4 ms·

I'm not sure I follow how it's possible that machines can be conscious but a sufficiently powerful Von Neumann machine cannot. Perhaps it's because I'm missing

I'm not sure I follow how it's possible that machines can be conscious but a sufficiently powerful Von Neumann machine cannot. Perhaps it's because I'm missing details (like what actually is integrated information and how do we think it relates to consciousness.) Perhaps someone with more insight can explain how this isn't a violation of the Church-Turing thesis?

Trasmatta 6y ago
It's all based off of integrated information theory: https://en.wikipedia.org/wiki/Integrated_information_theory https://en.wikipedia.org/wiki/Integrated_information_theory I don't fully understand it myself, but based on the axioms in that theory, you can calculate the potential consciousness of a system.
tlb 6y ago
IIT is not a phenomenological measure. That is, two systems could behave identically to all external appearances and one could be conscious and the other not. The metric depends on how things are connected inside. There are a few versions of it, but generally systems with more direct wires between components are more conscious. I guess if you don’t think consciousness is measurable through behavior, you have to cook up some metric that depends on internal organization.
- pdonis 6y ago
  > IIT is not a phenomenological measure. Sure it is; brain imaging, which is how IIT tests for consciousness according to the article, is looking at phenomenology.
  - goatlover 6y ago
    It's looking at a correlation, not the conscious experience itself. Brain images are of neural activity, which is different form a thought, desire, or sensation. You don't watch someone's dreams from brain imaging.
    - pdonis 6y ago
      > You don't watch someone's dreams from brain imaging. First, while that's true now, there is no guarantee that it will always be true as technology advances. Second, neural activity is the physical basis for conscious experience, which, to a physicalist like me, means it is conscious experience, when the neural activity has the right properties (according to IIT, those properties are what "integrated information" is trying to capture). So I don't accept the distinction you are making between "the conscious experience itself" and neural activity.
      - goatlover 6y ago
        > First, while that's true now, there is no guarantee that it will always be true as technology advances. True, but what does that mean? We get to watch someone else's dream like a movie because neuroscientists will have figured out how to translate the neural activity into a format that can be outputted as video and audio. That won't work for every sensation. You can't feel that you're having the same dream as being that person's body. > Second, neural activity is the physical basis for conscious experience, which, to a physicalist like me, means it is conscious experience, This is similar to an identity theory of mind, with a focus on information. The problem is understanding how integrated information produces smells and pains. Saying the experience is identical is making a claim that seems mysterious. Why would some forms of information be conscious? Did God set that up? It sounds similar to Chalmers' position on information, except that Chalmers says there is a natural link between rich information and being conscious that is an additional law in addition to the physical ones.
        mannykannot 6y ago
        > It sounds similar to Chalmers' position on information, except that Chalmers says there is a natural link between rich information and being conscious that is an additional law in addition to the physical ones. I will apply Occam's razor to that additional law until there is evidence for it. More generally, there is a straightforward physical possibility for explaining the subjective and personal nature of qualia, and other aspects of consciousness: our brains are not connected in such a way that language processing is capable of initiating all of the state changes that other sensory inputs can, and so we cannot communicate those state changes directly. This not only accords with our actual experiences, but it is probably just as well, or else we might find it (even) hard(er) to distinguish reality from imagination (and by "reality", I mean concrete, "that bear just noticed me", reality.) One may, of course, take one's own Occam's razor to this idea, if you think physical constraints on information flow in a brain are less pausible than Chalmer's extra-physical link.
