[HN Gopher] AI Decodes the Calls of the Wild
___________________________________________________________________
AI Decodes the Calls of the Wild
Author : bookofjoe
Score : 82 points
Date : 2024-12-22 14:01 UTC (2 days ago)
(HTM) web link (www.nature.com)
(TXT) w3m dump (www.nature.com)
| firtoz wrote:
| I can totally understand an animal associating some of the sounds
| that it has heard in its lifetime, perhaps even a "name" for
| itself. There may also be possibilities for some animal
| populations e.g. crows if enough generations have passed with
| common sounds, to mean different things, especially when those
| things were encountered enough times during the lifetimes of
| them. So that could build to some kind of "for the birds of this
| region, this sound/combination may represent something specific".
|
| Same for dogs and so on.
|
| There may also be some kind of more subtle genetics/morphology
| based distinctions, beyond the obvious "snarl for most dogs means
| afraid, angry, aggressive, possessive, or in pain"...
|
| But... it seems that the collapse of tower of babel happens very
| frequently for non-humans, so unsure how "useful" an AI training
| may be for the majority of the cases.
| ripped_britches wrote:
| So your comment is just "I don't know how useful this new
| science will be"?
|
| What separates any science from corporate R&D is basically that
| exact statement.
|
| I wouldn't be surprised if we learn more about ourselves from
| this research, not to mention the animals it covers.
| firtoz wrote:
| Don't get me wrong, I'm excited for it, just trying to
| understand or predict the extents... Also I'm a layman in
| this field, so who knows!
| dumpsterdiver wrote:
| I imagine this line of research would be of interest to
| intelligence agencies. Especially when you start taking a
| multi-modal approach that incorporates sensors such as
| heart rate, perspiration, etc, then we're basically talking
| about pre-Judgment Day Terminators.
| vasco wrote:
| They are more sophisticated in their communication than you are
| portraying.
|
| > "Perhaps" even a name for itself
|
| No perhaps needed, dogs will respond to a name and know it
| refers to them. You might be interested in reading this study
| about dogs doing fMRI while responding to new and old words:
| https://www.frontiersin.org/journals/neuroscience/articles/1...
| jjtheblunt wrote:
| we have, besides a dog, cats, and they all know their
| individual names very clearly, repeatably, demonstrably. they
| also seem to know the names of one another, but of course
| cooperation is more eager in dogs than in cats, so harder to
| demonstrate that.
| bongodongobob wrote:
| My dog knows all the neighbor dogs names, and their owner's
| names. He's a smart breed but I feel like you're really
| underestimating animal cognition.
| jazzyjackson wrote:
| in my opinion, derived from just sitting in the woods being
| observant of the bird song, I don't think any individual bird
| has much to say (besides the requisite 'somebody screw me' and
| 'get the fuck away') but there is a lot of nuance'd information
| being transmitted in aggregate. How long its been since there
| was a predator, how soon is the next rain, are we in agreement
| that just about all the food is gone here (quorum sensing /
| collective decision making).
|
| Crows and other corvids of course are another level of
| sophistication. When they're gathering in trees (like when
| instead of leaves a tree has crows, that kind of gathering)
| somebody told me that there's elections going on, which
| individuals will be given decision making power are being
| chosen through bickering and persuasion.
|
| As for dogs (and a brief google doesn't recall my source so
| take it with a grain of salt, but,) I'll add that they're the
| only animal besides humans that have been observed making
| pacts, promises, "if you go I'll go" style. IANA-consensus-in-
| animal-populations however, and body language goes a long way
| in signaling submission, agreeableness, resistance, doubt etc.
| esperent wrote:
| It's a fair bet that anything intellectual that a dog can do,
| many other animals can do, too. If it appears that it's only
| dogs, that's almost certainly just because we get to observe
| them so much more than other creatures.
|
| https://english.elpais.com/science-tech/2023-09-13/pacts-
| amo...
