I have a thought experiment (video link: "Paradox of Subjective Bilateral Vision"16:00-28:00) that results in very strange situations if "high-level visual areas themselves are not sufficient for conscious vision, (or low/mid-level visual areas are necessary)", namely, that the neural mechanism of conscious vision, its verbal report and solving of perceptual visual tasks (e.g. bilateral symmetry detection) violates physics that we know of today. I would like to know if there is any experimental/theoretical evidence on this issue. Thanks in advance!
Thanks to the two contributors, the above question has developed into a discussion on how subjective vision gain simultaneous holistic access to spatially distributed neural codes. There have been claims that 'holistic access' should be considered as a serious constraint on the neural mechanism of subjective experience. In case of vision, the seamless and the unified nature of our bilateral percept can be thought as an indicator of our consciousness mechanism having holistic access to wide-spread neural representation.
Unlike many popular theories of consciousness, some scientists believe that holistic access should be solved by actual physical processes in the realm of established science. In other words, there should be some single 'entity' that has causal physical access with consequences, to all subjectively experienced information. Although, there are surprising small number of models on consciousness that actually implement such a mechanism.
I explain my "Chaotic Spatiotemporal Fluctuation" hypothesis in the linked video (40:00 - 50:00), where holistic access is implemented by deterministic chaos components in neural fluctuation. Here, I define holistic access as 'every local change in the distributed neural code evoking global system-level changes in neural fluctuation', which relies on the so-called 'butterfly effect' of deterministic chaos. For the sake of clarification, the link between 'holistic access' and 'subjective experience' goes beyond physics that we know of today.
I would very much appreciate comments on the first question too.
https://archive.org/details/Redwood_Center_2014_04_30_Masataka_Watanabe
Direct experimental evidence suggests that a particular kind of brain mechanism, theoretically modeled as the *retinoid system*, can explain conscious vision. For example, see on my RG page "Space, self, and the theater of consciousness" and, in particular, *Seeing-more-than-is-there*, pp. 324 - 325.
Direct experimental evidence suggests that a particular kind of brain mechanism, theoretically modeled as the *retinoid system*, can explain conscious vision. For example, see on my RG page "Space, self, and the theater of consciousness" and, in particular, *Seeing-more-than-is-there*, pp. 324 - 325.
Thank you, I find your theory very interesting. I always thought that 3D egocentric representation may be crucial for conscious vision, etc and wondered about its neural correlates.
In which visual areas would your 'retinoid system' potentially exist? Do you assume one egocentric contralateral mapping in each cortical hemisphere? If so, how would they communicate to form a seamless bilateral visual experience?
One important aspect of my thought experiment is the hierarchical properties of inter hemispheric cortical connectivity. According to classic corpus callosum bisection and degenerated cell staining experiments, low/mid level visual representation is only stitched together at the vertical meridian, whereas, high level visual areas have extra-vertical-meridian connectivity. (I explain this around 16:00 in the linked video)
My thought experiment claims that if visual consciousness require fine-grained visual information only available in low/mid-level areas, neural mechanism of solving visual tasks such as bilateral random-dot symmetry detection violates present day physics. Due to limited inter-hemispheric connectivity in these areas (only vertical meridian stitching connections), extra-parafoveal L/R hemifield dot coordinates cannot be put together and compared in the normal sense. In other words, perceptual neural mechanism of solving such bilateral tasks "magically" combines more L/R information than what is communicated between the two cortical hemispheres.
Dear Masataka,
You conjecture is correct. Moreover, I think one can say it is correct without recourse to complex modern neuroanatomical studies. There is a fundamental problem here: the need for experience to obey locality in the same way that other aspects of physics do. If you go to my RG site you will see a previous thread on this of mine. You can also connect to material on my UCL home page. The problem has been known for centuries and is well described by William James. It is, after all, the basis of Descartes choice of the pineal - as the only local option. A wide range of people have been thinking about this problem, particularly in the last two decades. I have a particular solution, which you can peruse. Arnold Trehub has a different solution, as does Alfredo Pereira. Others with ideas in this area include Sue Pocket and Johnjoe McFadden on EM fields, Walter Freeman and Bernard Baars on workspace theories etc etc. Stanislas Dehaene has a new book out with his view on pathways but he does not address the physical paradox.
I will look at your video.
OK Masataka, I have looked at the video. I see that you start with a very specific neurocomputational problem, which is an interesting one in its own right. I am not convinced that you have demonstrated a paradox, however. Information going up levels in the analysis hierarchy will be encoded as differentials. The fact that areas like V4 do not seem to have the resources for a high enough definition field to account for the accuracy of the symmetry match percept does not, I think, preclude that accuracy being transmitted in some encoded differential that might, for instance, bypass V4.
I think the conclusion has to be not that there is some 'extra-Shannon' effect or holism across physical non-locality but that our ways of interrogating the information in brain areas may be partial and misleading.
