Re: Implications of Tononi's IIT?
--- On Sun, 7/25/10, Brent Meeker wrote: From: Brent Meeker Subject: Re: Implications of Tononi's IIT? To: everything-list@googlegroups.com Received: Sunday, July 25, 2010, 7:10 PM On 7/24/2010 1:32 PM, Allen wrote: On 7/23/2010 3:03 PM, Brent Meeker wrote: I'd say the information comes from the surface of Mars - it is integrated (which means summed into a whole) by the Rover and acted upon. Tononi seems to be abusing language and using "integrated" when he actually means "generated". Whether there is information generated would depend on how you defined it and where you draw the boundaries of the system. Shannon information is a measure of the reduction in uncertainity - so if you were uncertain about what the Mars Rover would do, then you could say it's action generated information. But if you knew every detail of it's programming and memory and the surface scene it viewed you might say it didn't generate any information. Brent Thanks for replying. I hope my comments to Jason explain my difference in perspective here. I don't think the information is "integrated" in the way Tononi uses the term. I don't view this system as being "connected in such a way that information is generated by causal interactions among rather than within its parts." (Balduzzi D, Tononi G 2009) I think the physical structures of the computers involved in this example exclude the generation of additional information via re-entrant feedback between any of the components (I don't know the proper terms to use here). There's no component saying to its neighbour "I see you're not 'firing', which means possibilities p & q must be excluded", everyone just goes about their business independently. Isn't that how it works at the fine scale, where everything is binary? Nobody checks which of their neighbours are 0's and which are 1's? I think you're confused about Tononi's theory. He talks about generating "effective information" which he measures by the Kullback-Lieber difference between the potential information, what Shannon would call the bandwidth, and that which the mechanism actually realizes. So the effective information is greatest when the potential states are large and the actual ones are small. So the Mars Rover is generating a lot of "effective information" when it picks out a single action based on a whole range of potential inputs. For example, it choose to go around the rock - but it would have made the same choice if dozens of pixels in it's camera switched digits. It would have chosen to go around a hole as well as a rock. I would have chosen to go around the rock if it were night or day - even though the camera image would have been quite different. Brent Brent, For some reason this message didn't make it's way to my inbox until today (Or yesterday). I had been trying a new email client until yesterday. It was not a success. I was confused about Tononi's theory, when I read the specific portion of the text regarding effective information, I made an unfounded mental leap, putting something there that didn't belong there. Now that you've cleared it up, I can't even remember fully what that phantom was, I just know it wasn't what you've stated here. I have a few textbooks on information theory, most are beyond my ability and I've put them aside to read at a later date. I never believed I knew a lot about it, but now I see I know even less about information than I thought. Sorry to have taken so long to reply, but I do appreciate your clarification. I hope some of this is sensible. I've only ever read about these things, this is the first time trying to explain any of them, and the holes in my understanding have never been so blatantly obvious. -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en. -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en. -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/24/2010 1:32 PM, Allen wrote: On 7/23/2010 3:03 PM, Brent Meeker wrote: I'd say the information comes from the surface of Mars - it is integrated (which means summed into a whole) by the Rover and acted upon. Tononi seems to be abusing language and using "integrated" when he actually means "generated". Whether there is information generated would depend on how you defined it and where you draw the boundaries of the system. Shannon information is a measure of the reduction in uncertainity - so if you were uncertain about what the Mars Rover would do, then you could say it's action generated information. But if you knew every detail of it's programming and memory and the surface scene it viewed you might say it didn't generate any information. Brent Thanks for replying. I hope my comments to Jason explain my difference in perspective here. I don't think the information is "integrated" in the way Tononi uses the term. I don't view this system as being "connected in such a way that information is generated by causal interactions /among/ rather than /within/ its parts." (Balduzzi D, Tononi G 2009) I think the physical structures of the computers involved in this example exclude the generation of additional information via re-entrant feedback between any of the components (I don't know the proper terms to use here). There's no component saying to its neighbour "I see you're not 'firing', which means possibilities p & q must be excluded", everyone just goes about their business independently. Isn't that how it works at the fine scale, where everything is binary? Nobody checks which of their neighbours are 0's and which are 1's? I think you're confused about Tononi's theory. He talks about generating "effective information" which he measures by the Kullback-Lieber difference between the potential information, what Shannon would call the bandwidth, and that which the mechanism actually realizes. So the effective information is greatest when the potential states are large and the actual ones are small. So the Mars Rover is generating a lot of "effective information" when it picks out a single action based on a whole range of potential inputs. For example, it choose to go around the rock - but it would have made the same choice if dozens of pixels in it's camera switched digits. It would have chosen to go around a hole as well as a rock. I would have chosen to go around the rock if it were night or day - even though the camera image would have been quite different. Brent I hope some of this is sensible. I've only ever read about these things, this is the first time trying to explain any of them, and the holes in my understanding have never been so blatantly obvious. -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en. -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/25/2010 12:14 PM, Bruno Marchal wrote: On 24 Jul 2010, at 23:02, Allen wrote: On 7/23/2010 1:55 PM, Bruno Marchal wrote: I think this has nothing to do with technology. It is just that consciousness is not related to the activity of the physical machine, but to the logic which makes the person supported by the computation integrating that information. Thank you for your reply. You are welcome. I mention technology because, as of now, we don't seem to have any conscious artifacts. Do you agree? Yes I disagree. But this may be not important, especially at the beginning of the argument. I can argue that all universal purpose computer are conscious artifact. But I agree that they have no means to manifest their consciousness relatively to us. So it is only through theoretical computer science that you could eventually understand what I mean by that. So I don't intend to insist on this point at all. I will have to begin learning theoretical computer science, then. I don't mean to trivialize it by stating it so lightly, but my interest has been excited. If so, what do we need to construct conscious artifacts? We need only to open our own mind. It is difficult because we are ourselves not really programmed to do that. With this respect I think that the Greek