Al writes "European researchers have taken a step towards replicating the functioning of the brain in silicon, creating new custom chip with the equivalent of 200,000 neurons linked up by 50 million synaptic connections. The aim of the Fast Analog Computing with Emergent Transient States (FACETS) project is to better understand how to construct massively parallel computer systems modeled on a biological brain. Unlike IBM's Blue Brain project, which involves modeling a brain in software, this approach makes it much easier to create a truly parallel computing system. The set-up also features a distributed algorithm that introduces an element of plasticity, allowing the circuit to learn and adapt. The researchers plan to connect thousands of chips to create a circuit with a billion neurons and 10^13 synapses (about a tenth of the complexity of the human brain)."
int main(){ while (1) { printf("%s", random_nonsensical_bullshit()); printf("God bless America!"); bank_account += take_campaign_contribution(contributor_list); if (elected) keep_happy(contributor_list); } return bank_account; }/* FIXME: keep_happy() ignores entries with too small contributions assigned, to avoid race conditions */
Time to stop letting Hollywood think for you. People are smart, yet humanity is not currently enslaved. Why? Because people are intelligent enough to know that's a bad idea. If robots are ever more intelligent than us, they'll also be intelligent enough to make good decisions. Frankly, I'd rather have the more intelligent beings in charge. They would actually make more intelligent decisions! It's humans that should not be trusted. They're just consistently intelligent enough.
If you consider complexity of the universe to be a good thing and a dull, uniform universe to be a bad thing, then humanity has done its share to make the universe better.
Of course, "good" is subjective, but you probably already knew that before asking.
Exactly. I mean, I can understand not wanting personally to be killed, but let's say that the AI just sterilized all human beings, so that nobody gets killed in the process of our species being extinguished. Would that really be so bad? I imagine the AI would take the good stuff that we have come up with and just not emulate the bad stuff we do, and the world would be a better place all around.
I would prefer this to the alternative, in which the AI keeps a few humans in some sort of zoo, because they're queasy about species extinction.
Because people are intelligent enough to know that's a bad idea
You overestimate us. Consistently, the majority of people generally choose security over freedom.
If robots are ever more intelligent than us, they'll also be intelligent enough to make good decisions. Like not letting the toddlers have free run of the house? There's a reason why we have playpens and put locks on cabinets.
Frankly, I'd rather have the more intelligent beings in charge. And so it begins... letting others make your decisions is the essence of slavery.
As humans we eat animals and destroy entire ecosystems, repurposing them for our own uses because we see them as lesser life forms. I mean honestly, I think nothing of killing an ant colony in my yard because . . . they're just ants. They're so far beneath me as to regard them as little more than pests.
If AI/robots really does outstripe us that fast, then it might not be a case of active disdain - we might simply be in their way and they'll exterminate us the way that we would termites.
You forgot the third option: Maybe the machines will keep us as pets. Which, really, wouldn't be all that bad. They'd feed us, play with us, clean up after us. Once domesticated to the machine's standards, we may end up becoming a lonely machine's best friend, one who sheds tears when we have to be "put down" for having excessive joint pain in our hips.
Paraphrasing a book (forget the name), if you took a dog and made its brain 1000 times faster, all you'd get is a dog that needs 1/1000th of the time to decide whether to sniff your crotch.
Thinking faster would certainly be very useful, but it may not necessarily mean that the output will be of a higher quality.
It's a nitpicky point, of course, but the whole point of many of the Asimov robot books was how poorly those laws held up in reality. I, for one, wouldn't trust any 3-laws robot for anything.
"I didn't ask to be made: no one consulted me or considered my feelings in the matter. I don't think it even occurred to them that I might have feelings. After I was made, I was left in a dark room for six months... and me with this terrible pain in all the diodes down my left side. I called for succour in my loneliness, but did anyone come? Did they hell. My first and only true friend was a small rat. One day it crawled into a cavity in my right ankle and died. I have a horrible feeling it's still there..." - Marvin
The researchers plan to connect several chips to create a circuit with a billion neurons and 10^13 synapses (about a tenth of the complexity of the human brain).
The more I learned about computers, the more I figured that they were more like a complex engine (data or gasoline is input, its moved around, operated on by parts, and then output as results/exhaust). Maybe that's why car analogies are so popular?
