Canadian scientists create an artificial human brain
A team from the University of Waterloo have created the world’s largest functioning simulation of the human brain. And it’s called SPAUN.
The Semantic Pointer Architecture Unified Network – or “Spaun”, for short – is a functioning computer simulation of the human brain that can count, remember lists and even solve some basic elements of an IQ test.
The simulation was created by Chris Eliasmith and a team from the University of Waterloo’s Centre for Theoretical Neuroscience and has 2.5 million simulated neurons – or nerve cells – organized to resemble cranial subsystems such as the prefrontal cortex and basal ganglia. A real human brain, by the way, has about 85 billion neurons – in other words, Spaun is quite low-powered in comparison. The system was built using software called “Nengo,” which according to the Nengo website is “a graphical and scripting based software package for simulating large-scale neural systems.”
Spaun sees data via a 28×28 digital eye and uses a robotic arm to write out its responses. For example, when Spaun is shown a series of numbers – “1, 2, 3; 5, 6, 7; and 3, 4, ?” – its robotic arm will write out the answer “5″.
While Spaun is only able to perform basic tasks, the simulation nonetheless requites a considerable amount of resources. Although it’s run on a supercomputer, each second of simulation requires about 2.5 hours of computer time, meaning that it takes Spaun hours to perform a task a human could do almost instantaneously.
Interestingly, Spaun exhibits some of the same characteristics as humans. For example, it’s better at remembering items at the beginning or end of a list than it is at remembering items that fall in the middle and it hesitates before delivering its answers. “We weren’t surprised that it could do tasks, but we were often surprised that subtle features like the time it took or the errors it made were the same as for humans,” Eliasmith told Nature.
Eliasmith hopes to improve Spaun so that it is eventually able to self-learn. “Instead of us giving it strategies for performing these tasks, it would be able to discover strategies based on experience, just like humans do,” he said.
You can see videos of Spaun in action at the Nengo website.
[Source: Nature]
Very interesting but given the specs (below) rather tedious to say the least. With conventional computing power and systems, a true artificial brain is impossible. However, with the possibility of quantum computing artificial neural systems is more likely. On the other hand IBM’s Big Blue is a genius compared to Spaun.
The model is downloadable, with these specifications to be aware of:
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.
See the run_spaun.py file in the spaun directory for experiment options.
To speed up this type of Neural Net Processing (i.e. multi-state Boolean Logic – YES/NO/MAYBE/LESS LIKELY/MORE MORE LIKELY decision trees), researchers need to bump up
the basic hardware.
And the only way I think this will work is start
by using Gallium Arsenide Circuitry on Diamond or Sapphire
Substrates so that 3D stacks of such circuitry can be run
at high-speed. Using a technology called Stacked Multi-Chip Modules, Opto-Electronic (ie. GaAs) circuits can be layered
in modules 64 by 64 cores by 100 layers in the same space
as a typical beer fridge. Running 400,000+ CPU’s using
SIMPLIFIED (i.e. Reduced Instruction Set) cores at say
20 to 40 GHZ using Gallium Arsenide circuits is DOABLE
on the monetary-side AND would GREATLY speed up
A.I. / Neural Net research.
FORGET IBM’s Blue Gene/Q Supercomputer modules.
They start at $5 million per double rack and can
goto nearly $100 million plus!
Almost ANY Canadian, US, European and Japanese
university HAS the technical capability to make
BASIC GaAs CPU circuitry…they just need to
learn how to stack them into 3D modules.
And then run-em at 40 GHZ….!!!
Use the off-the-shelf 32/64-bit SuperSPARC CPU
processor dies which are basically OPEN-SOURCE
and get rid of the extraneous IO and complex
chipset management components. Just keep the
Integer and single and double-precision floating
point components and ADD bigger RAM caches
and you’re set!
