A very interesting read, which describes how for ~50 years the community was lead astray by a seemingly simple oversight: studying dead animals instead of living ones!
>Charles Darwin theorized that monkeys had the vocal anatomy needed for speech, but lacked the necessary neural mechanisms.
>This was the most popular hypothesis until 1969 [when] Lieberman studied the vocal anatomy of a monkey corpse and concluded that other primates could not produce as many vowels as humans because of the position of their larynxes [which] cemented the idea that a descended larynx is a prerequisite for speech.
>Towards the end of the 20th century, evolutionary biologist W. Tecumseh Fitch realized a crucial fact – all the existing evidence was based on the anatomy of dead primates. Surprised that this had not been done before, Fitch used x-ray imaging to study the vocal tracts of live animals while they were voicing sounds. He was amazed to observe that their larynxes at rest remained high and then descended during vocalization to a position very similar to the human larynx.
>All these studies indicate that apes have all the anatomical characteristics necessary for speech. The reason they don’t is purely neural. Humans have much better control of the larynx, not because of its position, but because of the neural connections that connect it to the brain. Parrots don’t even have a larynx, but they have wonderful control of their speech organ, which enables them to articulate intelligible words and phrases.
I'm pretty sure descended larynx is not prequisite for producing vowels.
Vowels are merely overtones which occurs because the waves generated in the vocal folds(with a charachtrized fundamental frequency) travel the vocal tract in 3d, which causes destructive interference for most frequencies except those with constructive interference (overtones / formants)
When humans move that tongue and lips we basically change the vocal tract shape which changes the overtones, the larynx is behind the lips and behind/in the back of the tongue anyway.
"require" is a fuzzy term. If you force your lips and tongue to stay in place, you can get partial intelligible vowel sounds from your larynx, and feel your larynx changing as vowels change.
Human exceptionalism has poisoned so many studies.
My pet-theory why primates do not vocalize is that the neural difference is not even that cemented, it has to be some difference at the language acquiring stage as babys and children, were one human pattern matching center for sound fires constantly, acquiring the language "keys" and theirs dont.
20th century hubris poisoned so many studies I would rather say.
I have this example but there are literally hundreds:
You know about the 5 μm particle size limit for it to become airborne? Washing hands for protecting against aerosols? It’s parroted bad research; and wrong.
The main infection path of upper respiratory diseases is through aerosols. The medical literature says it’s through hands that are touching the mouth and the nose but this is wrong. It was shown in the 1940s that viruses can spread through aerosols and that aerosols are particles under 100 μm. Research was done on tuberculosis. Tuberculosis is caused by bacteria that must infect the deep inside parts of the lung. For that it needs an aerosol size of 5 μm. Other diseases can do with aerosols of 5-100 μm.
Now, a meme was born that only 5 μm particles can get airborne.
And this was just accepted by the Western medical canon and the WHO.
Monkeys have a much easier time learning sign language than spoken, so there's probably something about vocal chords too. I suspect it's because humans and parrots independently developed music as a bonding mechanism and apes did not.
They did develop fine control in their eyebrows to appeal / communicate to humans. And cats developed mewling similar to human babies to draw their attention.
We've been building tech to translate dog barks to something meaningful for us to understand.
A few things we observed:
1. Dogs are more communicative once they believe you are correctly responsive to their vocalizations
2. Their barks are more differentiated over time and they start to introduce new types of vocalizations or sequences of vocalizations
3. Dogs don't have a structured vocal language as it starts of slightly differentiated and mostly to get people's attention
This isn't true for other animals (based on our reading of academic papers) and a commonality we've seen in the papers: as animals have more individuals they interact with across their life especially not directly related, their vocalizations are more structured. Within family units they tend to rely more on touch, body language, and some intermittent vocalizations. Though marine mammals are very vocal given they operate within acoustic environments in the ocean and it gets dark around 100m - 200m.