theologians, and many mystics (from East and West), and perhaps Salvia divinorum smoker can get some glimpse of what it could mean to be conscious, yet completely disconnected from our fives senses, and from time and space. Some yogic introspection technic can also leads to such an understanding. Some sleep experience too. But I don't expect most people to get the point without much more theorizing. I can appreciate this, but I have not yet really grasped it. If you don't agree, what has man constructed that is or may be conscious? Any concrete or abstract Löbian machine or theory. Such entities can reflex themselves entirely. Those are universal machine (thus conscious, but once Löbian, I would say that they are as conscious as ourselves). Or is my question nonsensical? It is juts a very difficult question, where we are deluded, by years of evolution together with 1500 years of Aristotelian brainwashing. The subject is really taboo. You have to be able to doubt about physicalism or materialism. It is better if your doubt are based on logic and observation, so that you can share them with others. I'm coming to realize I've been having problems here. In a sense it is just false to relate consciousness to any third person describable activity, and in fine, if we are machine, our consciousness, which is a first person notion, is related (not even defined by) all the possible computations going through the logical state of the machine. This entails that any machine looking at itself below its substitution level (the level at which it feels surviving an artificial digital substitution) will discover that the apparent material reality is multiple: matter relies on infinity of computations. This is retrospectively confirmed by quantum mechanics. In fine, matter is a construction of the mind, in the case we are digital machine. The brain does not makes consciousness, it filters it from infinities of first person histories. Tononi is a bit naïve, like many, on the mind-body (consciousness-reality) relationship. The integration does not rely on what a machine do, but on what an infinity of possible machines can do, and how consistent environment reacts to what the machine (person) decides. I don't want to ignore this portion, it's just more advanced than I am, I don't have a comfortable grasp on the concepts, so I can't make even an attempt at a response. Your honesty honor you. Note that I was summing up many years of solitary work in a highly counterintuive field, so it is not astonishing you have difficulties. My fault. Sorry. Thank you. I appreciate the explanations even if I do not understand them well right now. They're guidance, good for finding directions as to where I need to go. I want to ask a question about "The Origin of Physical Laws and Sensations". I don't understand it yet, I'll need to re-read the seventh step multiple times more before I figure it out comfortably. The fault is certainly my own ignorance, not your explanations. I'll be returning to it, and taking your advice on reading the List's archives. As to my question: At the third step, you wrote "Giving that Moscow and Washington are permutable without any noticeable changes for the experiencer, it is reasonable to ascribe a probability of ½ to the event 'I will be in Moscow (resp. Washington).'" I don't understand the probability here. If I am duplicated, won't there just be two Allens, AllenM (for Moscow) and AllenW (Washington)? To understand this it is useful to grasp the difference between third person descri
Re: Implications of Tononi's IIT?
On 7/25/2010 7:18 PM, Jason Resch wrote: I agree with Quentin's answer below. When information is processed recursively, iteratively, or hierarchically used to build upon results it can no longer be viewed as conveying the same meaning. An analogy is the meaning of a Book, which is built of chapters, which is build of paragraphs, sentences, words and letters. There is little to no meaning in individual letters, but when organized appropriately and combined in certain ways the meaning appears. Looking at individual operations performed by a machine is like focusing on individual letters in a book. Your analogy was very helpful. I misunderstood your previous post, sorry. Would you consider the firing or non-firing of a neuron to count as information? Yes, I would consider it to count as information, but I think it only counts as integrated information if it is connected to other neurons in a way that 'tells them' the neuron is either firing or not firing, because if it is firing, that excludes certain possibilities, and if it is not firing, that excludes certain possibilities, and so it is informative. I think of it as the neurons being so connected as to be able to 'watch' each other. Thank you for all of the other explanations you gave. I don't have a response to them, but I do appreciate them. You've cleared up some of my confusion. -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/25/2010 5:19 AM, Quentin Anciaux wrote: As Jason pointed out below, you must take the software + hardware as a whole like in the chinese room argument. If the chinese room room was running a conscious program, consciousness wouldn't be in the man acting on the symbols nor in the rules book... Like Bruno says, consciousness supervenes on all functionnaly equivalent computations (there is an infinity of them, ie: you have an infinity of possible implementations of the same computation). If consciousness was supervening on hardware your argument would stand... but it would work the same for a human brain. I'm sorry, I misunderstood Jason's message. Thank you for making me see that. Consciousness is a high level phenomena, it does not exists in parts taken separately. You won't find consciousness in a neuron nor you'll find it in an atom of a human being. Information integration like what you're talking about only exists at a high level of information processing... So likewise, if you have a conscious program, you won't find consciousness as a subroutine. Thank you for your reply. I see I've been confused about multiple things. Your explanations were helfpul! -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On Sat, Jul 24, 2010 at 3:17 PM, Allen wrote: > On 7/24/2010 12:55 AM, Jason Resch wrote: > > In the case of a digital camera, you could say the photodectors each map > directly to memory locations and so they can be completely separated and > their behavior remains the same. That isn't true with the Mar's rover, > whose software must evaluate the pattern across the memory locations to > identify and avoid objects. You cannot separate the mar's rover into > components which that behave identically in isolation. > > > Thank you for replying. > > Doesn't the rover's software run on hardware that is functionally > similar to the photodetectors, in that the memory locations could be > separated yet still behave the same? > > > I agree with Quentin's answer below. When information is processed recursively, iteratively, or hierarchically used to build upon results it can no longer be viewed as conveying the same meaning. An analogy is the meaning of a Book, which is built of chapters, which is build of paragraphs, sentences, words and letters. There is little to no meaning in individual letters, but when organized appropriately and combined in certain ways the meaning appears. Looking at individual operations performed by a machine is like focusing on individual letters in a book. > That quote reminds me of the Chinese Room thought experiment, in which a > person is used as the machine to do the sequential processing by blindly > following a large set of rules. I think a certain pattern of thinking about > computers leads to this confusion. It is common to think of the CPU reading > and acting upon one symbol at a time as the brains of the machine, at any > one time we only see that CPU acting upon one symbol, so it seems like the > it is performing