My more general comment is that if we want conscious experience to be a real physical event it has to be local and if we believe it is 'whole' or indivisible in some dynamic way then it has to relate to some dynamically indivisible event. That throws out all models that put experience in networks or populations of firing cells, like Tononi etc. It also means that if you insert a machine into the system there is no conscious experience in the machine if it is feeding in together with real brain signals. The experience has to be where **both of these arrive somewhere in the brain**. Signals do not exist as Shannon signals until they arrive, so arrival is necessary for the functional definition. Arrival is the only thing that can be totally necessary and sufficient, so experience must be arrival. The great mistake from Chalmers to Tononi is to put experience in the input-output, or even in the output of cells. This contradicts information theory - something known has to be an input.
You mention that the scale of representation seems to be over centimetres. However, that hides a non-sequitur assumption that it is a single representation. If experience is a representation and there are millions of them scattered all over the anterior cortex then representation extends over centimetres but each representation may extend over no more than 100 microns. This solves dancing qualia at a stroke because you lose a phenomenal representation each time you replace a tiny domain. Unfortunately, as John Searle said, your ingenious test will not work because it is just as zombie-deflatable as all the others.
The cloud of chaotic fluctuations is not 'bound' or 'together' or dynamically 'whole' in anything but an arbitrary 'systems theory' sense, which physics does not recognise. Moreover, these fluctuations are not causally related to inputs in a way that could be construed as 'access'. We have to have a causal chain, not just jumping about, with chaotic or Crick/Singer synchronic jumping about. The only way to get dynamic indivisibility is to find a place where a complex input encoding a representation interacts at a fundamental mode level with some dynamic unit. This treads into quantum territory but can be accommodated at a classical biophysical level I believe. Just in functional terms we can see there is only one possibility for this - post synaptic integration in individual dendritic trees. That is functionally indivisible at least in neuroinformational terms. Tononi's hope for a bigger phi has no physical basis.
So your question is right but the answer has to be causal. Whether it fits with known physics is a secondary issue but it does have to be causal - and I think that makes consciousness a complex input to some functionally indivisible event. See what you think.
Thank you Jonathan for pointing me to interesting links and discussion! The following is a reponse to your first comment. I will study your second comment and respond. (In short, like I say in the video, I am aware that the paradox is just a possibility and therefore my original question)
My take on the neural mechanism of holistic conscious access would be the "Chaotic Spatialtemporal Fluctuation Hypothesis (video at 40:30)".
Using it as a working hypothesis, my intention is to bring down the "fluffy ideas of qualia" to "down to earth experimentally testable grounds", as in the final "realistic experiment (video at 51:30)". In doing so, I believe that the key is to work with neural mechanisms of "solving perceptual tasks", instead of "qualia".
I am aware of the philosophical zombie debate and blind sight, so my hidden assumption would be that "neural mechanism of consciousness is also crucial for holistic access of distributed neural codes, and without it, we lose the behavioral ability to conduct precision perceptual tasks together with qualia".
Do you know of any other examples where people claim that neural mechanisms of solving visual tasks, not just qualia, may violate physics that we know of today?
As a counter example, the neural mechanism of Global Workspace Theory does not violate physics, although the claim that "fame in the brain is experienced as qualia" is metaphysical.
Thanks in advance!
Jonathan: "Arnold Trehub has a different solution, as does Alfredo Pereira. Others with ideas in this area include Sue Pocket and Johnjoe McFadden on EM fields, Walter Freeman and Bernard Baars on workspace theories etc etc."
There are many intuitions concerning the physical nature of consciousness, but empirical evidence trumps intuition. According to the retinoid model, subjectivity is the hallmark of consciousness. In this view, all competing candidate models of consciousness must account for the existence of subjectivity. Our pursuit of a standard theoretical model of consciousness would profit if other proposed theories were formulated in sufficient detail to explain how subjectivity is a natural consequence of their theoretical features. We should also expect a candidate model of consciousness to be described in a way that enables us to propose empirical tests of its theoretical implications. See "Where Am I? Redux" and "A Foundation for the Scientific Study of Consciousness" on my RG page.
Jonathan: Thanks for the great comments! I would like to study your ideas in detail. What would be the best single paper to read?
Apart from that, let me ask a question related to my paradox. I didn't exactly understand,
'The fact that areas like V4 do not seem to have the resources for a high enough definition field to account for the accuracy of the symmetry match percept does not, I think, preclude that accuracy being transmitted in some encoded differential that might, for instance, bypass V4. '
So are you claiming that high-level areas like TE TEO may recieve input directly from "lower than V4" areas, and maintain sufficient information on its own for conscious vision, using coding strategies other than the simple retinopic place coding we see in V1-V4? I totally believe in this possibility, but I am curious whether there is such direct evidence.
BTW, I am not necessary claiming that physics should be violated by neural mechanisms of subjective bilateral vision and solving of bilateral visual tasks. Just that, there is a chance, according to common-sense neuroscience regarding invariance, interhemispheric connectivity and psychophysics (linked video:16:00-28:00).