  - tlb 6y ago
    It’s not, because a Turing machine simulating conscious behavior could also be programmed to emit brain waves or NMR spectra similar to a real brain, without being conscious.
    - pdonis 6y ago
      > a Turing machine simulating conscious behavior If it can simulate conscious behavior well enough to fool conscious observers (like us), how do you know it wouldn't be conscious?
mindyourai 6y ago
This whole theory seems like metaphysics because of this: it implies that among two systems with absolutely identical external behavior in all respects, one can have conscious and the other not have it. Fundamentally, we can simulate neuromorphic computer (or even a quantum computer, or even the entire human brain) on Von Neumann machine (with performance degradation). According to our current understanding of physics, the behavior of these systems should be identical in all respects except the simulation will be slower. However integrated information theory says that a neuromorphic computer may be highly conscious and the brain is most definitely highly conscious, but their simulation isn't. What's still unclear to me is whether this is indeed just metaphysics (i.e., the simulation and the real thing are absolutely identical and but we should still treat them differently from ethical standpoint), or is it a hypothesis about our physics (i.e., the brain somehow fundamentally cannot be simulated on a Von Neumann machine) as well as computational theory (i.e., neuromorphic or quantum computer cannot be simulated either).
- ggggtez 6y ago
  It doesn't help that quantum computers can be simulated with standard transistors (albeit much slower). Are we really arguing that speed of computation is now a pre-requisite for consciousness?
- quantumsequoia 6y ago
  "it implies that among two systems with absolutely identical external behavior in all respects, one can have conscious and the other not have it." That statement can be true without being metaphysical. You can't treat a human as a black box, and then use inputs and outputs to definitively draw conclusions about what's going on inside the black box. You could potentially have two black boxes that give identical outputs for the same inputs, yet have completely different mechanisms inside for arriving at those outputs. For example, say you have someone who's very good at arithmetic able to multiply two numbers. You put him inside one black box and a pocket calculator inside another. You give each box two numbers and they both output the product. Both black boxes will give you identical outputs for any given input. You know the box with the person inside is conscious, but this is not enough information to conclude the other black box is conscious. It's quite possible that there is a way to simulate human behavior through some other non-conscious mechanism. And just because we can't currently prove whether something is conscious doesn't mean it's something metaphysical we'll never be able to prove. For example, if two people tell you they're experiencing some pain, but one is lying about it, a few hundred years ago you'd be unable to prove it. That didn't mean pain is something metaphysical. Today, we know it might be possible to put both people in an MRI and prove whether they're actually feeling pain
quantumsequoia 6y ago
I'm not an expert on the subject, but from my understanding a Turing machine is insufficient to simulate physics. [Bell's theorem](https://en.wikipedia.org/wiki/Bell%27s_theorem https://en.wikipedia.org/wiki/Bell%27s_theorem) proves that quantum physics is truly random, as opposed to being a deterministic system that we just can't measure precisely enough. A Turing machine is unable to compute a truly random number so therefore there could be quantum phenomenon that a Von Neumann machine cannot simulate. It's not proven that consciousness can be simulated by a Turing machine. There's a possibility consciousness requires some hypercomputation to simulate

