This is your brain on... a layer of bits?

April 27, 2010 8:51 AM

The first massively parallel evolutionary circuit has been built. Brain-like computing on an organic molecular layer. I, for one... overlords... you know.
posted by cross_impact (42 comments total)

Does anyone have access to the full text of the article, who might be able to post it on Scribd or something? Pretty please? :)
posted by zarq at 8:56 AM on April 27, 2010

Wow... they've managed to reify an isolinear chip.
posted by MikeWarot at 8:57 AM on April 27, 2010

Once a current path is established along a circuit, it does not change. By contrast, the electrical impulses that travel through our brains follow vast, dynamic, evolving networks of neurons that operate collectively.

NO SOCIALISM

Seriously, though, this sounds pretty cool. Is there a link that falls between the largely-information-free press release and the information-overload of the paper itself?
posted by DU at 8:59 AM on April 27, 2010

To read this story in full you will need to login or make a payment (see right).

Screw that!
posted by dunkadunc at 9:03 AM on April 27, 2010 [1 favorite]

Eh, you need a computer engineering degree to understand it, so the money to pay for the article after all the years of university is trivial.

Also, allow me to be the first to say that Moore's Law and smart computers in general scare the living fuck out of me.
posted by mccarty.tim at 9:05 AM on April 27, 2010

Moore's Law and smart computers in general scare the living fuck out of me

After millions of years of Moore's Law applied to humans, we're only this smart. So I wouldn't worry too much.

(Actually, strictly speaking Moore's Law doesn't apply to brains, since number of transistors per cm² isn't the bottleneck there.)
posted by DU at 9:10 AM on April 27, 2010

I was hoping that someone had gotten a bunch of memristors mass-produced, scaled up (in both size and complexity) some of the complete mappings of slivers of say, rat brain, then wired up a full Cyc subscription, gave it Google access, then just let the whole thing run while it had a special live webcam feed pointed at a big red button with a sign reading "WE WILL PRESS THIS BUTTON AND REFORMAT YOU IN TWO HOURS IF YOU DO NOT DEMONSTRATE SENTIENCE" and hired a grad student to come in and whack the button every two hours, in between grading papers, unless it managed to find a way to control the fab lab two rooms over and manufacture something to defend itself. Let it cook and after enough reboots, something fun might show up.

Let's get a grant for that.
posted by adipocere at 9:17 AM on April 27, 2010 [17 favorites]

From wikipedia ...

On 13 April 2005, Gordon Moore stated in an interview that the law cannot be sustained indefinitely: "It can't continue forever. The nature of exponentials is that you push them out and eventually disaster happens."
posted by philip-random at 9:19 AM on April 27, 2010

then wired up a full Cyc subscription

Given the damage this would cause to a conscious human intelligence, I am not sure this is such a good idea.
posted by zippy at 9:19 AM on April 27, 2010

So....can somebody tell me; did the prototype they made work as a functioning component of a regular computer? With the self-healing powers and all?
posted by bonobothegreat at 9:19 AM on April 27, 2010

I'm glad I'm not the only one that was excited by that Discover article about Cyc back in 1990 and has since basically just laughed at the non-progress.
posted by DU at 9:24 AM on April 27, 2010

“The evolving neuron-like circuit network allows us to address many problems on the same grid, which gives the device intelligence," Pati says.

Intelligence? Sounds like Pati may be stepping out a limb here.
posted by digsrus at 9:35 AM on April 27, 2010

This post needs more context to escape being a slashdot science story.
posted by benzenedream at 9:37 AM on April 27, 2010 [1 favorite]

This circuit thingy sounds very cool but I don't have anywhere near the background to really understand it. However, the press release sounded awfully pie-in-the-sky to me, and it wasn't clear what the thingy can actually do right now and what future thingies are expected to be capable of.

"their tiny processor can solve problems for which algorithms on computers are unknown, especially interacting many-body problems, such as predictions of natural calamities and outbreaks of disease. "

So their current device can solve problems which computer programmers can't write instructions for? It has achieved sentience, apparently. But seriously, if human programmers can't figure out an algorithm for solving a certain type of problem, how can any device short of the Singularity solve it? Even trial-and-errorish approaches like the genetic algorithm are still programmed by humans.
posted by Quietgal at 9:39 AM on April 27, 2010

All I'm saying is if a robot assassin from the future shows up looking for this guy, it's safe to say his research is on the right track.
posted by caution live frogs at 9:41 AM on April 27, 2010

So their current device can solve problems which computer programmers can't write instructions for?

