7 ms·

The Brain in the Machine: IBM's Compass

48 points by drnewman 14y ago

Claudus 14y ago
How will the world change when you have hundreds of virtual human brains, overclocked to think hundreds of times faster than a normal human? Create the biggest MMO ever made, in a week?
- themgt 14y ago
  "And it was good"
- tejaswiy 14y ago
  Brain uploads living in simulated universes are I think are our surest bet at immortality. @cstross has a highly entertaining book called Accelerando which deals with this. Check it out if you haven't yet already. If you push is a bit, you can also branch the human brain at a certain point in your life and merge back to master whenever you see fit, and you can experience more than you ever could in a single life. EDIT: I realize space is a big deal, but I only wish some organization working with the funds that NASA has that can explore the human brain and consciousness. This could initially lead to a variety of psychological hacks coming out which could be used by marketers and ultimately a deeper understanding of intelligence. Like OP said, if we could simulate a brain in a computer, there's no reason we couldn't make it better. If we could make it better, and this is pushing it even further, provide it with the right "memories" as a starting point, you could have the equivalent of the greatest minds of humanity living simultaneously. Think about what we could achieve then ?
- D_Alex 14y ago
  Well, here is one theory: http://en.wikipedia.org/wiki/Technological_singularity http://en.wikipedia.org/wiki/Technological_singularity
- anigbrowl 14y ago
  They'll just go mad faster. I'm a proponent of strong AI, but it ain't gonna be a cakewalk.
- nealabq 14y ago
  The overclocked super-brains would camp at HN and always comment first. They'd hog all the karma!
aristidb 14y ago
If the goal is to learn about how thinking works, and maybe find ways to adapt specific methods to software problems, simulating brains makes a lot of sense, of course. But it bewilders me how so many people see the "holy grail" in simulating whole human brains. Two things: 1. Human brains are extraordinarily power efficient - I don't think we can beat them anytime soon, as long as we are power constrained. I believe the difference in efficiency for tasks the human brain is good at (such as visual pattern recognition) is a significant number of orders of magnitude. 2. Computers are at their best when they complement us, when they solve tasks the human brain is NOT good at.
adriand 14y ago
This article, and the author's critique of Ray Kurzweil's latest book, which was also on HN last week, makes me think about a powerful point made by David Deutsch in his book "The Beginning of Infinity". He says (excuse the paraphrasing) that explanation-less theories are useless and bad science. Devoid of an explanation, data are meaningless and can be interpreted in countless nonsensical ways. When it comes to consciousness the problem is not that we haven't built a big enough simulation with enough neurons, but rather that we lack an explanation for consciousness and its associated complex phenomena, like qualia. When we can explain consciousness, he suspects it will be easy to program. Conversely, if you can't program it, you don't understand it.
- thirdtruck 14y ago
  The notion of "qualia" confuses me. Tracking down such a thing sounds like trying to find the essential "Tuxness" of a given Linux installation. If I install one kernel patch, does that make the system "less Linux-y"? What if that patch was a bug fix by Torvalds himself? Would that have the opposite effect? I know the answer to the questions above: they're nonsense questions, so answers don't even enter the equation. We have identified specific modules of the brain that, among other things, let us distinguish faces from mere shapes, friends from strangers (even when it comes to the same individual), and language processing itself. Our own sense of self seems to reduce to a distinct, toggle-able neurological phenomenon. In short, the very idea of qualia reduces to just various artifacts of our brain's multiple systems of abstraction.
mememememememe 14y ago
Google has Big Brains. It's expecting open source either the end of this year or early next year.
- anigbrowl 14y ago
  Link?
  - mememememememe 14y ago
    http://www.youtube.com/watch?v=gb3Wed9M25A http://www.youtube.com/watch?v=gb3Wed9M25A click to 08:50
mercuryrising 14y ago