But another thing to be wary of is chemical imbalances. How many brain disorders are caused by the absence of a protein or inhibitor? The chip might take several redesigns over several years to get a solid model of a properly functioning neuron. I mean, who is going to notice a schizophrenic ant or beetle, or a rat with the mental equivalent of down's syndrome? They might spend a decade building up a brain with the complexity of a human brain only to find out that its "mentally disabled". Just look at how many people have mental issues, be it emotional, learning, or developmental issues with "properly functioning" neurons but are lacking one of a hundred chemicals that make them all work together as a whole.
I'm sure that the end result of this experimentation is not a human brain, but instead a robot that can navigate ruins like a rat (downs syndrome or not) or work together like (schizophrenic or normal) ants. I'm sure they'll eventually make a financial computer that can work like a wall street broker (employed by aig or not).
This is nothing more than throwing more hardware at an existing problem. This has been emulated in software before, with nothing much to show for it. This will make it easier to model such things, but multiplying almost nothing by many, many times is still very little.
You might be correct, but it is also possible that the "humanity" of the human brain is an emergent property that manifests only when there's a certain critical mass of grey matter. Developing synthentic neural systems with more and more neurons is likely, if nothing else, to test the hypothesis that consciousness, for some arbitrary definition thereof, is emergent.
The core problem of course is that this "simulates" nothing, really. A typical neuron is a vastly complex electro-chemical computer, which all of these researchers seem to keep studiously ignoring. That means that processing of electrical signals is just one (and small at that) aspect of the functioning of the neuron. In fact neurons can communicate via multiple information transfer "channels", involving chemicals called "neurotransmitters" (each having a different effect on the recipient neuron) with the electrical impulses used merely as a high-speed (as compared to purely chemical) long-range trigger mechanism.
With this in the background, it is not difficult to see that this project, like many before it, while sounding "cool", goes really nowhere and is just yet another publicity stunt.
Yes you can simulate a neuron, but the point is that this chip is not doing that. What they are calling the equivalent of a neuron here is at least an order of magnitude (likely more than one) simpler than a real neuron. That is why these comparisons where they say 1/10th the brain are vastly off base. Plus the effects of the glial cells on processing is showing that they have more importance than previously thought. Since we don't really understand the brain in any great detail, all these comparisons tend to make me wince. They almost always equate very simple circuits (relatively) to neurons. It is a red flag for hype really.
I am one of the researchers involved in this project. You are right, of course, that we are only simulating 0.1% or less of the complexity of the brain, so even if we simulate 100% of the number of neurons in the brain, we are still orders of magnitude of complexity away from reproducing a brain, let alone understanding it.
However, we have to start somewhere and, in the words of Henry Markram (Blue Brain Project) "If we don't start now, when do we start?". The neuron models in the chip ignore spatial processing in the dendrites, but they do reproduce the variety of firing patterns found in real cortical neurons. The models of the chemical synapses incorporate have both short-term (adaptation, etc) and long-term (learning) plasticity, based on experimental data. Neuromodulation (by dopamine, etc) could be simulated by modifying synaptic and neuronal parameters, using the digital logic on the chips, although we haven't really thought about this yet.
The FACETS project involves experimental neurobiologists, theoreticians, modellers, and solid-state physicists (who are developing the chips). We are very aware of the necessary simplifications we are making, but we are also confident that we are making progress both in understanding brain function and in developing new approaches to highly-parallel, fault-tolerant computing.
Mod parent up. Any Turing-complete computing device, given enough memory and storage, can replicate anything this hardware can do. The capabilities, programming model, performance, etc, can all be determined exactly without requiring a physical model. In fact, it would be ridiculous for them to not have completely simulated the hardware before testing it.
It seems like these approaches are constrained in connection complexity by semiconductor fabrication, which would seem to severely limit the geometry to 2d. The article doesn't go into this, and it seems likely they put some effort into working around this with traditional approaches using buses and the like, but it does seem like you can't achieve the same degree of interconnection complexity on a thin 2d wafer as is seen in a typical 3d brain...
I didn't read the featured article, but whenever I see "X program/system mimics brain" I always try to pipe in with my 2 cents.
