Squawk bots: Can generative AI lead us to understanding animals?
cloud.google.com
cloud.google.com
But still, we already had lots of evidence how some animals like cetacean and primate are potentially sentient. Many people just don't want to admit it and we can't convince them that human isn't anything special. In the end this may just be another piece of proof on a pile that get ignored.
Cant we then just wait for animals to develop an ai that translates to our languages?
Genesis 9:3 "Everything that lives and moves about will be food for you. Just as I gave you the green plants, I now give you everything."
Additionally, in Genesis 1:24-26, remesh is listed among other types of animals such as beasts of the land and birds of the sky, so it is pretty clear that it is not meant as all animals. In these phrases remesh appears in the English translation as "creeping thing".
I do, however, appreciate your point and agree with the spirit of your post, if not the exact content.
I definitely feel like I am two 'selves' - one of which is capable of rational thought, foresight, rational modeling of the world around me, etc. That self is probably running things 25% of the time. The other self is emotional, understands the world intuitively for all of the benefits and drawbacks that carries, and seeks short term satisfaction of needs above all else.
I don't mean this in a hand-wavey new age holistic way, I mean it in a very concrete sense. I can detect that change in myself when I switch from the 'monkey pilot' to the 'human pilot'. Mindfulness and concentration is required to bring the human back. I think I have below average executive functioning skills, and I spend a lot of my 'human pilot' time thinking about how I can stay in this mode longer but it's not trivial to make that happen.
I think my 'human pilot' is definitely sentient. My 'monkey pilot' might not be depending on the definition.
If anything, LLMs have exposed our inconsistency in how we ground sentience, consciousness, and subjective experience. At times we thought we were talking about the same things, but we've clearly have been tossing terms about.
LLMs have, at least/last, given us the opportunity to reexamine important social ontologies. Better late than never.
I believe I've seen an article touching this topic here, long before GPT. Will link if I find it
The only description I've ever seen of consciousness that makes any sense is that it's a kind of condition monitoring system for our minds, a cross-check evolved to catch some of the cognitive mistakes that the main machinery of our minds tend to make from time to time. And because our whole brain is linked into this system, including itself, we get that infinite regress of omphaloskepsis that we call "consciousness."
I don't do a lot of metacognition (thinking about my thinking) while I'm in a flow state, which maybe was your point. But I can think about my prior flow state thinking after the fact, and I'm aware of how much my mind was working, and what that felt like.
In other words, consciousness is an emergent phenomena that happens in a system of sending messages when eventually an agent starts sending messages to itself. "you" are the voices in your head not attributed to anyone else.
One way to see this is that sentience and language processing are just two aspects available to us in consciousness.
We are also not identical to our thoughts or feelings, but most people never pause to contemplate this.
I suspect the autopilot you describe is really closer to our default state, but it’s easier to notice when in a calm state and the “me-ness” of actions/thoughts doesn’t feel so strong.
https://www.lesswrong.com/posts/4AHXDwcGab5PhKhHT/humans-who...
Currently we tolerate the enslavement of humans but contrary to popular narratives about the US Civil War, we haven't fought any wars specifically to end slavery of other humans. We have had successful "slave rebellions" (e.g. Haiti, which continues to pay the debt imposed by their former slave masters for "damages and economic losses" to this day) but anti-slavery sentiments among those not in the demographic directly suffering from slavery never went quite that far. The anti-colonialist coup in Portugal was fueled more by the military's resentment over the forever-war required to suppress the colonies than any sympathy for their cause of liberation. The US Civil War was a reaction to the attempted secession of its then-member states rather than specifically their stance on slavery.