And if it is actually violated, which can be confirmed by my "minimal realistic experiement (linked video 51:30-)", the neural mechanism for "solving precision bilateral perceptual tasks" can be used as a tool to scientifically approach the mechanisms of conscious vision.
So let's say it all works. Although the results would already tell us that physics is violated at least by a purely objective behavioral mechanism, the problem of whether it was only violated by a zombie system or it was actually the mechanism of subjective vision that violated it, can be ultimately confirmed by doing experiments on ourselves.
Finally, the reason I focus on the violation of physics is that, like many other scientists/philosphers, I believe that the neural mechanism of consciousness is metaphysical. Given this assumption, I am hoping that the metaphysical nature bleeds out to behavioral neural mechanisms so that we can study it on laboratory animals with all the great tools.
Circumstantial evidence implying the "bleed out" is that all reported blind sight is heavlily limited in terms of task performance.
I am aware that it is not strong evidence, becuase brain lesions likely severe both mechanisms, subjective vision totally and visuomotor partially.
Dear Masataka,
I will respond in parts as well, since otherwise posts will get too long.
You are trying to address the key problem for perception but I worry that your framework is built on sand. To my mind contemporary philosophy of mind literature has lost the plot. Let me try to clarify what I mean by that.
Qualia are not ‘fluffy’. They are the read out of all science. Science is the business of trying to find the rules of dynamic relation that explain which qualia we get in a given context. The trouble with qualia is that they exist only as the ‘effect’ at the end of a casual chain, as the explanandum, and as such are useless as part of the explanans. We want a ‘currency converter’ that gives the rules of immediate correspondence of qualia and dynamics proximally in the brain, which would then allow us to make use of qualia in an analysis of a causal chain that ‘passes through’ those qualia to some other qualia belonging to a scientist investigating. But there are huge problems due to ascertainment and linguistic structure so we have to approach the problem using indirect inferences, rather as for inferring Higgs bosons, but for different reasons.
So you are absolutely right that we want to work with neural mechanisms rather than qualia, but not because qualia are fluffy, because neural mechanisms are the language of the explanans, while qualia remain only the explanandum. Holistic access to neural codes is indeed crucial for behaviour, but we need to make sure that we have a causal meaning for ‘holistic access’.
I worry about your use of ‘metaphysical’. Metaphysics is a level of discussion the deals with fundamental concepts in physics rather than specific instances of causality. There are no metaphysical processes because ‘metaphysical’ is merely a property of a mode of talk, not of things. If applied to things it can probably only mean ‘supernatural’ and I think that leads to nowhere.
The problem with the Global Workspace theory as proposed by people like Baars is that it talks of ‘fame’ but denies that anything is aware of the famous. It is a metaphor that fails on its own terms. It violates physics in the sense that it makes no causal sense. Fame does not occur until something is aware of the famous. The neurophysiological basis underpinning the theory may be fine, but it cannot be a theory of qualia as the effects of dynamics because it is a theory of cause without effect. I find this is difficult to get across but the theory just doesn’t make sense as it stands.
I think it is very possible that signals based on differentials used to detect right left symmetry might bypass V4. I also think they might go through V4 but this might not be apparent from the way people interrogate the function of V4. I do not know the detail of what has been done but I suspect both are options.
I am not sure what you mean by ‘retinotopic place coding’ in V4. The positions of the cells in V4 in space may be homotopic to the retina to a degree but I do not think this has anything to do with ‘coding’. I presume it is just due to morphogenetic convenience during embryogenesis. If the cells in V4 were all moved around at random it would make no difference to their function or to reportable experience as long as their connections stayed the same. The positions of the cells are no more important than the positions of fuses in a fuse box in a house. These may mirror the positions of the rooms in some way but that would be irrelevant to function.
The coding strategies for experienced images will have nothing to do with homotopy or spatiotopy. The coding has to be a causal coding. Positional relations are not causal unless they co-contribbute to a causal relation. So the relative positions of synapses in a dendritic tree might be involved in coding, because integration may depend on those positions, but relative positions of cell bodies are not causal.
Jonathan: This discussion is already helping me a lot in shaping my ideas for a manuscript. Thank you very much!
So I didn't exactly understand what you meant by,
'You are trying to address the key problem for perception but I worry that your framework is built on sand'
because I am trying to do exactly the opposite. Is this argument besides the fact that the paradox itself might not be true? Like I said, I totally admit the possibility, and it is all part of my framework, namely, trying to look for the "weakest point" to attack consciousness scientifcally. And of course, our first attempt might fail, but that is part of the framework too.
BTW, I am not claiming that "qualia is fluffy", but that the "ideas of qualia (and currently proposed approches to it) are fluffy'.
And you are right that I am mis-using the word "metaphysical". I should simply say, "beyond physics that we know of today".
