that advanced
[5:22] Speaker 1: civilizations intentionally avoid broadcasting their signals or their presence for
[5:28] Speaker 1: fear of attracting unwanted attention or causing conflicts with less
[5:33] Speaker 1: advanced civilizations. Or maybe they're just not interested in contacting us.
[5:38] Speaker 1: They might be too busy with their own problems, or
[5:40] Speaker 1: they simply don't see us as an interesting threat or ally,
[5:44] Speaker 1: so there's no reason to pick up the phone. And
[5:46] Speaker 1: then there's the possibility that they use extremely different communication
[5:52] Speaker 1: methods than we do, so different that we can't currently
[5:56] Speaker 1: understand them.
[5:57] Speaker 2: Or even know what we should be looking for.
[6:00] Speaker 1: So ultimately we don't know why we haven't heard from
[6:03] Speaker 1: anyone yet. There may be other reasons, or maybe multiple
[6:07] Speaker 1: of the reasons I mentioned, or all at play, but
[6:10] Speaker 1: for now we just have to live with the fact
[6:12] Speaker 1: that we haven't yet heard from anyone. So this was
[6:17] Speaker 1: part of the appeal of Star Trek. Every week there
[6:20] Speaker 1: spacewarping off to some new coordinates, and.
[6:22] Speaker 2: Everywhere they go they meet new civilizations.
[6:26] Speaker 1: Now, the thing that always struck me as funny and
[6:29] Speaker 1: the point of this episode is that each week they
[6:33] Speaker 1: end up meeting these new aliens. And often these aliens
[6:37] Speaker 1: look like a female movie star in a cool jumpsuit,
[6:40] Speaker 1: but with subtle differences like pointy ears and green skin.
[6:44] Speaker 1: But the key thing is that all these aliens speak English.
[6:49] Speaker 1: Usually it's a slightly broken English with a difficult to
[6:52] Speaker 1: discern accent, but nonetheless pretty easily understandable, which is of
[6:57] Speaker 1: course very lucky for these Star warsvoyagers who happen in
[7:01] Speaker 1: several hundred years from now to speak English themselves. Now,
[7:05] Speaker 1: why did the writers of Star Trek choose to make
[7:08] Speaker 1: everyone speak just like we do?
[7:10] Speaker 2: Well?
[7:11] Speaker 1: This is a basic constraint of storytelling. It's the only
[7:14] Speaker 1: thing that will work for telling a narrative that people
[7:18] Speaker 1: will tune into. It's hard to tell a story if
[7:22] Speaker 1: the alien is some kind of weird fungus thing that
[7:26] Speaker 1: doesn't speak or lives at a different timescale than we do,
[7:30] Speaker 1: like a tree. If we land on a planet of mute,
[7:34] Speaker 1: slow fungus creatures, it's not going to make good television.
[7:38] Speaker 1: So the stories we tell will always have aliens that
[7:41] Speaker 1: we can talk with and that serve as not so
[7:45] Speaker 1: distant reflections of ourselves. Okay, so no problem, that's what
[7:49] Speaker 1: storytelling requires. But in real life, it's much more likely
[7:55] Speaker 1: that we're going to have a very very difficult time
[7:58] Speaker 1: communicating much of anything to aliens when we find them.
[8:04] Speaker 1: You might think that we can get by with something
[8:06] Speaker 1: like take me to your leader or some sort of
[8:09] Speaker 1: hand signals, but in fact none of that's going to work. Now,
[8:13] Speaker 1: what's the reason that I say this? Why should we
[8:17] Speaker 1: think that communication is going to be so difficult. Well,
[8:20] Speaker 1: the aliens we find on other planets are going to
[8:23] Speaker 1: have a totally different evolutionary history. They may not be
[8:27] Speaker 1: based on DNA like all earthly creatures are, but instead
[8:31] Speaker 1: may have found a completely different way of encoding information
[8:36] Speaker 1: and managing replication and building societies. And it's possible they
[8:41] Speaker 1: won't even be carbon based like all creatures on Earth are,
[8:45] Speaker 1: but instead based on something like the element silicon, which
[8:48] Speaker 1: also has a tetrahedral structure and can make lots of
[8:51] Speaker 1: useful elements and so on. So there's a whole field
[8:55] Speaker 1: of this called astrobiology, where astro refers to star and
[9:00] Speaker 1: often known as exobiology, where exo means outside, and the
[9:05] Speaker 1: idea with this area of study is to search for
[9:08] Speaker 1: naturally evolved life in the universe, mostly on other planets
[9:13] Speaker 1: in the habitable zone, that's the Goldilocks zone of planets
[9:17] Speaker 1: who rotated just the right distance from their star so
[9:21] Speaker 1: it's not too hot not too cold. I'll also mention
[9:24] Speaker 1: there's a closely related field called zenobiology, which means alien
[9:29] Speaker 1: or foreign biology, and that term is usually reserved to
[9:33] Speaker 1: refer to biology that is synthetic, not found in nature
[9:38] Speaker 1: that science has no clue about yet. The general idea
[9:41] Speaker 1: is that astrobiologists try to detect and eventually analyze life
[9:47] Speaker 1: elsewhere in the universe, while xenobiologists will attempt to design
[9:52] Speaker 1: forms of life with a totally different biochemistry or different
[9:56] Speaker 1: genetic code than on planet Earth. Now, when we search
[10:00] Speaker 1: for life in the universe, I don't really see any
[10:03] Speaker 1: reason to imagine a distinction between naturally evolved life on
[10:08] Speaker 1: other planets and synthetic forms of life, because a planet
[10:12] Speaker 1: ten thousand years ahead of us would be well on
[10:15] Speaker 1: its way to building other species in the same way
[10:17] Speaker 1: that our species existence might simply serve as the spark
[10:22] Speaker 1: to build an artificial species that colonizes the Solar System.
[10:27] Speaker 1: And so for this reason we should always be thinking
[10:29] Speaker 1: about new biologies, new animals that could use a completely
[10:34] Speaker 1: different kind of biochemistry. On this planet, all we've ever
[10:38] Speaker 1: seen for coding information is DNA and RNA. We have
[10:43] Speaker 1: twenty amino acids from which we build all our proteins,
[10:47] Speaker 1: which are like little molecular machines. But instead of DNA,
[10:51] Speaker 1: we might find elsewhere what we'll call XNA zeno nucleic acid.
[10:56] Speaker 1: You might find a massively expanded genetic coat that uses
[11:00] Speaker 1: other amino acids to build totally new kinds of proteins,
[11:04] Speaker 1: or perhaps most likely, you could have something that's not
[11:07] Speaker 1: like our genetic code at all. Something to considers that
[11:10] Speaker 1: we only discovered the genetic code in nineteen fifty three,
[11:14] Speaker 1: which is not that long ago. I actually worked with
[11:16] Speaker 1: Francis Criek, the code discoverer of the structure of DNA,
[11:20] Speaker 1: So this is massively recent and we don't know what
[11:24] Speaker 1: we haven't thought of yet. So for this reason, I
[11:29] Speaker 1: think of the whole endeavor of space biology as zenobiology,
[11:34] Speaker 1: thinking about and looking for ways that life could be
[11:37] Speaker 1: built that we haven't yet imagined yet. And by the way,
[11:41] Speaker 1: it's thought exercises like this that make me wish I
[11:44] Speaker 1: could see the text books that our descendants will read
[11:49] Speaker 1: five hundred years from now or one thousand years from now.
[11:51] Speaker 1: There's going to be so much known that we currently
[11:56] Speaker 1: can't imagine. That's totally in our dark zone. Things we
[11:59] Speaker 1: have I haven't even realized that we don't know. Okay,
[12:03] Speaker 1: So all of this is to say life might be
[12:07] Speaker 1: massively different than what we have here on Earth, and
[12:10] Speaker 1: the question is will we figure out how to be
[12:14] Speaker 1: like Captain Kirk and beam onto their planet and have
[12:18] Speaker 1: a conversation with them.
[12:20] Speaker 2: Now, maybe all this talk.
[12:22] Speaker 1: About extraterrestrials seems abstract because we haven't discovered any yet.
[12:28] Speaker 1: But I want to point out that we are surrounded
[12:31] Speaker 1: by aliens. They are all around us, and they're making
[12:34] Speaker 1: constant sounds. We have measured this alien language with our recorders,
[12:41] Speaker 1: and at the moment we have no idea what they're saying.
