I, for one, welcome...

April 29, 2007 9:43 AM

[scifilter] Scientists use a supercomputer to simulate a biological neural structure "as big and as complex as half of a mouse brain"
posted by delmoi (51 comments total)

++ Out of Cheese Error. Redo From Start. ++
posted by Wolfdog at 9:45 AM on April 29, 2007 [6 favorites has favorites]

Supercomputer=Half a mouse brain.

They should use a mac instead, could perhaps do the whole brain.
posted by HuronBob at 9:47 AM on April 29, 2007

The vast complexity of the simulation meant that it was only run for ten seconds at a speed ten times slower than real life - the equivalent of one second in a real mouse brain.

*sees cat*

*thinks. must.......... run.......... awa.. chomp!*

program terminated early.
posted by three blind mice at 9:49 AM on April 29, 2007 [1 favorite has favorites]

the mice are still the ones experimenting on us, I'm afraid.
posted by oonh at 9:53 AM on April 29, 2007

three blind mice said "*sees cat*"

darn....that's funny...
posted by HuronBob at 10:01 AM on April 29, 2007

Boy, if ever a science article needed more information...

On other smaller simulations the researchers said they had seen "biologically consistent dynamical properties" emerge as nerve impulses flowed through the virtual cortex.

Really? What kind of "biologically consistent dynamical properties" are we talking about? We don't even get an attempt to describe them, just a mention of "a very short research note" from the scientists. Sounds like a very poor and premilinary simulation of "biological neural structure." I love this subject area, but I'm having a hard time seeing what's newsworthy here.
posted by mediareport at 10:04 AM on April 29, 2007

Boy, if ever a science article needed more information...

BBC science articles always leave me wanting a lot more. There's a little more info [pdf] on Modha's blog:

On other (smaller) models, we have observed biologically consistent dynamical properties such as spontaneous formation of neuronal groups and synchronous/asynchronous firing patters.

posted by blendor at 10:24 AM on April 29, 2007

Virtual cortex?
posted by Elmore at 10:29 AM on April 29, 2007

Pinky: "What are we gonna do tonight, Brain?"

The Brain: "The same thing we do every night, Pinky - Try to take over the world!"
posted by inconsequentialist at 10:34 AM on April 29, 2007 [3 favorites has favorites]

Mousenet became self-aware at 2:14am EDT August 29, 2007.
posted by mr_crash_davis at 10:43 AM on April 29, 2007

the early bird gets the worm, but the second mouse gets the cheese
posted by pyramid termite at 10:47 AM on April 29, 2007

This was not as complicated as half of a mouse brain. This simulation was neuron soup. I don't mean to trivialize it - what Henry Markram has done with Blue Gene/L is unprecedented. Nonetheless, a bowl of neuron soup that pays absolutely no importance to the differentiation and highly evolved connectivity of the brain will never achieve intelligence. It won't be until the next simulation that he tries to account for the actual structure of the mouse brain, so you can can the headline.

By the way, even though it's unprecedented, it's still not all that impressive. It takes this to simulate that? :)



posted by reflection at 11:08 AM on April 29, 2007

It's just a mouse.
posted by neustile at 11:10 AM on April 29, 2007

Call me when they simulate half an elephant brain.
posted by katillathehun at 11:21 AM on April 29, 2007

Mousesim 1.0 running....

running....

running...

calculating output...

42.
posted by Joe Invisible at 11:28 AM on April 29, 2007

The vast complexity of the simulation meant that it was only run for ten seconds

Why'd they only run it for 10 seconds? It's not like this was a supercollider or a fusion reaction that could've taken out the neighborhood if it got out of control - it was a computer simulation. Were they having trouble paying the electricity bill?
posted by rkent at 11:38 AM on April 29, 2007

iirc they simulated a vastly reduced neural network whose computation approximated 10 seconds of activity - the simulation lasted much longer than 10 seconds.

I suspect that the amount of data generated in approximating those 10 seconds of activity is vast so now they're going to poke around the data a bit before making adjustments to the program and running it again.

It's impressive, but again - another anger-inducing piece of science reporting.

If anyone wants to play around with simulating the interaction between a piddling few neurons, there's Neuron which lets you make assumptions about different parameters of the different neurons in a simple circuit (ie., membrane resistance, dendrite diamter, membrane potential, receptor #s, &c).
posted by porpoise at 11:48 AM on April 29, 2007

Why'd they only run it for 10 seconds?

More than ten seconds of even a very basic neural simulation might attract the attention of the Eschaton.

And rule one of AI is YOU DO NOT FUCK WITH THE ESCHATON.
posted by ROU_Xenophobe at 11:51 AM on April 29, 2007 [2 favorites has favorites]

Which half?
posted by armoured-ant at 11:53 AM on April 29, 2007

More than ten seconds of even a very basic neural simulation might attract the attention of the Eschaton.