Basically, as number of sustained individuals in a group goes up and the number of non-related individuals increases, vocalizations get more diverse. We see cows, pigs, goats, and other herd animals are also very vocal which is how scientists have been decoding their speech recently similar to prairie dogs.
Much of what I've written here is a super simplification but happy to get deeper into the weeds.
We had an old dog move in with us and it’s been interesting because it has developed a new dog culture based on my existing dog. It now barks when someone comes home, barks to go outside etc. It didn’t do any of that before but it has started doing it by watching the existing dog communicate and having people who are more attentive to its needs.
Everything I see in dogs suggests that they are sentient, they just don’t seem to need language. They speak when it is useful, like for getting attention from people in different rooms, but they don’t really need it beyond that. So they don’t go any further.
If you ever deliberately raise a puppy, you'll watch the formation of many communication patterns that you create both deliberately (like barking or ringing a bell to go outside) and by accident.
The more I learn about domesticated animals, the more impressed I am by their intelligence and attunement to humans, natural and learned.
> Everything I see in dogs suggests that they are sentient, they just don’t seem to need language. They speak when it is useful, like for getting attention from people in different rooms, but they don’t really need it beyond that. So they don’t go any further.
If you haven't check out whataboutbunny [1] on Instagram to get an indication of how far dogs can go when given an opportunity. Bunny is trained to use buttons to "speak". A lot of it is very simple and functional ("outside", "play") that could easily be handled by body language and the odd bark, but occasionally you get fairly complex conversations which seems to indicate introspection and fairly complex reasoning that as you say they "just don't seem to need" to be able to express in language in normal conditions.
I do chuckle at one of mine, who definitely has a range of meanings to her barks. One solitary "WOOF", repeated at an interval means, "I'm stuck in the paddock, please let me out." She has another for "I've got a raccoon up a tree," and a gruff huffing noise for, "get off your ass and take us for a walk!"
I'm pretty sure I'm at least as much the trainee as the trainer here though.
> We've been building tech to translate dog barks to something meaningful for us to understand.
Can most dog owners not already understand or differentiate their dogs' barks? I can tell from the quality and context of my dog's barks, whether he
- wants something (food, water, attention)
- is angry
- is scared
- wants to play
- is excited
- is voicing a territorial dispute
and I think most dog people can also be confident in interpreting a fair range of basic emotions and also perhaps unique behaviors they've accidentally trained or organically developed with their dogs.
What additional things can your tech detect or identify beyond stuff like that? Are you using AI with human coders on training data, and if so, do they need special training or is the wisdom of naive dog-adjacent crowds good enough here?
I thought the other info you shared about canine and other animal vocalizations was very cool and interesting. I'd love to learn more about that any time, from anyone!
While maybe not very useful, it would be fun to have a real translation of dogs (and other animals). I have a lot of questions that I'd like them to answer.
Speaking of Dogs and Human interaction: I remember a study stating that wolfs lack a certain muscle, which dogs have. A muscle above the eye; believed to be directly linked to dogs’ long time interaction with humans.
Strange that this article doesn't mention any of the current research into what the cause is. [Erich Jarvis](https://en.m.wikipedia.org/wiki/Erich_Jarvis) has done research into the neurological mechanisms and seems to have discovered that animals capable of complex speech patterns have some mutations that permit a neural connection to grow from the motor cortex to the larynx (or syrinx in birds) by way of the brain stem. In other words, we can make complex vocalizations in part because we have fine motor control over our voice box, whereas most other animals do not.
This is only part of the story — there's still the question of how auditory information makes it in as an input so that we can replicate sounds we've heard — but it's a pretty big part for TFA to omit.
Yet people can make perfectly intelligible speech without using the larynx at all - whispering. People can also speak using a vibrator pressed against the throat.
more importantly, people can make perfectly intelligible speech using a completely different set of symbols, for example writing, whistling^1, hand gestures (sign language), pictures, etc.