operations on the data, but in reality the past data has in > no small part led to this current state and position, in this sense the data > is defining and controlling the operations performed upon itself. > > For example, create a chain of cells in a spread sheet. Define B1 = > A1*A1, C1 = B1 - A1, and D1 = B1+2*C1. Now when you put data in cell A1, > the computation is performed and carried through a range of different memory > locations (positions on the tape), the CPU at no one time performs the > computation to get from the input (A1) to the output (D1), instead it > performs a chain of intermediate computations and goes through a chain of > states, with intermediate states determining the final state. To determine > the future evolution of the of the system (The Machine and the Tape) the > entire system has to be considered. Just as in the Chinese room thought > experiment, it is not the human following the rulebook which creates the > conscious, but the system as a whole (all the rules of processing together > with the one who follows the rules). > > > I'm sure I have confused patterns of thinking where computers are > concerned. I haven't spent very much time with the Chinese Room thought > experiment, either. I followed your instructions, with the spread sheet. > Still, I don't understand how this can explain consciousness. > > I was trying to show how multiple memory locations can be processed to generate a result. Extending this, multiple results can then be taken together and processed to make a more meaningful result, and so on. At the highest levels of these layers of processing are where conscious as we know it would appear. > > Forgive me for my lack of knowledge in the subject, but what is it > that neurons in the corticothalamic area of the brain that is different from > what other neurons do or can do? > > > I apologize, I really should have explained this in post you've quoted > from. Reading it back to myself now, it seems out of context. The mention > of it again comes from my understanding of Tononi's work. I have a very > brief overview of the thalamocortical region, which I believe applies just > as well (For illustrative purposes) to the corticothalamic system. (I think > the term "thalamo-cortico-thalamic system" refers to both as a single > entity.) > > "There are hundreds of functionally specialized thalamocortical areas, each > containing tens of thousands of neuronal groups, some dealing with responses > to stimuli and others with planning and execution of action, some dealing > with visual and others with acoustic stimuli, some dealing with details of > the input and others with its invariant or abstract properties. These > millions of neuronal groups are linked by a huge set of convergent or > divergent, reciprocally organized connections that make them all hang > together in a single, tight meshwork while they still maintain their local > functional specificity. The result is a three-dimensional tangle that > appears to warrant at least the following statement: Any perturbation in one > part of the meshwork may be felt rapidly everywhere else. Altogether, the > organization of
Re: Implications of Tononi's IIT?
On 24 Jul 2010, at 23:02, Allen wrote: On 7/23/2010 1:55 PM, Bruno Marchal wrote: I think this has nothing to do with technology. It is just that consciousness is not related to the activity of the physical machine, but to the logic which makes the person supported by the computation integrating that information. Thank you for your reply. You are welcome. I mention technology because, as of now, we don't seem to have any conscious artifacts. Do you agree? Yes I disagree. But this may be not important, especially at the beginning of the argument. I can argue that all universal purpose computer are conscious artifact. But I agree that they have no means to manifest their consciousness relatively to us. So it is only through theoretical computer science that you could eventually understand what I mean by that. So I don't intend to insist on this point at all. If so, what do we need to construct conscious artifacts? We need only to open our own mind. It is difficult because we are ourselves not really programmed to do that. With this respect I think that the Greek theologians, and many mystics (from East and West), and perhaps Salvia divinorum smoker can get some glimpse of what it could mean to be conscious, yet completely disconnected from our fives senses, and from time and space. Some yogic introspection technic can also leads to such an understanding. Some sleep experience too. But I don't expect most people to get the point without much more theorizing. If you don't agree, what has man constructed that is or may be conscious? Any concrete or abstract Löbian machine or theory. Such entities can reflex themselves entirely. Those are universal machine (thus conscious, but once Löbian, I would say that they are as conscious as ourselves). Or is my question nonsensical? It is juts a very difficult question, where we are deluded, by years of evolution together with 1500 years of Aristotelian brainwashing. The subject is really taboo. You have to be able to doubt about physicalism or materialism. It is better if your doubt are based on logic and observation, so that you can share them with others. In a sense it is just false to relate consciousness to any third person describable activity, and in fine, if we are machine, our consciousness, which is a first person notion, is related (not even defined by) all the possible computations going through the logical state of the machine. This entails that any machine looking at itself below its substitution level (the level at which it feels surviving an artificial digital substitution) will discover that the apparent material reality is multiple: matter relies on infinity of computations. This is retrospectively confirmed by quantum mechanics. In fine, matter is a construction of the mind, in the case we are digital machine. The brain does not makes consciousness, it filters it from infinities of first person histories. Tononi is a bit naïve, like many, on the mind-body (consciousness-reality) relationship. The integration does not rely on what a machine do, but on what an infinity of possible machines can do, and how consistent environment reacts to what the machine (person) decides. I don't want to ignore this portion, it's just more advanced than I am, I don't have a comfortable grasp on the concepts, so I can't make even an attempt at a response. Your honesty honor you. Note that I was summing up many years of solitary work in a highly counterintuive field, so it is not astonishing you have difficulties. My fault. Sorry. I want to ask a question about "The Origin of Physical Laws and Sensations". I don't understand it yet, I'll need to re-read the seventh step multiple times more before I figure it out comfortably. The fault is certainly my own ignorance, not your explanations. I'll be returning to it, and taking your advice on reading the List's archives. As to my question: At the third step, you wrote "Giving that Moscow and Washington are permutable without any noticeable changes for the experiencer, it is reasonable to ascribe a probability of ½ to the event 'I will be in Moscow (resp. Washington).'" I don't understand the probability here. If I am duplicated, won't there just be two Allens, AllenM (for Moscow) and AllenW (Washington)? To understand this it is useful to grasp the difference between third person description and first person description. In those thought experiment some simple definition can be used (without preventing a more thorough treatment later). Consider the self-duplication experiment. I suppose you have a diary, in which you write the result of some personal testing, like "where do I feel to be"?. So if you feel yourself to be in Brussels, you write "I am in Brussels" in your diary. An external observer can agree with you, here.