Wow, I didn't know Atrios caused that much fear in AI circles ... ;-)
posted by teece at 11:59 AM on April 29, 2007 [1 favorite has favorites]

Wake me up when they use half a mouse brain to simulate a super-computer.
posted by grobstein at 12:09 PM on April 29, 2007 [2 favorites has favorites]

I didn't the article itself was that good, but it wasn't as bad as some science writing can be.

Yeah, it could have been a hell of a lot more interesting if they'd tried to map out all the neurons in a real mouse brain, and so on, but this was the first I'd heard of anyone doing anything like this at that scale.
posted by delmoi at 12:22 PM on April 29, 2007

This is as good a time as any to link to a story I wrote a few years ago about computer-simulated brains.
posted by Tlogmer at 12:46 PM on April 29, 2007

I think the article says the simulation ran for 10 seconds and it modeled 1 second of mouse brain activity.

My guess is they can't go more than 10 seconds as they might run out of computer memory - as in chaos theory where every step forward creates exponential complexity of possibilities.
posted by stbalbach at 1:25 PM on April 29, 2007

Well, at least it's a reasonable alternative for running the country.

Maybe not HAL, but I can live with it for the time being.
posted by spiderwire at 1:28 PM on April 29, 2007

Well, at least it's a reasonable alternative for running the country.

yeah ... until we end up ass deep in cheese with the fbi confiscating every cat in the country
posted by pyramid termite at 1:43 PM on April 29, 2007

Yeah, in retrospect- why don't we hack out some mouse brains and try to interface with them? If we can get it from NEED CHEESE to n=cheese, we've got some really cheap supercomputers.
Man, why can't life be more like my ridiculous science fiction dreams?
posted by 235w103 at 1:46 PM on April 29, 2007

I went to a talk recently where a guy was talking about getting a mouse "wiring diagram" via electron microscopy of a sectioned, frozen mouse brain. That's still a few years off, though. It's was done with the C. elegans nervous system in the 70s, or maybe early 80s, but those worms only have a few hundred neurons, and despite this simplicity, no one's been able to turn that wiring diagram into a plausible simulation of a worm nervous system, as far as I know. So they're kind of guessing when they compare the complexity of their system to that of a mouse brain, and it's likely to be a long time before anyone actually manages to accurately simulate even a portion of one.
posted by Coventry at 1:50 PM on April 29, 2007

It will not be long before we can accurately simulate zombie mice. BRANEZ! MMMM, GOOD!
posted by five fresh fish at 2:10 PM on April 29, 2007

I'm writing a paper right now on the problems (and possibilities) of machine intelligence. I take a break to check MetaFilter and what do I end up reading? An article about... err... simulating a brain in a computer. Fuck this, I'm going back to playing FreeCell.
posted by OverlappingElvis at 2:24 PM on April 29, 2007

Did they name it Alger?
posted by shmegegge at 2:24 PM on April 29, 2007

This seems like it would pair pretty well with the work discussed in this sensational Popular Science article. where a researcher has been able to send and receive impulses from a chip to a brain slice. Given the brain's ability to self-heal by shifting functions away from a damaged area, I wonder if it would possible to:

1) Create big undifferentiated sim-neuron soup

2) Connect neuron soup to functional wetware brain

3) Give brain time to adapt to using the newly-available "neurons"

4) shave off a bit of the original brain

5) repeat steps 3 & 4

I wonder how much brain function, if any, could be transferred to silicon in this way? Then we can .rar it, hand it off to some warez kidz and tell them it's Duke Nukem Forever. We'll have the secrets of consciousness decoded and available on p2p within a week. As long as we're fantasizing, we might as well dream big amirite?
posted by contraption at 2:56 PM on April 29, 2007 [2 favorites has favorites]

this story is a little inaccurate. the sim in question is undifferentiated. the real brain has a very specific structure to it, and is very differentiated. also I suspect that the neurons used were very simplistic compared to the processes that go on in real neurons.
posted by bhouston at 3:11 PM on April 29, 2007

>Why'd they only run it for 10 seconds?

Who knows. Maybe thats all the time they could get for their money. I can't imagine the use of that supercomputer is going to be cheap. Or it may be that's all they needed for the experiement. They got their virutal snapshot of whatever they wanted.
posted by damn dirty ape at 3:27 PM on April 29, 2007

Yeah, without simulating the structures it seems relatively pointless. Also, it seems to me that you are not going to get a very good sense of how real brains work without simulating development.

most of the wiring in actual brains seems to take place via exposure to experience so if you are starting with something "pre set" in some way, it doesn't seem like it would be very likely to be much like a real brain.

it's also probably easier to start with a fetal mouse brain simulation..
posted by Maias at 4:03 PM on April 29, 2007

coventry - thanks for bringing that up. I, too, am surprised they didn't go for C. elegans - the wiring schematic is well known, there's only 103 (?) neurons and a paltry 8000 synapses, and people are already working on computer modeling. (And the worm people are one of the most open and sharing communities in biological sciences.)