Trasmatta 6y ago

It's all based off of integrated information theory: https://en.wikipedia.org/wiki/Integrated_information_theory https://en.wikipedia.org/wiki/Integrated_information_theory I don't fully understand it myself, but based on the axioms in that theory, you can calculate the potential consciousness of a system.

tlb 6y ago

IIT is not a phenomenological measure. That is, two systems could behave identically to all external appearances and one could be conscious and the other not. The metric depends on how things are connected inside. There are a few versions of it, but generally systems with more direct wires between components are more conscious. I guess if you don’t think consciousness is measurable through behavior, you have to cook up some metric that depends on internal organization.

pdonis 6y ago
> IIT is not a phenomenological measure. Sure it is; brain imaging, which is how IIT tests for consciousness according to the article, is looking at phenomenology.
- goatlover 6y ago
  It's looking at a correlation, not the conscious experience itself. Brain images are of neural activity, which is different form a thought, desire, or sensation. You don't watch someone's dreams from brain imaging.
  - pdonis 6y ago
    > You don't watch someone's dreams from brain imaging. First, while that's true now, there is no guarantee that it will always be true as technology advances. Second, neural activity is the physical basis for conscious experience, which, to a physicalist like me, means it is conscious experience, when the neural activity has the right properties (according to IIT, those properties are what "integrated information" is trying to capture). So I don't accept the distinction you are making between "the conscious experience itself" and neural activity.
    - goatlover 6y ago
      > First, while that's true now, there is no guarantee that it will always be true as technology advances. True, but what does that mean? We get to watch someone else's dream like a movie because neuroscientists will have figured out how to translate the neural activity into a format that can be outputted as video and audio. That won't work for every sensation. You can't feel that you're having the same dream as being that person's body. > Second, neural activity is the physical basis for conscious experience, which, to a physicalist like me, means it is conscious experience, This is similar to an identity theory of mind, with a focus on information. The problem is understanding how integrated information produces smells and pains. Saying the experience is identical is making a claim that seems mysterious. Why would some forms of information be conscious? Did God set that up? It sounds similar to Chalmers' position on information, except that Chalmers says there is a natural link between rich information and being conscious that is an additional law in addition to the physical ones.
      - mannykannot 6y ago
        > It sounds similar to Chalmers' position on information, except that Chalmers says there is a natural link between rich information and being conscious that is an additional law in addition to the physical ones. I will apply Occam's razor to that additional law until there is evidence for it. More generally, there is a straightforward physical possibility for explaining the subjective and personal nature of qualia, and other aspects of consciousness: our brains are not connected in such a way that language processing is capable of initiating all of the state changes that other sensory inputs can, and so we cannot communicate those state changes directly. This not only accords with our actual experiences, but it is probably just as well, or else we might find it (even) hard(er) to distinguish reality from imagination (and by "reality", I mean concrete, "that bear just noticed me", reality.) One may, of course, take one's own Occam's razor to this idea, if you think physical constraints on information flow in a brain are less pausible than Chalmer's extra-physical link.
- tlb 6y ago
  It’s not, because a Turing machine simulating conscious behavior could also be programmed to emit brain waves or NMR spectra similar to a real brain, without being conscious.
  - pdonis 6y ago
    > a Turing machine simulating conscious behavior If it can simulate conscious behavior well enough to fool conscious observers (like us), how do you know it wouldn't be conscious?

mindyourai 6y ago

This whole theory seems like metaphysics because of this: it implies that among two systems with absolutely identical external behavior in all respects, one can have conscious and the other not have it. Fundamentally, we can simulate neuromorphic computer (or even a quantum computer, or even the entire human brain) on Von Neumann machine (with performance degradation). According to our current understanding of physics, the behavior of these systems should be identical in all respects except the simulation will be slower. However integrated information theory says that a neuromorphic computer may be highly conscious and the brain is most definitely highly conscious, but their simulation isn't. What's still unclear to me is whether this is indeed just metaphysics (i.e., the simulation and the real thing are absolutely identical and but we should still treat them differently from ethical standpoint), or is it a hypothesis about our physics (i.e., the brain somehow fundamentally cannot be simulated on a Von Neumann machine) as well as computational theory (i.e., neuromorphic or quantum computer cannot be simulated either).

ggggtez 6y ago
It doesn't help that quantum computers can be simulated with standard transistors (albeit much slower). Are we really arguing that speed of computation is now a pre-requisite for consciousness?
quantumsequoia 6y ago
"it implies that among two systems with absolutely identical external behavior in all respects, one can have conscious and the other not have it." That statement can be true without being metaphysical. You can't treat a human as a black box, and then use inputs and outputs to definitively draw conclusions about what's going on inside the black box. You could potentially have two black boxes that give identical outputs for the same inputs, yet have completely different mechanisms inside for arriving at those outputs. For example, say you have someone who's very good at arithmetic able to multiply two numbers. You put him inside one black box and a pocket calculator inside another. You give each box two numbers and they both output the product. Both black boxes will give you identical outputs for any given input. You know the box with the person inside is conscious, but this is not enough information to conclude the other black box is conscious. It's quite possible that there is a way to simulate human behavior through some other non-conscious mechanism. And just because we can't currently prove whether something is conscious doesn't mean it's something metaphysical we'll never be able to prove. For example, if two people tell you they're experiencing some pain, but one is lying about it, a few hundred years ago you'd be unable to prove it. That didn't mean pain is something metaphysical. Today, we know it might be possible to put both people in an MRI and prove whether they're actually feeling pain

quantumsequoia 6y ago

I'm not an expert on the subject, but from my understanding a Turing machine is insufficient to simulate physics. [Bell's theorem](https://en.wikipedia.org/wiki/Bell%27s_theorem https://en.wikipedia.org/wiki/Bell%27s_theorem) proves that quantum physics is truly random, as opposed to being a deterministic system that we just can't measure precisely enough. A Turing machine is unable to compute a truly random number so therefore there could be quantum phenomenon that a Von Neumann machine cannot simulate. It's not proven that consciousness can be simulated by a Turing machine. There's a possibility consciousness requires some hypercomputation to simulate