Re: Implications of Tononi's IIT?
2010/7/24 Allen > On 7/24/2010 12:55 AM, Jason Resch wrote: > > In the case of a digital camera, you could say the photodectors each map > directly to memory locations and so they can be completely separated and > their behavior remains the same. That isn't true with the Mar's rover, > whose software must evaluate the pattern across the memory locations to > identify and avoid objects. You cannot separate the mar's rover into > components which that behave identically in isolation. > > > Thank you for replying. > > Doesn't the rover's software run on hardware that is functionally > similar to the photodetectors, in that the memory locations could be > separated yet still behave the same? > > As Jason pointed out below, you must take the software + hardware as a whole like in the chinese room argument. If the chinese room room was running a conscious program, consciousness wouldn't be in the man acting on the symbols nor in the rules book... Like Bruno says, consciousness supervenes on all functionnaly equivalent computations (there is an infinity of them, ie: you have an infinity of possible implementations of the same computation). If consciousness was supervening on hardware your argument would stand... but it would work the same for a human brain. > > That quote reminds me of the Chinese Room thought experiment, in which a > person is used as the machine to do the sequential processing by blindly > following a large set of rules. I think a certain pattern of thinking about > computers leads to this confusion. It is common to think of the CPU reading > and acting upon one symbol at a time as the brains of the machine, at any > one time we only see that CPU acting upon one symbol, so it seems like the > it is performing operations on the data, but in reality the past data has in > no small part led to this current state and position, in this sense the data > is defining and controlling the operations performed upon itself. > > For example, create a chain of cells in a spread sheet. Define B1 = > A1*A1, C1 = B1 - A1, and D1 = B1+2*C1. Now when you put data in cell A1, > the computation is performed and carried through a range of different memory > locations (positions on the tape), the CPU at no one time performs the > computation to get from the input (A1) to the output (D1), instead it > performs a chain of intermediate computations and goes through a chain of > states, with intermediate states determining the final state. To determine > the future evolution of the of the system (The Machine and the Tape) the > entire system has to be considered. Just as in the Chinese room thought > experiment, it is not the human following the rulebook which creates the > conscious, but the system as a whole (all the rules of processing together > with the one who follows the rules). > > > I'm sure I have confused patterns of thinking where computers are > concerned. I haven't spent very much time with the Chinese Room thought > experiment, either. I followed your instructions, with the spread sheet. > Still, I don't understand how this can explain consciousness. > > Consciousness is a high level phenomena, it does not exists in parts taken separately. You won't find consciousness in a neuron nor you'll find it in an atom of a human being. Information integration like what you're talking about only exists at a high level of information processing... So likewise, if you have a conscious program, you won't find consciousness as a subroutine. Regards, Quentin > > Forgive me for my lack of knowledge in the subject, but what is it > that neurons in the corticothalamic area of the brain that is different from > what other neurons do or can do? > > > I apologize, I really should have explained this in post you've quoted > from. Reading it back to myself now, it seems out of context. The mention > of it again comes from my understanding of Tononi's work. I have a very > brief overview of the thalamocortical region, which I believe applies just > as well (For illustrative purposes) to the corticothalamic system. (I think > the term "thalamo-cortico-thalamic system" refers to both as a single > entity.) > > "There are hundreds of functionally specialized thalamocortical areas, each > containing tens of thousands of neuronal groups, some dealing with responses > to stimuli and others with planning and execution of action, some dealing > with visual and others with acoustic stimuli, some dealing with details of > the input and others with its invariant or abstract properties. These > millions of neuronal groups are linked by a huge set of convergent or > divergent, reciprocally organized connections that make them all hang > together in a single, tight meshwork while they still maintain their local > functional specificity. The result is a three-dimensional tangle that > appears to warrant at least the following statement: Any perturbation in one > part of the meshwork may be felt
Re: Implications of Tononi's IIT?