It's interesting to compare this "sound-talk" that parrots can easily accomplish with the obviously greater intelligence of apes, who can't reproduce human sounds, with the debate over whether chatGPT is intelligent or not. Note that Koko was obviously a greater communicator than any parrot ever was.
Is mimicking human behaviour a necessary, or even sufficient, criteria for demonstrating cognition and intelligence?
Koko's language skills were misrepresented in a pretty extreme way. Very few recordings of Koko communicating anything more than short phrases were ever released, and those that were never demonstrated much more than word association. I've met dogs and pigs that were able to go fetch the "carrot" versus the "bunny" stuffed animals based on verbal commands, but I'm not going to claim that constitutes mastery of language.
Koko primarily communicated via repeating individual words that she'd been able to correlate with getting immediate effects. "Food", "Water", etc. When she did use multi-word phrases, they had no semblance of grammar: human babies will quickly work out the difference between "Me bite him" and "him bite me", whereas Koko never did. Frankly, despite being billed as a decisive blow against the Chomsky school of linguistics, Koko probably stands as strong evidence for it.
IMO, the massive complexity gap between human-level language compared to the relatively tight pack of merely possessing vocabulary animals, pretty directly correlated with levels of problem solving skills and tool use, should signal that there's something to the idea that our cognitive and language abilities are pretty closely linked. Mimicking human behavior may not be necessary for intelligence, but some kind of language likely is.
The short story "The Great Silence" by Ted Chaing feels very relevant here.
> The humans use Arecibo to look for extraterrestrial intelligence. Their desire to make a connection is so strong that they’ve created an ear capable of hearing across the universe. But I and my fellow parrots are right here. Why aren’t they interested in listening to our voices? We’re a nonhuman species capable of communicating with them. Aren’t we exactly what humans are looking for?
I think the dark forest theory rings very true here. The risk/reward equation for communication is skewed so far towards the risk end that, as the joke goes, maybe the surest sign that there’s intelligent life is that none of it has contacted us.
Or Peter Watts’ much more succinct take: if they do exist, they’ll be mean.
TL;DW - Data was VERY cherry-picked to make it look like she could communicate via sign language. At best, she was responding via rote memorization and had no idea what she was actually communicating.
The part that starts at 32:00 to around 38:00 is quite interesting, they programmed a PDP-8 to function as an icon-based chat mechanism. I always remembered the "grooming" part.
I think Koko's trainer really did believe her more extravagant claims. So did Clever Hans's. But there was a lot of wishful thinking along the lines of "Koko's making a weird gorilla joke that doesn't make sense to human minds" when a more likely explanation would be "Koko really doesn't know what she is saying".
I also personally asked GPT-4 to solve an easy CodeForces problem dated March 2023 and its solution worked perfectly the first time. (I haven’t yet tested it on a harder Codeforces problem.) Although the problem is relatively easy, it’s still much harder than the Fizz buzz test.
Thanks, the explanation of jokes was very interesting, although both GPTs made the wrong interpretation of the goat/toga party joke. The joke is that the speaker thought it was a "goat party", not a "toga party" (rather than that they dressed as a goat in a toga party).
Similarly for the intelligence joke; the joke there is that the speaker can only get an unintelligent girl, rather than that he seeks an unintelligent girl.
I was more impressed by the glove and ball interpretation, to be honest.
Apes are not more intelligent than parrots, despite not having hands like apes parrots use tools with the same sophistication as chimps, and Crows surpass all great apes except humans in tool use and complexity. Parrots have been able to solve complex puzzles of equal or greater complexity to what a Chimp could solve. And when it comes to social behavior parrots/corvids are far more sophisticated and complex.
Honestly if I had to bet which animal would outsmart which in a real world game of wits I would bet on the parrot or crow over a chimp every time. I apes have hands which are perfect for tool use and making, but crows make more complex tools using their beak.