We're just starting to unlock the secrets of the brain, but for all those hoping that this is the one that does it, I will take the risk of saying "It won't happen this time". We don't know what the information transfer mechanisms are in the brain, nor do we know how information is stored. We do know that action potentials are the observable and recordable signs of potential brain activity, but that's all we can see. If you have something as complicated [1] as the brain, assuming that one of the first things we found about the brain is the way it lets me think of art, reason about the past, and adapt to my surroundings is plain stupid. Since we discovered neurons [2], there have been monumental changes in how we understand the world. But this still isn't enough. We cannot explain the things that happen in brains with our current understanding of physics. Minds are more connected than one may initially assume. I'll tell you a story about why I know this to be true, and what it means for our attempts to recreate a brain. For as long as I can remember, I have been falling asleep with some sort of music or movie playing. Whenever I would drift off to sleep, I would hear this noise in the background. It let me sleep easily, giving me something to focus on. When I went to Norway this past summer, I didn't have anything to listen to, only my mind. I noticed when I was about to fall asleep, something weird would happen. I would hear things that weren't thoughts I had, they were thoughts in other languages. They sounded like what I was hearing in Norway, but they could have been Chinese for all I know, but what I do know is that they were sounds I never recognized, and they were in languages that weren't my own. Whose thoughts were these, and why were they in my brain when I was falling asleep? Weird things happen when we sleep, but before all contact from the real world shuts off, the transition into sleep may be one of the most important for figuring out what is really happening in our brains. How would this information transfer boundaries between my brain and another person's brain, or another region in space time? This would be easy to pass off as misinterpreted noise, but this happened for many times while in Norway, but when I got back home, it tapered off (I lost my peace of mind when I returned home). Our current understanding of physics is not advanced enough to allow us to understand the brain. We have done cool things with technology, but I think we are farther from understanding what is a person is in the true sense of the question. It's interesting to think about the 'hard problems' that persist over life times. Almost all the problems that we are solving today are not 'hard problems'. They are challenging, fulfilling, rewarding, and helpful in the short term. I once went for a long bike ride, and it had been a while so I was pretty excited. I told my dad about my achievement. I said "Dad, I went for a 50 mile bike ride today!". He said "Yeah, but did you get anywhere?". We are solving problems, but the solutions are not advancing the human race. We are riding around in a circle. I'll stop now, and hopefully start a blog so I can put out my long winded rants there. [1] http://www.cerebromente.org.br/n17/history/neurons2_i.htm http://www.cerebromente.org.br/n17/history/neurons2_i.htm
fragsworth 14y ago
So many questions about the future. Will there be movements to give legal rights to our virtual creations based on the notion that they are conscious? Will this cause humans to lose control of things as the computer minds grow more powerful? Will we have laws that force us to imbue emotions in them? So that they can only be created with empathy for human beings? Will everyone obey these laws when creating computer minds? Will there be black markets of unempathic, cold-hearted superminds that blindly obey their self-serving owners? Will we all be out of jobs? Will we even need or want money?
- ams6110 14y ago
  And, do we think that if we succeed in recreating the complexity and capability of the human brain with computers, that the computer brains won't be just as error-prone and individually unpredictable as the real thing?
  - orangecat 14y ago
    Well, consider the advantages of running Windows in a VM. (Which admittedly has much fewer ethical issues).
- anonymfus 14y ago
  Because human moral is based on emotions, answer to first question will depend on applications of such virtual creations. We have movements to give legal rights to dolphins because they are cute, but we don't have such movements for chimpanzees or parrots because their human-like behavior is inside our uncanny valley. So probably we will have movements to give legal rights for toys and chatbots with AI and no such movements for medical or scientific AI's.
  - javert 14y ago
    Morality is properly based on reason. Many people do use a primitive kind of morality that is based on emotion.
- javert 14y ago
scrumper 14y ago
This kind of work has applications in neuroscience to model specific parts of the brain. Blue Brain is just such an effort. I'd also be confident in saying this thing isn't going to become intelligent any time soon. We also don't have any idea how the physical structure of the brain maps to its functional characteristics (pattern recognition, consciousness.) This work may help us explore that. It's a brain simulation, maybe, but it's definitely not a 'mind' simulation at all.
geuis 14y ago