On 7/23/2010 1:55 PM, Bruno Marchal wrote: I think this has nothing to do with technology. It is just that consciousness is not related to the activity of the physical machine, but to the logic which makes the person supported by the computation integrating that information. Thank you for your reply. I mention technology because, as of now, we don't seem to have any conscious artifacts. Do you agree? If so, what do we need to construct conscious artifacts? If you don't agree, what has man constructed that is or may be conscious? Or is my question nonsensical? In a sense it is just false to relate consciousness to any third person describable activity, and in fine, if we are machine, our consciousness, which is a first person notion, is related (not even defined by) all the possible computations going through the logical state of the machine. This entails that any machine looking at itself below its substitution level (the level at which it feels surviving an artificial digital substitution) will discover that the apparent material reality is multiple: matter relies on infinity of computations. This is retrospectively confirmed by quantum mechanics. In fine, matter is a construction of the mind, in the case we are digital machine. The brain does not makes consciousness, it filters it from infinities of first person histories. Tononi is a bit naïve, like many, on the mind-body (consciousness-reality) relationship. The integration does not rely on what a machine do, but on what an infinity of possible machines can do, and how consistent environment reacts to what the machine (person) decides. I don't want to ignore this portion, it's just more advanced than I am, I don't have a comfortable grasp on the concepts, so I can't make even an attempt at a response. It is a subtle matter, which necessitate to revise the fundamental status of physics. No amount of third person description will ever define what is consciousness, and this for reason related to Mechanism and discoveries in computer science/mathematical logic. You may look at my url for more if interested. Of you can find sum up and explanation in the archive of the list. Best, - Bruno Marchal http://iridia.ulb.ac.be/~marchal/ I want to ask a question about "The Origin of Physical Laws and Sensations". I don't understand it yet, I'll need to re-read the seventh step multiple times more before I figure it out comfortably. The fault is certainly my own ignorance, not your explanations. I'll be returning to it, and taking your advice on reading the List's archives. As to my question: At the third step, you wrote "Giving that Moscow and Washington are permutable without any noticeable changes for the experiencer, it is reasonable to ascribe a probability of ½ to the event 'I will be in Moscow (resp. Washington).'" I don't understand the probability here. If I am duplicated, won't there just be two Allens, AllenM (for Moscow) and AllenW (Washington)? When a probability becomes involved, doesn't it seem like you're saying that there is an entity "I" who is the real Allen, and that "I" may be AllenM or AllenW, but "I" will not be the other one. The other one has some "other I". Am I misunderstanding, and - since it's very likely - to what extent? I don't believe in I's, I think, for lack of a better phrase, that consciousness is all one. How do you feel about this? P.S. For anyone to answer: Is this acceptable to reply to three separate posts with three separate posts of my own, all within such a short time? I figured one would be quite lengthy, and maybe more confusing. So I split them into replies according to who I was replying to. -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/23/2010 3:03 PM, Brent Meeker wrote: I'd say the information comes from the surface of Mars - it is integrated (which means summed into a whole) by the Rover and acted upon. Tononi seems to be abusing language and using "integrated" when he actually means "generated". Whether there is information generated would depend on how you defined it and where you draw the boundaries of the system. Shannon information is a measure of the reduction in uncertainity - so if you were uncertain about what the Mars Rover would do, then you could say it's action generated information. But if you knew every detail of it's programming and memory and the surface scene it viewed you might say it didn't generate any information. Brent Thanks for replying. I hope my comments to Jason explain my difference in perspective here. I don't think the information is "integrated" in the way Tononi uses the term. I don't view this system as being "connected in such a way that information is generated by causal interactions /among/ rather than /within/ its parts." (Balduzzi D, Tononi G 2009) I think the physical structures of the computers involved in this example exclude the generation of additional information via re-entrant feedback between any of the components (I don't know the proper terms to use here). There's no component saying to its neighbour "I see you're not 'firing', which means possibilities p & q must be excluded", everyone just goes about their business independently. Isn't that how it works at the fine scale, where everything is binary? Nobody checks which of their neighbours are 0's and which are 1's? I hope some of this is sensible. I've only ever read about these things, this is the first time trying to explain any of them, and the holes in my understanding have never been so blatantly obvious. -Allen -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/24/2010 12:55 AM, Jason Resch wrote: In the case of a digital camera, you could say the photodectors each map directly to memory locations and so they can be completely separated and their behavior remains the same. That isn't true with the Mar's rover, whose software must evaluate the pattern across the memory locations to identify and avoid objects. You cannot separate the mar's rover into components which that behave identically in isolation. Thank you for replying. Doesn't the rover's software run on hardware that is functionally similar to the photodetectors, in that the memory locations could be separated yet still behave the same? That quote reminds me of the Chinese Room thought experiment, in which a person is used as the machine to do the sequential processing by blindly following a large set of rules. I think a certain pattern of thinking about computers leads to this confusion. It is common to think of the CPU reading and acting upon one symbol at a time as the brains of the machine, at any one time we only see that CPU acting upon one symbol, so it seems like the it is performing operations on the data, but in reality the past data has in no small part led to this current state and position, in this sense the data is defining and controlling the operations performed upon itself. For example, create a chain of cells in a spread sheet. Define B1 = A1*A1, C1 = B1 - A1, and D1 = B1+2*C1. Now when you put data in cell A1, the computation is performed and carried through a range of different memory locations (positions on the tape), the CPU at no one time performs the computation to get from the input (A1) to the output (D1), instead it performs a chain of intermediate computations and goes through a chain of states, with intermediate states determining the final state. To determine the future evolution of the of the system (The Machine and the Tape) the entire system has to be considered. Just as in the Chinese room thought experiment, it is not the human following the rulebook which creates the conscious, but the system as a whole (all the rules of processing together with the one who follows the rules). I'm sure I have confused patterns of thinking where computers are concerned. I haven't spent very much time with the Chinese Room thought experiment, either. I followed your instructions, with the spread sheet. Still, I don't understand how this can explain consciousness. Forgive me for my lack of knowledge in the subject, but what is it that neurons in the corticothalamic area of the brain that is different from what other neurons do or can do? I apologize, I really should have explained this in post you've quoted from. Reading it back to myself now, it seems out of context. The mention of it again comes from my understanding of Tononi's work. I have a very brief overview of the thalamocortical region, which I believe applies just as well (For illustrative purposes) to the corticothalamic system. (I think the term "thalamo-cortico-thalamic system" refers to both as a single entity.) "There are hundreds of functionally specialized thalamocortical areas, each containing tens of thousands of neuronal groups, some dealing with responses to stimuli and others with planning and execution of action, some dealing with visual and others with acoustic stimuli, some dealing with details of the input and others with its invariant or abstract properties. These millions of neuronal groups are linked by a huge set of convergent or divergent, reciprocally organized connections that make them all hang together in a single, tight meshwork while they still maintain their local functional specificity. The result is a three-dimensional tangle that appears to warrant at least the following statement: Any perturbation in one part of the meshwork may be felt rapidly everywhere else. Altogether, the organization of the thalamocortical meshwork seems remarkably suited to integrating a large number of specialists into a unified response." (From the book "A Universe of Consciousness: How Matter Becomes Imagination" written by Gerald M. Edelman and Giulio Tononi.) The important aspect of this, from my perspective with IIT in mind, is that it produces a great deal of what Tononi calls effective information, measured with the Kullback-Leibler divergence. If you haven't read Tononi's work, I think this sums up that part I'm referring to very well: "Informally speaking, the integrated information owned by a system in a given state can be described as the information (in the Theory of Information sense) generated by a system in the transition from one given state to the next one as a consequence of the causal interaction of its parts above and beyond the sum of information generated independently by each of its parts." (Alessandro Epasto, Enrico Na
Re: Implications of Tononi's IIT?