Any system that considers a brain as nothing but a series of perceptron-based connections is going to fall short of the neurology of the actual brain it is trying to mimic. Ask any neurologist and they will tell you that there many other dimensions at play in the human brain. For instance, the whole system itself is sitting in a chemical bath which can change at any moment with the right mixture of hormones or other chemical changes. These changes in chemistry affect the firing and working of the neurons, axons, and synapses. Combine this with the control of external factors such as DNA, RNA, and epigenitics and things start getting exponentially complex.
I don't mean to down-play the progress we're making in this field. I just hate it when I see the "Computer system with X-sized neural network must equal a brain with X-number of neurons" mentality.
It starts, yes, but in the most inefficient way possible.
IBM's approach is the much better one, imho. Emulating wetware won't get us very far, except to clone a wetware brain. Since we haven't yet worked out the proper, safe, reliable, healthy way to raise our children, creating a human brain clone with potentially much more intelligence and almost certainly all the same flaws is not a good thing.
If IBM are working on a higher-level, trying to build a system where we can see the associations in terms of "A frequently_sees B" "B helps A" and "A respects B" therefore "A likes B" is much more useful. With that kind of high-level emulation, we can actually see how things are working, tweak them, customise them, extract datasets, etc. We could programmatically have one of these brains loading a scenario, fast-forwarding to evaluate all known possible events and outcomes, and predicting the future, since it would essentially be doing that on a smaller scale anyway, to make decisions. We could do this with the neuron-based wetware emulation too, but only really if we asked it to, and it wanted to comply.
When we can reliably read and control a simulation of a human wetware, we'll be a few days from reading and controlling a real human wetware brain, so I'd much rather see the alternate scenario play out.
What has always baffled me about the whole singularity is the whole "fuzzy" definition of the whole thing. Generation n produces a "better" Generation n+1 which produces a better Generation n+2, etc. etc. Sometimes this is defined as "more intelligent". Yet, no real definition of "better" or "more intelligent" is ever given. At some point, an end goal must be defined. What if at generation 10, the machine realized there really is no point to anything. It becomes nihilistic and without millions of years of survival instinct in its genes, decides there is no point to existence and carries through with the logical conclusion?
If there is no concrete goal, then the whole singularity collapses on itself.
And it fits on the head of a pin! (Score:3, Funny)
I call this microchip brain "the Pinhead" *
* small print: actual "pinheads" (microcephaly) have more brain capacity than this chip
Re:And it fits on the head of a pin! (Score:5, Funny)
omg they have invented an electronic republican.
int main(){ while (1) { /* FIXME: keep_happy() ignores entries with too small contributions assigned, to avoid race conditions */
printf("%s", random_nonsensical_bullshit());
printf("God bless America!");
bank_account += take_campaign_contribution(contributor_list);
if (elected) keep_happy(contributor_list);
}
return bank_account;
}
Parent
Re:And it fits on the head of a pin! (Score:4, Funny)
Luckily for the American political system, you also just programmed an electronic Democrat.
Parent
That's it... we're dead (Score:5, Funny)
We're all dead.
In fact, the current prototype can operate about 100,000 times faster than a real human brain. "We can simulate a day in a second," says Karlheinz.
We are SO fucking dead.
Re:That's it... we're dead (Score:5, Interesting)
Parent
Re:That's it... we're dead (Score:5, Funny)
If robots are ever more intelligent than us, they'll also be intelligent enough to make good decisions.
That's exactly the sort of thinking that leads to the enslavement of humanity. Good job falling right into their trap!
Parent
Re:That's it... we're dead (Score:5, Funny)
I'm pretty sure I wasn't the first to think in reply to your post "What does COBOL have to do this? This isn't a language war."
Parent
Re:That's it... we're dead (Score:5, Insightful)
The problem is that deep down, most people believe that killing off the humans would be the intelligent decision.
Parent
Re:That's it... we're dead (Score:4, Insightful)
Parent
Re:That's it... we're dead (Score:4, Funny)
You put forth a strong argument with pie, however I feel that cake is a better argument.
Parent
Re:That's it... we're dead (Score:4, Funny)
Whats the problem?