So if you can provide a concrete positive example of a framework "not built on sand", it would make things easier to understand.
Together I am still trying to understand your theory. Does it provide a causal meaning for ‘holistic access’, at a different level compared to my hypothesis? If so, can you explain?
Thanks in advance!
Masataka, if consciousness is a biological event, why should we expect the "laws" of physics to inform us about consciousness?
Dear Masataka,
I appreciate that you are trying to build a theory on rock, and making use of a specific empirical discrepancy, or apparent discrepancy, at the neuronal level, which is ideal. However, I have given above various reasons why I think your current approach is founded on sand. If what I have said is not clear I am not very optimistic that I can make it clearer. It is essentially the problem posed by William James in his account of the Mind Dust Theory in Chapter 6 of Principles of Psychology. In simple terms a 'neural population vector' is for him 'not a physical fact'. James does not explain why but to a number of people it is immediately apparent what he means. Unfortunately, to others it is not. My own view is that the problem is that it does not exist as a unitary causal relation. Some spikes feed some cells, others feed others. Nothing could be aware of this 'whole' except God, if you have one.
There is a very peculiar problem here. Some people see this issue and others do not. I am a biomedical scientist. I trained in neuroscience in the lab where Hodgkin, Huxley, Adrian, and others worked (they were still around at dinner sometimes). But I spent my career in another area of complex cellular signalling - the immune system. My objections here are not metaphysical or philosophical so much as cold blooded physical and causal. What I find so odd is that some many neuroscientists, including very eminent ones I know well, do not appear to see the problem. You are not the only person to equate consciousness with 'cell activity'. In fact almost everyone does that. But it makes no sense. Consciousness is awareness, which is some sort of access to signals or receiving of signals, so we have to say what is receiving the signals and if we want the signals to function as a whole they are going to have to all be received together by something. Nothing else makes sense. So both Steve Sevush, who is an academic neurologist in Miami, and I, have concluded that what is receiving the signals has to be the dendritic tree of an individual cell - not a unique cell, but one of millions. In answer to your previous question, the paper to read from me is the one on consciousness as a property of individual cells on my RG page. The one from Steve is in J Theoretical Biology maybe 2006 - Google Sevush and consciousness. Steve is much better on anatomy than I. My UCL page gives all sorts of ways of trying to make the bsaic argument but either people see it or they do not.
So, finally, yes, my theory does provide a causal relation to explain holistic access - that is what the theory is about in toto. The paper on my site is now nine years old and has some reference to Bohm's interpretation of quantum theory that I would now see as naive. The standard formulation of quantum theory works much better, and particularly in relation to Goldstone theorem, which sounds obscure but is relevant to most phenomena in our daily lives, like sound, reflection of light, making ice cream and all sorts. And I would strongly resist trying to form a theory of consciousness 'beyond physics that we know today'. I think today's physics is fine, it is just that most people have not caught up with how condensed matter physics works these days - it is very counterintuitive to begin with but then becomes much more familiar once you see the results.
The software seems to be playing up. The last paragraph is :So, finally, yes, my theory does provide a causal relation to explain holistic access - that is what the theory is about in toto. The paper on my site is now nine years old and has some reference to Bohm's interpretation of quantum theory that I would now see as naive. The standard formulation of quantum theory works much better, and particularly in relation to Goldstone theorem, which sounds obscure but is relevant to most phenomena in our daily lives, like sound, reflection of light, making ice cream and all sorts. And I would strongly resist trying to form a theory of consciousness 'beyond physics that we know today'. I think today's physics is fine, it is just that most people have not caught up with how condensed matter physics works these days - it is very counterintuitive to begin with but then becomes much more familiar once you see the results.
Arnold, sorry for the confusion. I referred to physics as the basic microscopic building block of all sciences. So we may say instead "violating known science", and in my specific case, it can be read "the neural mechanism of solving precision bilateral matching tasks have access to more information than what is communicated between the two hemispheres".
Jonathan,
Your single dendrite theory of consciousness seems to be inspired by what you take to be James' intuition that the parallel activity of a specified population of neurons is "not a physical fact". Would you deny that the pattern of parallel activity over a specified population of rods and cones constitutes an image on the retina? Would you say that such a retinotopic image is not a biophysical fact?
Jonathan, so thanks, I guess we are getting into some juicy territory now.
'It is essentially the problem posed by William James in his account of the Mind Dust Theory in Chapter 6 of Principles of Psychology. In simple terms a 'neural population vector' is for him 'not a physical fact'. James does not explain why but to a number of people it is immediately apparent what he means.'
I am totally on the same page that "neuronal population vectors” themselves are not good enough and this is the whole reason I am proposing the "Chaotic Spatiotemporal Fluctuation Hypothesis".
My defintion of holistic access would be
"every small local change in the neuronal population vector evokes a global change in the neuronal fluctuation due to chaotic dynamics"
and in that sense,
"chaotic spatiotemporal fluctuation has holistic access to the spatially distributed neuronal population vector".
