SPAUN of the living
December 8, 2012 6:09 AM Subscribe
The simulated brain - "First computer model to produce complex behaviour performs almost as well as humans at simple number tasks." [1,2,3,4,5,etc.]
For those curious, here's a wiki list of animals ranked by neuron count. SPAUN is like two cockroach brains taped together.
posted by codacorolla at 6:35 AM on December 8, 2012 [3 favorites]
posted by codacorolla at 6:35 AM on December 8, 2012 [3 favorites]
Semantic Pointer Architecture Unified Network, which is short for "we wanted a cool acronym".
posted by Wolfdog at 6:36 AM on December 8, 2012 [5 favorites]
posted by Wolfdog at 6:36 AM on December 8, 2012 [5 favorites]
SPAUN is like two cockroach brains taped together.
I didn't realize they were able to replicate [insert politician name]'s brain already!
posted by blue_beetle at 6:46 AM on December 8, 2012 [2 favorites]
I didn't realize they were able to replicate [insert politician name]'s brain already!
posted by blue_beetle at 6:46 AM on December 8, 2012 [2 favorites]
Alright, so, how am I revising my Singularity Calendar? 50 years now? 40?
posted by Mooseli at 7:13 AM on December 8, 2012 [1 favorite]
posted by Mooseli at 7:13 AM on December 8, 2012 [1 favorite]
The trouble is that several technological singularities have actually occurred in recent memory. Who guessed right about the explosion in cell phone usage and social networks? Not even DARPA anticipated what a big deal the internet would be.
posted by LogicalDash at 7:26 AM on December 8, 2012 [1 favorite]
posted by LogicalDash at 7:26 AM on December 8, 2012 [1 favorite]
Maybe it's just my reflexive anti-robot prejudices showing, but I totally thought the headline said the electrobrain was performing 'simple murder tasks'.
posted by FatherDagon at 7:29 AM on December 8, 2012 [1 favorite]
posted by FatherDagon at 7:29 AM on December 8, 2012 [1 favorite]
Alright, so, how am I revising my Singularity Calendar? 50 years now? 40?
Well, it's been "40 years away" for several decades now, so that seems like a good bet.
posted by IjonTichy at 7:45 AM on December 8, 2012 [1 favorite]
Well, it's been "40 years away" for several decades now, so that seems like a good bet.
posted by IjonTichy at 7:45 AM on December 8, 2012 [1 favorite]
We are now just a few decades away from the Butlerian Jihad because of this...
posted by JibberJabber at 8:14 AM on December 8, 2012 [5 favorites]
posted by JibberJabber at 8:14 AM on December 8, 2012 [5 favorites]
SPAUN is like two cockroach brains taped together.
Metafilter: It's like two cockroach brains taped together.
posted by StrangerInAStrainedLand at 9:42 AM on December 8, 2012 [1 favorite]
Metafilter: It's like two cockroach brains taped together.
posted by StrangerInAStrainedLand at 9:42 AM on December 8, 2012 [1 favorite]
Yeah, so if you happen to have a beefy computer and lots of free time, you can download SPAUN and run it yourself.
posted by LogicalDash at 10:33 AM on December 8, 2012 [1 favorite]
posted by LogicalDash at 10:33 AM on December 8, 2012 [1 favorite]
SPAUN is like two cockroach brains taped together.
So basically they have a computer that obsesses about crumbs, breeds like nobody's business, and just disappears when you turn on the light?
posted by Grangousier at 10:46 AM on December 8, 2012 [2 favorites]
So basically they have a computer that obsesses about crumbs, breeds like nobody's business, and just disappears when you turn on the light?
posted by Grangousier at 10:46 AM on December 8, 2012 [2 favorites]
- This model requires a machine with at least 24GB of RAM to run the full implementation.
Estimated run times for a quad-core 2.5GHz are 3 hours per 1 second
of simulation time.
Think I'm gonna need an upgrade.
posted by BungaDunga at 11:02 AM on December 8, 2012 [3 favorites]
Estimated run times for a quad-core 2.5GHz are 3 hours per 1 second
of simulation time.
Think I'm gonna need an upgrade.
posted by BungaDunga at 11:02 AM on December 8, 2012 [3 favorites]
So the scientists behind this model did a AMA (ask me anything) interview over at reddit a couple of days ago.
Probably one of the best ones ever, IMO.
Linky
posted by psycho-alchemy at 12:10 PM on December 8, 2012 [1 favorite]
Probably one of the best ones ever, IMO.