Going to mouse, I guess, is sexy but it's like trying to exit Earth's gravity when you only know how to build a kite or glider.

contraption - interesting link. A (very - I've been drinking) cursory search of pubmed doesn't pull up any papers on Ted Berger's technique over the past couple of years.

Mike Colicos, however, has published techniques to stimulate specific neurons in culture, though, and create long-term activity patterns with a relatively simple setup.

As for gradually siliconizing the brain... we don't really know how the brain works well enough to know if that'll work. Not only is the connectivity important, but the strengths of those connections are too - I wonder if Berger's setup can also take into account the strengths of the synaptic connection. There's also very likely other signalling other than ("electric") depolarization that goes on in neurons (soluble second messengers, pre-synaptic receptors, &c), which may not be able to be reproduced by silicone.

Also, the brain doesn't work by sending linear signals; timing has a lot to do with cognition and brain states. I wonder if lagginess in the silicone would be a problem (if it's faster than a brain cell, I suppose you can program in lag).

I'm guessing the largest problem is that the silicone doesn't change over time; neurons aren't static and will break off connectivity or strengthen connectivity or change the types/amounts of proteins they make or change the kinds of modifications they make to proteins in response to different kinds of activity (or lack of activity) and activity, again, isn't limited to just electrochemical depolarization (such as in response to hormones and growth factors [which do more than simply promote growth/survival], &c).

That's not even taking into account that there is very likely ongoing neurogenesis from neuronal stem cells - some, but not all, of these cells seem to integrate into the existing brain by some as yet undetermined mechanism.

So... silicone brains are still quite a far ways away unless there's a breakthrough in decoding the "language of the brain."
posted by porpoise at 4:05 PM on April 29, 2007

...gravity well...
posted by porpoise at 4:06 PM on April 29, 2007

So... silicone brains are still quite a far ways away
But when we get there, the breasts will be sentient!

Three cheers for breast augmentation!
posted by wierdo at 4:31 PM on April 29, 2007 [1 favorite has favorites]

yeah ... until we end up ass deep in cheese with the fbi confiscating every cat in the country

actually, i prefer those policy priorities over our current ones.
posted by spiderwire at 4:32 PM on April 29, 2007 [1 favorite has favorites]

porpoise, do you mean silicon or silicone?
posted by spiderwire at 4:33 PM on April 29, 2007

whoops. Aliens must have replaced my brain with a silicone implant.

I wish it was a saline one instead. What if it breaks?
posted by porpoise at 4:40 PM on April 29, 2007

And the worm people are one of the most open and sharing communities in biological sciences.

They have to be, looking like they do.
posted by Sparx at 4:57 PM on April 29, 2007 [1 favorite has favorites]

There's also very likely other signalling...

IIRC we had a recent FPP on someone who thinks it's all about soundwaves, not electrical impulses.
posted by five fresh fish at 6:31 PM on April 29, 2007

it could all just be a timecube.
posted by b1tr0t at 7:51 PM on April 29, 2007

it could all just be a timecube.

You say you want the simulated neurons to be differentiated and have structure, but really you just want it EDUCATED STUPID.
posted by grobstein at 8:03 PM on April 29, 2007

Metafilter: from a chip to a brain slice.
posted by CynicalKnight at 8:14 PM on April 29, 2007

Sorry grobstein, but that's just what they want you to think.
posted by pompomtom at 9:53 PM on April 29, 2007

Did they name it Alger?
posted by shmegegge at 2:24 PM on April 29

Non, monsieur, ceci n'est pas une souris.
posted by eritain at 3:10 AM on April 30, 2007

There's also very likely other signalling...

IIRC we had a recent FPP on someone who thinks it's all about soundwaves, not electrical impulses.

And this dude speculates that the electrical impulse is the medium, not the message; the message, in his hypothesis, is which longitudinally-connected chain of ion channels is responsible for the action potential. After that his stuff gets way wacky, and I haven't had time to check it out yet, but at least read the first two or three posts through.

That stuff may or may not be gorp itself; but structure is a fact, and these guys are neglecting that wholesale. Second the motion to do it on a properly-modeled C. elegans network, with the real deal alongside in a dish for comparison.
posted by eritain at 3:18 AM on April 30, 2007

And why only ten seconds? Well, when I was at BYU, the absolute slowest way to get anything done was to do it on the supercomputers. There's an allocation committee *shudder*, and you're always competing with other users who are also ... simulating rodents.
posted by eritain at 3:32 AM on April 30, 2007

By the way, even though it's unprecedented, it's still not all that impressive. It takes this to simulate that? :)

Which actually makes the mouse brain the impressive part...
posted by wildcrdj at 4:38 PM on April 30, 2007