I recall that the origin of the word "barbarian" comes from the ancient Greeks, who interpreted foreign languages as simply "bar bar bar". So, other animals may be communicating vocally, it's just that all we hear is "woof woof woof".
Or it might be that because we aren’t teaching them the way we teach babies. With a baby you are constantly reinforcing the language and encouraging them to match you. They obviously don’t have a human voice in there but they may be capable of a richer dog language than they currently use if they were taught it from early life.
> There was also a communications problem. Though they had an equivalent IQ of sixty, and could understand several hundred words of English, they were unable to talk. It had proved impossible to give useful vocal chords either to apes or monkeys, and they therefore had to express themselves in sign language
It would be impossible to give them useful vocal cords. Just not because of the vocal cords.
There should be a category specifically for invalidated science fiction. Things found to be unpossible in any timeline, assuming consistent laws of nature.
I believe this category is referred to as "uplifting" [0]. And while it certainly hasn't been proved to be viable so far, are you really sure it's time to rule it out as a possibility?
Just change the emphasis - "...give useful vocal chords...".
Important note on context - IIR, the "they" in that story were a spare-no-expense domesticated hybrid "working breed" of apes. Designed to replace expensive human astronauts doing repetitive, menial "housekeeping" jobs on space ships. One can read this as creative and edgy in context (1973, Civil Rights, "race riots", etc.) - but having the story's housekeeping apes able to talk might have gotten a "too big a risk" veto from the author's editor.
On the other side: Parrots cannot recognize themselves in a mirror https://talkieparrot.com/behavior/can-parrots-recognize-them..., while monkeys can https://en.wikipedia.org/wiki/Mirror_test . I consider the mirror test a better test for a higher form of intelligence than simply being able to simulate talking. So the real question nowadays would be: Can GPT-4 pass the mirror test? Can it detect it‘s own output, if you echo it back to it?
The problem is also that intelligence isn't on some kind of a line where a specific test tells you that you're closer to a base amount of intelligence needed for sentience. It's a whole bunch of things combined in some yet to be defined way that ends up allowing for the development of intelligence.
As for GPT-4 - it can't even recognize for itself whether data is good or bad, and it can't even recognize that it doesn't know something; IMO right now it's still basically just a black box that takes an input and produces a statistically matching output. A very good one, but still.
Readit News
>Charles Darwin theorized that monkeys had the vocal anatomy needed for speech, but lacked the necessary neural mechanisms.
>This was the most popular hypothesis until 1969 [when] Lieberman studied the vocal anatomy of a monkey corpse and concluded that other primates could not produce as many vowels as humans because of the position of their larynxes [which] cemented the idea that a descended larynx is a prerequisite for speech.
>Towards the end of the 20th century, evolutionary biologist W. Tecumseh Fitch realized a crucial fact – all the existing evidence was based on the anatomy of dead primates. Surprised that this had not been done before, Fitch used x-ray imaging to study the vocal tracts of live animals while they were voicing sounds. He was amazed to observe that their larynxes at rest remained high and then descended during vocalization to a position very similar to the human larynx.
>All these studies indicate that apes have all the anatomical characteristics necessary for speech. The reason they don’t is purely neural. Humans have much better control of the larynx, not because of its position, but because of the neural connections that connect it to the brain. Parrots don’t even have a larynx, but they have wonderful control of their speech organ, which enables them to articulate intelligible words and phrases.
Fascinating
My pet-theory why primates do not vocalize is that the neural difference is not even that cemented, it has to be some difference at the language acquiring stage as babys and children, were one human pattern matching center for sound fires constantly, acquiring the language "keys" and theirs dont.
And it all cascades from there.
https://en.wikipedia.org/wiki/Language_deprivation
It would be really fascinating, to help them acquiring language and uplift them. Maybe its even mammal universal applyable.
I have this example but there are literally hundreds:
You know about the 5 μm particle size limit for it to become airborne? Washing hands for protecting against aerosols? It’s parroted bad research; and wrong.