It's plausible that we'd see acts we'd consider terrorism in the name of animal liberation (arguably some of the ALF's past actions fit into this category) much like we have seen with other liberation movements. But as this is a struggle entirely led by those not themselves in the group of the oppressed, the dynamics are different from most civil rights struggles. Arguably even the abolition of formal slavery in most countries had less to do with zealous support from non-slaves and more with economic factors making slavery less appealing. And even then the countries only got away with slavery in their own mainland while still happily continuing to benefit from overseas slavery, much like we now mostly hide animal suffering in farms, closed trucks and meat processing factories rather than back alleys and town squares.
https://en.wiktionary.org/wiki/sentient:
Adjective
sentient (comparative more sentient, superlative most sentient)
1. Experiencing sensation, thought, or feeling.
2. Able to consciously perceive through the use of sense faculties.
3. (loosely, chiefly science fiction) Possessing human-like awareness and intelligence. Synonyms: sapient; see also Thesaurus:self-aware
https://en.wiktionary.org/wiki/sapient:
Adjective
sapient (comparative more sapient, superlative most sapient)
1. Attempting to appear wise or discerning.
2. (dated) Possessing wisdom and discernment; wise, learned.
3. (chiefly science fiction) Of a species or life-form, possessing intelligence or self-awareness. Synonyms: sentient; see also Thesaurus:self-aware
https://www.dictionary.com/browse/sentient:
adjective
1. having the power of perception by the senses; conscious.
2. characterized by sensation and consciousness.
https://www.dictionary.com/browse/sapient:
adjective
1. having or showing great wisdom or sound judgment.
2. having or showing self-awareness
A lot of it is based on smell, body language, and signals we don't pay attention to. I don't think any animal has been shown to tell stories, but obviously they share information in other ways. Only a handful of individuals have demonstrated some kind of ability to understand a story (macaws and koko the gorilla).
For anybody interested in animal intelligence, that's the focus of my newsletter.
The dog is out in the wild, away from it's pack. Ever see your dog smell something, or find something it finds interesting? Often, they will rub on it, or roll around in it. They do that so they have a record they can take with them.
When said dog returns to their pack, their body holds these records. Their movements communicate their mood and the entire thing is a tale of the day.
Another example I had happen was my cat.
I would come home and the cat would want to get in the car. Then, if I got up to move, she would put a paw out, like "hold on a moment" and then she would inspect me everywhere, sniffing, looking at me, etc... The cat wanted my story! Where I went, was it the same strange animals and people, or different, when I last ate, and all sorts of stuff.
In the evenings, that same cat would come to me, want to cuddle up in the evening and would be near my face, so I tried the same thing and guess what? It's a damn solid story! I now can smell my cat, watch it's movements and get a good sense of it's activity for the day.
Back to the dogs. Having returned to the pack, the dog could tell the story of animals it encountered, possible food sources, enemies and other things by the scents it carried home to share with the others.
In my view, our biggest struggle understanding animals comes down to both how much they are like us, just in simpler form, and how we fail to see that due to our higher complexity taking us away from basics that make more sense than we know. And with some animals, they are just very different from us, whales, dolphins, octopus and the like.
All that said, I agree with you! Humility and an open mind are key to understanding animals better. Frankly, the same tools help us understand one another better.
Having discovered this dynamic through careful observation of my animals, story seems appropriate. A session with the cat yields:
Who: Me
What: Trip away from pack
When: Today, more days maybe
Where: Various places, work, car, outside places
Why: Who knows what the cat thinks. She does differentiate a trip away from a food run
How: in that moving thing
And me with the cat:
Who: Cat
What: romp in flower beds or play with squirrels (they leave a different scent and the moss on the trees helps) Fight with that other cat.
When: recently
Where: grassy yard, trees, across the street
Why: Cat bored playful finds squirrels. Enemy cat entered our territory, etc...
How: On foot, maybe through fence.
Given some time observing, smelling, listening, one can get a pretty great sense of the day!
One other thing I should say is animals are like us, just simpler. Their every move communicates! They lack the layers of abstraction needed to make it otherwise.
I can gather a lot from a few minutes spent close, feeling, watching, smelling, watching the cat move and vocalize.
I find the whole thing remarkable! And it was the cat who educated me!
