4 ms·

The State of Whole Brain Emulation in 2015

20 points by goblin89 12y ago

mjfl 12y ago
"human brains are turing complete" Are we sure we can really support that statement?
- cervice 12y ago
  Would a not turing complete human brain be able to conceive turing completeness?
  - JadeNB 12y ago
    That sounds like saying "Would a not-Turing complete markup language be able to record the definition of Turing completeness?" (If you argue that there's a difference between conceiving something and recording it, then I can soup up the markup language with a mild genetic algorithm equipped with a stop condition that'll let it generate the definition on its own.)
  - actsasbuffoon 12y ago
    My brain can imagine what telepathy would be like, but that doesn't make it telepathic.
- JadeNB 12y ago
  > Are we sure we can really support that statement? I assume you mean "isn't it possible that the human brain would run out of memory?", but I think that it depends on what you mean by Turing complete. If I hand you the specification of a universal Turing machine, then you can certainly execute it, by hand, on a limitless supply of paper—making the paper your external memory. I think that this is a reasonable way of claiming that your brain is Turing complete. (Or perhaps you worry that one must include the rest of the body to achieve Turing completeness?)
- Chinjut 12y ago
  Well, augmented with a theoretical infinite memory device, and granted immortality (protection against senility, etc.), sure. Do you really doubt humans can carry out "Look up within this finite table what to do, based on the current state and current symbol at the cursor. Update the state and manipulate the cursor accordingly. Repeat."?
  - kybernetikos 12y ago
    Humans are pretty bad at following instructions reliably and correctly, so maybe more like a turing machine with some probabilistic error rate.
    - TheLoneWolfling 12y ago
      We're assuming arbitrary amounts of time - in which case one can check calculations an arbitrary number of times. (Quick idea: recheck all calculations done so far at all computation steps that are powers of 2. So recheck everything after the first step, the second step, the fourth step, the eighth step, the sixteenth step, etc.)
- murbard2 12y ago
  People have already pointed out why. In addition, let me point out why it is also no more powerful than a Turing machine. If you accept physicalism, then thoughts arise as the result of electrical, mechanical, and chemical processes. As far as we know, those processes are governed by laws of physics which are all computable to arbitrary precision. Assuming that our thought processes are not infinitely sensitive to random perturbations, then the human brain can be simulated by a Turing machine. Thus, everything we know about nature indicates that the human brain is no more powerful than a Turing machine.
  - TheLoneWolfling 12y ago
    > As far as we know, those processes are governed by laws of physics which are all computable to arbitrary precision. Look at, for example, electron shot noise for a counterexample. Given the state of a low-output lamp and a detector, one cannot compute the state of the detector after a period of time to arbitrary precision. We can say that the detector will have x% chance of triggering within the time period - but that is all. We cannot say definitely "the detector will trigger" or "the detector will not trigger", no matter how exactly we know the initial state of the system. And we already know that the human brain can be affected by similar processes (If nothing else, our eyes start registering at full night vision when <10 photons hit within 100ms, which is within the realm of shot noise.)
- stegosaurus 12y ago
  Other posters seem to be attacking this from the angle of providing examples of non-turing-complete systems which can describe or 'imagine' turing-complete systems. I would attack this by claiming that we have designed turing complete systems using no external input. That is to say - we can _produce_ a turing complete system, which implies we are a turing complete system given an IO port (limbs, for example). Fundamentally I could sit and perform calculations with pen and paper. I'm ignoring the unlimited criterion here, I assume colloquial usage. The only real way I could fathom otherwise is by deciding to isolate the brain and consider it as a single entity. But in that case a CPU isn't really turing-complete because it can't do anything on its' own.
ericjang 12y ago
Seems like nothing really exceptionally different in 2015. All the projects or ideas the author mentioned have been around for quite some time, at least since the early 2000's. Also, Turing completeness is not a strict requirement for intelligence. That line of thinking has fallen out of favor in recent years, in the same way that comparing the brain to a steam engine has. I think we are increasingly realizing that the brain is not a magnetic pin and the world is not a piece of tape. Intelligence could be solved via artificial neural networks or statistical techniques, so it's not clear to me that a continuous conscious experience requires swapping out one neuron at a time.
jimrandomh 12y ago
This is a decent overview, but I feel like I should bring up one thing it didn't mention: safety. We're a significant distance away from being able to emulate a whole brain, and... that's a good thing. Imagine your brain were scanned, and you found yourself running as a computer program. Consider the powers that would give you. You could make copies of yourself, obviously. You could tune some parameters, to try to increase your intelligence. You could A/B test your copies with identical memories in identical situations, to see what effect that tuning has, and you could incorporate enhancements that wouldn't work biologically. So there's a pretty high chance that you'd end up with a superhuman level intelligence, or at least John von Neumann level intelligence. Plus a large number of copies. At the same time, the emulation might be imperfect, the modifications and enhancements might have unanticipated side effects, there's a whole lot of opportunities for sanity damage. That's on top of the base rate of humans being evil or turning evil when given power, which is itself uncomfortably high. And finally, there's one more wrinkle. Suppose brain-scanning and whole-brain emulation technology are in the wild, and a bunch of different people scan their brains and start running on computers. In this world, if you hack into someone's hardware and halt their brain program, you can repurpose it to run copies of yourself. While computer security is probably better in that future world than it is today, I seriously doubt that it will be better by nearly enough. If you're thinking of getting into brain-emulation research, please consider also putting some work into safety research. And in the unlikely event your work should pass the level of treating mundane ailments, and moves into the territory of breakthroughs, please remember that you are playing with fire.