The main infection path of upper respiratory diseases is through aerosols. The medical literature says it’s through hands that are touching the mouth and the nose but this is wrong. It was shown in the 1940s that viruses can spread through aerosols and that aerosols are particles under 100 μm. Research was done on tuberculosis. Tuberculosis is caused by bacteria that must infect the deep inside parts of the lung. For that it needs an aerosol size of 5 μm. Other diseases can do with aerosols of 5-100 μm.
Now, a meme was born that only 5 μm particles can get airborne.
And this was just accepted by the Western medical canon and the WHO.
…then goes on to describe a human exception.
Deleted Comment
Dead Comment
A few things we observed: 1. Dogs are more communicative once they believe you are correctly responsive to their vocalizations 2. Their barks are more differentiated over time and they start to introduce new types of vocalizations or sequences of vocalizations 3. Dogs don't have a structured vocal language as it starts of slightly differentiated and mostly to get people's attention
This isn't true for other animals (based on our reading of academic papers) and a commonality we've seen in the papers: as animals have more individuals they interact with across their life especially not directly related, their vocalizations are more structured. Within family units they tend to rely more on touch, body language, and some intermittent vocalizations. Though marine mammals are very vocal given they operate within acoustic environments in the ocean and it gets dark around 100m - 200m.
Basically, as number of sustained individuals in a group goes up and the number of non-related individuals increases, vocalizations get more diverse. We see cows, pigs, goats, and other herd animals are also very vocal which is how scientists have been decoding their speech recently similar to prairie dogs.
Much of what I've written here is a super simplification but happy to get deeper into the weeds.
Everything I see in dogs suggests that they are sentient, they just don’t seem to need language. They speak when it is useful, like for getting attention from people in different rooms, but they don’t really need it beyond that. So they don’t go any further.
The more I learn about domesticated animals, the more impressed I am by their intelligence and attunement to humans, natural and learned.
If you haven't check out whataboutbunny [1] on Instagram to get an indication of how far dogs can go when given an opportunity. Bunny is trained to use buttons to "speak". A lot of it is very simple and functional ("outside", "play") that could easily be handled by body language and the odd bark, but occasionally you get fairly complex conversations which seems to indicate introspection and fairly complex reasoning that as you say they "just don't seem to need" to be able to express in language in normal conditions.
[1] https://www.instagram.com/whataboutbunny/
I'm pretty sure I'm at least as much the trainee as the trainer here though.
Can most dog owners not already understand or differentiate their dogs' barks? I can tell from the quality and context of my dog's barks, whether he
and I think most dog people can also be confident in interpreting a fair range of basic emotions and also perhaps unique behaviors they've accidentally trained or organically developed with their dogs.What additional things can your tech detect or identify beyond stuff like that? Are you using AI with human coders on training data, and if so, do they need special training or is the wisdom of naive dog-adjacent crowds good enough here?
I thought the other info you shared about canine and other animal vocalizations was very cool and interesting. I'd love to learn more about that any time, from anyone!
Then we can just ask the LLM to translate the same whale song to english.
(I would dismiss trying this with dogs as in my experience they communication channels revolve more around touch and smells than barking)
This is only part of the story — there's still the question of how auditory information makes it in as an input so that we can replicate sounds we've heard — but it's a pretty big part for TFA to omit.
Of course language is purely neural!
^1 https://en.wikipedia.org/wiki/Whistled_language
https://www.mirror.co.uk/science/monkeys-brain-power-talk-la...
Is mimicking human behaviour a necessary, or even sufficient, criteria for demonstrating cognition and intelligence?
Koko primarily communicated via repeating individual words that she'd been able to correlate with getting immediate effects. "Food", "Water", etc. When she did use multi-word phrases, they had no semblance of grammar: human babies will quickly work out the difference between "Me bite him" and "him bite me", whereas Koko never did. Frankly, despite being billed as a decisive blow against the Chomsky school of linguistics, Koko probably stands as strong evidence for it.
