5 ms·

That stuff is all fun to think about, but ultimately it's not a rigorous model for anything - just a lot of assumptions and leaky abstractions. The core proble

by GuiA 10y ago

That stuff is all fun to think about, but ultimately it's not a rigorous model for anything - just a lot of assumptions and leaky abstractions. The core problem with trying to reduce reality to computation is that we have no clue how laws of the physical world around us map to computation. Are computable functions enough to model the mathematical properties of quantum particles? You can cross your fingers all you want, without this all of your reasoning becomes meaningless. The answer might be yes, but then again it might be no. If you could prove either in a rigorous and academic way, I suspect you would have a few prestigious awards waiting for you.

logicallee 10y ago
This is like asking whether the machine learning model built into AlphaGo by Google really learned Go, or whether we have any reason to think that it is. It really doesn't matter - it just matters that it beat the best human player. Yes, it plays go. When we take a very slightly different approach and put in computations that mimic human neural setups, and those computations (obviously - there are only like 80 billion neurons, they're slow and biological) report feeling pain, then of course they'll feel pain. It's absurd to even imply otherwise. The stuff you said about the laws of physics and computation is a complete and utter red herring. It'd like saying "we can't genetically engineer crops, we don't even know how plants work!! Computers can't model that stuff." It doesn't really matter if they can model it or not, we're doing it. People will be doing this very shortly. The kind of convolutions that you would have to come up with are just insane. Here, I'll give you an insane model that supports your hypothesis, could be reasonably believed, and is insane: -> The human and animal nervous systems are as a matter of fact directly controlled by God to take on their emergent behavior, and God does this by means of making tiny perturbations to the random outcomes of quantum effects. So, when a quantum effect is in a laboratory, it behaves one way: but when it is part of a biological system, God influences individual quantum outcomes such that overall the object reports feeling pain. This is just the input/output between us and God. Now, separately, God also runs this "souls-r-us" place in the Ethereal plane (where we have no contact). That's where souls actually feel pain. So the architecture is: SOuls-R-Us <-> God -> Quantum perturbations -> Physical laws -> Output says it is feeling pain. But that pain is really felt in the Souls-R-Us plane, not our worldly plane. So what happens, is if you model the biological creature, the model looks like this: Physical laws -> Output without the quantum perturbations that let God influence the physical laws to make the output report pain. So, under the above insane model, you're right, computing states would never cause actual pain to exist, since there's no Souls-R-Us involved, we didn't actually build one. But that's insane and not worth discussing. The actual state of reality is quite simple: if you build it, it will report feeling pain, because it will actually feel pain. That's it. All of your other stuff is just noise -- like arguments for why heavier than air flight is impossible. It's a distraction. Don't waste time with that distracting line of thought. It's as big of a waste of time as adding Souls-R-Us to our model. It's just not productive. Also this isn't just "fun to think about", a server farm full of servers can do more computation than a single human brain, today. This stuff is not theoretical, it's a current issue that is actually happening. fine, a decade or two out, whatever. we aren't anywhere near a VR with a human set of neurons declaring that it's feeling pain. But we are getting news every day of virtual neural nets surpassing humans in various human computational tasks, and in many cases with similar architectures (in the sense that they're neural nets). sure it'll be a while before pain emerges - but it could also be accidental. this stuff isn't "fun to think about", it's happening right now. You're like if Tim Berners Lee were saying, when he invented the web, "sure it's fun to think about someone one day controlling a tea kettle with it." well a couple of decades later, we're stuck with those design choices. These things need to be done today. It's not hand-wavy, it's not "fun to think about" - it's happening right now. As a matter of fact it's not completely out of the question that some neural net machine learning project happens to already actually feel pain but just doesn't report it. this is possible today. unlikely, but possible. it would be an absolutely shocking news report. but not as shocking as faster than light communication. Not even as shocking as an announcement that someone has built a working fusion reactor. Let that sink in: there's an active project, ITER, which is currently building a fusion reactor: https://hu.wikipedia.org/wiki/ITER https://hu.wikipedia.org/wiki/ITER Someone else announcing that they've done it, would be more shocking than the discovery that some heavy machine earning project running on a few million cores happens to be feeling pain. The latter is similar to news reports out of machine learning that we we are reading every single day. so you need to radically realign your perspective. your comment would have made a lot of sense in 1999 - not today.