20 points by goblin89 12y ago

mjfl 12y ago

"human brains are turing complete" Are we sure we can really support that statement?

cervice 12y ago
Would a not turing complete human brain be able to conceive turing completeness?
- JadeNB 12y ago
  That sounds like saying "Would a not-Turing complete markup language be able to record the definition of Turing completeness?" (If you argue that there's a difference between conceiving something and recording it, then I can soup up the markup language with a mild genetic algorithm equipped with a stop condition that'll let it generate the definition on its own.)
- actsasbuffoon 12y ago
  My brain can imagine what telepathy would be like, but that doesn't make it telepathic.
JadeNB 12y ago
> Are we sure we can really support that statement? I assume you mean "isn't it possible that the human brain would run out of memory?", but I think that it depends on what you mean by Turing complete. If I hand you the specification of a universal Turing machine, then you can certainly execute it, by hand, on a limitless supply of paper—making the paper your external memory. I think that this is a reasonable way of claiming that your brain is Turing complete. (Or perhaps you worry that one must include the rest of the body to achieve Turing completeness?)
Chinjut 12y ago
Well, augmented with a theoretical infinite memory device, and granted immortality (protection against senility, etc.), sure. Do you really doubt humans can carry out "Look up within this finite table what to do, based on the current state and current symbol at the cursor. Update the state and manipulate the cursor accordingly. Repeat."?
- kybernetikos 12y ago
  Humans are pretty bad at following instructions reliably and correctly, so maybe more like a turing machine with some probabilistic error rate.
  - TheLoneWolfling 12y ago
    We're assuming arbitrary amounts of time - in which case one can check calculations an arbitrary number of times. (Quick idea: recheck all calculations done so far at all computation steps that are powers of 2. So recheck everything after the first step, the second step, the fourth step, the eighth step, the sixteenth step, etc.)
murbard2 12y ago
People have already pointed out why. In addition, let me point out why it is also no more powerful than a Turing machine. If you accept physicalism, then thoughts arise as the result of electrical, mechanical, and chemical processes. As far as we know, those processes are governed by laws of physics which are all computable to arbitrary precision. Assuming that our thought processes are not infinitely sensitive to random perturbations, then the human brain can be simulated by a Turing machine. Thus, everything we know about nature indicates that the human brain is no more powerful than a Turing machine.
- TheLoneWolfling 12y ago
  > As far as we know, those processes are governed by laws of physics which are all computable to arbitrary precision. Look at, for example, electron shot noise for a counterexample. Given the state of a low-output lamp and a detector, one cannot compute the state of the detector after a period of time to arbitrary precision. We can say that the detector will have x% chance of triggering within the time period - but that is all. We cannot say definitely "the detector will trigger" or "the detector will not trigger", no matter how exactly we know the initial state of the system. And we already know that the human brain can be affected by similar processes (If nothing else, our eyes start registering at full night vision when <10 photons hit within 100ms, which is within the realm of shot noise.)
stegosaurus 12y ago
Other posters seem to be attacking this from the angle of providing examples of non-turing-complete systems which can describe or 'imagine' turing-complete systems. I would attack this by claiming that we have designed turing complete systems using no external input. That is to say - we can _produce_ a turing complete system, which implies we are a turing complete system given an IO port (limbs, for example). Fundamentally I could sit and perform calculations with pen and paper. I'm ignoring the unlimited criterion here, I assume colloquial usage. The only real way I could fathom otherwise is by deciding to isolate the brain and consider it as a single entity. But in that case a CPU isn't really turing-complete because it can't do anything on its' own.

ericjang 12y ago

Seems like nothing really exceptionally different in 2015. All the projects or ideas the author mentioned have been around for quite some time, at least since the early 2000's. Also, Turing completeness is not a strict requirement for intelligence. That line of thinking has fallen out of favor in recent years, in the same way that comparing the brain to a steam engine has. I think we are increasingly realizing that the brain is not a magnetic pin and the world is not a piece of tape. Intelligence could be solved via artificial neural networks or statistical techniques, so it's not clear to me that a continuous conscious experience requires swapping out one neuron at a time.

jimrandomh 12y ago

This is a decent overview, but I feel like I should bring up one thing it didn't mention: safety. We're a significant distance away from being able to emulate a whole brain, and... that's a good thing. Imagine your brain were scanned, and you found yourself running as a computer program. Consider the powers that would give you. You could make copies of yourself, obviously. You could tune some parameters, to try to increase your intelligence. You could A/B test your copies with identical memories in identical situations, to see what effect that tuning has, and you could incorporate enhancements that wouldn't work biologically. So there's a pretty high chance that you'd end up with a superhuman level intelligence, or at least John von Neumann level intelligence. Plus a large number of copies. At the same time, the emulation might be imperfect, the modifications and enhancements might have unanticipated side effects, there's a whole lot of opportunities for sanity damage. That's on top of the base rate of humans being evil or turning evil when given power, which is itself uncomfortably high. And finally, there's one more wrinkle. Suppose brain-scanning and whole-brain emulation technology are in the wild, and a bunch of different people scan their brains and start running on computers. In this world, if you hack into someone's hardware and halt their brain program, you can repurpose it to run copies of yourself. While computer security is probably better in that future world than it is today, I seriously doubt that it will be better by nearly enough. If you're thinking of getting into brain-emulation research, please consider also putting some work into safety research. And in the unlikely event your work should pass the level of treating mundane ailments, and moves into the territory of breakthroughs, please remember that you are playing with fire.