Deterministic chaos assures causality.
It’s totally understandable, if you don't believe in such a mechanism, but I hope you understand that I am on the same page in terms of arguing against bare-bone "neuronal population vectors" as conscious mechanisms.
As a side note, the expression "'neural population vector' is for William James 'not a physical fact'", would lead to a lot of unnecessary mis-understandings, taken out of context. Then again, we cannot rely on Tononi's CCD pixel metaphor, because even if the neural representation is 'integrated' as in their IIT theory, it is still 'not a physical fact'. Is this right? If so, I am wondering what would be the best way to discribe this. Have any idea?
So the question would be, would you still have a problem in my logic/framework, even if my minimal realistic experiment (linked video 51:30-) ended up showing positive results? That is, the rat doing better than what is communicated between the two hemispheres with a bilateral matching task, but perfomance going down to chance when fluctuation components are filtered out?
Jonathan,
So let me try to briefly summarize your theory.
All necessary information is gathered down onto a single dendritic tree. And a quantum mechanism has holistic accesses, since now, things are close enough for quantum effects, etc. (I apologize if there are any mis-understandings)
My question would be, are you not going beyond physics if you claim that
“information accessed by the quantum mechanism is subjectively experienced”?
I don’t see a whole big difference between my claim,
“information accessed by chaotic spatiotemporal fluctuation is subjectively experienced”.
because even in your case, the information is still spatially distributed, although in a much smaller space, but then again, is absolute distance critical?
How I view it is that, we are in same grounds in terms of proposing an actual entity that has holistic access. And by doing so, I believe that we are going one step further than simply claiming that "barebone neuronal population vector, integrated or not" is consciousness. But I would say, the final link to subjective experience is beyond physics.
Although, we don't have to argue about this because its probably just a matter of definition, I hope.
Dear Arnold,
My single cell theory was derived from my knowledge of neurophysiology. As indicated in my original paper, after submission a helpful referee pointed out that both Leibniz and James had considered the same constraints on perception and come to a related, although not identical, conclusion. (James throws in the towel in the end.)
I would certainly deny that the pattern of parallel activity in retnal cells constitutes an 'image' that is a 'physical fact' in James's terms. I am very unclear what 'image' would mean here and my understanding is that a lot of neuropsychologists would want to avoid the word. It would add nothing to the physics of the individual cellular events so in that sense would not constitute a physical fact. James's term is loose but his well argued context I think provides a useful meaning. Nothing unifies the causal relations in the bank of receptors until at some downstream point they all converge on some pathway.
Dear Masataka,
I fear that you have missed my point, which is much broader. Your local fluctuations may have knock-on effects on the rest of the brain (as in any neurophysiological model) but those effects will be later. Fluctuations cannot be consciousness by dint of what might happen later, can they? And why is this different from an ordered effect on all brain areas rather than chaotic? As far as I can see evoking chaos simply allows one to avoid making any precise predictions and that makes the theory scientifically useless.
The key point, however, is that even in a chaotic theory all causal relations are local. Their knock-on effects may spread wiely but that does not constitute 'global access' in any sense that I can see helpful. What I think may be a specific stumbling block is the use of 'access'. I can get access to the books in a library or I can get access to a coconut shy. In the former this is access in a sensory sense - I am able to be influenced by the works of many authors. In the latter it is access in a motor sense. I can knock down some coconuts. What we interested in in consciousness - in particular in your symmetry problem, is sensory access. How does whatever is conscious get sensory access to distances that seem not to be specified at the point where sensory access channels are available?
This does not seem to be helped by your 'motor access' by which a fluctuation accesses the global state later. In fact I am not sure that this counts as access since it is indirect, through many local events. More importantly, if it chaotic then it will be no good for information purposes. Chaotic behaviour is typified by high noise levels. A butterfly flap might give rise to a storm once in a million years but mostly it will not. So if you see a storm coming you cannot know that a butterfly flapped. In terms of informational access a truly chatic model is no good.
Or put simply, you cannot cheat on Shannon, nor can you cheat on the locality of phsyical relations. We all want to go 'beyond current physics' in the sense of saying what known physical relations consciousness corresponds to but you should ot mix that up with violating physics!
So Masataka, we both want to define the sort of physical relations that constitute consciousness: indeed. The difference is that in my model there is a single causal relation of influence of sensory data on some physical entity, whereas I do not see this in yours. 'Chaotic fluctuation' has no boundaries in space and time. Moreover, it is not'reading' anything in your model. The signals are the fluctuations. Something cannot read itself. You have a non-causal relation. I don't know how to put it more clearly but I cannot find a causal story in your model that would give holistic sensory access to anything anywhere. The point of the dendritic tree is that this is precisely where holistic sensory (input) access always occurs - in Whitehead's terms the PSPs are 'compresent' in a single event or 'occassion'. The reason for invoking Goldstone theorem is that it provides a specific boundary for such an event - a macroquantum boundary if you like but even without quantum theory it would count as a single energy bearing mode in classical statistical thermodynamics. One of the beauties of Goldstone modes is that they are required by classical thermodynamics just as much as quantum theory - that is how they came to be posited. So this is not really a 'quantum mechanism' so much as an energetic mode mechanism.