[12:52] Speaker 2: Now, that's the.
[12:53] Speaker 1: Sound of whales on our planet, and lots of animals
[12:59] Speaker 1: make noise, And the question is, are these animals that
[13:03] Speaker 1: we're surrounded.
[13:04] Speaker 2: With speaking a language.
[13:07] Speaker 1: These are the aliens that we don't have to travel
[13:10] Speaker 1: very far to listen to, And the question is can
[13:14] Speaker 1: we understand them?
[13:15] Speaker 2: So maybe so, maybe not?
[13:17] Speaker 1: After all, these sounds could be like belches or like grunting,
[13:22] Speaker 1: the way you do when you're home alone and you
[13:24] Speaker 1: bang your knee the night before and you're going up
[13:26] Speaker 1: the stairs and you make a sound, but it's not
[13:29] Speaker 1: really meant for anyone in particular.
[13:31] Speaker 2: It's just a noise that you're making.
[13:33] Speaker 1: So how would we know if animals are actually implementing
[13:37] Speaker 1: language and communicating meaning to one another. Well, this is
[13:43] Speaker 1: an unanswered question right now, and it probably differs species
[13:47] Speaker 1: by species. One thing biologists look at to try to
[13:50] Speaker 1: address this is things like turn taking. Does one animal
[13:54] Speaker 1: make some sound and then the other animal goes, and
[13:57] Speaker 1: then the first one again, and then the second. That
[13:59] Speaker 1: feels more like language, right something had said, there's some response.
[14:04] Speaker 1: It feels like there's at least the possibility for some
[14:07] Speaker 1: real meaningful conversations that way. But there are a lot
[14:10] Speaker 1: of questions here. Even if we found that some species
[14:14] Speaker 1: were speaking language, would we be able to understand the meaning?
[14:20] Speaker 1: And I don't mean this in terms of this call
[14:23] Speaker 1: means this thing, but in terms of what does that
[14:26] Speaker 1: thing mean for a human? Imagine if a bee is
[14:31] Speaker 1: talking about some experience that you really have to see
[14:33] Speaker 1: in ultraviolet to understand, or your dog is experiencing something
[14:38] Speaker 1: about smell that we couldn't possibly get from our experience,
[14:42] Speaker 1: or a dolphin is talking about the joy of that
[14:45] Speaker 1: moment where there's no more little fish, and so the
[14:49] Speaker 1: whole pod suddenly turns upward and rockets through their world
[14:53] Speaker 1: and breaks through some surface where everything is different. Might
[14:57] Speaker 1: it be the case that there's simply no way we
[15:00] Speaker 1: could totally understand what they mean. There may be things
[15:03] Speaker 1: we can identify that they are talking about, but I
[15:06] Speaker 1: think there's a spectrum of how close we would actually
[15:10] Speaker 1: be in our interpretation. And by the way, I'll just
[15:14] Speaker 1: note this is true with our fellow humans as well.
[15:17] Speaker 1: Someone might tell you about an experience with hang gliding
[15:21] Speaker 1: or stamp collecting, or psychedelic drugs or whatever, and even
[15:26] Speaker 1: though you gnawd and you say, oh, I gotcha, I
[15:28] Speaker 1: can relate that to my own experience. There's a spectrum
[15:33] Speaker 1: of how closely you are actually capturing what they are describing.
[15:37] Speaker 1: Sometimes you might have an analogous experience that puts you close,
[15:41] Speaker 1: and sometimes your assumptions may be pretty distant. So back
[15:45] Speaker 1: to animals, how much could we even understand in a translation,
[15:50] Speaker 1: given that animals have such different sensory windows on the world,
[15:55] Speaker 1: and so their concepts might be very different from ours. Now,
[16:00] Speaker 1: even with all these caveats, how amazing would it be
[16:03] Speaker 1: if we could get even a low dimensional, blurry glimpse
[16:08] Speaker 1: of what they were talking about. People have always wanted
[16:12] Speaker 1: to understand animals, and they've tried in the past. But
[16:15] Speaker 1: we are at an amazing moment in history where I
[16:19] Speaker 1: think the time is right around the corner. And I'm
[16:22] Speaker 1: not saying this is a general claim. I'm saying this
[16:24] Speaker 1: for two specific reasons. First, we have incredible technology now
[16:30] Speaker 1: which makes it possible to do biologging. What is biologging.
[16:34] Speaker 1: This is about collecting data from an animal with a small,
[16:39] Speaker 1: lightweight device.
[16:40] Speaker 2: These are called biologgers.
[16:42] Speaker 1: You hook these up to an animal and that way
[16:44] Speaker 1: you can collect data for long windows of time without
[16:48] Speaker 1: humans being around. You can record sounds, and measure physiology
[16:53] Speaker 1: and track movements, and this is how you get a
[16:56] Speaker 1: secret window into an animal's world. Any thing is that
[17:00] Speaker 1: it gives new, rich data that you just can't get
[17:04] Speaker 1: otherwise about animals in their natural environment. Now, it's not
[17:08] Speaker 1: always easy to attach to the biologgers to animals, especially
[17:12] Speaker 1: if they're small and fast. But the main problem is
[17:15] Speaker 1: that the data collected by the biologgers can be really
[17:19] Speaker 1: complicated and difficult to analyze. So we humans have amassed
[17:24] Speaker 1: a ton of rich data that we're sitting on. And
[17:27] Speaker 1: that leads to the second specific reason why we're just
[17:31] Speaker 1: on the verge of something amazing, and that's artificial intelligence.
[17:51] Speaker 1: AI has been used for years to translate and decode
[17:56] Speaker 1: human languages, and now we have this incredible opportunity to
[18:00] Speaker 1: leverage it for understanding animal communication. And I'm lucky enough
[18:05] Speaker 1: to be friends with one of the people at the
[18:07] Speaker 1: lead of the effort to decrypt these alien languages that
[18:10] Speaker 1: were surrounded with So how does our modern technology give
[18:14] Speaker 1: us hope for decoding animal language.
[18:19] Speaker 3: So the insight originally came in twenty thirteen when I
[18:23] Speaker 3: was listening actually to NPR and there was a researcher
[18:27] Speaker 3: describing to a lot of monkeys.
[18:29] Speaker 1: That's Asa Raskin. He's a writer and entrepreneur and inventor,
[18:33] Speaker 1: and for the purposes of today, he founded the Earth
[18:36] Speaker 1: Species Project ESP, which is a nonprofit focused on using
[18:42] Speaker 1: AI to decode non human communication.
[18:50] Speaker 3: And these animals are They're credible. They live in the
[18:53] Speaker 3: Ethiopian highlands. They have like huge maines, red patches on
[18:57] Speaker 3: their chests. And what I did realize is that, according
[19:01] Speaker 3: to the researcher, they had one of the largest vocabularies
[19:04] Speaker 3: of any primate except for US humans. In fact, the
[19:08] Speaker 3: researchers swear that these animals talk about them behind their back.
[19:13] Speaker 3: And so the thought back then was like, well, if
[19:15] Speaker 3: people are just out there. The researchers are just out
[19:17] Speaker 3: there trying to understand what these beings are saying using
[19:23] Speaker 3: like a hand recorder in hand transcribing. Couldn't there be
[19:25] Speaker 3: a better way? Couldn't we use like machine learning AI
[19:28] Speaker 3: large scale microphone arrays. But of course, in twenty seventeen,
[19:33] Speaker 3: that wasn't yet possible because machine learning couldn't do something
[19:36] Speaker 3: that human beings couldn't already do. They couldn't translate a
[19:40] Speaker 3: language without a Rosetta stone, without any examples. And that
[19:45] Speaker 3: really changed in twenty seventeen when the machine learning community
[19:49] Speaker 3: there were two papers that came up back to back
[19:51] Speaker 3: and the Way this Thing Happens, that showed that you
[19:53] Speaker 3: could translate between any two human languages without the need
[19:58] Speaker 3: for examples or Rosetta stones. And I can dive into
[20:01] Speaker 3: how that works, but that was the moment that we
[20:03] Speaker 3: said we should get going and along the journey, I
[20:07] Speaker 3: just should say, like, all right, is there even a
[20:09] Speaker 3: there there? Like we say animal language, What does that mean?