There are many thing that science doesn't yet know conclusively about neural architectures. Arguing that a machine such as described in the article will get us nowhere massively misses the point. We live in the age of simulation in science. Computation has reached the point where, in many disparate systems such as weather models, particle physics, protein folding, and neural architecture, building complex simulations lets investigators freely play with hypothesis and compare the results to real-world data. Hypothesis are proved or disproved, science advances, and we know that much more about how such complex systems indeed work. The author of this article derides this approach, saying that we just don't have enough information to do anything useful with the machine. I call bullshit. One of the entire points of the machine is to allow experimenters perform experiments that are either impossible or morally unethical on biological creatures. In a few years, suppose a research decides to change the algorithm about how simulated neurons exchange a neurotransmitter or an electrical pulse. The simulation then produces patterns very similar or identical to an FMRI or electrocardiogram of a living brain. Biologists can then use that as an unexpected avenue to exist living neurons to see if they react in the same way. Its an instance of simulation allowing experimentation that illuminates the real deal.
- Xcelerate 14y ago
  > ...the machine is to allow experimenters to perform experiments that are either impossible or morally unethical on biological creatures Ignoring the fact that whether this ever happens is up for debate, why would it be moral to torture a simulated animal as long as you're not torturing a "real" one? > In a few years, suppose a research decides... You vastly overestimate what research will produce in a few years.
  - javert 14y ago
    why would it be moral to torture a simulated animal as long as you're not torturing a "real" one? Morality is a human concept that should serve human ends--individually and socially. As such, performing experiments on a dog to gain more knowledge is moral.
    - 14y ago
6ren 14y ago
Simulating a brain using the same number of neurons is like simulating a body using the same number of atoms. Also relevant, Chomsky vs Norvig, on AI http://www.tor.com/blogs/2011/06/norvig-vs-chomsky-and-the-fight-for-the-future-of-ai http://www.tor.com/blogs/2011/06/norvig-vs-chomsky-and-the-f... Which boils down to "make everything as simple as possible, but no simpler". Chomsky wants to understand what's happening. Norvig says it's too complicated. I think we can usually carve off an aspect or layer of phenomena that - in isolation - is simple enough to understand. But I don't see any guarantee that this will necessarily be the case; models are possible that aren't decomposable into simpler parts (because they must interact with everything else, informationally). However (1) understanding is what we seek in Science; and (2) it's always been possible before - for whatever reason, the universe does seem to be decomposable hierarchically... at least, the bits that we've understood, so far...
fudged71 14y ago
I found a comment on the article to be more concise and interesting than the article itself.
javert 14y ago
The author has committed a serious epistemological error by repeatedly conflating neruons with simulated neurons. For example: "In sheer scale, it’s far more ambitious than anything previously attempted, and it actually has almost ten times as many neurons as a human brain." No, it has ten times as many simulated neurons. It does not have any neurons. Unless it's a big wet biological computer. This kind of sloppy thinking is not too surprising to me, given that this is The New Yorker, which I consider to be exemplary of sloppy thinking almost as a rule.
kghose 14y ago
http://www.youtube.com/watch?v=q4KrhDZQ088 http://www.youtube.com/watch?v=q4KrhDZQ088 A popular debate about what knowing the brain's wiring diagram will bring us
mck- 14y ago
Once technology advances to a level at which we succesfully replicate the human brain and its interactions all the way down to the nano/quantum level, would that artificial neural network then gain consciousness?
- D_Alex 14y ago
  Nobody really knows, but I think the informed consensus is that consciousness would probably happen long before we "replicate the human brain and its interactions all the way down to the nano/quantum level".
briggers 14y ago
It's important to note that while this simulates approximately the number of synapses in a human brain, it does so 1542 times slower than realtime.
VexXtreme 14y ago
Some theorists say this is the first step to achieving technological singularity. Once it becomes possible to replicate human mind/consciousness in computer memory (by running a brain simulation), the line between man and machine will become very thin. What happens once you upload your consciousness into a computer? Are you creating a new instance of yourself? What happens once such creations are able to harness the power of underlying infrastructure and use high powered processors to perform calculations and do reasoning? I remember reading somewhere that humans might well become a completely digital species if this ever occurs. Interesting stuff to say the least...
joeyo 14y ago
Before you get too excited, (from the IBM article): It is important to clarify that we have not built a biologically realistic simulation of the complete human brain. Rather, we have simulated a novel modular, scalable, non-von Neumann, ultra-low power, cognitive computing architecture at the DARPA SyNAPSE metric of 10^14 synapses that itself is inspired by the number of synapses in the human brain.