2010/7/23 Jason Resch > I am very familiar with Tononi's definition of information integration, > but if it is something that neurons do it is certainly something computers > can do as well. > > > Sorry, I meant to say that "I am *not *very familiar..." Jason -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On Fri, Jul 23, 2010 at 10:17 AM, Allen wrote: > Thank you both for replying! > > > On 7/22/2010 8:47 PM, Brent Meeker wrote: > > Sure. Consider a Mars Rover. It has a camera with many pixels. The > voltage of the photodetector of each pixel is digitized and sent to a > computer. The computer processes the data and recognizes there is a rock in > its path. The computer actuates some controller and steers the Rover around > the rock. So information has been integrated and used. Note that if the > information had not been used (i.e. resulted in action in the environment) > it would be difficult to say whether it had been integrated or merely > transformed and stored. > > Brent > > > Isn't this the same as the digital camera sensor chip? Aren't the > functions you're describing built on this foundation of independent, minimal > repertoires, all working independently of each other? I can see how, from > our external point of view, it seems like one entity, but when we look at > the hardware, isn't it functionally the same as the sensor chip in the quote > from Tononi? That is, even the CPU that is fed the information from the > camera works in a similar way. Tononi, in *"Qualia: The Geometry of > Integrated Information"*, says: > > "Integrated information is measured by comparing the actual repertoire > generated by the system as a whole with the combined actual repertoires > generated independently by the parts." > > So, what I mean is, the parts account for all the information in the > system, there is no additional information generated as integrated > information (Which Tononi refers to as "phi" Φ.) > > In the case of a digital camera, you could say the photodectors each map directly to memory locations and so they can be completely separated and their behavior remains the same. That isn't true with the Mar's rover, whose software must evaluate the pattern across the memory locations to identify and avoid objects. You cannot separate the mar's rover into components which that behave identically in isolation. > > > On 7/23/2010 12:15 AM, Jason Resch wrote: > > A Turing machine can essentially do anything with information that can be > done with information. They are universal machines in the same sense that a > pair of headphones is a universal instrument, though practical > implementations have limits (a Turing machine has limited available memory, > a pair of headphones will have a limited frequency and amplitude range), > theoretically, each has an infinite repertoire. > > > I hope no one will be offended if I borrow a quote I found on > Wikipedia: > > "At any moment there is one symbol in the machine; it is called the scanned > symbol. The machine can alter the scanned symbol and its behavior is in part > determined by that symbol, but the symbols on the tape elsewhere do not > affect the behavior of the machine." (Turing 1948, p. 61) > > I'm sure none of you needed the reminder, it's only so that I may > point directly to what I mean. Now, doesn't this - the nature of a Turing > machine - fundamentally exclude the ability to integrate information? The > computers we have today do not integrate information to any significant > extent, as Tononi explained with his digital camera example. Is this a > fundamental limit of the Turing machine, or just our current technology? > > > That quote reminds me of the Chinese Room thought experiment, in which a person is used as the machine to do the sequential processing by blindly following a large set of rules. I think a certain pattern of thinking about computers leads to this confusion. It is common to think of the CPU reading and acting upon one symbol at a time as the brains of the machine, at any one time we only see that CPU acting upon one symbol, so it seems like the it is performing operations on the data, but in reality the past data has in no small part led to this current state and position, in this sense the data is defining and controlling the operations performed upon itself. For example, create a chain of cells in a spread sheet. Define B1 = A1*A1, C1 = B1 - A1, and D1 = B1+2*C1. Now when you put data in cell A1, the computation is performed and carried through a range of different memory locations (positions on the tape), the CPU at no one time performs the computation to get from the input (A1) to the output (D1), instead it performs a chain of intermediate computations and goes through a chain of states, with intermediate states determining the final state. To determine the future evolution of the of the system (The Machine and the Tape) the entire system has to be considered. Just as in the Chinese room thought experiment, it is not the human following the rulebook which creates the conscious, but the system as a whole (all the rules of processing together with the one who follows the rules). >There is no conceivable instrument whose sound could not be reproduced > by an ideal pair
Re: Implications of Tononi's IIT?