[20:13] Speaker 3: What is a rich complex communication structure? What would that
[20:15] Speaker 3: look like? I just want to give a couple examples
[20:18] Speaker 3: for your listeners. So off the coast of Norway, every
[20:21] Speaker 3: year there's a group of false killer whales that all
[20:26] Speaker 3: phenomenologically speak one way, and a group of dolphins that
[20:29] Speaker 3: all speak another way, and they come together and they
[20:33] Speaker 3: hunt in a superpod. And when they do this, they
[20:37] Speaker 3: speak a third different way, which is just sort of crazy,
[20:43] Speaker 3: right like. And it turns out that whales, you know,
[20:45] Speaker 3: have a culture extending back thirty four million years. They
[20:48] Speaker 3: have dialects that sort of split off, which they can
[20:51] Speaker 3: understand each other with, and that can split all the
[20:53] Speaker 3: way into mutually unintelligible languages. Another example is I learned
[21:00] Speaker 3: this in twenty fourteen that for campell monkeys, hawk for
[21:03] Speaker 3: them means eagle, crack means leopard, and hawk ooh means
[21:06] Speaker 3: predator that's up, crack ooh means predator that's down. So
[21:09] Speaker 3: now we have a simple syntax. And then one of
[21:12] Speaker 3: my favorite studies is from the University of Hawaii in
[21:16] Speaker 3: nineteen ninety four, and here they taught dolphins two gestures,
[21:19] Speaker 3: and the first gesture was do something you've never done before,
[21:24] Speaker 3: which is sort of a crazy thing to be able
[21:25] Speaker 3: to communicate, but the dolphins will do it. And to
[21:27] Speaker 3: remember to do that, that means they have to remember
[21:30] Speaker 3: every single thing they've done before that session, understand the
[21:33] Speaker 3: concept negation not one of those things. Then invent whole cloth,
[21:36] Speaker 3: some new thing that they've never done before, but they
[21:39] Speaker 3: can do it. And then they'll teach the dolphins a
[21:41] Speaker 3: second gesture, do something you haven't done before. Together, and
[21:45] Speaker 3: they'll say, at the same time, do something you haven't
[21:47] Speaker 3: done before. Together. The dolphins go down, exchange sawnic information,
[21:51] Speaker 3: come up and do the same trick they've never done
[21:53] Speaker 3: before at the same time. And while that doesn't prove
[21:57] Speaker 3: representational language, it certainly places I think Auckham's razor. On
[22:01] Speaker 3: the other foot, it certainly seems that way. How do
[22:04] Speaker 3: you know, though, when you're approaching these that there exists languages?
[22:08] Speaker 3: For example, you said that whale culture or civilization is
[22:11] Speaker 3: thirty four million years old, But how do we know
[22:14] Speaker 3: that they're speaking a language that has the kind of
[22:16] Speaker 3: structure that we have that is capable of passing on
[22:20] Speaker 3: a culture or civilization. Yeah, great question, And of course
[22:25] Speaker 3: it's hard if you can't listen in and understand. But
[22:29] Speaker 3: you know, there are two hallmarks of language that human
[22:33] Speaker 3: beings have, one of which is to be able to
[22:36] Speaker 3: talk about something that isn't here and something that isn't now,
[22:40] Speaker 3: can you refer to things that are in a different time,
[22:42] Speaker 3: in a different place. And we can actually see already
[22:46] Speaker 3: from the research that the answer is. It appears to
[22:49] Speaker 3: be the case that at least some animals can do this.
[22:52] Speaker 3: So Adrian Lumero is a researcher on great apes, and
[22:56] Speaker 3: he's discovered in the last year or so that are
[23:00] Speaker 3: tangs do have a version of a past tense. They
[23:03] Speaker 3: can talk about things that are not now. And then
[23:06] Speaker 3: dolphins have names that they called each other by, and
[23:11] Speaker 3: Ian Yannick in twenty sixteen discovered that they will use
[23:14] Speaker 3: those names in the third person. They can talk about
[23:17] Speaker 3: one of their own that is not here. So now
[23:20] Speaker 3: we have two of the hallmarks not here not now. Now,
[23:25] Speaker 3: when we say language, we only have one example of
[23:28] Speaker 3: a species that speaks what we call language humans and
[23:31] Speaker 3: almost always like if you only know one color and
[23:34] Speaker 3: then you learn a second color, you discover an entire
[23:36] Speaker 3: rainbow in between. Like when we say language, you know
[23:40] Speaker 3: we're using that to be a catch all for rich
[23:43] Speaker 3: communication systems that can pass on cultural information.
[23:47] Speaker 2: Yeah, agreed.
[23:49] Speaker 1: Let me jump back to the orangutans for one second.
[23:51] Speaker 1: Is there any evidence that they have future tents.
[23:54] Speaker 3: That's a great question. We do not know. That's what's
[24:00] Speaker 3: so exciting about this field right now, is I think
[24:04] Speaker 3: of this it's sort of like the invention of the
[24:06] Speaker 3: Hubble telescope, right, It's like, and if you remember, I
[24:10] Speaker 3: think it was back in nineteen ninety five they pointed
[24:14] Speaker 3: the Hubble telescope and an empty patch in sky and
[24:16] Speaker 3: what they discovered was the most galaxies that have ever
[24:19] Speaker 3: existed in one spot. That's essentially what we're discovering here
[24:23] Speaker 3: is that we just haven't had the tools to look
[24:27] Speaker 3: and when we do, what we're discovering is much more
[24:31] Speaker 3: than everything we discovered. What we're discovering is everything.
[24:33] Speaker 1: Yeah, exactly right, the deep field experiment. So how do
[24:36] Speaker 1: we actually do it? How do we apply all the
[24:40] Speaker 1: modern tools of science to see if we can decode
[24:43] Speaker 1: a language?
[24:44] Speaker 2: Yeah?
[24:44] Speaker 3: Great question. So I'm going to start with this twenty
[24:47] Speaker 3: seventeen technology, and I just want your audience to remember
[24:50] Speaker 3: that twenty seventeen is essentially the Stone Age in AI.
[24:55] Speaker 3: But I think it's a really useful conceptual tool to
[24:57] Speaker 3: understand how it might work. So, how do you translate
[25:00] Speaker 3: between languages that don't have Rosetta stones. And it turns
[25:04] Speaker 3: out what you can ask AI to do is build
[25:07] Speaker 3: a shape that represents a language. So you say, feed
[25:09] Speaker 3: in all of Wikipedia, a whole bunch of text, and
[25:12] Speaker 3: the AI generates a shape that represents a language. Imagine
[25:16] Speaker 3: a galaxy where every star is a word, and words
[25:20] Speaker 3: that mean similar things are placed near each other, and
[25:23] Speaker 3: then words that share a sort of conceptual relationship get
[25:28] Speaker 3: turned into sharing a geometric relationship. What does that mean?
[25:31] Speaker 3: That means if you imagine king is to man as
[25:35] Speaker 3: woman is to queen, then in this shape, king is
[25:37] Speaker 3: the same distance direction to man as woman is to queen.
[25:42] Speaker 3: And so you actually just subtract king minus man. That
[25:44] Speaker 3: gives you a distance of direction. You add that to
[25:47] Speaker 3: boy and that'll equal prince. You add that to girl
[25:49] Speaker 3: eqal princess. You add that to woman and equal queen.
[25:53] Speaker 3: And so if you think about all of the relationships,
[25:56] Speaker 3: the internal relationships of a language, they think about the
[25:58] Speaker 3: word dog. Dog has relationship to man and to howl
[26:01] Speaker 3: and to wolf and to fer. If you it sort
[26:04] Speaker 3: of fixes in a point in space, and if you
[26:07] Speaker 3: solve this massive multi dimensional Sudoku puzzle of how every
[26:11] Speaker 3: concept relates to every other concept that gets turned into
[26:14] Speaker 3: a geometry, and out pops a rigid structure that represents
[26:18] Speaker 3: a language. Now the computer doesn't know what anything it means.
[26:20] Speaker 3: It just knows how they all relate to each other.
[26:22] Speaker 3: The shape represents all of the internal relationships of a language,
[26:26] Speaker 3: which is of course just a model of the world.
[26:29] Speaker 3: All right, So you have this shape for English, and
[26:32] Speaker 3: this is what the machine learners asked in twenty seventeen.