If I pull up in the car and do not roll my window down, I get the look. If I do, then she is happy to jump in and exchange events of the day, essentially adding me to her story.
The question sometimes happens. Usually, it is that paw on my chest when I get up to leave. Looks for all the world like, "hold on, I haven't got it all yet, or "this is new, WTF?
Don't get me wrong. Not all animals are that high functioning. We have a couple and in my experience the more we try to interact and the less inhibited we are about all the ways, rubbing, smelling, etc... the more a higher functioning animal will respond with their own efforts.
I only wish I had this insight earlier in life! My animal interactions are far more rich than they were before.
Yes, let us flex these things a little and appreciate these beings here with us.
https://www.amazon.com/Are-Smart-Enough-Know-Animals/dp/0393...
what's going on there?
Also, since chatbots just output "meaningful" strings of tokens based on a string of input tokens, you can build a chatbot that can "talk" to animals without us necessarily being able to understand it. But sure how useful that would be, but interesting nonetheless.
No, but they have parents. I am sure they have taught languages. Just ask any cat lady who will swear their cats understand them.
I'd be more concerned that different regions have different languages and we'd need to control for that.
There could of course be regional languages or dialects in more intelligent social animals just like humans.
Was this written in a parallel universe where OpenAI never released an LLM?
I had the same reaction to the idea of legal personhood for natural features of the landscape.
Am I too jaded?
Not at all. You can construct tests to see if the animal's behaviour is consistent with what the network say they are "saying". You can go and check if what information you receive through it is consistent with the observed reality.
You can test that behavioural differences observed between individuals show up in the translation. For example a dog who you know is hurt, or old is probably less chipper than one who just chased a squirrel. Observers of animals often notice that some individuals are more timid while others are more curious/adventurous. One would expect that their vocalisation also reflects of this difference in personalities.
It's not like we just slap a network together and we call it done. If the translator says Lassie is saying that Timmy has fallen down the well you might as well check the well, but if Timmy is not there then this might be a translator failure.
> Am I too jaded?
Yes.
But I don't see what that has to do with my comment. Are you explicitly a radical behaviorist, or do you think animals have internal experience?
Suppose that instead of a dog, we were talking about a human with dementia in a nursing home. Do you think that not being able to witness your own condition to a doctor would put you at a disadvantage?
In the absence of a feedback mechanism, the virtue and the intelligence of people who are in control of a system do not count for much.
You will have to use smaller words with me, or explain your vocabulary, if we are to have a discussion.
> But I don't see what that has to do with my comment
You said these things in your original comment: "If we decide that it does" (here I presumed "does" mean AI let us understanding animals) "then conveniently there's no way for the animals to dispute it."
That is what I'm disagreeing with. There are many ways animals can dispute us understanding them.
Imagine a scenario that we have a gadget which is supposedly translating horse body language and vocalisations to human language. You step up to a big black mare and use the gizmo. It tells you that the horse wants scratches behind its ear. You reach out, but oh horror it was a complete mistranslation. The horse was in distress by you approaching and is completely spoked by you reaching for their ear. It reacts, bites your hand, kicks you and drags you through the mud. That is how you realise that the gizmo's understanding of horse communication is less than perfect. That is the feedback mechanism.
You said there is no way, and here is one way they can do it.
> Suppose that instead of a dog, we were talking about a human with dementia in a nursing home. Do you think that not being able to witness your own condition to a doctor would put you at a disadvantage?
Sure. But animals are not dementia patients and being at a doctor/vet is not the only scenario we can test our understanding of them.
Sorry, I was alluding to B.F. Skinner, but I used the term "radical behaviorist" because I didn't want to get drawn into a discussion of him in particular.
I think it's a known term, although I'm not sure whether anybody self-identifies as such or if it's more of a caricature by opponents.
I threw it out there, thinking if you did recognize it, and identified yourself as one or as not one, I would learn more about how it is used.
>But animals are not dementia patients and being at a doctor/vet is not the only scenario we can test our understanding of them.