Jonathan, Thanks for the great comments!
I can kind of see where you are getting to. My point would be all theories of subjective experience needs to introduce some sort of "beyond physics that we know of today" assumptions at some point, because in the end, known science has nothing to say about subjective experience.
A prime example of what I mean by "beyond physics" would be Chalmer's claim that "all information has its subjective side and therefore even thermostats are minimally conscious"
Your theory to my view is no exception. And if you cannot close the loop completely, you are ultimately in the same category.
And if so, I believe that its crucial that theories are testable, at least in the near future, if we want it to be taken seriously by the scientific community and more importantly, help us understand consciousness. This is because, like I said, unfortunately in the case of subjective experience, you cannot close the loop completely with pure theory/logic.
So I would like to listen to your answer to the two critical questions that I have raised earlier, in regard to the above point.
1) Would you still have a problem in my logic/framework, even if my minimal realistic experiment (linked video 51:30-) ended up showing positive results?
2) Are you not going beyond physics if you claim that
“information accessed by the quantum mechanism is subjectively experienced”?
BTW, I am not claiming that your theory is not testable, and I am curious if you have any ideas to test it experimentally.
Jonathan,
Also, I would say your comment,
'As far as I can see evoking chaos simply allows one to avoid making any precise predictions and that makes the theory scientifically useless.',
together with other comments you make on chaos is quite wrong and you have to dive into the theories of deterministic chaos and understand what it is all about if we want to get into more juicy discussion on the critical difference between your theory and mine.
For example, Edelman and Tononi makes a very interesting point in the possible functional roles of "edge of chaos" or "class IV dynamics" to consciousness in their now classic book, "A Universe of Consciousness".
Dear Masataka,
I think much of the problem is that contemporary academic discourse tends to occur within a remarkably blinkered naive framework of what physics is about. You say beyond physics we know of today - but who's physics is that? What is wrong with Leibniz's physics of 300 years ago for instance - Leibniz was one of the main founders of modern physics and he explained how physics and consciousness relate. And when you say that known science has nothing to say about subjective experience I would suggest considering what Russell and Eddington said - that ALL science is about trying to explain subjective experience. Subjective experience is all we have to go on - ever. Physics is just working out the rules of how that experience will change in different contexts. People like Chalmers are working from a very blinkered standpoint that started up with the Australian 'materialists' of the mid twentieth century. I think we can now see that was a particularly impoverished period in academic thought. So I would suggest that maybe if one reads Descartes, Locke, Hume, Leibniz, Kant (if only to see how Kant dumbed things down) and Schopenhauer Chalmers's hard problem looks a bit silly and certainly unoriginal. Leibniz tells us that 'all information has its subjective side' and gives us a very robust framework. (You might look at my Monadology project on the UCL site.) Leibniz closes the loop completely and I have tried to fil in the modern biophysical detail.
My theory is testable in two ways. It can be tested by what we already know of neurobiology - and it is consistent in a causal framework in a way that I do not know of any other theory that is. It is also testable in that it predicts the involvement of acoustic modes in post synaptic integration, with a tentative prediction of a frequency of 1-10 MHz and the prediction that consciousness would be lost if this frequency were 'detuned' by e.g. anaesthetics, and more critically that loss of consciousness should have a better correlation with detuning than other parameters known to be influenced by anaesthetics. Not easy to do, but probably easier than your wiring up of rats I think!
And I fear your minimal realistic experiment, as you admit on the video, is going to be more or less impossible to set up in an interpretable way. I think there are problems with the premise that higher level areas 'are not sensitive to detailed location' becasue the information at this point may be carried in a complex parallel form that cannot be interrogated the way Hubel and Weisel did. Even Hubel and Wiesel only worked out what their data meant because of a chance glitch in their protocol! You might find the paper on my UCL site 'Response to Poeppel and Embick' interesting in this regard.
The software seems to be playing up again, so a bit more here.
I do understand that chaotic behaviour, as modelled, is deterministic and therefore in principle predictable. However, it cannot out-Shannon Shannon and because it involves patterns of correlation cannot even match Shannon. The reality is that in the literature this sort of 'global dynamics' suggestion is never formulated in a way that allows precise prediction, but is used to imply that you can get 'more out than you put in' in some way.
So I think you are right to consider that information is being passed in some form that is not showing up on the standard methods of interrogation of the system. However, I think we would entirely expect that to be the case, since even getting as far as Hubel and Wiesel did was a bit of a fluke. Once you get to higher levels of abstraction by differentiation anything like homotopy is lost long ago, so the reverse engineering needed to interpret findings becomes impossible - to my mind at least.