[26:35] Speaker 3: Is it possible? They said that the shape which is
[26:38] Speaker 3: English might be similar to or the same as, the
[26:41] Speaker 3: shape which is German. And if you ask anthropologists, they'd
[26:45] Speaker 3: be like, no, that's a silly thing to think, like.
[26:47] Speaker 3: They have different ways of viewing the world, different cosmologies.
[26:50] Speaker 3: But the machine is like, whatever, let's give it a try.
[26:53] Speaker 3: And it turns out that it works. You can take
[26:55] Speaker 3: the shape which is English and the shape which is German,
[26:58] Speaker 3: and literally rotate one shape on top of the other.
[27:01] Speaker 3: And even though there are words in one language that
[27:03] Speaker 3: don't appear in the other, if you blew your eyes,
[27:05] Speaker 3: the shapes are roughly the same, and the point, which
[27:09] Speaker 3: is dog ends up in the same in both. Now
[27:12] Speaker 3: you might be saying okay, but that's because English and
[27:14] Speaker 3: German are very similar languages. But it turns out this
[27:18] Speaker 3: works for Finnish, which is a really weird language, Turkish, Aramaic, Urdu.
[27:22] Speaker 3: Pretty much every human language fits in a kind of
[27:26] Speaker 3: universal human meaning shape, and the point, which is dog
[27:31] Speaker 3: ends up in the same spot in all of them,
[27:34] Speaker 3: and this lets you do translation without the need for
[27:37] Speaker 3: any examples. And this is I think, such a beautiful,
[27:42] Speaker 3: profound realization that there is a hidden structure underlying all
[27:48] Speaker 3: of us that unites our way of seeing. So that
[27:52] Speaker 3: was the sort of the core insight that said, well,
[27:56] Speaker 3: maybe now it's time to start building that shape for
[27:59] Speaker 3: animal communication, which by the way, is very hard because
[28:02] Speaker 3: it takes denoising and working with many partners to collect
[28:05] Speaker 3: like the years with the data that's required. But that's
[28:09] Speaker 3: sort of what we started to do. Now I'll pausit
[28:12] Speaker 3: for a second, but there are a couple other techniques
[28:14] Speaker 3: that can add to the top of this.
[28:15] Speaker 1: Great so let me jump in for one second. So
[28:17] Speaker 1: the fact that all the human languages have a similar
[28:20] Speaker 1: structure to them is in part because humans radiated out
[28:25] Speaker 1: of Africa sort of yesterday and as a result, you know,
[28:29] Speaker 1: we all have the same brain and it's not so surprising.
[28:33] Speaker 1: And the question is what do we expect when we're
[28:37] Speaker 1: looking at animal languages, which I'll come back to some
[28:40] Speaker 1: more questions on that a second, But what do we
[28:41] Speaker 1: expect in terms of the similarity there given that animals
[28:45] Speaker 1: are picking up on different signals from the world they're
[28:48] Speaker 1: umvelt is different the signals they can get and their
[28:52] Speaker 1: concepts might be very different.
[28:53] Speaker 2: How do you think about that?
[28:55] Speaker 3: This is a great question. It's just to repeat what
[28:57] Speaker 3: you're saying, is that the censorium, the way that animals
[29:01] Speaker 3: perceive the world, like what it is like to be
[29:03] Speaker 3: a bat, may be so completely different than what it
[29:06] Speaker 3: is like to be a human because they're seeing in
[29:08] Speaker 3: three D sound that we can never translate anything. And
[29:11] Speaker 3: that may turn out to be the case, but you know,
[29:16] Speaker 3: I think there's reason to believe that there may be
[29:19] Speaker 3: some kind of overlap with our experience. And to just
[29:22] Speaker 3: give a couple examples. You know, lemurs, for example, are
[29:29] Speaker 3: known to bite down on centipedes, literally to take a
[29:33] Speaker 3: hit off of centipedes to get high. They enter this
[29:36] Speaker 3: very trance like state, they get super cuddly. It looks
[29:40] Speaker 3: sort of like a scene from Burning Man. Dolphins too,
[29:43] Speaker 3: are known to intentionally inflate pufferfish to get high after
[29:48] Speaker 3: their venom and then pass them around literally puff pass.
[29:52] Speaker 3: Great apes are known to like hang off of vines
[29:54] Speaker 3: and spin to get dizzy. There is something about a
[29:58] Speaker 3: transcendent state of conscious altering our state that is at
[30:03] Speaker 3: least shared amongst the mammals, and so if they're communicating,
[30:07] Speaker 3: they may well communicate about that, and that's something we'd share.
[30:10] Speaker 3: Another example is something known as the mirror test. This
[30:14] Speaker 3: is a test where you take an animal, you paint
[30:18] Speaker 3: a dot on them where they can't see it. You
[30:20] Speaker 3: give them a mirror. They look in the mirror, they
[30:23] Speaker 3: see the dot, and they turn to the dot and
[30:25] Speaker 3: they try to brush it off of themselves or investigate it.
[30:28] Speaker 3: And in order for an animal to do that, they
[30:30] Speaker 3: have to associate the image that's in the mirror with themselves.
[30:33] Speaker 3: They have to look in the mirror and say like
[30:35] Speaker 3: that's me. So that means there's a rich sense of interiority,
[30:39] Speaker 3: like a self awareness. Dolphins past this test, elephants past
[30:44] Speaker 3: the tests. A number of other species pass this test.
[30:47] Speaker 3: So even the concept so profound as me self awareness
[30:51] Speaker 3: that seems to be shared. You know, examples of people
[30:56] Speaker 3: showing orangutangs magic tricks and they go crazy. It's worth
[31:01] Speaker 3: just looking them up on YouTube to see these kinds
[31:03] Speaker 3: of videos. Pilot whales carry they're dead young for three
[31:08] Speaker 3: four weeks, like grief is a shared part of the experience.
[31:12] Speaker 3: So if you imagine these shapes, where one of the
[31:15] Speaker 3: shapes is like human language, is one of these is
[31:17] Speaker 3: animal communication, I think we should expect to see some
[31:20] Speaker 3: part of those shapes overlap, and that should be the
[31:23] Speaker 3: part we should do direct translation. But then there's going
[31:27] Speaker 3: to be a huge portion of the shape that can
[31:29] Speaker 3: never be directly translated to human experience, and you'd sort
[31:31] Speaker 3: of expect that to be sticking out, like where we
[31:33] Speaker 3: can see complexity, but we don't know how to translate it.
[31:36] Speaker 3: And I still don't know which one of these two
[31:38] Speaker 3: is going to be more fascinating, the part where we
[31:39] Speaker 3: can directly translate the part we don't, because, as I'm saying,
[31:43] Speaker 3: human beings have been communicating vocally for one hundred thousand
[31:46] Speaker 3: to three hundred thousand years, passing up culture. Whales and
[31:49] Speaker 3: dolphins have been doing this for thirty four million years,
[31:53] Speaker 3: and that which is oldest correlates with that which is wisest.
[31:56] Speaker 3: So for something to survive thirty four million years, there
[32:00] Speaker 3: has to be some deep kernel of adaptive truth in there.
[32:04] Speaker 3: And whatever it is that is the solution to humanities problems, like,
[32:07] Speaker 3: it's not in our imagination, because if it is, we'd
[32:10] Speaker 3: probably be trying to do it. So this is a
[32:12] Speaker 3: way of starting to get the first polaroid sort of
[32:15] Speaker 3: blurry image pictures of that which is beyond our imagination.
[32:20] Speaker 2: Now, let me ask you this. If we're just looking
[32:23] Speaker 2: at the auditory.
[32:24] Speaker 1: Information that we get from animals, we can do this
[32:27] Speaker 1: kind of technique where we're looking to match one galaxy
[32:30] Speaker 1: of stars to the other galaxy of stars and see
[32:32] Speaker 1: what parts are sticking out and so on. But we
[32:35] Speaker 1: may well need much more than just the audio right
[32:39] Speaker 1: to understand the context of what the animal is saying
[32:42] Speaker 1: in a particular situation. So how are people pursuing that?
[32:47] Speaker 3: That's a great question. Even for humans, we know that
[32:51] Speaker 3: so much of the information that we convey. If you've
[32:53] Speaker 3: ever had to try to order food in a country
[32:56] Speaker 3: where you don't speak the language and somehow you can
[32:58] Speaker 3: do it, you can have communication without words the same
[33:03] Speaker 3: thing may be true for animals. So in fact, chimpanzees
[33:08] Speaker 3: are known to have sixty plus hand and feet gestures,
[33:10] Speaker 3: which seems to be at least as far as we know,
[33:12] Speaker 3: their predominant form of more symbolic communication.