On 7/23/2010 8:17 AM, Allen wrote: Thank you both for replying! On 7/22/2010 8:47 PM, Brent Meeker wrote: Sure. Consider a Mars Rover. It has a camera with many pixels. The voltage of the photodetector of each pixel is digitized and sent to a computer. The computer processes the data and recognizes there is a rock in its path. The computer actuates some controller and steers the Rover around the rock. So information has been integrated and used. Note that if the information had not been used (i.e. resulted in action in the environment) it would be difficult to say whether it had been integrated or merely transformed and stored. Brent Isn't this the same as the digital camera sensor chip? Aren't the functions you're describing built on this foundation of independent, minimal repertoires, all working independently of each other? I can see how, from our external point of view, it seems like one entity, but when we look at the hardware, isn't it functionally the same as the sensor chip in the quote from Tononi? That is, even the CPU that is fed the information from the camera works in a similar way. Tononi, in /"Qualia: The Geometry of Integrated Information"/, says: "Integrated information is measured by comparing the actual repertoire generated by the system as a whole with the combined actual repertoires generated independently by the parts." So, what I mean is, the parts account for all the information in the system, there is no additional information generated as integrated information (Which Tononi refers to as "phi" ?.) I'd say the information comes from the surface of Mars - it is integrated (which means summed into a whole) by the Rover and acted upon. Tononi seems to be abusing language and using "integrated" when he actually means "generated". Whether there is information generated would depend on how you defined it and where you draw the boundaries of the system. Shannon information is a measure of the reduction in uncertainity - so if you were uncertain about what the Mars Rover would do, then you could say it's action generated information. But if you knew every detail of it's programming and memory and the surface scene it viewed you might say it didn't generate any information. Brent On 7/23/2010 12:15 AM, Jason Resch wrote: A Turing machine can essentially do anything with information that can be done with information. They are universal machines in the same sense that a pair of headphones is a universal instrument, though practical implementations have limits (a Turing machine has limited available memory, a pair of headphones will have a limited frequency and amplitude range), theoretically, each has an infinite repertoire. I hope no one will be offended if I borrow a quote I found on Wikipedia: "At any moment there is one symbol in the machine; it is called the scanned symbol. The machine can alter the scanned symbol and its behavior is in part determined by that symbol, but the symbols on the tape elsewhere do not affect the behavior of the machine." (Turing 1948, p. 61) I'm sure none of you needed the reminder, it's only so that I may point directly to what I mean. Now, doesn't this - the nature of a Turing machine - fundamentally exclude the ability to integrate information? The computers we have today do not integrate information to any significant extent, as Tononi explained with his digital camera example. Is this a fundamental limit of the Turing machine, or just our current technology? There is no conceivable instrument whose sound could not be reproduced by an ideal pair of headphones, just as there is no conceivable physical machine whose behavior could not be reproduced by an ideal Turing machine. This implies that given enough memory, and the right programming a Turing machine can perfectly reproduce the behavior of a person's Brain. If an ideal Turing machine cannot integrate information, then the brain is a physical machine whose behavior can't be reproduced by an ideal Turing machine. No matter how much memory the Turing machine has, it's mechanism prevents it from integrating that information, and without integration, there is no subjective experience. Does this make the Turing machine conscious? If not it implies that someone you know could have their brain replaced by Turing machine, and that person would in every way act as the original person, yet it wouldn't be conscious. It would still claim to be conscious, still claim to feel pain, still be capable of writing a philosophy paper about the mysteriousness of consciousness. If a non-conscious entity could in every way act as a conscious entity does, then what is the point of consciousness? There would be no reason for it to evolve if it served no purpose. Also, what sense would it make for non-conscious entities to contemplate and write e-mails abou
Re: Implications of Tononi's IIT?
On 23 Jul 2010, at 17:17, Allen wrote: I'm sure none of you needed the reminder, it's only so that I may point directly to what I mean. Now, doesn't this - the nature of a Turing machine - fundamentally exclude the ability to integrate information? The computers we have today do not integrate information to any significant extent, as Tononi explained with his digital camera example. Is this a fundamental limit of the Turing machine, or just our current technology? I think this has nothing to do with technology. It is just that consciousness is not related to the activity of the physical machine, but to the logic which makes the person supported by the computation integrating that information. In a sense it is just false to relate consciousness to any third person describable activity, and in fine, if we are machine, our consciousness, which is a first person notion, is related (not even defined by) all the possible computations going through the logical state of the machine. This entails that any machine looking at itself below its substitution level (the level at which it feels surviving an artificial digital substitution) will discover that the apparent material reality is multiple: matter relies on infinity of computations. This is retrospectively confirmed by quantum mechanics. In fine, matter is a construction of the mind, in the case we are digital machine. The brain does not makes consciousness, it filters it from infinities of first person histories. Tononi is a bit naïve, like many, on the mind-body (consciousness-reality) relationship. The integration does not rely on what a machine do, but on what an infinity of possible machines can do, and how consistent environment reacts to what the machine (person) decides. It is a subtle matter, which necessitate to revise the fundamental status of physics. No amount of third person description will ever define what is consciousness, and this for reason related to Mechanism and discoveries in computer science/mathematical logic. You may look at my url for more if interested. Of you can find sum up and explanation in the archive of the list. Best, - Bruno Marchal http://iridia.ulb.ac.be/~marchal/ -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
Thank you both for replying! On 7/22/2010 8:47 PM, Brent Meeker wrote: Sure. Consider a Mars Rover. It has a camera with many pixels. The voltage of the photodetector of each pixel is digitized and sent to a computer. The computer processes the data and recognizes there is a rock in its path. The computer actuates some controller and steers the Rover around the rock. So information has been integrated and used. Note that if the information had not been used (i.e. resulted in action in the environment) it would be difficult to say whether it had been integrated or merely transformed and stored. Brent Isn't this the same as the digital camera sensor chip? Aren't the functions you're describing built on this foundation of independent, minimal repertoires, all working independently of each other? I can see how, from our external point of view, it seems like one entity, but when we look at the hardware, isn't it functionally the same as the sensor chip in the quote from Tononi? That is, even the CPU that is fed the information from the camera works in a similar way. Tononi, in /"Qualia: The Geometry of Integrated Information"/, says: "Integrated information is measured by comparing the actual repertoire generated by the system as a whole with the combined actual repertoires generated independently by the parts." So, what I mean is, the parts account for all the information in the system, there is no additional information generated as integrated information (Which Tononi refers to as "phi" ?.) On 7/23/2010 12:15 AM, Jason Resch wrote: A Turing machine can essentially do anything with information that can be done with information. They are universal machines in the same sense that a pair of headphones is a universal instrument, though practical implementations have limits (a Turing machine has limited available memory, a pair of headphones will have a limited frequency and amplitude range), theoretically, each has an infinite repertoire. I hope no one will be offended if I borrow a quote I found on Wikipedia: "At any moment there is one symbol in the machine; it is called the scanned symbol. The machine can alter the scanned symbol and its behavior is in part determined by that symbol, but the symbols on the tape elsewhere do not affect the behavior of the machine." (Turing 1948, p. 61) I'm sure none of you needed the reminder, it's only so that I may point directly to what I mean. Now, doesn't this - the nature of a Turing machine - fundamentally exclude the ability to integrate information? The computers we have today do not integrate information to any significant extent, as Tononi explained with his digital camera example. Is this a fundamental limit of the Turing machine, or just our current technology? There is no conceivable instrument whose sound could not be reproduced by an ideal pair of headphones, just as there is no conceivable physical machine whose behavior could not be reproduced by an ideal Turing machine. This implies that given enough memory, and the right programming a Turing machine can perfectly reproduce the behavior of a person's Brain. If an ideal Turing machine cannot integrate information, then the brain is a physical machine whose behavior can't be reproduced by an ideal Turing machine. No matter how much memory the Turing machine has, it's mechanism prevents it from integrating that information, and without integration, there is no subjective experience. Does this make the Turing machine conscious? If not it implies that someone you know could have their brain replaced by Turing machine, and that person would in every way act as the original person, yet it wouldn't be conscious. It would still claim to be conscious, still claim to feel pain, still be capable of writing a philosophy paper about the mysteriousness of consciousness. If a non-conscious entity could in every way act as a conscious entity does, then what is the point of consciousness? There would be no reason for it to evolve if it served no purpose. Also, what sense would it make for non-conscious entities to contemplate and write e-mails about something they presumably don't have access to? (As Turing machines running brain software necessarily would). I wonder if this is what the vast majority of AI work done so far is working towards: philosophical zombies. We can very likely, and in the not-too-distant future, build artifacts that are so life-like they can trick some of us into believing they are conscious, but until hardware has been constructed that can function in the same manner as the neurons in the corticothalamic area of the brain, or surpass them, we won't have significantly conscious artifacts. No amount of computational modeling will make up for the physical inability to integrate information. - Allen -- You received this message becau
Re: Implications of Tononi's IIT?