- andbberger 10y ago
  'The stuff you said about the laws of physics and computation is a complete and utter red herring.' You can't argue with math. There are functions that are simply not computable on Turing machines. The canonical example is the Halting function, that in finite time reports whether another function will halt on a given input. It is easy to show that this function cannot be computed. There is no justification a priori for discarding the possibility that simulation of the substrate of qualia is outside of the realm of computable functions. Or to discard the hypothesis that qualia itself may be not Turing computable. Intriguingly, Cabessa and Siegelman (https://pdfs.semanticscholar.org/a112/4aca43b60b1e5777ae9a9c9a34695f247a30.pdf https://pdfs.semanticscholar.org/a112/4aca43b60b1e5777ae9a9c...) proved that recurrent neural networks with rational time evolving weights are super-Turing. Whether or not synaptic weights are actually rational is another story... but the point is you can't just state that it's possible to simulate qualia. We simply don't know. On the other hand, since I (and you presumably) experience qualia, it's clear the physical substrate supports it and that it is in theory possible to create something that experiences qualia. And of course given our shared lineage and shared physical substrate it's ridiculous to assume that shrimp don't experience pain. You have to at least consider that they might have sensations; all of the ingredients are there. Not to mention animals even closer to us - it seems almost certain that your dog does experience and almost as richly as you too. Finally - shrimp don't have a central nervous system. So chopping their head off probably doesn't block their sensation, whatever form that may take.
  - logicallee 10y ago
    your discussion of non-computable functions is a distracting red herring. it derails the discussion. Remember: when you talk about qualia not being computable, you are like someone saying "heavier than air flight is impossible", meanwhile birds are flying around merrily.... "well yeah but they're like biological, not metal." Not much of an argument for why flight is impossible. So what. Qualia are an emergent property of the behavior of quite well-defined physical systems. They can be simulated. they don't have a connection with something that can't be simulated. end of discussion. it's just a waste of our time to talk about it and distracts us from addressing what to do when someone makes a pain-sensing simulation, especially if that simulation has human neural structure and certainly if it reports having experiences. that is far-fetched today but will happen without any chance otherwise. It's simply impossible that it wouldn't. when I was a child I told someone that checkers will "obviously" be solved completely (i.e. is it a win for black, light-colored, or a draw) -- since there were just so few possible states, it was well within what would be computed very shortly. They called me crazy, I gave an upper bounds on number of states, and if you look at the math and compare it to the computation that was available it was simply impossible that it wouldn't be solved. it was brute-forced, and we know the outcome for perfect play. (i.e. I was "right".) simulation of a whole brain that reports qualia (the same as you do or I do) is obviously within what will be computed. It's simply something that will be done. I don't want to waste a second of my life entertaining any red herrings to the contrary. Your point about shrimp and central nervous systems is well-placed, I missed that. Here is an article on Wikipedia that discusses this in detail: https://en.wikipedia.org/wiki/Pain_in_crustaceans https://en.wikipedia.org/wiki/Pain_in_crustaceans That is a good article and mentions many of the points. However, on the balance I would be inclined to think that if something clearly behaves and indeed communicates as though it were feeling pain, then we should assign a probability that it likely does experience it, given that evolutionarily our qualia (which every person has an intuitive grasp of, since they experience it) has the same structural source. (I mean the exact wiring.) While today's reality characters might also "behave as though" they experience qualia (I mean flailing around on screen and so forth), they do not have the same structural evolutionary source (no neural net or anything), so for this reason I would not assign the idea of pain to them with certainty today. later it will be problematic (if there's machine learning for getting video game characters to flail around in pain, realistically), but that's not the case today. The end-game however is very problematic, as it will for sure be human. there is no question about this. We have the complete source code!!! Like, absolutely, we have the complete source code for how humans are wired (in the form of DNA, which has been sequenced), and in terms of emulating it with some simplifications (instead of on a molecular/atomic level) we know that the end-game is just not that many cells in the human body or the brain. the conclusion is set in stone. it's a waste of time to even talk about anything else. the question is what this means for us and how we deal with this when it happens. What do we do if someone makes a neural net and reports that they have created pain? It will be a huge problem and issue.