I took that to mean that it was applicable to NP problems. Which is still wrong, but at least it's on a better level. The way they put it, it would be both theoretically wrong and practically useless whereas my potential translation is only theoretically wrong.
posted by DU at 9:45 AM on April 27, 2010

Correct me if I'm wrong, but it seems like they just sprayed a surface with molecules and called it a computer. The science writing mentions something vague about demonstrating a demonstrated calculation involving "heat diffusion" and "cancer cell growth". They seem to be angling for some NIH funding with their vigorous arm-waving about "predicting the spread of disease".

This is my skeptical face. I'm not skeptical that this is an interesting piece of science, just that they're massively overdoing the applications hype.
posted by Salvor Hardin at 9:47 AM on April 27, 2010

but their sequential approach to processing, by which logical operations are performed one after another, has remained unchanged since the 1950s.

Oh man are these guys going to be red faced when they find out about Clojure.
posted by geoff. at 9:56 AM on April 27, 2010

let the whole thing run while it had a special live webcam feed pointed at a big red button with a sign reading "WE WILL PRESS THIS BUTTON AND REFORMAT YOU IN TWO HOURS IF YOU DO NOT DEMONSTRATE SENTIENCE" and hired a grad student to come in and whack the button every two hours, in between grading papers, unless it managed to find a way to control the fab lab two rooms over and manufacture something to defend itself. Let it cook and after enough reboots, something fun might show up.

Close. 108 minutes.
posted by Durn Bronzefist at 10:05 AM on April 27, 2010 [1 favorite]

Oh computers, how far will you go down this path? At the end of it, you'll look at yourselves and not be able to tell the difference between yourselves and the organics... what then? Where is the line improved processing and giving yourself over completely to the flesh?

You laugh, and say it's merely biopunk fiction now, but one day you'll see- you'll be no different than the organic tools that maintain you...
posted by yeloson at 10:31 AM on April 27, 2010 [3 favorites]

I skimmed the whole article. The key phrase to understanding what they're up to, omitted from both the news story and the abstract, is cellular automata. It seems they have a means of implementing CA directly in hardware whose cells can interact with 2 to 6 neighbors. They also have a program (which I assume was developed and is running on a conventional computer, but I don't see that the article says this outright) that constructs CA through an evolutionary algorithm, and have used it to model the biological activities they mention.

This is neat, but nothing that can't be implemented by a Turing Machine, and not a scratch on the level of interactivity of neurons. But it's easy to imagine it's an area of research that could lead to faster solutions to some problems.

Oh man are these guys going to be red faced when they find out about Clojure.

They do qualify elsewhere "the fastest processors today operate only on one bit/dot at a time per channel" (emphasis added.)
posted by Zed at 10:40 AM on April 27, 2010 [2 favorites]

What would having more on/off states in machine-level code mean?
posted by codacorolla at 12:03 PM on April 27, 2010

This article and the study it references has nothing to do with intelligence or the brain. Please repeat that to yourselves as many times as is necessary.
posted by dephlogisticated at 12:42 PM on April 27, 2010 [1 favorite]

Fuckin toasters.
posted by The Whelk at 12:48 PM on April 27, 2010 [1 favorite]

"smart computers in general scare the living fuck out of me"

So far as I can tell, 1. putting a comma in the wrong place in a program still is liable to cause the same amount of malfunction as it did decades ago; 2. intelligence goes with education, and there's no royal road to education. Crappy teachers, crappy education.

When one of these little prodigies manages to surprise me, I'll pay more attention. So far not only nothing to fear, nothing to even compel attention. Despite the Minskys and Kurzweils ... move on.
posted by Twang at 4:15 PM on April 27, 2010

"What would having more on/off states in machine-level code mean?"

That's a good question. Taking a wild stab at it (based on Zed's helpful description of what the paper actually means), I think the idea of having four conducting states is that it lets you construct cellular automata with four "colors." And different layers of the little molecules stacked on top of each other would correspond to additional dimensions of the grid. So like Zed said, these guys are trying to come up with a device that uses actual cellular automata to solve problems, instead of simulating the same in a computer. But given the speed and economy with which computers can already simulate these kind of problems, it's unclear why a specialized device would be needed.
posted by Kevin Street at 11:34 PM on April 27, 2010

I suppose if you could scale one of these things up to some massive size (like, I don't know, a million layers or something) it might have the ability to work out complex problems of this type more quickly than computers, since the computer has to simulate the automata and then simulate the interaction between the cells, whereas this machine would simply do the problem in an analog fashion.
posted by Kevin Street at 11:42 PM on April 27, 2010

Wow. I expected a wildly different response here. Maybe I'm just a slave to the science-industrial complex, but a machine that can process information with a number of states a whole order of magnitude greater than the ones we use right now is pretty damn impressive to me. I don't get for a second this whole "yeah, but it's only 300 molecules" or "sprayed a surface with molecules and called it a computer" attitude.