Jonathan,
So we seem to have a very interesting disagreement again.
You say, "known science do have something to say about subjective experience". Maybe there was a misunderstanding due to my poor explanation, but what I meant by known science is "established scientific theories which have been validated by numerous experiments so that we may apply, beyond question, as solid building blocks of further hypotheses"
Would you still say yes? If so, can you show me a concrete example? What would be the solid building block that bridges "holistic access by quantum mechanisms" and "subjective experience"?
So I would say, even if it does not explain everything, theories like "Quantum Mechanics", "Relativity Theory" and even "Newtonian Physics" can be applied as solid building blocks, at least as first order approximations, given a specific domain. But I would claim this is not true for "String Theory". It has never been successfully tested and they say its nearly impossible to test the current versions.
But thanks for the great discussion! Let me respond to your other questions later.
I really appreciate your comments and all the time you spent on this thread. It not only introduced me to interesting ideas, but also inspired me a lot in further shaping my own ideas. So I am even thinking of acknowledging you in my manuscript if you allow me to do so.
Jonathan, if you prefer to avoid the term "image", do you still claim that a particular pattern of excitation of retinal rods and cones is not a physical fact? I certainly agree that the detection and recognition of the pattern requires an additional mechanism that "funnels" the multi-unit excitation pattern Into a single pathway that tokens the pattern. But if the excitation pattern is not itself a physical fact, it seems obvious to me that it could not be detected.
Jonathan
'It is also testable in that it predicts the involvement of acoustic modes in post synaptic integration, with a tentative prediction of a frequency of 1-10 MHz and the prediction that consciousness would be lost if this frequency were 'detuned' by e.g. anaesthetics'
I am aware that Penrose and others are claiming this method as to test their theory. I think you share the "quantum" aspects with them, but the most shocking part of your theory is that "all necessary information (of what?) converges onto a single dentritic tree".
How would you test this aspect of your theory?
Jonathan
'And I fear your minimal realistic experiment, as you admit on the video, is going to be more or less impossible to set up in an interpretable way. I think there are problems with the premise that higher level areas 'are not sensitive to detailed location' becasue the information at this point may be carried in a complex parallel form that cannot be interrogated the way Hubel and Weisel did'
My experiment will first test the premise itself in a more harsh manner; after rewiring high-level visual areas in the two hemispheres with all other interhemispheric connectivity severed, whether "the neural mechanism of solving precision bilateral matching tasks have access to more information than what is communicated between the two hemispheres". In other words, "Does the subject do better, in terms of behavioral performance, than what is communicated between the two hemispheres".
What is communicated is not limited in format, therefore, it includes 'complex parallel form that cannot be interrogated the way Hubel and Weisel did'.
Do you still see a problem?
I wanted to initially ask, given that the results are non-ambiguously positive, putting aside for now the technical challenges, and if you still think there is one, difficulties in interpretation, is my framework/logic still built on sand?
Jonathan,
I didn't admit anything about 'more or less impossible to set up in an interpretable way' in my video. I said 'establishing sufficient artificial connectivity might be challenging'.
So if it works, there is no problem at all in the interpretation. The limitation is, if it does not work, we don't gain anything, because it might be simply due to insufficient rewiring. So it's just like in the case where you can prove the existence of a unicorn by showing one, but it is nearly impossible to disprove it.
Also let me remind you that when you referred to John Searle in your previous comments, that's not what he meant at all. There is no way you know what he meant, because I never talked to anyone in detail of the very interesting discussion I had with him.
Let me just tell you that we reached a basic agreement. He told me that he is battling against well known scientists claiming that 'neuronal population vectors' and/or 'Shannon information' is good enough for consciousness.
Masataka,
In my SMTT experiment, subjects had a vivid conscious experience (a hallucination) of a triangle in lateral motion when, in fact, there was no such object in their visual field. Any observer, looking over a subject's shoulder experiences the same hallucination. The content of this hallucinatory experience was predicted by the neuronal structure and dynamics of the retinoid model of consciousness.
It therefore seems that retinoid theory can claim to explain consciousness as a complementary relationship between patterns of autaptic-cell activity in retinoid space, a specialized neuronal brain mechanism, and conscious experience, in the same sense that the double-slit experiment in physics justifies our understanding of light as a complementary relationship between particle and wave.
What is critical evidence in this experiment is the induction of a conscious experience in N independent observers that corresponds to the conscious experience of the subject in the experiment. Could you produce the same kind of evidence in your proposed experiment?