48 points by drnewman 14y ago

How will the world change when you have hundreds of virtual human brains, overclocked to think hundreds of times faster than a normal human? Create the biggest MMO ever made, in a week?

themgt 14y ago
"And it was good"
tejaswiy 14y ago
Brain uploads living in simulated universes are I think are our surest bet at immortality. @cstross has a highly entertaining book called Accelerando which deals with this. Check it out if you haven't yet already. If you push is a bit, you can also branch the human brain at a certain point in your life and merge back to master whenever you see fit, and you can experience more than you ever could in a single life. EDIT: I realize space is a big deal, but I only wish some organization working with the funds that NASA has that can explore the human brain and consciousness. This could initially lead to a variety of psychological hacks coming out which could be used by marketers and ultimately a deeper understanding of intelligence. Like OP said, if we could simulate a brain in a computer, there's no reason we couldn't make it better. If we could make it better, and this is pushing it even further, provide it with the right "memories" as a starting point, you could have the equivalent of the greatest minds of humanity living simultaneously. Think about what we could achieve then ?
D_Alex 14y ago
Well, here is one theory: http://en.wikipedia.org/wiki/Technological_singularity http://en.wikipedia.org/wiki/Technological_singularity
anigbrowl 14y ago
They'll just go mad faster. I'm a proponent of strong AI, but it ain't gonna be a cakewalk.
nealabq 14y ago
The overclocked super-brains would camp at HN and always comment first. They'd hog all the karma!

aristidb 14y ago

If the goal is to learn about how thinking works, and maybe find ways to adapt specific methods to software problems, simulating brains makes a lot of sense, of course. But it bewilders me how so many people see the "holy grail" in simulating whole human brains. Two things: 1. Human brains are extraordinarily power efficient - I don't think we can beat them anytime soon, as long as we are power constrained. I believe the difference in efficiency for tasks the human brain is good at (such as visual pattern recognition) is a significant number of orders of magnitude. 2. Computers are at their best when they complement us, when they solve tasks the human brain is NOT good at.

adriand 14y ago

This article, and the author's critique of Ray Kurzweil's latest book, which was also on HN last week, makes me think about a powerful point made by David Deutsch in his book "The Beginning of Infinity". He says (excuse the paraphrasing) that explanation-less theories are useless and bad science. Devoid of an explanation, data are meaningless and can be interpreted in countless nonsensical ways. When it comes to consciousness the problem is not that we haven't built a big enough simulation with enough neurons, but rather that we lack an explanation for consciousness and its associated complex phenomena, like qualia. When we can explain consciousness, he suspects it will be easy to program. Conversely, if you can't program it, you don't understand it.

thirdtruck 14y ago
The notion of "qualia" confuses me. Tracking down such a thing sounds like trying to find the essential "Tuxness" of a given Linux installation. If I install one kernel patch, does that make the system "less Linux-y"? What if that patch was a bug fix by Torvalds himself? Would that have the opposite effect? I know the answer to the questions above: they're nonsense questions, so answers don't even enter the equation. We have identified specific modules of the brain that, among other things, let us distinguish faces from mere shapes, friends from strangers (even when it comes to the same individual), and language processing itself. Our own sense of self seems to reduce to a distinct, toggle-able neurological phenomenon. In short, the very idea of qualia reduces to just various artifacts of our brain's multiple systems of abstraction.

mememememememe 14y ago

Google has Big Brains. It's expecting open source either the end of this year or early next year.

anigbrowl 14y ago
Link?
- mememememememe 14y ago
  http://www.youtube.com/watch?v=gb3Wed9M25A http://www.youtube.com/watch?v=gb3Wed9M25A click to 08:50