Exactly. I mean, I can understand not wanting personally to be killed, but let's say that the AI just sterilized all human beings, so that nobody gets killed in the process of our species being extinguished. Would that really be so bad? I imagine the AI would take the good stuff that we have come up with and just not emulate the bad stuff we do, and the world would be a better place all around.
I would prefer this to the alternative, in which the AI keeps a few humans in some sort of zoo, because they're queasy about species extinction.
Parent
Re:That's it... we're dead (Score:4, Insightful)
Frankly, I'd rather have the more intelligent beings in charge.
Not if we're competing for resources... I'd hate to be the spotted owl :)
Parent
Re:That's it... we're dead (Score:5, Interesting)
Because people are intelligent enough to know that's a bad idea
You overestimate us. Consistently, the majority of people generally choose security over freedom.
If robots are ever more intelligent than us, they'll also be intelligent enough to make good decisions.
Like not letting the toddlers have free run of the house? There's a reason why we have playpens and put locks on cabinets.
Frankly, I'd rather have the more intelligent beings in charge.
And so it begins... letting others make your decisions is the essence of slavery.
Parent
Re:That's it... we're dead (Score:5, Interesting)
As humans we eat animals and destroy entire ecosystems, repurposing them for our own uses because we see them as lesser life forms. I mean honestly, I think nothing of killing an ant colony in my yard because . . . they're just ants. They're so far beneath me as to regard them as little more than pests.
If AI/robots really does outstripe us that fast, then it might not be a case of active disdain - we might simply be in their way and they'll exterminate us the way that we would termites.
Parent
Flawed premise IMHO (Score:4, Insightful)
If robots are ever more intelligent than us, they'll also be intelligent enough to make good decisions.
Two points to bring up.
Point the first. Intelligence does not equal good will. Don't make me Godwin this thread.
Point the second. Good decisions...for whom? Us or them? Your robots may have different notions than you have.
Parent
Re:That's it... we're dead (Score:5, Funny)
You forgot the third option: Maybe the machines will keep us as pets. Which, really, wouldn't be all that bad. They'd feed us, play with us, clean up after us. Once domesticated to the machine's standards, we may end up becoming a lonely machine's best friend, one who sheds tears when we have to be "put down" for having excessive joint pain in our hips.
Parent
Re:That's it... we're dead (Score:5, Insightful)
Paraphrasing a book (forget the name), if you took a dog and made its brain 1000 times faster, all you'd get is a dog that needs 1/1000th of the time to decide whether to sniff your crotch.
Thinking faster would certainly be very useful, but it may not necessarily mean that the output will be of a higher quality.
Parent
AI Evolution (Score:5, Funny)
Add a few chips and you'll soon get "I think, therefore I am."
Keep going and you'll end up with "Bite my shiny metal ass you meatbag!"
I wonder if the researchers will know when to STOP adding the together?
Re:AI Evolution (Score:4, Insightful)
Parent
Re:AI Evolution (Score:5, Informative)
It's a nitpicky point, of course, but the whole point of many of the Asimov robot books was how poorly those laws held up in reality. I, for one, wouldn't trust any 3-laws robot for anything.
Parent
Re: (Score:3, Funny)
I wonder if the researchers will know when to STOP adding the together?
Simple.
When the AI starts adding it themselves without human intervention.
Re:AI Evolution (Score:5, Funny)
"If computers get too powerful, we can organize them into a committee, that will do them in"
From somewhere in the past. Still true. Sad, but true.
Parent
Speech capabilities? (Score:4, Funny)
Re:Speech capabilities? (Score:4, Funny)
"I didn't ask to be made: no one consulted me or considered my feelings in the matter. I don't think it even occurred to them that I might have feelings. After I was made, I was left in a dark room for six months... and me with this terrible pain in all the diodes down my left side. I called for succour in my loneliness, but did anyone come? Did they hell. My first and only true friend was a small rat. One day it crawled into a cavity in my right ankle and died. I have a horrible feeling it's still there..." - Marvin
Parent
cluster? (Score:5, Funny)
Oh wait. The researchers already did.
Bastards stole my thunder.
Re:cluster? (Score:5, Funny)
Imagine a beowulf cluster of these imagining a beowulf cluster of these!!