[33:16] Speaker 1: Plus, we have indefinite references to things all the time, right,
[33:19] Speaker 1: So when I say she, I might be talking about
[33:22] Speaker 1: Marie Curie, or I might be talking about Michelle Obama
[33:25] Speaker 1: or something. But once I've introduced too I'm talking about
[33:28] Speaker 1: I can just use the word she. But somebody trying
[33:30] Speaker 1: to decode when I'm saying who doesn't speak English might
[33:34] Speaker 1: have a hard time understanding what the references to.
[33:37] Speaker 3: Yeah, that's exactly right, and what you're speaking to is
[33:41] Speaker 3: the importance of context. In order to understand what someone
[33:45] Speaker 3: is saying or what an animal is meaning, we have
[33:48] Speaker 3: to understand the context. Otherwise, like the same grunt may
[33:51] Speaker 3: mean like that monkey, or it may mean like I'm upset,
[33:54] Speaker 3: and it all sort of depends on social context. So
[33:57] Speaker 3: a lot of what we do now is multi mode.
[34:00] Speaker 3: That is to say, we work with biologists that have
[34:06] Speaker 3: tags on animals, that record often video, audio, and motion,
[34:13] Speaker 3: and that lets us begin to translate between all of
[34:16] Speaker 3: these different modalities, and in fact, with some of the
[34:19] Speaker 3: species to work with, these tags are on multiple animals
[34:22] Speaker 3: in the same group, so we can get social context.
[34:26] Speaker 3: And I want to return for a second to this
[34:30] Speaker 3: really interesting question you pose. You're like, well, maybe all
[34:33] Speaker 3: human languages fit in the same shape because we share
[34:36] Speaker 3: the same physical substrate, the same brains, in the same ears,
[34:39] Speaker 3: and the same eyes. But there's something deeper going on
[34:43] Speaker 3: in machine learning than just the ability to match the
[34:47] Speaker 3: shapes of languages. Maybe your audience has heard of or
[34:51] Speaker 3: seeing Dolly or mid journey or image diffusion where you
[34:56] Speaker 3: type in text and outcomes an image that is never
[35:00] Speaker 3: been seen before. How does that work? Well, these shapes
[35:05] Speaker 3: are actually really helpful to have in your mind to
[35:07] Speaker 3: understand how it works. So let's build now a shape
[35:10] Speaker 3: on human faces, and once again you end up with
[35:13] Speaker 3: a galaxy where every star now isn't a word, but
[35:16] Speaker 3: is a human face. Faces that share similar relationships share
[35:21] Speaker 3: geometric relationships. So if I take a picture of your face, David,
[35:25] Speaker 3: and then I take a picture of your face that's smiling,
[35:28] Speaker 3: there's this distance in direction that takes me between your
[35:30] Speaker 3: face and your faces just smiling. I subtract those two
[35:32] Speaker 3: as I get smilingness as a relationship. I can now
[35:35] Speaker 3: add that to any other face in the shape and
[35:39] Speaker 3: I'll get the smiling version of that face.
[35:41] Speaker 2: Right.
[35:41] Speaker 3: So now there's a direction that represents smiling, there's a
[35:43] Speaker 3: direction that represents frowning, that represents age, that represents gender,
[35:48] Speaker 3: more male, more female. You end up with a map
[35:51] Speaker 3: of all the semantic relationships, and you can now do
[35:55] Speaker 3: that not just for faces, you can do that for
[35:57] Speaker 3: all of images. And now you have a shape that
[35:59] Speaker 3: are presents images, a shape that represents languages. You look
[36:02] Speaker 3: at image caption pairs on the Internet and you can
[36:05] Speaker 3: align these two shapes. So now you have a way
[36:09] Speaker 3: of translating between text, language and images. So now you
[36:14] Speaker 3: just type in something like image I don't know, like
[36:17] Speaker 3: portrait of the country Chile as a woman. It goes
[36:22] Speaker 3: into the language space, gets translated to the image space,
[36:25] Speaker 3: the computer generates the image that's there, and you get that.
[36:27] Speaker 3: So that's how that technology works. There's something really deep
[36:31] Speaker 3: actually happening because it's not just working on language. It
[36:34] Speaker 3: seems to work on almost any modality out there. And
[36:39] Speaker 3: I think just like there is the unreasonable effectiveness of
[36:43] Speaker 3: mathematics where it's seems very strange. You go out on
[36:46] Speaker 3: some branch of abstract mathematics, it seems like it has
[36:49] Speaker 3: nothing to do with the world, and then it has
[36:51] Speaker 3: something profound to say about the world. You invent complex
[36:54] Speaker 3: numbers somehow that describes everything you're going to need to
[36:57] Speaker 3: deal with electricity. Who knew that's going on? In deep learning,
[37:02] Speaker 3: where there's an unreasonable effectiveness of deep learning, where the
[37:04] Speaker 3: same techniques are working across every modality from DNA to
[37:11] Speaker 3: fMRIs to language to audio to video to images to
[37:16] Speaker 3: computer code. There's nothing that says that had to have worked,
[37:20] Speaker 3: and yet it seems to be working. So we're learning
[37:23] Speaker 3: something I think profound about the structure of our universe.
[37:27] Speaker 3: But what that means for us specifically is that that
[37:30] Speaker 3: means we can build these kinds of shapes and embed
[37:32] Speaker 3: and translate between how an animal behaves and how it
[37:37] Speaker 3: sounds and what its body post is. So we can
[37:39] Speaker 3: say we're not quite there yet, but we're moving towards it.
[37:42] Speaker 3: Generate me the audio of two elephants coming together, and
[37:46] Speaker 3: that's going to view distribution of calls, some of which
[37:48] Speaker 3: might mean like hello, some of which might mean, this
[37:50] Speaker 3: is my name, some of which might mean I've missed you,
[37:52] Speaker 3: you don't know, but has something to do with affiliation.
[37:54] Speaker 3: Then you say, okay, now generate me the audio of
[37:56] Speaker 3: two elephants coming together, but where one of them's flapping
[37:58] Speaker 3: its ears and the other one's running quickly. What kind
[38:00] Speaker 3: of sounds does that make? And you can see that
[38:02] Speaker 3: this becomes a laboratory that lets you very quickly iterate
[38:07] Speaker 3: to understand what animals are saying. When you get to
[38:10] Speaker 3: do this in combination with the incredible biologists that are
[38:14] Speaker 3: out in the field and have already built up a
[38:15] Speaker 3: lot of that context from time, blood, sweat, and tears.
[38:20] Speaker 1: Excellent, and I think the biologging is becoming more sophisticated,
[38:24] Speaker 1: even right where they're looking at temperature and weather patterns
[38:28] Speaker 1: and gyroscopes and accelerometers and so on, where you get
[38:31] Speaker 1: all of this data from the animals, and in theory,
[38:35] Speaker 1: there's no limit to how much we can biologue, as
[38:37] Speaker 1: long as we can make it small and portable and
[38:39] Speaker 1: gets it on the animals, and then we're able to discover, hey,
[38:44] Speaker 1: these are the contextual cues that the animal is responding
[38:47] Speaker 1: to with the language.
[38:50] Speaker 3: Yeah, that's exactly right. And in fact, a big shift
[38:53] Speaker 3: that's happened in biology and conservation and ethology in the
[38:58] Speaker 3: last five years is because of cell phones driving the
[39:04] Speaker 3: cost of sensors lower. Biologists have gone from a world
[39:08] Speaker 3: where they're often data starved to where they're data drowned,
[39:11] Speaker 3: where they have access to terabytes of data, but they
[39:15] Speaker 3: don't yet have the tools to understand them. And so
[39:20] Speaker 3: our goal is to decode non human communication, translate animal language,
[39:26] Speaker 3: and use that to transform our relationship with the rest
[39:29] Speaker 3: of nature. But it's sort of like you're trying to
[39:31] Speaker 3: go to the moon along the way you invent velcro.