mercuryrising 14y ago

We're just starting to unlock the secrets of the brain, but for all those hoping that this is the one that does it, I will take the risk of saying "It won't happen this time". We don't know what the information transfer mechanisms are in the brain, nor do we know how information is stored. We do know that action potentials are the observable and recordable signs of potential brain activity, but that's all we can see. If you have something as complicated [1] as the brain, assuming that one of the first things we found about the brain is the way it lets me think of art, reason about the past, and adapt to my surroundings is plain stupid. Since we discovered neurons [2], there have been monumental changes in how we understand the world. But this still isn't enough. We cannot explain the things that happen in brains with our current understanding of physics. Minds are more connected than one may initially assume. I'll tell you a story about why I know this to be true, and what it means for our attempts to recreate a brain. For as long as I can remember, I have been falling asleep with some sort of music or movie playing. Whenever I would drift off to sleep, I would hear this noise in the background. It let me sleep easily, giving me something to focus on. When I went to Norway this past summer, I didn't have anything to listen to, only my mind. I noticed when I was about to fall asleep, something weird would happen. I would hear things that weren't thoughts I had, they were thoughts in other languages. They sounded like what I was hearing in Norway, but they could have been Chinese for all I know, but what I do know is that they were sounds I never recognized, and they were in languages that weren't my own. Whose thoughts were these, and why were they in my brain when I was falling asleep? Weird things happen when we sleep, but before all contact from the real world shuts off, the transition into sleep may be one of the most important for figuring out what is really happening in our brains. How would this information transfer boundaries between my brain and another person's brain, or another region in space time? This would be easy to pass off as misinterpreted noise, but this happened for many times while in Norway, but when I got back home, it tapered off (I lost my peace of mind when I returned home). Our current understanding of physics is not advanced enough to allow us to understand the brain. We have done cool things with technology, but I think we are farther from understanding what is a person is in the true sense of the question. It's interesting to think about the 'hard problems' that persist over life times. Almost all the problems that we are solving today are not 'hard problems'. They are challenging, fulfilling, rewarding, and helpful in the short term. I once went for a long bike ride, and it had been a while so I was pretty excited. I told my dad about my achievement. I said "Dad, I went for a 50 mile bike ride today!". He said "Yeah, but did you get anywhere?". We are solving problems, but the solutions are not advancing the human race. We are riding around in a circle. I'll stop now, and hopefully start a blog so I can put out my long winded rants there. [1] http://www.cerebromente.org.br/n17/history/neurons2_i.htm http://www.cerebromente.org.br/n17/history/neurons2_i.htm

fragsworth 14y ago

So many questions about the future. Will there be movements to give legal rights to our virtual creations based on the notion that they are conscious? Will this cause humans to lose control of things as the computer minds grow more powerful? Will we have laws that force us to imbue emotions in them? So that they can only be created with empathy for human beings? Will everyone obey these laws when creating computer minds? Will there be black markets of unempathic, cold-hearted superminds that blindly obey their self-serving owners? Will we all be out of jobs? Will we even need or want money?

ams6110 14y ago
And, do we think that if we succeed in recreating the complexity and capability of the human brain with computers, that the computer brains won't be just as error-prone and individually unpredictable as the real thing?
- orangecat 14y ago
  Well, consider the advantages of running Windows in a VM. (Which admittedly has much fewer ethical issues).
anonymfus 14y ago
Because human moral is based on emotions, answer to first question will depend on applications of such virtual creations. We have movements to give legal rights to dolphins because they are cute, but we don't have such movements for chimpanzees or parrots because their human-like behavior is inside our uncanny valley. So probably we will have movements to give legal rights for toys and chatbots with AI and no such movements for medical or scientific AI's.
- javert 14y ago
  Morality is properly based on reason. Many people do use a primitive kind of morality that is based on emotion.
javert 14y ago

scrumper 14y ago

This kind of work has applications in neuroscience to model specific parts of the brain. Blue Brain is just such an effort. I'd also be confident in saying this thing isn't going to become intelligent any time soon. We also don't have any idea how the physical structure of the brain maps to its functional characteristics (pattern recognition, consciousness.) This work may help us explore that. It's a brain simulation, maybe, but it's definitely not a 'mind' simulation at all.

geuis 14y ago

There are many thing that science doesn't yet know conclusively about neural architectures. Arguing that a machine such as described in the article will get us nowhere massively misses the point. We live in the age of simulation in science. Computation has reached the point where, in many disparate systems such as weather models, particle physics, protein folding, and neural architecture, building complex simulations lets investigators freely play with hypothesis and compare the results to real-world data. Hypothesis are proved or disproved, science advances, and we know that much more about how such complex systems indeed work. The author of this article derides this approach, saying that we just don't have enough information to do anything useful with the machine. I call bullshit. One of the entire points of the machine is to allow experimenters perform experiments that are either impossible or morally unethical on biological creatures. In a few years, suppose a research decides to change the algorithm about how simulated neurons exchange a neurotransmitter or an electrical pulse. The simulation then produces patterns very similar or identical to an FMRI or electrocardiogram of a living brain. Biologists can then use that as an unexpected avenue to exist living neurons to see if they react in the same way. Its an instance of simulation allowing experimentation that illuminates the real deal.