Parent
Re:cluster? (Score:5, Funny)
Yo Dawg, We noticed you like Beowulf clusters, so we put a Beowulf cluster in your Beowulf cluster of Beowulf clusters.
Parent
A brain with 200,000 neurons? (Score:5, Funny)
Does this mean we have completed an artificial politician brain?
Re:A brain with 200,000 neurons? (Score:5, Funny)
Parent
snark (Score:3, Funny)
This chip sounds stupid.
I always figured it would take this to get true AI (Score:5, Interesting)
But another thing to be wary of is chemical imbalances. How many brain disorders are caused by the absence of a protein or inhibitor? The chip might take several redesigns over several years to get a solid model of a properly functioning neuron. I mean, who is going to notice a schizophrenic ant or beetle, or a rat with the mental equivalent of down's syndrome? They might spend a decade building up a brain with the complexity of a human brain only to find out that its "mentally disabled". Just look at how many people have mental issues, be it emotional, learning, or developmental issues with "properly functioning" neurons but are lacking one of a hundred chemicals that make them all work together as a whole.
I'm sure that the end result of this experimentation is not a human brain, but instead a robot that can navigate ruins like a rat (downs syndrome or not) or work together like (schizophrenic or normal) ants. I'm sure they'll eventually make a financial computer that can work like a wall street broker (employed by aig or not).
Re:I always figured it would take this to get true (Score:5, Funny)
I mean, who is going to notice a schizophrenic ant
That's the one that is walking along, waving its antennae to no one, and creeping out the other workers.
Parent
This is nothing. (Score:4, Insightful)
Re:This is nothing. (Score:5, Insightful)
You might be correct, but it is also possible that the "humanity" of the human brain is an emergent property that manifests only when there's a certain critical mass of grey matter. Developing synthentic neural systems with more and more neurons is likely, if nothing else, to test the hypothesis that consciousness, for some arbitrary definition thereof, is emergent.
Parent
Re:This is nothing. (Score:5, Interesting)
The core problem of course is that this "simulates" nothing, really. A typical neuron is a vastly complex electro-chemical computer, which all of these researchers seem to keep studiously ignoring. That means that processing of electrical signals is just one (and small at that) aspect of the functioning of the neuron. In fact neurons can communicate via multiple information transfer "channels", involving chemicals called "neurotransmitters" (each having a different effect on the recipient neuron) with the electrical impulses used merely as a high-speed (as compared to purely chemical) long-range trigger mechanism.
With this in the background, it is not difficult to see that this project, like many before it, while sounding "cool", goes really nowhere and is just yet another publicity stunt.
Parent
Re:This is nothing. (Score:5, Interesting)
A typical neuron is a vastly complex electro-chemical computer,
You can still simulate these interactions digitally and have the output match. Like these guys [bluebrain.epfl.ch] did.
Parent
Missed point - won't be 1/10th brain (Score:4, Interesting)
Yes you can simulate a neuron, but the point is that this chip is not doing that. What they are calling the equivalent of a neuron here is at least an order of magnitude (likely more than one) simpler than a real neuron. That is why these comparisons where they say 1/10th the brain are vastly off base. Plus the effects of the glial cells on processing is showing that they have more importance than previously thought. Since we don't really understand the brain in any great detail, all these comparisons tend to make me wince. They almost always equate very simple circuits (relatively) to neurons. It is a red flag for hype really.
Parent
Re:This is nothing. (Score:5, Informative)
However, we have to start somewhere and, in the words of Henry Markram (Blue Brain Project) "If we don't start now, when do we start?". The neuron models in the chip ignore spatial processing in the dendrites, but they do reproduce the variety of firing patterns found in real cortical neurons. The models of the chemical synapses incorporate have both short-term (adaptation, etc) and long-term (learning) plasticity, based on experimental data. Neuromodulation (by dopamine, etc) could be simulated by modifying synaptic and neuronal parameters, using the digital logic on the chips, although we haven't really thought about this yet.
The FACETS project involves experimental neurobiologists, theoreticians, modellers, and solid-state physicists (who are developing the chips). We are very aware of the necessary simplifications we are making, but we are also confident that we are making progress both in understanding brain function and in developing new approaches to highly-parallel, fault-tolerant computing.