by GuiA 10y ago

logicallee 10y ago

This is like asking whether the machine learning model built into AlphaGo by Google really learned Go, or whether we have any reason to think that it is. It really doesn't matter - it just matters that it beat the best human player. Yes, it plays go. When we take a very slightly different approach and put in computations that mimic human neural setups, and those computations (obviously - there are only like 80 billion neurons, they're slow and biological) report feeling pain, then of course they'll feel pain. It's absurd to even imply otherwise. The stuff you said about the laws of physics and computation is a complete and utter red herring. It'd like saying "we can't genetically engineer crops, we don't even know how plants work!! Computers can't model that stuff." It doesn't really matter if they can model it or not, we're doing it. People will be doing this very shortly. The kind of convolutions that you would have to come up with are just insane. Here, I'll give you an insane model that supports your hypothesis, could be reasonably believed, and is insane: -> The human and animal nervous systems are as a matter of fact directly controlled by God to take on their emergent behavior, and God does this by means of making tiny perturbations to the random outcomes of quantum effects. So, when a quantum effect is in a laboratory, it behaves one way: but when it is part of a biological system, God influences individual quantum outcomes such that overall the object reports feeling pain. This is just the input/output between us and God. Now, separately, God also runs this "souls-r-us" place in the Ethereal plane (where we have no contact). That's where souls actually feel pain. So the architecture is: SOuls-R-Us <-> God -> Quantum perturbations -> Physical laws -> Output says it is feeling pain. But that pain is really felt in the Souls-R-Us plane, not our worldly plane. So what happens, is if you model the biological creature, the model looks like this: Physical laws -> Output without the quantum perturbations that let God influence the physical laws to make the output report pain. So, under the above insane model, you're right, computing states would never cause actual pain to exist, since there's no Souls-R-Us involved, we didn't actually build one. But that's insane and not worth discussing. The actual state of reality is quite simple: if you build it, it will report feeling pain, because it will actually feel pain. That's it. All of your other stuff is just noise -- like arguments for why heavier than air flight is impossible. It's a distraction. Don't waste time with that distracting line of thought. It's as big of a waste of time as adding Souls-R-Us to our model. It's just not productive. Also this isn't just "fun to think about", a server farm full of servers can do more computation than a single human brain, today. This stuff is not theoretical, it's a current issue that is actually happening. fine, a decade or two out, whatever. we aren't anywhere near a VR with a human set of neurons declaring that it's feeling pain. But we are getting news every day of virtual neural nets surpassing humans in various human computational tasks, and in many cases with similar architectures (in the sense that they're neural nets). sure it'll be a while before pain emerges - but it could also be accidental. this stuff isn't "fun to think about", it's happening right now. You're like if Tim Berners Lee were saying, when he invented the web, "sure it's fun to think about someone one day controlling a tea kettle with it." well a couple of decades later, we're stuck with those design choices. These things need to be done today. It's not hand-wavy, it's not "fun to think about" - it's happening right now. As a matter of fact it's not completely out of the question that some neural net machine learning project happens to already actually feel pain but just doesn't report it. this is possible today. unlikely, but possible. it would be an absolutely shocking news report. but not as shocking as faster than light communication. Not even as shocking as an announcement that someone has built a working fusion reactor. Let that sink in: there's an active project, ITER, which is currently building a fusion reactor: https://hu.wikipedia.org/wiki/ITER https://hu.wikipedia.org/wiki/ITER Someone else announcing that they've done it, would be more shocking than the discovery that some heavy machine earning project running on a few million cores happens to be feeling pain. The latter is similar to news reports out of machine learning that we we are reading every single day. so you need to radically realign your perspective. your comment would have made a lot of sense in 1999 - not today.