[39:33] Speaker 3: We're building the foundational tools that every biologist needs to
[39:39] Speaker 3: understand the data that they have now. And our hope
[39:43] Speaker 3: is that by building those foundational tools or nonprofit or
[39:47] Speaker 3: open source, we try to give back as much as
[39:49] Speaker 3: we can that that can broad scale accelerate all of
[39:53] Speaker 3: like conservation science, which we hope can also accelerate conservation itself.
[39:57] Speaker 1: Now, I have a technical question, which is if you
[40:00] Speaker 1: are looking at human languages and making this high dimensional
[40:05] Speaker 1: space of all the words when you're listening, when you're
[40:09] Speaker 1: eavesdropping on whales or lemurs or whatever species. How do
[40:13] Speaker 1: you know what a word is, what a unit of
[40:16] Speaker 1: meaning is.
[40:18] Speaker 3: Yeah, a hard problem and there's no one easy solution.
[40:23] Speaker 3: But one of the things that we can ask the
[40:26] Speaker 3: AI to do is try chopping up the audio in many, many,
[40:32] Speaker 3: many different ways and then see which one of those
[40:36] Speaker 3: ends up making good predictions for what comes next. And
[40:40] Speaker 3: so you can see if you're trying and varying and
[40:43] Speaker 3: you're not saying, well, which thing contains meaning, but which
[40:46] Speaker 3: things help make good predictions. When you try this on humans,
[40:49] Speaker 3: you end up with phonemes that you get out and
[40:54] Speaker 3: then those are then combinatorily built into words. So we're
[40:57] Speaker 3: playing with those kinds of techniques, but we don't have
[41:00] Speaker 3: like one surefire away yet.
[41:02] Speaker 1: And when you're thinking about predictions, one of the ways
[41:05] Speaker 1: that you could test a prediction is with playback, right,
[41:08] Speaker 1: So tell us about that.
[41:11] Speaker 3: Yeah, So this is the classic way that you test
[41:15] Speaker 3: your predictions in the field, where I'll just go out
[41:18] Speaker 3: and they will play a sound, often from the animal,
[41:22] Speaker 3: and they'll see whether the animal looks and for how long.
[41:26] Speaker 3: What we are starting to be able to do is
[41:29] Speaker 3: just like you can build a chat bot in text
[41:33] Speaker 3: that speaks Chinese without needing to speak Chinese. We are
[41:37] Speaker 3: on the cusp of being able to build these kinds
[41:40] Speaker 3: of chat bots, but that just directly speak in the
[41:44] Speaker 3: language of animals. So it's sort of like, imagine you
[41:47] Speaker 3: had a superpower, and your superpower was to go out
[41:50] Speaker 3: meet someone whose language you don't understand. You sort of
[41:52] Speaker 3: cock your head to the side and you listen for
[41:54] Speaker 3: a little bit and you're like, I don't know what
[41:56] Speaker 3: anything means, but I see that this sound pattern follows
[41:58] Speaker 3: this sound platted and this context. You just start to
[42:01] Speaker 3: babble and you have no idea what you're saying, but
[42:04] Speaker 3: the other person's like crosses the arm, like yeah, wow,
[42:07] Speaker 3: that's so meaningful. And at the end the person walks away.
[42:10] Speaker 3: I think they've had a great conversation. You're like, I
[42:12] Speaker 3: have no idea what I just said. I was just babbling.
[42:15] Speaker 3: But that's what that's literally what Chatchipdi does, and that's
[42:18] Speaker 3: the kind of thing that we are going to be
[42:20] Speaker 3: able to build in the next you know, like twelve months.
[42:24] Speaker 3: And what does that mean?
[42:25] Speaker 1: Just so it's clear, can you give an example of
[42:28] Speaker 1: playback and the kind of things that people are doing
[42:30] Speaker 1: right now with that yeah.
[42:33] Speaker 3: So an example of a playback might be one of
[42:36] Speaker 3: our partners, Michelle Fournet, and you can, actually your listeners
[42:40] Speaker 3: can go watch her incredible documentary Fathom. She was trying
[42:45] Speaker 3: to determine how do you say hello to humpback whale
[42:49] Speaker 3: and possibly include their name, So to say hello in
[42:53] Speaker 3: hump back, it turns out, is something like poop And
[42:57] Speaker 3: to test this, she recorded many different they're called whoop calls,
[43:02] Speaker 3: but they're hellos, many different whoop calls, and then went
[43:06] Speaker 3: out to Alaska, set up speakers underwater and would play
[43:11] Speaker 3: the hellos in a very controlled condition and would see
[43:14] Speaker 3: do the humpbacks respond? And the answer is yes, yes
[43:19] Speaker 3: they do. When she said hello, they would respond in
[43:23] Speaker 3: greater number of saying hello back. So that's an example
[43:27] Speaker 3: of a playback experiment.
[43:44] Speaker 1: I know you've thought a lot about the ethics involved
[43:48] Speaker 1: in this so far what we've been talking about sounds amazing,
[43:51] Speaker 1: and the question is what are the ethical things that
[43:53] Speaker 1: we need to keep an eye on.
[43:55] Speaker 3: Yeah, that is a great question because you know we
[44:00] Speaker 3: are going to be crossing this barrier very very soon,
[44:05] Speaker 3: which is I mean, this is the plot twist that
[44:08] Speaker 3: we will be able to communicate before we fully understand
[44:12] Speaker 3: what we're saying. That's again very surprising. I would have
[44:17] Speaker 3: not have guessed this if we rewound the clock three
[44:20] Speaker 3: or four years. What does this mean. This means that
[44:26] Speaker 3: if you're working with a species which has vocal learning,
[44:31] Speaker 3: well you might inject something that they say that then
[44:35] Speaker 3: changes their culture. So, to give an example, humpback whales
[44:39] Speaker 3: off the coast of Australia. For whatever reason, the Australian
[44:43] Speaker 3: humpbacks seem to be like the K pop singers, and
[44:46] Speaker 3: because they can sing halfway across an ocean basin and
[44:49] Speaker 3: they migrate across the world, often the songs that are
[44:54] Speaker 3: sung off the coasts of Australia will catch on and
[44:57] Speaker 3: be sung by much of the world population within a
[45:00] Speaker 3: couple of seasons. So it's, you know, the ultimate pop tune. Now,
[45:04] Speaker 3: we don't know as humans what truly the function of
[45:09] Speaker 3: humpback whale song is and how that culture works. So
[45:12] Speaker 3: if we just create a synthetic whale that sings, we
[45:17] Speaker 3: may infect a thirty four million year old wisdom tradition,
[45:20] Speaker 3: you know, create some kind of viral meme a whale QAnon.
[45:25] Speaker 3: We just don't know. So we have to be very
[45:30] Speaker 3: careful as we approach this new responsibility of what does
[45:35] Speaker 3: it mean to truly communicate with the other cultures of Earth?
[45:40] Speaker 3: And that means we should not go out and start
[45:43] Speaker 3: playing like two way communication real time. We should not
[45:47] Speaker 3: do those kinds of experiments with wild populations that vocally learn.
[45:51] Speaker 3: We have to think about what is it to have
[45:53] Speaker 3: like a prime direct of a Geneva convention for cross
[45:57] Speaker 3: species communication. And this is of course terrifying, and I
[46:00] Speaker 3: should say, everything that our species does we do with
[46:03] Speaker 3: biology partners and institutions. We are starting to work on
[46:08] Speaker 3: what are the ground rules before we even have the
[46:11] Speaker 3: technology for knowing when and how it is okay to
[46:17] Speaker 3: have these prime directive first contact moments because first contexts
[46:21] Speaker 3: have often not gone well for the beings being first contacted.
[46:25] Speaker 3: So I think the change in the relationship for how
[46:30] Speaker 3: we relate to nature is the point of our species
[46:33] Speaker 3: and it's exciting that we're getting to the place where
[46:36] Speaker 3: that becomes a necessity.
[46:38] Speaker 1: And what's your prediction for how long it'll be, what
[46:42] Speaker 1: year will have a meaningful conversation back and forth with
[46:46] Speaker 1: the species? And which do you think will be the
[46:50] Speaker 1: first species?
[46:52] Speaker 2: Yeah?
[46:52] Speaker 3: I mean This is science, so it's always very hard
[46:54] Speaker 3: to make predictions like this, and different people on my
[46:58] Speaker 3: team have different predictions, so I can just say mine,
[47:01] Speaker 3: but know that answers very I think certainly by twenty
[47:06] Speaker 3: thirty we will have had two way back and forth
[47:09] Speaker 3: to what degree we understand unknown, But I think we
[47:14] Speaker 3: will have a really good handle on it by then.
