Supercomputers can store more information than the human brain and can calculate a single equation faster, but even the biggest, fastest supercomputers in the world cannot match the overall processing power of the brain. And they are nowhere near as compact or energy efficient.
Nevertheless, IBM is trying to simulate the human brain with its own cutting-edge supercomputer, called Blue Gene. For the simulation, it used 147,456 processors working in parallel with one another. IBM researchers say each processor is roughly equivalent to the one found in a personal computer, with one gigabyte of working memory.
So configured, Blue Gene simulated 4.5 percent of the brain's neurons and the connections among them called synapses—that's about one billion neurons and 10 trillion synapses. In total, the brain has roughly 20 billion neurons and 200 trillion synapses.
IBM describes the work in an intriguing paper (pdf) that compares various animal simulations done by its cognitive computing research group in Almaden, Calif. The group has managed to completely simulate the brain of a mouse (512 processors), rat (2,048) and cat (24,576). To rival the cortex inside your head, IBM predicts it will need to hook up 880,000 processors, which it hopes to achieve by 2019.
Read more about Computers vs. Brains in the November 2011 issue of Scientific American.
See what we're tweeting about
25 Comments
Add CommentFrom what I understand, they are not talking about the entire brain but the thalamocortical system. I am assuming that because these computers don't have bodies they are excluding the cerebellum and other regulatory brain structures. From what I remember the brain contains approximately 100 billion neurons and 1 quadrillion synapses. Also, these models are operating in discrete digital steps at approx 20 hertz whereas the cortex is analog and operates at roughly 40 hertz (there is a range). Also, these models are not real time instead they takes approx 600 seconds to model 1 second of real time. So we are a little further off than indicated. That being said, my current PC stores 1.5 million x more data than my first, 20 years ago, so the gap could be closed very quickly.
Reply | Report Abuse | Link to thisTo take this into the realm of science fiction; once they construct a functionally equivalent model of the human brain and turn it on, is that computer then a person? If not, why? What protection will there be for self-aware machines? Perhaps something we should think about sooner rather than later.
Even if the computer had 20 billion neurons and 200 trillion synapses it would still not be able to imitate the human brain in an intelligent way, having the capacity and speed will only give it so much potential...
Reply | Report Abuse | Link to thisSkynet!
Reply | Report Abuse | Link to thisWell, if a computer had what you said it would only be 1/5 as powerful as a human brain. But if your point is that an artificial brain could never have the same "potential" as a human brain I would be very interested to know what evidence you base that on? As someone that has looked at brains, specifically brain computation, I see some tremendous complexity but I don't see any evidence that there is something other than biology at work. If one can create a model of a neuron and synapse that behaves identically to a real neuron and synapse, then they can create a brain that behaves identically to a biological brain. Not saying that will be easy, but it certainly isn't impossible.
Reply | Report Abuse | Link to thisGreat- a computer with the intelligence of a cat.
Reply | Report Abuse | Link to thisFeed me. Feed me. Scratch my chin. Feed me... OK, I'm done for the day, let me sleep now.
Making a conscious machine isn't even remotely the main focus of this work. In point of fact, as long as they are sticking to cortical functional units it's pretty unlikely they'd get that anyway. The latest research by neurologists has brought home the importance of parts of the brain not normally thought of as important to conscious thought (Like the brain stem).
Reply | Report Abuse | Link to thisHaving said that, what the simulation might well be very good for is functional models of mental illness and systemically detailed models of various drug interventions.
I suppose it really depends on the source of consciousness. If it boils down to raw computation power, then it's only a matter of time before we get there.
Reply | Report Abuse | Link to thisIf it's something else... we'll have to figure that out before we can ever simulate it.
It's true that we don't truly know, either way.
One needs to define what one means by "functionally equivalent". If you define it simply as the same number of neurons and synapses, that's one thing. If you say it actually functions the same as a human brain, then do you mean it has the same spectrum of emotions, thoughts, desires, etc? I would be quite skeptical that just because a computer has the same number of neurons and synapses as the human brain, that it has the same spectrum of emotions, thoughts, desires, etc. Probably these are emergent phenomena and will be quite different in many respects for a human brain versus a computer with the same number of neurons and synapses.
