Pigeons Problem-Solve Similarly To Artificial Intelligence, Research Shows (theguardian.com) 54
According to a new study published in iScience, the way pigeons problem-solve matches artificial intelligence. The Guardian reports: In the study, 24 pigeons were given a variety of visual tasks, some of which they learned to categorize in a matter of days, and others in a matter of weeks. The researchers found evidence that the mechanism that pigeons used to make correct choices is similar to the method that AI models use to make the right predictions. "Pigeon behavior suggests that nature has created an algorithm that is highly effective in learning very challenging tasks," said Edward Wasserman, study co-author and professor of experimental psychology at the University of Iowa. "Not necessarily with the greatest speed, but with great consistency."
On a screen, pigeons were shown different stimuli, like lines of different width, placement and orientation, as well as sectioned and concentric rings. Each bird had to peck a button on the right or left to decide which category they belonged to. If they got it correct, they got food, in the form of a pellet; if they got it wrong, they got nothing. "Pigeons don't need a rule," said Brandon Turner, lead author of the study and professor of psychology at Ohio State University. Instead they learn through trial and error. For example, when they were given a visual, say "category A", anything that looked close to that they also classified as "category A", tapping into their ability to identify similarities.
Over the course of the experiments, pigeons improved their ability to make right choices from 55% to 95% of the time when it came to some of the simpler tasks. Presented with a more complex challenge, their accuracy went up from 55% to 68%. In an AI model, the main goal is to recognize patterns and make decisions. Pigeons, as research shows, can do the same. Learning from consequences, when not given a food pellet, pigeons have a remarkable ability to correct their errors. Similarity function is also at play for pigeons, by using their ability to find resemblance between two objects. "With just those two mechanisms alone, you can define a neural network or an artificial intelligent machine to basically solve these categorization problems," said Turner. "It stands to reason that the mechanisms that are present in the AI are also present in the pigeon."
On a screen, pigeons were shown different stimuli, like lines of different width, placement and orientation, as well as sectioned and concentric rings. Each bird had to peck a button on the right or left to decide which category they belonged to. If they got it correct, they got food, in the form of a pellet; if they got it wrong, they got nothing. "Pigeons don't need a rule," said Brandon Turner, lead author of the study and professor of psychology at Ohio State University. Instead they learn through trial and error. For example, when they were given a visual, say "category A", anything that looked close to that they also classified as "category A", tapping into their ability to identify similarities.
Over the course of the experiments, pigeons improved their ability to make right choices from 55% to 95% of the time when it came to some of the simpler tasks. Presented with a more complex challenge, their accuracy went up from 55% to 68%. In an AI model, the main goal is to recognize patterns and make decisions. Pigeons, as research shows, can do the same. Learning from consequences, when not given a food pellet, pigeons have a remarkable ability to correct their errors. Similarity function is also at play for pigeons, by using their ability to find resemblance between two objects. "With just those two mechanisms alone, you can define a neural network or an artificial intelligent machine to basically solve these categorization problems," said Turner. "It stands to reason that the mechanisms that are present in the AI are also present in the pigeon."
Prior work in the field (Score:5, Funny)
Google was working on this over 20 years ago. Remember PigeonRank? [archive.google]
Re: (Score:2)
It weirds me out that my understudy at work was born after the pigeonrank page even came out.
I suspect theres more than a few slashdotters that would relate to that sentiment.
Re: (Score:3)
Interesting. When I think of PigeonRank, all I can think about is this [staticflickr.com].
Do they hallucinate too? (Score:2, Funny)
Alternative interpretation (Score:5, Interesting)
Large Language Models are only about at good at problem solving as a really stupid bird.
On the plus side, LLMs don't defecate all over our statues.
Re:Alternative interpretation (Score:5, Informative)
And LLMs aren't made to be smart, they are designed to blather on endlessly, auto-generate youtube video scripts, converse with customers on phone lines and as chat bots, and more stupid stuff like this like.
Re: (Score:2, Troll)
Pigeons aren't stupid, like most birds they are pretty intelligent animals.
No they are not. If they enter your chicken pen and want to catch them, just scare them and wait till they have tired themselves. That takes about 3 minutes. The only "intelligence" they show is that they lose their tail feathers when you catch them, just like a lizard loses its tail.
Re: (Score:3)
If they enter your chicken pen ...
Just decorate your chicken pen in a way that "scares" other birds [theprairiehomestead.com].
Re: (Score:2)
No they are not.
Yes they are. Pigeons can learn to use public transport, among other things.
If they enter your chicken pen and want to catch them, just scare them and wait till they have tired themselves.
So? Most animals including humans will panic when trapped and threatened.
Re: Alternative interpretation (Score:2)
Wait isn't public transportation designed to be easy and accessible. As a human, if you make it to the bus stop exactly on time do you think you've accomplished some get mental feat? Probably not. Also you have shown is you have a decent sense of time and distance.
Re: (Score:2)
Nothing in this research has anything to do with LMMs.
Its assuming a rather primative vision model not unlike early image classifier NNs from about 15 years ago.
Re: (Score:2)
Nothing in this research has anything to do with LMMs.
LLM is the hot word to attract grant money nowadays. Even if your research is only very tangentially related to LLMs, be sure to mention them in your paper, and the $$$ starts rolling in.
Re:Alternative interpretation (Score:5, Funny)
On the plus side, LLMs don't defecate all over our statues.
Indeed, LLMs defecate all over our Internet [theverge.com] instead.
Re: (Score:2)
The AI being described in this summary and study are not LLMs. They are classical AIs. LLMs do not operate this way at all.
Re: (Score:2)
LLMs do not operate this way at all.