[47:19] Speaker 3: It's just so exciting. My personal favorite is Belugas, and
[47:24] Speaker 3: again everyone has their own personal favorite. But when you
[47:27] Speaker 3: listen to Belugas communicate, it sounds like an alien modem.
[47:31] Speaker 3: It sounds digital. There are lots of whistles in there.
[47:35] Speaker 3: It turns out that you know dolphins, they say their name,
[47:37] Speaker 3: their signature, whistle in a whistle. It's like a single band.
[47:41] Speaker 3: This is like full modem pack. It encodes their name,
[47:44] Speaker 3: it encodes their clan identity. And doctor Valeria Vergaro, with
[47:49] Speaker 3: whom we work on Beluga communication, she's sort of like
[47:52] Speaker 3: one of the preeminent scholars. It was her work that
[47:55] Speaker 3: showed that they have names that they call each other by.
[47:57] Speaker 3: And what blew my mind is that when she talks
[48:00] Speaker 3: about her data, she's like she had to throw away
[48:04] Speaker 3: ninety seven percent of her data in those studies because
[48:08] Speaker 3: she couldn't tell which beluga was speaking or disentangle them.
[48:12] Speaker 3: And that's because they are like forty belugas in a
[48:14] Speaker 3: tight mass that are moving around super fast. It's very
[48:17] Speaker 3: hard from a computational perspective. But that's where your listener's
[48:22] Speaker 3: ears should perk up, because here we have the most
[48:24] Speaker 3: vocal underwater species with the largest vocabulary that we know of,
[48:29] Speaker 3: and the super majority of data, like ninety seven percent
[48:32] Speaker 3: is unknown the ocean is what five percent explored Bluga
[48:35] Speaker 3: communication or at least this data sets are three percent explored.
[48:38] Speaker 2: Like this is.
[48:38] Speaker 3: Where you get like brand new discoveries. This is the
[48:41] Speaker 3: next frontier.
[48:42] Speaker 1: And do Belugas do turn taking, by the way, which
[48:45] Speaker 1: is one of the signatures of an actual language as
[48:47] Speaker 1: opposed to just broadcasting noise.
[48:50] Speaker 3: Yeah, a number of species to do turn taking, from
[48:53] Speaker 3: parrots to gelatas to many of the whale species.
[48:58] Speaker 2: Yeah, I know.
[48:59] Speaker 1: This is one of the signs that people look at
[49:02] Speaker 1: to try to figure out, how would we know if
[49:04] Speaker 1: this is actually a language versus they're just singing songs
[49:07] Speaker 1: or they're doing whatever, but they're not listening back and forth,
[49:10] Speaker 1: which leads to this question about if we find alien species,
[49:15] Speaker 1: eventually we find life on other planets. The question is
[49:19] Speaker 1: how much do we have to share with another species
[49:22] Speaker 1: for us to have some meaningful interpretation of the language,
[49:27] Speaker 1: Because fundamentally we're trapped in our internal model and a
[49:31] Speaker 1: species that's so different, we will impose an interpretation on
[49:36] Speaker 1: what they must mean by it. But I wonder when
[49:40] Speaker 1: we find an alien species, how we will ever be
[49:44] Speaker 1: able to know whether we understand enough of their language
[49:48] Speaker 1: to say that we have a meaningful interpretation of it.
[49:52] Speaker 3: I mean, when you say that, it just makes me wonder,
[49:54] Speaker 3: how do we ever know that when we're communicating with
[49:57] Speaker 3: each other's as humans, that we truly understand each other.
[50:00] Speaker 3: There's almost this undeniably huge and yet invisible gulf, like
[50:05] Speaker 3: the myth of communication is that it ever happened in
[50:08] Speaker 3: the first place. We never truly know. We can only
[50:11] Speaker 3: have clues that we are getting closer, that we're approaching knowing.
[50:17] Speaker 3: I think it's really important to call out how much
[50:20] Speaker 3: of our language is built on the metaphor of bodies.
[50:23] Speaker 3: Almost all of it is body and space, right Like,
[50:28] Speaker 3: even the things that we might think are really abstract,
[50:30] Speaker 3: like cursor on your computer. What is the root of
[50:33] Speaker 3: cursor in Latin, It's cursor the one who runs. It's
[50:36] Speaker 3: the man who runs impeded, impeded against foot. It's like
[50:42] Speaker 3: the deeper you look into language, the more you realize.
[50:46] Speaker 3: And George Lakeoff does an incredible job in a book
[50:48] Speaker 3: called Metaphors. We live by really deconstructing all of the
[50:53] Speaker 3: ways that what we think of as our most abstract
[50:56] Speaker 3: ideas can be traced back to a root of us
[50:59] Speaker 3: having bodies and talking about our bodies in a physical world.
[51:01] Speaker 2: And particular bodies, particular bodies. That that is true.
[51:05] Speaker 1: What I mean is when we find an alien species,
[51:07] Speaker 1: let's say they're more like slime, mold or something that
[51:09] Speaker 1: might make it very difficult for us to understand their metaphors.
[51:14] Speaker 3: That is exactly right. And I think the hope here
[51:18] Speaker 3: is that because we are conditioned on and live in
[51:22] Speaker 3: a physical world, that to the extent that there is
[51:24] Speaker 3: an outside world in which we share, and that that
[51:29] Speaker 3: will give the kind of grounding that's needed to do
[51:33] Speaker 3: some kind of translation. But I think it would be
[51:36] Speaker 3: wrong to say that the translations are going to look
[51:39] Speaker 3: like Google Translate. You're going to get word forward to
[51:41] Speaker 3: English it might end up looking like a translation is
[51:43] Speaker 3: more like a piece of art or poetry, where the
[51:47] Speaker 3: translation is very ambiguous, but if you spend enough time
[51:51] Speaker 3: with it, you start to get a felt sense of
[51:53] Speaker 3: what it's like. Or maybe you're right, maybe it'll be
[51:55] Speaker 3: so different that we'll never There's just some things we
[51:57] Speaker 3: will never be able.
[51:58] Speaker 1: To work right in between, will and pose an interpretation
[52:01] Speaker 1: even though it will be incorrect.
[52:04] Speaker 3: Yeah, that is true, And at least here on Earth.
[52:08] Speaker 3: There are sort of two failure modes. One is anthropomorphizing,
[52:11] Speaker 3: which is what you're talking about, is assuming that we can, well,
[52:15] Speaker 3: we can only relate to the experience of others through
[52:17] Speaker 3: our own experience. That's the only way to ever happened.
[52:19] Speaker 3: It's very simple to say, but it's actually profound when
[52:21] Speaker 3: you think about it. So there's a over projection of
[52:25] Speaker 3: ourselves onto others. And then the other side is human exceptionalism,
[52:29] Speaker 3: where we assume that our experiences are completely unique to
[52:32] Speaker 3: ushen we share nothing with other animals. And obviously the
[52:35] Speaker 3: answer that the truth is in between the two. And
[52:38] Speaker 3: then we have to have the self honesty to understand
[52:42] Speaker 3: and the way of asking questions that lets us determine
[52:45] Speaker 3: when we are over projecting.
[52:47] Speaker 1: Yes, exactly, And I'm really interested when and this might
[52:50] Speaker 1: not happen in our lifetimes, but when we discover completely
[52:54] Speaker 1: alien life, I mean as in living on other planets
[52:59] Speaker 1: that might be so different.
[53:00] Speaker 2: Maybe they don't have DNA, maybe they.
[53:02] Speaker 1: Have a different coding system, maybe they have very different bodies.
[53:07] Speaker 1: The question is how much is Earth exceptionalism true? You know,
[53:13] Speaker 1: in Star Trek they go around and they communicate with
[53:16] Speaker 1: all these aliens and they have a good time, and
[53:18] Speaker 1: you know, really understand each other to some degree. And
[53:22] Speaker 1: the question is whether that will be the case or not.