andbberger 10y ago
'The stuff you said about the laws of physics and computation is a complete and utter red herring.' You can't argue with math. There are functions that are simply not computable on Turing machines. The canonical example is the Halting function, that in finite time reports whether another function will halt on a given input. It is easy to show that this function cannot be computed. There is no justification a priori for discarding the possibility that simulation of the substrate of qualia is outside of the realm of computable functions. Or to discard the hypothesis that qualia itself may be not Turing computable. Intriguingly, Cabessa and Siegelman (https://pdfs.semanticscholar.org/a112/4aca43b60b1e5777ae9a9c9a34695f247a30.pdf https://pdfs.semanticscholar.org/a112/4aca43b60b1e5777ae9a9c...) proved that recurrent neural networks with rational time evolving weights are super-Turing. Whether or not synaptic weights are actually rational is another story... but the point is you can't just state that it's possible to simulate qualia. We simply don't know. On the other hand, since I (and you presumably) experience qualia, it's clear the physical substrate supports it and that it is in theory possible to create something that experiences qualia. And of course given our shared lineage and shared physical substrate it's ridiculous to assume that shrimp don't experience pain. You have to at least consider that they might have sensations; all of the ingredients are there. Not to mention animals even closer to us - it seems almost certain that your dog does experience and almost as richly as you too. Finally - shrimp don't have a central nervous system. So chopping their head off probably doesn't block their sensation, whatever form that may take.
- logicallee 10y ago
  your discussion of non-computable functions is a distracting red herring. it derails the discussion. Remember: when you talk about qualia not being computable, you are like someone saying "heavier than air flight is impossible", meanwhile birds are flying around merrily.... "well yeah but they're like biological, not metal." Not much of an argument for why flight is impossible. So what. Qualia are an emergent property of the behavior of quite well-defined physical systems. They can be simulated. they don't have a connection with something that can't be simulated. end of discussion. it's just a waste of our time to talk about it and distracts us from addressing what to do when someone makes a pain-sensing simulation, especially if that simulation has human neural structure and certainly if it reports having experiences. that is far-fetched today but will happen without any chance otherwise. It's simply impossible that it wouldn't. when I was a child I told someone that checkers will "obviously" be solved completely (i.e. is it a win for black, light-colored, or a draw) -- since there were just so few possible states, it was well within what would be computed very shortly. They called me crazy, I gave an upper bounds on number of states, and if you look at the math and compare it to the computation that was available it was simply impossible that it wouldn't be solved. it was brute-forced, and we know the outcome for perfect play. (i.e. I was "right".) simulation of a whole brain that reports qualia (the same as you do or I do) is obviously within what will be computed. It's simply something that will be done. I don't want to waste a second of my life entertaining any red herrings to the contrary. Your point about shrimp and central nervous systems is well-placed, I missed that. Here is an article on Wikipedia that discusses this in detail: https://en.wikipedia.org/wiki/Pain_in_crustaceans https://en.wikipedia.org/wiki/Pain_in_crustaceans That is a good article and mentions many of the points. However, on the balance I would be inclined to think that if something clearly behaves and indeed communicates as though it were feeling pain, then we should assign a probability that it likely does experience it, given that evolutionarily our qualia (which every person has an intuitive grasp of, since they experience it) has the same structural source. (I mean the exact wiring.) While today's reality characters might also "behave as though" they experience qualia (I mean flailing around on screen and so forth), they do not have the same structural evolutionary source (no neural net or anything), so for this reason I would not assign the idea of pain to them with certainty today. later it will be problematic (if there's machine learning for getting video game characters to flail around in pain, realistically), but that's not the case today. The end-game however is very problematic, as it will for sure be human. there is no question about this. We have the complete source code!!! Like, absolutely, we have the complete source code for how humans are wired (in the form of DNA, which has been sequenced), and in terms of emulating it with some simplifications (instead of on a molecular/atomic level) we know that the end-game is just not that many cells in the human body or the brain. the conclusion is set in stone. it's a waste of time to even talk about anything else. the question is what this means for us and how we deal with this when it happens. What do we do if someone makes a neural net and reports that they have created pain? It will be a huge problem and issue.