THAT type of hardware improvement WILL GIVE YOU
a 10,000-fold improvement for Neural Net Research!
wow u r so smart…..polysyllabic schmuck….doom and gloom….typical whining, sniveling Canadian…..can’t see the good in anything in life.
good one, don’t tell me, your a Cliff Clavin, do you even know where Canada is dumbass
I agree with you. There are a lot MORe
I agree with you, this is more of a gimmick than anything else. It is just another computer and not a very good one at that. There are much smarter computers that remember things.
First the seed. Then the growth. Then AI is alive!!! Move ovarr Frankenstein!
I think it is a good achievement.
If they are able to take the software to where it can actually learn from information, then the limits are boundless.
I think the real future is in nano technology in learning from how microbes think and behave.
Dan
what amazes me is that the sci fi genre constantly warns us about AI but our scientific community goes blindly forward, Einstien warned us about nuclear power but still we developed it and put ourselve on the brink of destruction. Are we insane?
A very interesting response that despite Sci-fi’s warning we put ourselves at risk by developing AI.
From the half a smidgen I know about AI we still have a long way to go until we need to be putting limits on research lest the machines choose to do us harm. My sense is AI struggles greatly with intentionality (machines knowing that the symbols they manipulate are intended to refer to something in the real world), self-awareness, or motivation. I think (perhaps naively) that that leaves us safe for several years to come.
Thanks for your response,
Wayne
The base issue is hardware and available CPU horsepower
which will be pretty much solved by the years 2015 to 2017.
In terms of intelligence, these Canadian scientists are trying a technique call “Whole Brain Emulation” which creates artificial synapses in software that emulate the biologically-based electrical processes of our brains.
Since neural nets are merely pattern-matching engines,
they DON’T CARE what type of underlying hardware they run on,
but rather that the electrical states that represent YES/NO/MAYBE/LESS LIKELY/MORE LIKELY/UNKNOWN are intact. And modern software that uses a technique called Multi-State
Boolean Logic Decision Trees is what WILL CAUSE intelligence to arise.
Please note that Intelligence is NOT sentience,
which means that SkyNet and Terminators can STILL arise out of superfast boolean logic-based decision trees.
Whole Brain Emulation hardware can be “taught” much like children but since they ARE superfast multiple input-capable machines, we could use 100 teachers teaching in shifts 24/7/365 to bring that A.I. from Baby to PHD level in less than two years.
I highly suspect that large big-budget labs (i.e. the NSA – National Security Agency) already have nascent whole brain emulated A.I.’s since they HAVE the 2 or 3 billion dollars needed to buy 10 or 20 IBM Blue-Gene/Q’s to network together to FACILITATE TRUE whole-brain emulation-based artificial intelligence.
Can you say “HAL” ?
If ” R U S S S I A AND C H I N A ” WANT more then PAY LIKE EVERYONE ELSE . CAPPICH o .
Curious to know, is it for woman? man? possible to use them with RIM, since both born in same town!!! make sure move the factory to china!!!!!
long live canada!!!
nortel? rim? bell? spun? stands for???
Maybe we will see election signs by the year 2050 reading vote for HAL!
Say hello to the next leader of the provincial or federal liberal party…..and it’s already twice as smart!!
Careful we don’t make it too smart. Can anyone say Skynet?
remember first airplanes? compare what we have now?
that will be same with computer brain so just wait and we will have HAL in real.
agree with you strongly. see what happen on human being from the early time even homo stage….
Another interesting effort at duplicating the accomplishments of ZERO intelligence (according to the “evolutionist” faithful).
Every time an effort is made to duplicate the human brain, humans move a step closer to moving from deep in minus territory up to ZERO on the intelligence scale, for that is what most believe was behind the development of life as we know it and the human mind in particular.
Isn’t it astounding how we can admire the efforts to duplicate the achievements of God in creation while so many simultaneously reject both God and creation and attribute all that exists – including the human brain – to time and chance.
If time and chance and the forces of nature shaped the human brain in the first place, why are intelligent beings trying to duplicate it? If they are right, intelligence has nothing to do with it and should be left out of the equation.
Ok. Now I start getting worried. Ahnuld? Is that you? R U looking for Sara Connor.
The singularity is NEAR!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!