| dools wrote:
| "Donning his new canine decoder, Professor Schwartzman becomes
| the first human being on Earth to hear what barking dogs are
| actually saying"
| Terr_ wrote:
| "Hey!"
|
| These were the memes that we brought to and from school uphill
| both ways, kids.
| codingwagie wrote:
| We are about to watch alot of mysteries unravel
| vasco wrote:
| There recently was a 2025 predictions thread, my prediction just
| got adjusted to "the Department of Defense will bribe birds to
| spy for them by talking to them through AI".
| Terr_ wrote:
| > The ducks in St. James' Park are so used to being fed bread
| by secret agents meeting clandestinely that they have developed
| their own Pavlovian reaction. Put a St. James' Park duck in a
| laboratory cage and show it a picture of two men--one usually
| wearing a coat with a fur collar, the other something somber
| with a scarf--and it'll look up expectantly. The Russian
| cultural Attaches black bread is particularly sought after by
| the more discerning duck, while the head of M19's soggy Hovis
| with Marmite is relished by the connoisseurs.
|
| _Good Omens_ , by Terry Pratchett and Neil Gaiman
| mlsu wrote:
| Total speculation, but I think it may be possible for transformer
| models to internalize a deep understanding of animal language,
| and possibly generate animal language. For some reason I also
| believe that they may never be able to translate it for us, and
| animals and humans will be unable to communicate just as much as
| today.
| sourcepluck wrote:
| What makes you think this? On the face of it, what you're
| saying seems to be absolutely nonsensical, so I'm curious.
| Jensson wrote:
| Because animals doesn't use words, they use the equivalent of
| laughter, crying, body language, shrieks etc. So you already
| understand your dog and cat, as it isn't that hard to read
| these signals in them.
| arcastroe wrote:
| I agree, transformer models trained solely on animal language
| may be able to predict the next "token" and generate coherent
| animal "sentences", even if we don't understand what any of
| those sounds/vocalizations mean.
|
| However, I do think there is a path to some sort of translation
| to human languages. An English dictionary may seem pointless
| because it defines every word using only other English words.
| But the meaning is contained in the _relationship_ between the
| words.
|
| The _relationship_ between certain animal-language tokens may
| look close enough to the _relationship_ between certain human-
| language tokens to bridge the gap, even if both languages are
| entirely disconnected and we've never seen a direct translation
| between them.
| hun3 wrote:
| See also: word embedding and sentence embedding models
| hackinthebochs wrote:
| Just correlate the tokens with the state of the environment
| and surrounding behavior and the meaning should reveal
| itself. The extent that vocalizations predict subsequent
| environment states, or the environment predicts subsequent
| words, is the meaning captured by those vocalizations.
| 082349872349872 wrote:
| to what extent do ethologists already (manually) look for
| these correlations?
| Sparkyte wrote:
| Animals will probably evolve to speak well before AI figures
| out what animals are saying.
| xandrius wrote:
| Given the evolution of AI vs the evolution of animal
| language, the probability is actually heavily leaning towards
| the opposite.
| consp wrote:
| I would really like a _very_ detailed explanation of how AI
| is evolving instead of just having more computation power
| and shortcuts applied.
| spacebanana7 wrote:
| Why is much detail required for the pace of change
| comparison?
|
| Animal language doesn't change much. Dogs sound pretty
| much the same as they did 10 years ago.
|
| But AI generated sounds have changed rapidly in the past
| couple of years. In terms of music generation, sound to
| sound translation, and audio to text transcription.
| Sparkyte wrote:
| Actually I believe this is false. If dogs are not
| socialized with other dogs there is a distinct lack of
| language, body language varies incredibly on a per dog
| basis. You can say most are adopting human like
| mannerisms to communicate with us better.
|
| Smart glasses might be able to leverage the process of
| LLMS to decipher a gesture or two of dog like behavior,
| but it also has to learn it from the that dog's own
| unique vocabulary.