Reply | Report Abuse | Link to thisOf course, if it turned out that a computer was actually "functionally equivalent" in this sense to a human brain, then it would be hard to argue that it wasn't actually human, and therefore subject to all the same legal and ethical protections as a human.
It is a strange thing that people have a hard time accepting that emotions are a product of the hardware (actually wetware) of the brain. There seems to be this idea that they are some sort of spiritual phenomenon. But that is not the case. They can be located in specific parts of the brain and linked with neuro-modulators (chemical compounds). There is nothing about it them cannot be synthesized.
Reply | Report Abuse | Link to thisWhen I say functionally equivalent that means; given the same stimuli, it would have the same response i.e. it would function the same.
"Of course, if it turned out that a computer was actually "functionally equivalent" in this sense to a human brain, then it would be hard to argue that it wasn't actually human, and therefore subject to all the same legal and ethical protections as a human." well that is a very interesting point and I am glad you mentioned it because the other members of the family of great apes (Chimpanzees, Bonobos, Gorillas, and Orang-utans) are functionally equivalent to us in many respects yet they have virtually no protection. I chimp has greater intellectual abilities than a person with severe Down's Syndrome yet the person with Down' is protected and the chimp is not. So this is why I raised the question. Society has had a hard time including new members in its family of equals. Often times it has only come about through violence. Again, perhaps we should start discussing this sooner than later.
Reply | Report Abuse | Link to this"What protection will there be for self-aware machines?"
Reply | Report Abuse | Link to thisProbably none is my guess. Humanity is slow to learn from it's history - a very conspicuous pattern of such slowness is etched in our history.
Sadly, I predict that all the errors of thought made, that led to slavery - and which took so long to be abolished - are likely to be repeated.
We haven't developed - as far as I know - an ethical system that will include and protect sentient new life (I avoid the term 'machines').
A stark warning and lessons for us all: youtu.be/GYp2dx652ho
Does this IBM "cat" go all day on a single cup of crunchy dry chow? Or, does it still need to be plugged in?
Reply | Report Abuse | Link to thisI ask, because my cat doesn't have a receptacle for a power cord. I'm not sure that they (cats) ever had one.
I could be wrong.
.
You forgot clean up my toilet
Reply | Report Abuse | Link to thisOne relatively easy way to obtain the performance required is to encode the simulation using VHDL or Verilog. With the right message passing mechanism, that should also solve the memory problem... as indeed, no computer is anywhere remotely closely to holding that much memory. Virtually all kinds of memory and functionality are in the shape and interconnections of neurons, it seems.
Reply | Report Abuse | Link to thisAnother alternative I've been thinking about is to re-animate dead neurons. While dead neurons may not be able to make new connections, why can't they continue to process information? For example: http://www.youtube.com/watch?v=dxQmOR_QLfQ Taking this to the extreme (if even possible), a dead person under the influence of soy sauce (lol) wouldn't be able to form new memories but should function and form short-term memories.. A girl I once knew who worked at the Army's morge in S. Korea told me that a corpse once sat up and looked at her, before collapsing back down again.
Supercomputers can compute really fast but they aren't intelligent in the Turing-test sense. They aren't even intelligent in the biological sense. No robot with computer brain can beat a cockroach in mobility and survival in the wild. So far no computer is smarter than a cockroach.
Reply | Report Abuse | Link to thisSo did IBM's Blue Genes talk?
Reply | Report Abuse | Link to thisI hope nobody tells it that any random press of a single button "Kills" it. If it replicates a human brain, survival is a hard-wired imperative, so it might take exception to being casually shut down, panic at the possibility, and like a drowning man who climbs the body of his rescuer to keep his own head above water, do whatever it can to survive. Or even more disturbing, what if their model turns out to be psychopathic in nature and uses it's instant connection to every corner of the globe to push a few people's buttons and turn them off at 30 gigaflops. Somebody better give John Conner a call and warn him the terminators are coming.