Linky
posted by psycho-alchemy at 12:10 PM on December 8, 2012 [1 favorite]
Next step: hooking it up to a cup of tea.
Next next step: hooking it up to a quadcopter.
posted by the painkiller at 12:29 PM on December 8, 2012
Next next step: hooking it up to a quadcopter.
posted by the painkiller at 12:29 PM on December 8, 2012
Singularity ... that's where all the stuff collapses into a black pile at the bottom of the pit and stops emitting light ... becoming nothing more than a danger to anything passing nearby ... and only creates light when it rips something apart. Right?
And we're supposed to look forward to this?
posted by Twang at 6:16 PM on December 8, 2012
And we're supposed to look forward to this?
posted by Twang at 6:16 PM on December 8, 2012
I would not be hoping for a "singularity" just yet. If we simulate the brain, we might just discover that the reason why humans forget, are emotionally unstable and suffer from mental disorders is just part and parcel of how our brains are built.
In short, I have always thought that our higher purpose on this world might just as well be to help create a superior intelligence, but also that if we just replicate ourselves, we will not end up with superinteligent robots, but rather artificial humans who would refuse to cooperate if not motivated properly, etc.
Also, an integral part of the artificial brain's function is the range of peripherals. Unless they match ours, the artificial brain will not be able to comprehend humans. But I have little doubt that those peripherals are also coming.
Eventually, we will find the right balance of memory, obedience and physical capability to create a sentient, obedient servant who'll not only grow our food, but help us solve problems.
In light of all this, I'd like to raise my glass of brandy and say that I, for one, welcome our new robot slaves!
posted by Laotic at 11:36 PM on December 8, 2012
In short, I have always thought that our higher purpose on this world might just as well be to help create a superior intelligence, but also that if we just replicate ourselves, we will not end up with superinteligent robots, but rather artificial humans who would refuse to cooperate if not motivated properly, etc.
Also, an integral part of the artificial brain's function is the range of peripherals. Unless they match ours, the artificial brain will not be able to comprehend humans. But I have little doubt that those peripherals are also coming.
Eventually, we will find the right balance of memory, obedience and physical capability to create a sentient, obedient servant who'll not only grow our food, but help us solve problems.
In light of all this, I'd like to raise my glass of brandy and say that I, for one, welcome our new robot slaves!
posted by Laotic at 11:36 PM on December 8, 2012
So Currently it can handle 2.5 million neurons, and the human brain has 86 billion? That's only a 40,000. 16 more doublings and it will be able to handle it. If mores law keeps up, that's 20 years.
On the other hand, you can go out and buy 40,000 PCs today - and theoretically have that much computing power right now, and of course people have tried it.
Anyway, the thing to keep in mind is that human brain cells are much more complicated then the simple level-trigger "on off" switches that people typically use in simulating neural networks. There could be things going on inside the cells, not just in terms of their connections.
Guessing the next number in a sequence like "n = n-1 + n-2" or "n = n-1 + 1" but testing lots and lots of potential combinations very quickly.
Maybe it's the first "Neural Network" model to be able to do so.
posted by delmoi at 12:17 AM on December 9, 2012 [1 favorite]
On the other hand, you can go out and buy 40,000 PCs today - and theoretically have that much computing power right now, and of course people have tried it.
Anyway, the thing to keep in mind is that human brain cells are much more complicated then the simple level-trigger "on off" switches that people typically use in simulating neural networks. There could be things going on inside the cells, not just in terms of their connections.
The simulated brain - "First computer model to produce complex behaviour performs almost as well as humans at simple number tasks."That can't be right. Can a human brain solve "(234234068984 + 724320025843)√2 = x" faster than a computer? Or is that somehow not a "simple number task"?
Guessing the next number in a sequence like "n = n-1 + n-2" or "n = n-1 + 1" but testing lots and lots of potential combinations very quickly.
Maybe it's the first "Neural Network" model to be able to do so.
posted by delmoi at 12:17 AM on December 9, 2012 [1 favorite]
Delmoi, you should go read that AMA linked further up. It answers a lot of your questions.
posted by simen at 6:34 AM on December 9, 2012
posted by simen at 6:34 AM on December 9, 2012
Meaning on the Brain: How Your Mind Organizes Reality
posted by kliuless at 10:49 AM on December 28, 2012
posted by kliuless at 10:49 AM on December 28, 2012
« Older "The other fork of the road — the one less... | Happy holidays, everyone! Newer »
This thread has been archived and is closed to new comments
posted by jeffburdges at 6:16 AM on December 8, 2012