IMO, the massive complexity gap between human-level language compared to the relatively tight pack of merely possessing vocabulary animals, pretty directly correlated with levels of problem solving skills and tool use, should signal that there's something to the idea that our cognitive and language abilities are pretty closely linked. Mimicking human behavior may not be necessary for intelligence, but some kind of language likely is.
> The humans use Arecibo to look for extraterrestrial intelligence. Their desire to make a connection is so strong that they’ve created an ear capable of hearing across the universe. But I and my fellow parrots are right here. Why aren’t they interested in listening to our voices? We’re a nonhuman species capable of communicating with them. Aren’t we exactly what humans are looking for?
Or Peter Watts’ much more succinct take: if they do exist, they’ll be mean.
And like most birds they also have superior spatial navigation skills.
Sadly, I don't think this is actually true, despite the popular belief.
See https://youtu.be/e7wFotDKEF4
TL;DW - Data was VERY cherry-picked to make it look like she could communicate via sign language. At best, she was responding via rote memorization and had no idea what she was actually communicating.
I remember seeing this on TV when I was a kid.
https://youtu.be/UwZWWsc5xDk
The part that starts at 32:00 to around 38:00 is quite interesting, they programmed a PDP-8 to function as an icon-based chat mechanism. I always remembered the "grooming" part.
The vast majority of people would say that no unintelligent, or parrot-level-intelligent, entity can explain a joke like this: https://www.freethink.com/robots-ai/gpt-4-jokes
I also personally asked GPT-4 to solve an easy CodeForces problem dated March 2023 and its solution worked perfectly the first time. (I haven’t yet tested it on a harder Codeforces problem.) Although the problem is relatively easy, it’s still much harder than the Fizz buzz test.
Geoffrey Hinton has updated his AGI timeline significantly in the last 2 years. He is definitely worth listening to: https://mobile.twitter.com/JMannhart/status/1641764742137016...
Similarly for the intelligence joke; the joke there is that the speaker can only get an unintelligent girl, rather than that he seeks an unintelligent girl.
I was more impressed by the glove and ball interpretation, to be honest.
Apes are not more intelligent than parrots, despite not having hands like apes parrots use tools with the same sophistication as chimps, and Crows surpass all great apes except humans in tool use and complexity. Parrots have been able to solve complex puzzles of equal or greater complexity to what a Chimp could solve. And when it comes to social behavior parrots/corvids are far more sophisticated and complex.
Honestly if I had to bet which animal would outsmart which in a real world game of wits I would bet on the parrot or crow over a chimp every time. I apes have hands which are perfect for tool use and making, but crows make more complex tools using their beak.
Deleted Comment
> Is mimicking human behaviour a necessary, or even sufficient, criteria for demonstrating cognition and intelligence?
Depends on how you define them. We love to anthropomorphize everything. If an animal shows language comprehension, that's bigger than speech.
Deleted Comment
- Rendezvous with Rama (1973)
Guess it's more fiction than science fiction now.
There should be a category specifically for invalidated science fiction. Things found to be unpossible in any timeline, assuming consistent laws of nature.
[0] https://en.wikipedia.org/wiki/Uplift_(science_fiction)
Important note on context - IIR, the "they" in that story were a spare-no-expense domesticated hybrid "working breed" of apes. Designed to replace expensive human astronauts doing repetitive, menial "housekeeping" jobs on space ships. One can read this as creative and edgy in context (1973, Civil Rights, "race riots", etc.) - but having the story's housekeeping apes able to talk might have gotten a "too big a risk" veto from the author's editor.
As for GPT-4 - it can't even recognize for itself whether data is good or bad, and it can't even recognize that it doesn't know something; IMO right now it's still basically just a black box that takes an input and produces a statistically matching output. A very good one, but still.