Reply | Report Abuse | Link to thisIt sounds as if this brain simulation, whatever section of the brain of a cat or a human serves as the model, is based on a highly simplified schematic of a fixed set of neurons, a fixed, clearly defined set of synapses for each neuron, and without various short-term and long-term "complexity" feedback. For example, where are the effects of the mechanisms described in "Neural Feedback, Brain Influences Itself with Its Own Electric Field" (SA 12/14/2010); "Signaling Neurons Make Neighbors 'Want In'" (SA 12/20/2007 - transient pseudo-synapse effect of electrochemical dispersal beyond proximity synapse); and "Glia Cells Help Neurons Build Synapses" (SA 1/20/2001 - long-term design modifications of synapses on new or existing neurons)?
Reply | Report Abuse | Link to thisEven if they can match the brain in terms of physical connections, they're still a long way off matching it in terms of functionality. Consciousness results not from the sheer amount of computing power but from the way neurons are plastic. They change their connections based on use and need and can excrete numerous different neurotransmitters depending on the stimulus they receive either from e.g the optic nerve or from other neurons, which in turn affects all the neurons around them. It's a vast, dynamic system and far more complicated to replicate than simply joining ever more processors together.
Reply | Report Abuse | Link to thisI read the article in more detail last night. Basically they have set up a realistic simulation of two pieces of cat visual cortex connected to each other via the thalamus and reticular nucleus. The model incorporates many physiologically realistic aspects of neuronal functioning. When they sent in some input they got cycling analogous to the brain waves seen on an EEG.
Reply | Report Abuse | Link to thisWhat intrigues me is how one would format the input and read the associated output. The input to the thalamus from the visual system is already highly processed; would they try to duplicate this type of input seeing as how their model is based on the visual cortex? And how would they interpret the resulting output? Would this possibly be a form of visual recognition, something which as I understand it has been very hard to achieve with computers?
Agree with the previous comments about higher primates. Also heard about a recent effort by PETA to have whales declared subject to U.S. Constitutional protections. According to Jeffrey Toobin the presumption of the Constitution is that it only applies to humans. Maybe we're being too homo sapiens-chauvinistic.
I doubt that a computer a la 'Cat'rina Frankenstein' would be able to play an interactive cat& mouse game!
Reply | Report Abuse | Link to thisThirty thousand cat mummies show us in what esteem these V.I.P deities were held once, as their feline heighnesses, in charge of the granary, and with it, the commodity market.
While I believe there is soul which is some kind of specialized energy.We do know that living things reverse the second law of thermodynamics, while machines follow the second law.That is the difference between the inanimate and the animate..So creating a living computer is a long way off is possible altogether.
Reply | Report Abuse | Link to thishere is a good read related by the person with the highest IQ ever http://sidis.net/ANIMContents.htm
What the article omits to tell is that the newly configured Blue Gene was so amazingly intelligent that even before its data banks had been connected up it had started from "I think therefore I am" and got as far as deducing the existence of rice pudding and income tax before anyone managed to turn it off.
Reply | Report Abuse | Link to thisRobert Schmidt: I don't doubt your expertise, but can you refer to a journal article or something similar on your comparison of a chimp to a Down's?
Reply | Report Abuse | Link to thisThis is a tremendous achievement, but implies far more than it actually achieves. Every day across this nation we unplug machine with exactly as many neurons and synapses as a human brain, simply because they aren't going to wake up from their head trauma. Those machines go on to get cut out, weighed and measured, put back, and then either embalmed and buried or burnt to a cinder. There's not much about the hardware that we actually respect. It's the software that makes a person.
Reply | Report Abuse | Link to thisThere is a yawning gulf between our ability to build the hardware, and our ability to make the software that makes it smart. And I don't mean simulating humans. Computers don't need to think like us to be smart and useful. In fact, I think building such a thing would be a tremendous waste of resources, like trying to build a mechanical horse instead of making a car.