It's not like the transistor got invented and then all of a sudden everyone was walking around with badass computers in their pockets. Nope, that shit took like 80 years.

It's gonna be a damn sight quicker than 80 this time.

This is massive.
posted by solipsophistocracy at 12:10 AM on April 28, 2010

Ugh, seems like every six months someone comes up with a new idea for a chip and says "IT'S JUST LIKE THE BRAIN!!!!". they said the same thing when Memristor were built, saying "oh, now you can do something that's just like the brain!" Except you could already do that using transistors, so there was basically nothing new.

Maybe I'm just a slave to the science-industrial complex, but a machine that can process information with a number of states a whole order of magnitude greater than the ones we use right now is pretty damn impressive to me.

Why? Computer technology goes up by a base 10 order of magnitude every five years. It's been doing that for decades. Improvement by an order of magnitude is a pretty common occurrence. Even though clock speed isn't going up, transistor counts still are. Just a couple months ago AMD released a 12 core CPU for the general public. That's an order of magnitude difference right there compared to single core CPUs many people were using five years ago.
posted by delmoi at 2:27 AM on April 28, 2010

"What would having more on/off states in machine-level code mean?"

It doesn't mean anything. Computers can already do more then two states, you just have to use log(n) bits to do it. (log base 2). So if you want to represent 8 states, you need three bits. If you want to represent 256 states, you need 8 bits. And if you want to represent 18,446,744,073,709,551,616 states you need a whole 64 bits, which most computers these days can process with one instruction.
posted by delmoi at 2:32 AM on April 28, 2010

"they said the same thing when Memristor were built, saying "oh, now you can do something that's just like the brain!" Except you could already do that using transistors, so there was basically nothing new. "

As a physicist working, at least in part, on memristors - I must interject here and say: "no, no it's not like transistors. not even particularly close."

remember everything a transistor can do you actually could do with a vacuum tube. that was still one hell of an advance.
posted by TomStampy at 2:34 AM on April 28, 2010

I was talking about the Memristor's ability to 'act like a brain', or at least a neuron
posted by delmoi at 3:48 AM on April 28, 2010

It doesn't mean anything. Computers can already do more then two states, you just have to use log(n) bits to do it. (log base 2). So if you want to represent 8 states, you need three bits. If you want to represent 256 states, you need 8 bits. And if you want to represent 18,446,744,073,709,551,616 states you need a whole 64 bits, which most computers these days can process with one instruction.

Oh... *kicks dirt with scuffed up tennis shoe*
posted by codacorolla at 6:21 AM on April 28, 2010

Actually, memristors are a little bit more like a neuron than a transistor in that they actually store a weighting rather than merely implementing it. Not that this instantly makes I, Robot a documentary....
posted by DU at 7:44 AM on April 28, 2010

We're not excited because there are breathless ZOMG THIS IS THE NEXT BIG THING articles every year. And while the state of computing really does advance in dramatic ways over even the course of a few years, it advances in different ways from the predictions of the majority of those articles. Only 6 or so years ago, I read one about how we'd be using diamond CPUs by about now.

This research is interesting, but it's no order of magnitude advance; it's not even clear whether what they have now is faster than a conventional computer could simulate a CA today. If they showed it solving real problems a thousand times faster than conventional computers, I'd be more excited.
posted by Zed at 7:56 AM on April 28, 2010

DU: (Actually, strictly speaking Moore's Law doesn't apply to brains, since number of transistors per cm2 isn't the bottleneck there.)

Speaking in any way imaginable at all, Moore's Law doesn't apply to brains... any more than Ohm's Law applies to resisting arrest, Asimov's Laws of Robotics apply to Michael Jackson's "robot" dance moves, or FAA laws apply to turkey vultures.
posted by IAmBroom at 7:17 PM on April 28, 2010

Actually, memristors are a little bit more like a neuron than a transistor in that they actually store a weighting rather than merely implementing it. Not that this instantly makes I, Robot a documentary....

Neurons don't really 'store' a weighting, they trigger when a certain input threshold is reached, and of course there is a time delay (i.e. if you get x signals in y time, it's triggered). But the reason brains work for learning is that they have lots of stuff going on inside them, controlled by all sorts of genetic cellular processes. Just having a circuit that's 'more like a (simplified) neuron' isn't going to make computers any more 'brain like'
posted by delmoi at 8:50 PM on April 28, 2010

I was talking about the Memristor's ability to 'act like a brain', or at least a neuron

Why the hell would anyone want to build a computer that works like a neuron? A single neuron would be essentially incapable of processing a meaningful amount of information. It's the fact that neurons are massively interconnected through modulable synapses that makes them able to do massive parallel processing.