|
| It is different if a group of animals is socialized
| between themselves for a long duration and time. Why?
| https://en.wikipedia.org/wiki/Language_deprivation You do
| not genetically pass down language. It has to be learned
| or made up.
| victorbjorklund wrote:
| Why would animals ever have an evolutionary need to learn to
| speak english?
| noman-land wrote:
| So we can sell them things, obviously.
| Sparkyte wrote:
| Sometimes to convey needs better. Some species of dogs
| vocalize almost human like sounds. However there has to be
| selective pressures to encourage DNA change that produces a
| dog that can talk. Without breeding happening to produce
| those selective features it will not happen. This probably
| won't happen for 200K+ years.
|
| However we've been training dogs to press buttons to
| express their needs and this is likely to come well before
| AI manages to establish LLMs around dog mannerisms.
| madaxe_again wrote:
| It's an interesting thought - even with a translator, could
| their cognition and world be so alien to us as to be
| incompressible?
|
| I would think largely not - particularly for mammals - it's
| easy to think you have nothing in common with a shrew or a
| whale, but we are quite recent cousins, and much of our
| cognitive architecture and sensorium is the same.
|
| That said, there could well be challenges and revelations
| beyond the obvious - for instance, can domesticated members of
| a species communicate with their wild cousins, or have they
| lost their culture/language? Cetaceans - do they talk in
| holograms? These are perhaps still obvious questions, but if
| it's something we figure out, we may get answers to things
| we've never even considered asking.
| soulofmischief wrote:
| Language requires a prior shared context. Communication is then
| optimized around relationships between those concepts. Without
| situational context, we might not ever determine the "meaning"
| for a lot of animal language. But we can still likely
| understand the relationships between different expressions,
| which may be of some use within itself to at least establish an
| emotional baseline for some expressions.
|
| There are already research groups who have been attempting to
| use embeddings and transformers to build out a vector space of
| specific animal languages. I think they will find some success
| and improve our current understanding, the question is how much
| context we will be able to derive. It's possible for example,
| that end-to-end multimodal AI would be able to train not just
| on animal sounds, but the body language and environmental and
| situational context they are in via video and other captured
| environmental data.
| piedra wrote:
| I think each animal has its own language, for example my dog make
| a noise with his noise when he is frustrated (expected something
| better that what he got), and move the tail sideways when he
| feels like to play or put his tail up when he is happy, so it is
| a body language. He can't learn other dogs language because there
| is no consensus on what words mean and so there is not feed-back.
| Sparkyte wrote:
| Exactly it is all expressive.
| bregma wrote:
| Q: What do you get when you cross artificial "intelligence" with
| "dumb" animals?
|
| A: I don't know. Further research required.
| nopromisessir wrote:
| We are translating language from a ton of other species. Now we
| can see that they are using names and probably sharing
| information.
|
| The brightest minds now want to use machine learning to tease out
| more low level features of language itself, across various
| species.
|
| We are already calling monkeys and elephants by their native
| names y'all... And they are responding to this...
|
| Think about it... This research is unlikely to stall. We're
| barely scratching the surface. The animals were having more
| nuanced conversation than we thought. They've been doing this for
| millions of years...
|
| Condescension toward nature... Usually a mistake. I'm not saying
| they are all philosophers, but it seems very likely they've been
| having some pretty advanced conversations surrounding their own
| affairs. If we can call them by their name... Probably we're
| going to unlock some pretty interesting communication.
| sofayam wrote:
| If we could talk to animals, then, on the positive side we could
| explain things to them like "watch for traffic when you cross
| this road" but we could also deceive them. A lot of what hunters
| and farmers have done since time immemorial has used our superior
| - or maybe just different - intelligence to exploit or trap
| animals, but imagine the chaos we could wreak if we could
| literally argue them into behaving against their best interests.
| Not everyone would use this ability responsibly, especially if
| there was money to be made.
___________________________________________________________________
(page generated 2024-12-24 23:01 UTC)