"...is the best we can do" is what I see you arguing, and I'm neither affirming nor denying it. It's not relevant to my interests or the intent of my commenting, is all.
I seem to recall a (possibly recurring) news item about the poultry industry wherein it explains that non-egg laying chicks are being disposed of in huge shredders, alive. I am certain an industry expert would say that is the best alternative. What does anyone do all day but try to figure out the best way to do things at work?
If someone suggests that being shredded might be unpleasant, and you coincidentally work in the industry, you might be concerned that you are being judged as immoral, or that they are on a crusade to stop people eating meat, or otherwise react to an attack on your lifestyle, ethics, and self-image.
But maybe (in my case) I just have a bit of a pre-occupation with acknowledging unpleasant aspects of reality, because I worry that people who rationalize treating living creatures that way might shade over to denial of facts and develop faulty intuition about other situations.
>being at a doctor/vet is not the only scenario we can test our understanding of them.
The reason I wrote "to a doctor" was just that I was reaching for a viscerally urgent or critical situation in my example, not because communicating one's mental state is uniquely relevant in medical situations. It could just as well be "saying goodbye to a loved one".
One conception of empathy is feeling what someone (or some creature) is feeling, that is, accurately modeling what they feel.
On the other hand, in reality, all anyone can know for sure is if they get positive reinforcement for using their model of others' feelings.
In any situation of unequal power and/or imperfect communication, there is a danger that the person with power gets positive reinforcement that is detached from reality, without the possibility of correction.
It goes through the extraordinarily diverse sensory systems of all animals. In the push for at-the-edge, exteme-low-latency, low-resource "intelligent" agents, mimicking some of these system may lead to cool systems.
" We need another and a wiser and perhaps a more mystical concept of animals. Remote from universal nature and living by complicated artifice, man in civilization surveys the creature through the glass of his knowledge and sees thereby a feather magnified and the whole image in distortion. We patronize them for their incompleteness, for their tragic fate for having taken form so far below ourselves. And therein do we err. For the animal shall not be measured by man. In a world older and more complete than ours, they move finished and complete, gifted with the extension of the senses we have lost or never attained, living by voices we shall never hear. They are not brethren, they are not underlings: they are other nations, caught with ourselves in the net of life and time, fellow prisoners of the splendour and travail of the earth. "
Henry Beston, The Outermost House: A Year of Life On The Great Beach of Cape Cod
Wouldn't this be totally pointless? What's the use of a ChatGPT for elephants if it doesn't allow us to communicate with them?
There is also a phenomenon in LLM research where a comparatively tiny amount of a language confers unexpectedly high fluency to a model which is otherwise intelligent in its main language and able to transfer that knowledge over. For example, an LLM needs to be trained on a lot of text to become intelligent in the ways modern LLMs are, which means English text because there's so much of it. But if you then train that LLM on a comparatively tiny amount of another language, like Catalan, it doesn't behave like a model that was trained on that tiny sample of Catalan alone, even when it's only speaking in Catalan. Instead it is nearly as intelligent as it is in English, just in a different language. What that means is that we don't necessarily have an impossible task in front of us in order to get enough human-translatable "language" from any given animal. We may only need to get a surprisingly small amount that we understand for an LLM to be able to generalise and come closer to 'fluency' in that language, which is going to expand the amount of data we have access to by a lot. It's an interesting area of research and I hope it gets expanded on
Has anything similar been done for human vocalizations (aka spoken language)? Text is obviously so much more efficient, and so much more abundant, so even for human speech this sounds like a challenge.
Same as it already handles "speak like a pirate" and "in the style of Shakespeare" and "Python" and "Hungarian". From the perspective of the AI these are all just "dialects and accents".
As long as it will have enough data containing the different accents and dialects it will be able to handle it no problem.
I can't see any obstacle to an LLM ingesting a database of squawks and emitting squawks which may or may not be in relevant patterns consistent with how a particular flock of birds communicate, but I can't see the birds advising us on whether it's using them idiomatically or not.