[53:26] Speaker 3: Yeah, I mean if you start thinking about I think
[53:30] Speaker 3: in Star Trek they have the crystalline entity, which is
[53:33] Speaker 3: a giant being the size of a whole planet. And
[53:36] Speaker 3: at that point, I think I'd come a little more
[53:39] Speaker 3: along your lines that the scale of which that being
[53:42] Speaker 3: is feeling and sensing is so broad. We probably share
[53:45] Speaker 3: very little but anything of roughly our size. And if
[53:48] Speaker 3: they have family structures, like then there is like hunger,
[53:53] Speaker 3: there's being tired, there's like safety, there's like familiar relationships,
[54:01] Speaker 3: there is gossip, and those things are probably conserved across many,
[54:05] Speaker 3: many different types of beings.
[54:11] Speaker 1: So we're entering a really exciting time, but the challenges
[54:15] Speaker 1: are real and there are still a lot of question marks.
[54:18] Speaker 1: For example, in the scientific literature, there's an ongoing debate
[54:22] Speaker 1: about which species might have languages. Some researchers listen to
[54:27] Speaker 1: a particular species and say that seems like that could
[54:30] Speaker 1: be language, and others listen and they say, no, way,
[54:33] Speaker 1: that's not language because there's no turn taking, and also
[54:37] Speaker 1: because the order of the sounds doesn't seem to make
[54:40] Speaker 1: any difference. And these are all valid debates because we
[54:43] Speaker 1: don't actually know what qualifies as a language and what doesn't.
[54:48] Speaker 1: Some species, for example, some songbirds do what's called dueting,
[54:53] Speaker 1: where they're singing at the same time. Does this mean
[54:56] Speaker 1: they're not doing language or is it possible? Are very
[55:00] Speaker 1: different ways of doing language. I'll give you a concrete
[55:04] Speaker 1: example of a different way of doing language, which is
[55:08] Speaker 1: sign language. It turns out that the temporal order doesn't
[55:12] Speaker 1: matter very much. In sign language. You can switch up
[55:14] Speaker 1: the order of the words and it can still mean
[55:17] Speaker 1: the same thing. And there are aspects of it that
[55:20] Speaker 1: are spatial. So, for example, an American sign language, you
[55:24] Speaker 1: can indicate that something happened in the past by doing
[55:27] Speaker 1: the signs slightly to your left, and if you make
[55:30] Speaker 1: the same signs over on your right side, that means
[55:33] Speaker 1: you're talking about the future. So it's the same signs
[55:37] Speaker 1: with this subtly different spatial position, and it can mean
[55:41] Speaker 1: different things. And I call it subtle because if someone
[55:44] Speaker 1: didn't know to watch for a slight spatial change, they
[55:48] Speaker 1: wouldn't even notice it. And even a language that only
[55:51] Speaker 1: uses sounds can be very difficult to decode because so
[55:56] Speaker 1: much of it depends on shared assumptions about meaning. So
[56:00] Speaker 1: just as an example, if I'm talking about someone named Aviva,
[56:04] Speaker 1: I use her name once, and in the next sentence
[56:07] Speaker 1: I just say her, and you know who I'm talking about.
[56:10] Speaker 1: I'm referencing Aviva. But if you are an alien working
[56:14] Speaker 1: to decode my language, you might be confused because one
[56:17] Speaker 1: minute later you hear me use the same utterance her,
[56:21] Speaker 1: but now I'm referring to someone else entirely.
[56:23] Speaker 2: I'm now talking about Sarah, but I.
[56:25] Speaker 1: Still use the word her, So the same word can
[56:29] Speaker 1: refer to totally different things. And the alien would be
[56:32] Speaker 1: very confused if it had concluded that her was the
[56:37] Speaker 1: word for Aviva, and in the same way when we
[56:40] Speaker 1: hear a whale make the same sound that we always hear,
[56:43] Speaker 1: it might be talking about something totally different than the
[56:46] Speaker 1: last time that used that sound. The context matters, and
[56:51] Speaker 1: this issue of context, in other words, what's going on
[56:54] Speaker 1: around the animal. This is why biologuers are interested in
[56:58] Speaker 1: collecting things beyond and just the audio data. Good biologuing
[57:03] Speaker 1: now uses video and gyroscope and altimeter and GPS and
[57:08] Speaker 1: any other measure they can get their hands on. And
[57:11] Speaker 1: this matters because so much of communication is about context,
[57:16] Speaker 1: and by the way, a lot of it is nonverbal.
[57:19] Speaker 1: Consider how you pick up stress from someone else even
[57:22] Speaker 1: without words, body language, the tightness of their facial muscles,
[57:27] Speaker 1: the way they're walking, and so on, and animals presumably
[57:31] Speaker 1: have many equivalents to this. Just think about smells and pheromones.
[57:37] Speaker 1: Take a close look at your dog the next time
[57:39] Speaker 1: you're on a walk. It's obvious that a lot of
[57:42] Speaker 1: your dog's language is happening silently. So all this is
[57:47] Speaker 1: to say that language can be complicated, and much of
[57:51] Speaker 1: it can be nonverbal, and this is why the challenge
[57:55] Speaker 1: of decoding animal language is a big one. And you know,
[57:59] Speaker 1: one of the things that I'm always on the lookout
[58:02] Speaker 1: for is whether we can see any evidence that animals
[58:05] Speaker 1: engage in something like storytelling. One of the classes that
[58:09] Speaker 1: I teach at Stanford is the Brain and Literature, and
[58:13] Speaker 1: I teach how weird it is that we go to
[58:15] Speaker 1: the theater or the movies or a lecture and someone
[58:18] Speaker 1: speaks and whosh, we get immediately transported into a different
[58:23] Speaker 1: space and time.
[58:24] Speaker 2: It's like a guided dream.
[58:27] Speaker 1: I'm going to do an episode on this issue soon,
[58:29] Speaker 1: but for now, I just want to point out that
[58:31] Speaker 1: we don't see bears congregating like hundreds of them on
[58:35] Speaker 1: a Saturday night listening to one bear grunt along. And
[58:40] Speaker 1: I'm not certain that we see that in any species,
[58:42] Speaker 1: but I don't know. But these are the kinds of
[58:45] Speaker 1: clues we would look for as we move forward. These
[58:48] Speaker 1: are the questions of not just do they have some
[58:51] Speaker 1: simple language, but what they can do with their language.
[58:54] Speaker 1: This is a tougher problem and one that we need
[58:57] Speaker 1: to keep our eye on. So plenty of remaining question
[59:01] Speaker 1: marks all around us, but what's clear is that technology
[59:05] Speaker 1: like biologuers and artificial intelligence are leveling us up into
[59:10] Speaker 1: a very exciting time. Not all species are going to
[59:14] Speaker 1: have something interesting to say, but many might. And if
[59:18] Speaker 1: we find we can decode animal language due to the
[59:22] Speaker 1: labors of Aseraskin and his co founders Katie and brit
[59:26] Speaker 1: and dozens of other people in this exciting field of
[59:28] Speaker 1: animal communication, that will give us a very different view
[59:33] Speaker 1: of ourselves and our species on this planet. Our grandchildren
[59:38] Speaker 1: will grow up and they'll feel amazed that we considered
[59:42] Speaker 1: ourselves the only ones and it wasn't even necessarily because
[59:46] Speaker 1: of species chauvinism, but instead because we can only hear
[59:51] Speaker 1: our own voices, and therefore we thought we were the
[59:53] Speaker 1: only ones in the room. And with enough time, maybe
[59:57] Speaker 1: we'll have enough technology and practice at decoding animal languages
[1:00:02] Speaker 1: that eventually, in the more distant future, we can tackle
[1:00:07] Speaker 1: extra planetary communication, and our great great grandkids will be
[1:00:12] Speaker 1: amazed that there was a time when we thought we
[1:00:15] Speaker 1: were the only ones in the galaxy.
[1:00:18] Speaker 2: We maybe look back upon.
[1:00:19] Speaker 1: As the era of loneliness, surrounded by voices of all
[1:00:25] Speaker 1: types that we just didn't know how to hear.
[1:00:34] Speaker 2: Please join me at eagleman.
[1:00:36] Speaker 1: Dot com, slash podcasts more information and links to various
[1:00:40] Speaker 1: animal communication projects and further reading. Send me an email
[1:00:44] Speaker 1: at podcasts at eagleman dot com with questions or discussion,
[1:00:47] Speaker 1: and I'll be making an episode soon in which I
[1:00:50] Speaker 1: address those. Until next time, I'm David Eagleman, and this
[1:00:54] Speaker 1: is Inner Cosmos.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.