Smart spider and AI

[Mr Skeptic]

Maybe this belongs in biology, but I think that it has applications to AI. There is a spider, Portia labiata, that is shockingly intelligent (for a spider, anyways). Portia is a spider that hunts spiders, often tackling pray much more powerful than itself, and apparently relying on its brainpower for much of its success. Portia has excellent vision, surpassing some birds and mammals, but in a very limited field. However, its good eyes can swivel to scan an area, and this seems to be part of how it can solve problems. Interestingly, some of Portia's maneuvering can last hours and continue despite losing sight of its prey (for Portia, out of sight is not out of mind). Also, Portia seems to have some capability to learn.

 

Main story: http://www.freerepublic.com/focus/f-chat/1640513/posts

Abstract, concerning the relation between eye movements and path-finding.

 

Anyhow, Portia has about 600,000 neurons, compared to 100 billion for a human, 70 million for a mouse, 1 million for a honeybee, and 250,000 for a housefly.

 

Anyhow, it occurs to me that if this spider is smart, we ought to be able to make an AI that is smarter than that. How hard could it be to simulate 600,000 neurons?

[iNow]

One thing I'm honestly not sure about, but which I think is a relevant question is... Do "number" of neurons really matter, or is the important bit how they are organized and conected?

[Mokele]

Smart animals with few neurons is far from unusual. Octopi are astonishingly smart, but don't have terribly high neuron counts either. Less extreme, New Guinea tree monitor lizards have a brain the size of a large lima bean but are *frighteningly* intelligent, and Cuban crocodiles, with a brain the size of a small cigar, display pack-hunting behavior.

 

A lot of the human brain is sensory, after all, in part due to our phenomenally good vison (seriously, primates are up there with hawks at the top of the list for vision)

[ydoaPs]

Smart animals with few neurons is far from unusual. Octopi are astonishingly smart, but don't have terribly high neuron counts either. Less extreme, New Guinea tree monitor lizards have a brain the size of a large lima bean but are *frighteningly* intelligent, and Cuban crocodiles, with a brain the size of a small cigar, display pack-hunting behavior.

 

A lot of the human brain is sensory, after all, in part due to our phenomenally good vison (seriously, primates are up there with hawks at the top of the list for vision)

 

I thought the important part of our brain(AI-wise) is only as big as a dinner napkin and as thick as six playing cards.

[Mr Skeptic]

One thing I'm honestly not sure about, but which I think is a relevant question is... Do "number" of neurons really matter, or is the important bit how they are organized and conected?

 

Well, the number of neurons limits your computing capabilities, so it does matter to some extent. And of course, an awful lot of the brain is going to be for perception and body control, which needn't be part of intelligence. More to the point with respect to AI, the number of neurons limits how hard it would be to simulate that brain. For example, the blue brain project simulated a rat neocortical column, with 10,000 neurons. They're not too far off from being able to simulate the entire brain of Portia, which may be more interesting but I don't think they will try (they're focusing on mammal neocortex).

---

Merged post follows:

Consecutive posts merged

I thought the important part of our brain(AI-wise) is only as big as a dinner napkin and as thick as six playing cards.

 

So they say, but I doubt that it would be much use without the other parts of the brain. I'm pretty sure you use your visual cortex when you think visually, for example.

[ydoaPs]

Well, the number of neurons limits your computing capabilities, so it does matter to some extent. And of course, an awful lot of the brain is going to be for perception and body control, which needn't be part of intelligence. More to the point with respect to AI, the number of neurons limits how hard it would be to simulate that brain. For example, the blue brain project simulated a rat neocortical column, with 10,000 neurons. They're not too far off from being able to simulate the entire brain of Portia, which may be more interesting but I don't think they will try (they're focusing on mammal neocortex).

---

Merged post follows:

Consecutive posts merged

 

 

So they say, but I doubt that it would be much use without the other parts of the brain. I'm pretty sure you use your visual cortex when you think visually, for example.

The visual cortex is part of the neocortex, isn't it?

[iNow]

Yessir. The visual cortex (aka the "occipital lobe") is classified as part of the neocortex.

 

 

 

 

Mr S - I'd encourage you to validate your assumption that number of neurons is the key factor. That does, of course, play a role, but it's the number of dendritic connections and the routing which matters most.

[Xittenn]

Anyhow, it occurs to me that if this spider is smart, we ought to be able to make an AI that is smarter than that. How hard could it be to simulate 600,000 neurons?

 

At first I thought hmmmmmmm no problem standard off the shelf computer from Future Shop plus or minus a generation. And then I thought about it a bit @100fps you get like 100 clock cycles per neuron per frame. That's like an average of 15-25 JMP FE/MOV FF etc. machine commands per neuron per frame. So I Wikied the generalities on mathematical representations of a neuron and got this............So I place my bets on an SGI machine with like 1000 dual(not quad) core Itanium processors. That's my best guess!

 

I was hoping to make a comparison between this and the worlds smallest organism which was modeled on the atomic level in real time a few years ago but I can't find anything on it.............

 

oooooo found it..........

 

so 50nanoseconds of real time(Quantum Physicists definition on this; I used this term not them) simulation took 50 days.............I think my guess should hold!

Edited March 8, 2009 by buttacup

[Dudde]

Apparently we've been studying that spider for quite some time now, but I've never even heard of similarities - I'm not an entomologist by any means, but I've always found spiders fascinating

 

It's a very interesting read, I've been studying the Portia genus all day (I'm stuck at work, and apparently all storage devices work on saturday.)

[bascule]

One thing I'm honestly not sure about, but which I think is a relevant question is... Do "number" of neurons really matter, or is the important bit how they are organized and conected?

 

Both matter, just as they would with transistors or any other sort of component.

 

However, the largest part of the human brain in terms of neurons, the cerebellum, isn't even necessary for humans to function (although any human without one is going to be a bit of a klutz)