Multiple neural nets working together and pretrained with data. Not exactly a wholly different concept.
Although you say classical AI which is really more heuristics and pre-NN and I don't think related at all to the pigeons.
Re: (Score:2)
I don't really grock what you are saying, but the AI being described in this paper is a classic RB classification. You don't use LLMs for this kind of task, and you wouldn't even use a neural network you'd use something like naive bayes or K nearest.
To do this task with an LLM first you would need a base model then you would feed that model with instructions on the goals of what it was trying to do and start giving it encodings of different pictures, and ask it to choose. And when it made a mistake you woul
Re: (Score:2)
I've never classified retinoblastomas, but it does sound really hard to get it right.
The AI being described in the paper is a type of NN classifier.
EXACTLY (Score:3, Funny)
Re: EXACTLY (Score:2)
They'll be wasted in marketing.
Animals and humans learn many different ways... (Score:3)
Wrong headline (Score:2, Troll)
Wrong way round (Score:4, Insightful)
In fact, AI problem-solves similarly to pigeon brains.
"nature created an algorithm" they write, conveniently ignoring that AI algorithms are based on actual neural networks.
By their reasoning the headline could have been "Human brains problem-solve similarly to AI"
Re: (Score:3)
"It turns out, biological neural nets do function a lot like the artificial ones we modeled after them"
Re: (Score:2)
Stands to reason? (Score:5, Insightful)
"It stands to reason that the mechanisms that are present in the AI are also present in the pigeon." What reason is that, precisely? You see similar behavior and somehow this means the causes must be similar? There are myriad ways to solve some math problems, it stand to no reason that they are similar. At best the similar behavior in pigeons is an hypothesis.
history (Score:3)
Where's Tom Lehrer when you need him?
Re: (Score:2)
Or Georg Kreisler, who, curiously enough, created an almost identical song at almost the same time in German. Both claim they didn't crib.
pigeon existential threat (Score:3, Funny)
I for one welcome our new pigeon singularity overlords
Re:pigeon existential threat (Score:5, Funny)
Once General Pigeon Intelligence appears, all bets are off! They will be so good at getting humans to feed them that humanity won't care about anything else. Not eating, not sleeping, just feeding pigeons. And eventually this pigeon-obsessed humanity will die off. Not because the General Pigeon Intelligence hates us, but because it doesn't care about anything else but being fed.
You may object that the chance of that is really low. But human extinction has such an extreme impact that we must safeguard against it even when multiplied by infinitesimal probability. Expected value (and Blaise Pascal) tell us so.
Therefore, never feed pigeons. Get regulations written ASAP. And never ever create a machine in pigeon's image.
Re: (Score:2)
All the warnings about Skynet didn't say anything about all the feces.
Re: (Score:2)
I'm not worried about birds, cats already own us...
This checks out (Score:5, Funny)
I too have noticed AI is about as smart as a pigeon.
Re: (Score:2)
BF Skinner invented AI! (Score:3)
Jfc, this is first week freshman psychology class material from when I was a kid.
BF Skinner did amazing cutting edge work doing positive reinforcement training with various animals to get them to do all sorts of stuff,
My favorite was the pig on its hind legs pushing the vacuum cleaner. But did the pig have any concept of what a vacuum is or what it was accomplishing by pushing it around? No. It just knew that it was rewarded for following a series of steps.
What next? Bitcoin and pig slop have the same economic value? Oh wait....
Re: (Score:2)
What next? Bitcoin and pig slop have the same economic value? Oh wait....
That post was magnificent! Started out slowly, gathered speed, then Bam!
Well played sir, well played indeed.
Re: (Score:2)
Lol, thank you, thank you! But I must admit that was a last second addition. :-)
Re: (Score:2)
My favorite was the pig on its hind legs pushing the vacuum cleaner.
But did the pig ever want to break free out of this stupid experiment?
Re: (Score:2)
did the pig have any concept of what a vacuum is or what it was accomplishing by pushing it around? No. It just knew that it was rewarded for following a series of steps.
Not entirely unlike an 18-month old pushing around a toy vacuum. Except the reward is intrinsic because the human brain has a self-training reward system.
Bigger brains have bigger concepts. They can often be manipulated the same way, though.
That is a mile stone (Score:1)
I'm Not Happy (Score:2)
I'm not happy to hear that I will be ruled and constrained by a system whose intelligence is merely that of a pigeon.
I deserve a more sophisticated overlord!
*Classical* Artificial Inteligence (Score:2)
It is important to note that this is similar to *classical* artificial intelligence. Large Language Models do not work this way at all.
Oh great (Score:2)
So we're going to have racist sexist pigeons now?
Or maybe we already do. They seem to perch on statues of white men more than they do on statues of people of color, or women.
Re: (Score:2)
So we're going to have racist sexist pigeons now? Or maybe we already do. They seem to perch on statues of white men more than they do on statues of people of color, or women.
And on the odd occasion that they do land on a statue of a person of color. They try to paint it white.
I knew it (Score:4, Funny)
So it was true, pigeons aren't real, they have a little AI chip inside their robotic body.
Re: (Score:2)
pipe dreaming (Score:2)
Utter nonsense. No one knows how pigeons work, and no one really even knows how neural-net based AI works, either, in the sense that decisions are opaque.
Re: (Score:2)
Hahaha, yes. But they need the current AI hype going longer to make more money of it. And there are enough humans that are missing general intelligence and have not noticed how bad at things LLMs are.
Comment removed (Score:3)
Seems they are lowering the bar now. A lot. (Score:2)
Because pigeons are fucking stupid.
Title. (Score:2)