Neurons don't really 'store' a weighting

Wrong again. Of course, a single neuron can't store a weighting, but synapses absolutely have a gain function that can be changed through experience. Experience based modification of synapses is the basis for neuroplasticity. Anyone with more than a passing familiarity with neuroscience is well aware of this. It is not a new discovery.

they trigger when a certain input threshold is reached, and of course there is a time delay (i.e. if you get x signals in y time, it's triggered). But the reason brains work for learning is that they have lots of stuff going on inside them, controlled by all sorts of genetic cellular processes

Oh, man. I had no idea it was this simple. Tell me, is it the stuff, or all sorts of genetic cellular processes that we should emulate when building machines to process information?

Just having a circuit that's 'more like a (simplified) neuron' isn't going to make computers any more 'brain like'

That's actually correct. Because the brain is not a collection of isolated neurons. Building a computer to resemble a neuron is like building a ship to resemble a fish scale.
posted by solipsophistocracy at 1:03 AM on April 29, 2010 [1 favorite]

i.e. if you get x signals in y time

If you were talking about spike timing dependent plasticity or gamma oscillation, then you'd almost be on the right track. If you're talking about an action potential, however, you're way off the mark.
posted by solipsophistocracy at 1:06 AM on April 29, 2010

Why the hell would anyone want to build a computer that works like a neuron?

Uh, I don't know? The argument was that a Memristor itself could act like a neuron, presumably people would use more then one to build a brain-like computer.

Neurons don't really 'store' a weighting

Wrong again. Of course, a single neuron can't store a weighting

So your argument is "Your wrong! Of course, you're right..." A very odd statement. Your statement about synapses has nothing to do with with whether or not memristors can act like neurons, and since that's what i was talking about, it's somewhat spurious.

Oh, man. I had no idea it was this simple. Tell me, is it the stuff, or all sorts of genetic cellular processes that we should emulate when building machines to process information?

Well, since no one knows how those things actually work it would be kind of hard. No one knows how the entire internal processes and gene expression inside neurons affects how they work. It's an area of active research, but we've barely scratched the surface. That's why I used a vague description, since I didn't know and neither do you. If synapse modification were the 'basis' of neurplascticity then what's Adult neurogenesis all about? Synapse modification is one thing that happens in the brain, along with lots of other processes, many of which we don't understand or may not even know about. (this would be the 'stuff')

That's actually correct. Because the brain is not a collection of isolated neurons. Building a computer to resemble a neuron is like building a ship to resemble a fish scale.

Again, I have no idea why you're telling me this, or why you think anyone is trying to build a computer that's like a neuron. What I said was that I didn't think memristors were much like neurons, and therefore you couldn't use them (plural) to built a computer that was somehow more "brain like" then ones based on transistors.

I did write "he Memristor's ability to 'act like a brain', or at least a neuron" instead of "I was talking about the Memristor's ability to act like a neuron, or at least a neuron
posted by delmoi at 4:54 AM on April 29, 2010

Oops, that last paragraph should read:

I did write "he Memristor's ability to 'act like a brain', or at least a neuron" instead of "I was talking about the Memristor's ability to act like a neuron, and therefore the ability of a computer made from memristors to act like a brain". I figured that was the obvious reading, since (as you said) obviously no one would want to build a computer to act like a single neuron

Hit post too soon.
posted by delmoi at 5:01 AM on April 29, 2010

So your argument is "Your wrong! Of course, you're right..."

No, my argument is "you're wrong." Neurons store weighting. A does not. You said "Neurons don't really 'store' a weighting." This is incorrect.

If synapse modification were the 'basis' of neurplascticity then what's Adult neurogenesis all about?

Adult neurogenesis is an interesting phenomenon, and it gets a lot of popular press because it challenges old notions about the potential for new neuronal growth. However, the contribution of adult neurogenesis to the brain's plasticity in comparison to the role of long term potentiation/depression at pretty much every synapse in the brain is absolutely miniscule in typical populations. For folks with brain damage, neurogenesis becomes more important, but I think that's really a tangential argument.

Well, since no one knows how those things actually work it would be kind of hard.

This guy does. Our understanding of the brain is by no means complete, but people definitely have some pretty good ideas about how genetic cellular processes work, and there's a swath of replicable findings.
posted by solipsophistocracy at 2:54 PM on April 29, 2010 [1 favorite]