On Thu, Jul 22, 2010 at 5:33 PM, Allen Kallenbach wrote: > > Considering this, can consciousness be Turing emulable? That is, can > a Turing machine integrate information? I want to expand my question here, > but I don't have the knowledge to do so without distracting from the main > question I'm asking. So, all I can say is, details greatly appreciated! > > - Allen > > A Turing machine can essentially do anything with information that can be done with information. They are universal machines in the same sense that a pair of headphones is a universal instrument, though practical implementations have limits (a Turing machine has limited available memory, a pair of headphones will have a limited frequency and amplitude range), theoretically, each has an infinite repertoire. There is no conceivable instrument whose sound could not be reproduced by an ideal pair of headphones, just as there is no conceivable physical machine whose behavior could not be reproduced by an ideal Turing machine. This implies that given enough memory, and the right programming a Turing machine can perfectly reproduce the behavior of a person's Brain. Does this make the Turing machine conscious? If not it implies that someone you know could have their brain replaced by Turing machine, and that person would in every way act as the original person, yet it wouldn't be conscious. It would still claim to be conscious, still claim to feel pain, still be capable of writing a philosophy paper about the mysteriousness of consciousness. If a non-conscious entity could in every way act as a conscious entity does, then what is the point of consciousness? There would be no reason for it to evolve if it served no purpose. Also, what sense would it make for non-conscious entities to contemplate and write e-mails about something they presumably don't have access to? (As Turing machines running brain software necessarily would). There is a concept in which any Turing machine can emulate any other. This is what allows for such technology as virtual machines, and game system emulators. An old Atari game running on an emulator has no way to tell whether it is running on a physical Atari game console or within an emulator program running on a modern desktop computer. In fact there is no way any program can determine the ultimate, or actual physical substrate on which it is running. Extending this principle, if a brain's behavior can be reproduced by software, such software will have no way of knowing whether it is running on a real brain or on a bunch of computer chips. If a person did feel different, for example by not experiencing anything, or experiencing consciousness differently, this would violate the idea that software can never know for certain what its hardware is. Jason -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.
Re: Implications of Tononi's IIT?
On 7/22/2010 3:33 PM, Allen Kallenbach wrote: Buongiorno, Everything List! I have been lurking here since mid-2009, and had hoped to have a better intellectual foundation to support me before I posted anything of my own, but I would really like to ask this question. Giulio Tononi's Integrated Information Theory (IIT) states that consciousness is integrated information. In "Consciousness as Integrated Information: a Provisional Manifesto" he writes, referring to the sensor chip in a digital camera: "In reality, however, the chip is not an^ integrated entity: since its 1 million photodiodes have no way^ to interact, each photodiode performs its own local discrimination^ between a low and a high current completely independent of what^ every other photodiode might be doing. In reality, the chip^ is just a collection of 1 million independent photodiodes, each^ with a repertoire of two states. In other words, there is no^ intrinsic point of view associated with the camera chip as a^ whole. This is easy to see: if the sensor chip were cut into^ 1 million pieces each holding its individual photodiode, the^ performance of the camera would not change at all." Considering this, can consciousness be Turing emulable? That is, can a Turing machine integrate information? I want to expand my question here, but I don't have the knowledge to do so without distracting from the main question I'm asking. So, all I can say is, details greatly appreciated! - Allen Consciousness as Integrated Information: a Provisional Manifesto (/Tononi/ G 2008): http://www.ncbi.nlm.nih.gov/pubmed/19098144 Qualia: The Geometry of Integrated Information (/Balduzzi/ D, /Tononi/ G 2009): http://www.ncbi.nlm.nih.gov/pubmed/19680424 Sure. Consider a Mars Rover. It has a camera with many pixels. The voltage of the photodetector of each pixel is digitized and sent to a computer. The computer processes the data and recognizes there is a rock in its path. The computer actuates some controller and steers the Rover around the rock. So information has been integrated and used. Note that if the information had not been used (i.e. resulted in action in the environment) it would be difficult to say whether it had been integrated or merely transformed and stored. Brent -- You received this message because you are subscribed to the Google Groups "Everything List" group. To post to this group, send email to everything-l...@googlegroups.com. To unsubscribe from this group, send email to everything-list+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/everything-list?hl=en.