Xcelerate 14y ago
> ...the machine is to allow experimenters to perform experiments that are either impossible or morally unethical on biological creatures Ignoring the fact that whether this ever happens is up for debate, why would it be moral to torture a simulated animal as long as you're not torturing a "real" one? > In a few years, suppose a research decides... You vastly overestimate what research will produce in a few years.
- javert 14y ago
  why would it be moral to torture a simulated animal as long as you're not torturing a "real" one? Morality is a human concept that should serve human ends--individually and socially. As such, performing experiments on a dog to gain more knowledge is moral.
  - 14y ago

6ren 14y ago

Simulating a brain using the same number of neurons is like simulating a body using the same number of atoms. Also relevant, Chomsky vs Norvig, on AI http://www.tor.com/blogs/2011/06/norvig-vs-chomsky-and-the-fight-for-the-future-of-ai http://www.tor.com/blogs/2011/06/norvig-vs-chomsky-and-the-f... Which boils down to "make everything as simple as possible, but no simpler". Chomsky wants to understand what's happening. Norvig says it's too complicated. I think we can usually carve off an aspect or layer of phenomena that - in isolation - is simple enough to understand. But I don't see any guarantee that this will necessarily be the case; models are possible that aren't decomposable into simpler parts (because they must interact with everything else, informationally). However (1) understanding is what we seek in Science; and (2) it's always been possible before - for whatever reason, the universe does seem to be decomposable hierarchically... at least, the bits that we've understood, so far...

fudged71 14y ago

I found a comment on the article to be more concise and interesting than the article itself.

javert 14y ago

The author has committed a serious epistemological error by repeatedly conflating neruons with simulated neurons. For example: "In sheer scale, it’s far more ambitious than anything previously attempted, and it actually has almost ten times as many neurons as a human brain." No, it has ten times as many simulated neurons. It does not have any neurons. Unless it's a big wet biological computer. This kind of sloppy thinking is not too surprising to me, given that this is The New Yorker, which I consider to be exemplary of sloppy thinking almost as a rule.

kghose 14y ago

http://www.youtube.com/watch?v=q4KrhDZQ088 http://www.youtube.com/watch?v=q4KrhDZQ088 A popular debate about what knowing the brain's wiring diagram will bring us

mck- 14y ago

Once technology advances to a level at which we succesfully replicate the human brain and its interactions all the way down to the nano/quantum level, would that artificial neural network then gain consciousness?

D_Alex 14y ago
Nobody really knows, but I think the informed consensus is that consciousness would probably happen long before we "replicate the human brain and its interactions all the way down to the nano/quantum level".

briggers 14y ago

It's important to note that while this simulates approximately the number of synapses in a human brain, it does so 1542 times slower than realtime.

VexXtreme 14y ago

Some theorists say this is the first step to achieving technological singularity. Once it becomes possible to replicate human mind/consciousness in computer memory (by running a brain simulation), the line between man and machine will become very thin. What happens once you upload your consciousness into a computer? Are you creating a new instance of yourself? What happens once such creations are able to harness the power of underlying infrastructure and use high powered processors to perform calculations and do reasoning? I remember reading somewhere that humans might well become a completely digital species if this ever occurs. Interesting stuff to say the least...

joeyo 14y ago

Before you get too excited, (from the IBM article): It is important to clarify that we have not built a biologically realistic simulation of the complete human brain. Rather, we have simulated a novel modular, scalable, non-von Neumann, ultra-low power, cognitive computing architecture at the DARPA SyNAPSE metric of 10^14 synapses that itself is inspired by the number of synapses in the human brain.