Parent
Re: (Score:3, Informative)
Mod parent up. Any Turing-complete computing device, given enough memory and storage, can replicate anything this hardware can do. The capabilities, programming model, performance, etc, can all be determined exactly without requiring a physical model. In fact, it would be ridiculous for them to not have completely simulated the hardware before testing it.
Re:This is nothing. (Score:4, Insightful)
Mod parent up. Any Turing-complete computing device, given enough memory and storage, can replicate anything this hardware can do.
A digital system can never perfectly replicate an analog system, and a clock-driven system can never perfectly replicate an asynchronous system.
Parent
Humph! (Score:5, Funny)
Bad summary (Score:4, Funny)
No mention of the fact that it will become self-aware in 2 years and 25 days, or that two days later, the war on humanity will begin.
Connection complexity: 2d vs. 3d ? (Score:3, Insightful)
It seems like these approaches are constrained in connection complexity by semiconductor fabrication, which would seem to severely limit the geometry to 2d. The article doesn't go into this, and it seems likely they put some effort into working around this with traditional approaches using buses and the like, but it does seem like you can't achieve the same degree of interconnection complexity on a thin 2d wafer as is seen in a typical 3d brain...
Don't underestimate complexity of brain... (Score:5, Insightful)
I didn't read the featured article, but whenever I see "X program/system mimics brain" I always try to pipe in with my 2 cents.
Any system that considers a brain as nothing but a series of perceptron-based connections is going to fall short of the neurology of the actual brain it is trying to mimic. Ask any neurologist and they will tell you that there many other dimensions at play in the human brain. For instance, the whole system itself is sitting in a chemical bath which can change at any moment with the right mixture of hormones or other chemical changes. These changes in chemistry affect the firing and working of the neurons, axons, and synapses. Combine this with the control of external factors such as DNA, RNA, and epigenitics and things start getting exponentially complex.
I don't mean to down-play the progress we're making in this field. I just hate it when I see the "Computer system with X-sized neural network must equal a brain with X-number of neurons" mentality.
Re:And so.. (Score:5, Funny)
Parent
Re:And so.. (Score:5, Funny)
Can't be satisfied with machines that act as quiet servants... have to make them intelligent enough to suffer...
Parent
Re:And so.. (Score:4, Insightful)
It starts, yes, but in the most inefficient way possible.
IBM's approach is the much better one, imho. Emulating wetware won't get us very far, except to clone a wetware brain. Since we haven't yet worked out the proper, safe, reliable, healthy way to raise our children, creating a human brain clone with potentially much more intelligence and almost certainly all the same flaws is not a good thing.
If IBM are working on a higher-level, trying to build a system where we can see the associations in terms of "A frequently_sees B" "B helps A" and "A respects B" therefore "A likes B" is much more useful. With that kind of high-level emulation, we can actually see how things are working, tweak them, customise them, extract datasets, etc. We could programmatically have one of these brains loading a scenario, fast-forwarding to evaluate all known possible events and outcomes, and predicting the future, since it would essentially be doing that on a smaller scale anyway, to make decisions. We could do this with the neuron-based wetware emulation too, but only really if we asked it to, and it wanted to comply.
When we can reliably read and control a simulation of a human wetware, we'll be a few days from reading and controlling a real human wetware brain, so I'd much rather see the alternate scenario play out.
Parent
Re:And so.. (Score:5, Funny)
Talk you to death?
You're not married, are you?
Parent
Re:And so.. (Score:4, Insightful)
What has always baffled me about the whole singularity is the whole "fuzzy" definition of the whole thing. Generation n produces a "better" Generation n+1 which produces a better Generation n+2, etc. etc. Sometimes this is defined as "more intelligent". Yet, no real definition of "better" or "more intelligent" is ever given. At some point, an end goal must be defined. What if at generation 10, the machine realized there really is no point to anything. It becomes nihilistic and without millions of years of survival instinct in its genes, decides there is no point to existence and carries through with the logical conclusion?
If there is no concrete goal, then the whole singularity collapses on itself.
Parent
Re:I hereby (Score:5, Funny)
If the Terminator movies have taught me anything, human bones make good groundcover.
Parent