The software seems to want bite size posts and maybe that's a good thing! I will respond to Arnold first. William James's use of the term 'physical fact' is cast in a very particular sense, which, as I indicated, is given by the context of his chapter 6 section on the mind stuff theory. What I think he means is a causal relation, and a single causal relation. He is worried about the 'combination problem'. If a physical fact is a causal relation then within the retina there are many many causal relations. Groups of similar causal relations we can consider as belonging to a common type, relevant to a particular process like vision. But they cannot be bundled together as a single causal relation or physical fact. Causal relations are not addable. If a influences B and C influences D we cannot say A+B influences C+D. So the point is that the excitations are a group of a large number of phsyical facts that considered in themselves have no causal relation to each other. The practical point is that there is nothing in the retina that could be aware of all the excitations. There is no C+D entity we can offer to take this role.
Jonathan: "The practical point is that there is nothing in the retina that could be aware of all the excitations."
I certainly don't claim that awareness is an excitation pattern on the retina. All that I insist on is that a particular excitation pattern over the rods and cones on the retina is a physical fact with causal properties for the content of vision. Similarly, I claim that excitation patterns over the egocentrically organized autaptic neurons in the brain's retinoid space constitute our global conscious/phenomenal experience, and that this global subjective experience has causal properties for our perceptual content.
Mastaka,
Yes, we are talking about conventional science - in fact I said we should not try to venture beyond. String theory would not do.
Let's go back to Descartes. He said we must be sceptical but that the one thing we can be sure of is our expereince (cogito). We can then argue that beyond that is a network of relations of necessity or cause that we can believe truly exist because they are regular in the way arithmetic is. Physics attempts to define those regularities or rules. We like to think of 'matter' but Descartes is as aware as Leibniz that our idea of 'solid stuff' is illusory and that matter can only be defined in terms of dynamic relation. Descartes feels that the dynamic relations of human souls must be unique, although he admits to Hobbes that they might not be. Leibniz makes things simpler by saying that all dynamic relations are both 'physical causations' and relations of perception and sometimes knowing. (Complexities abound but I will simplify.)
So science is not actually about 'what the world is made of' but getting as full an understanding as we can of what determines our future experiences. We spend a lot of time on distal connections in machines, bodies, stars and whatnot but we have also learnt a lot about brains. So science tells us that our experiences must be the flip side of certain causal relations deep inside our brains. We know an enormous amount about relations inside brains but we need to pin down exactly which ones host the experiences we tak about. That is tough because ascertainment is blocked by the fact that brains like to work en bloc, and a whole lot of other issues. But we can do more.
So science has got us a very long way to knowing what must underlie consciousness. There is no great new puzzle. We just have a very tricky task whittling things down to the precise relations involved. My concrete example then is 'all of science' has helped.
The reason for tying subjective experience to the interaction between a notionally quantised mode and the field potentials in its domain is that we have to have an intrinsic definition of a 'single causal relation' or for James 'physical fact' to underpin an experience, otherwise we have a regress in space and time. Essentially we have to have locality and as far we know causal relation is distributed within the domain of a mode but it is never distributed amongst determinate events. Leibniz saw this and built his entire philosophy on it. In a sense experiences must be quantised, or individuated, rather than continuous across the universe. So we are constrained to only one option by a universally non-violable aspect of physics.
Arnold: Thanks!
I am very curuious about your experiment, because working with Nikos Logothesis, that kind of work is what I usually do (unless I am heaviliy misunderstanding your short description!). Let me study it first. Can you point me to your paper because I don't have academic internet address during the weekend.
I was slightly surprised when Stuart Hameroff claimed that 10MHz acoustic modes would suit his theory as well as mine. My prediction is based on the domain being a single cell whereas he has been talking of 1000 cells. His invocation of QM in consciousness is very different from mine in many respects, although he agrees that it must be in dendrites.
Why would it be shocking to suggest that all the information in a percept/experience comes in to a dendritic tree? The estimates for the number of bits needed for an experience are sometimes given as low as 50. I think it has to be 1,000 (active or nul signals) but a million certainly seems too rich. Neurons can have 50,000 inputs so we are comfortably in the right ball park. Dendritic trees are the only places where integrative causal relations occur in brains so there isn't actually an alternative for holistic sensory access. As indicated before you cannot add together neurons C and D and say that they 'share' their inputs A and B. This is causally incorrect. C gets A and D gets B, period. Stuart Hameroff is aware of this and suggest we actually have hypernerons in which 1000 dendritic trees are all functioning as one. My question is why bother - why have 50 million degrees of freddom when 50,000 will do? And it is far from clear how you provide rules for the 1000 outputs from the single unit!
More to come but I have a very eminent quantum phsycist coing to tea it seems - quite unexpected. Maybe my theory will be different this evening.
I think the big problem with your proposed experiment is the practicality. I agree that if you get behaviour that indicates that inferences have been drawn without the necessary cross-hemisphere passage of information you have proven that experience, and indeed physical causality can violate locality. I don't think that will happen but you can try. I am also a bit worried that if you cut the callosum then your fluctuations will not go global either so any surprising behavioural abilities would not be explained by your theory? Maybe I have misconstrued that.
Jonathan: "Why would it be shocking to suggest that all the information in a percept/experience comes in to a dendritic tree?"
It is not unreasonable to suggest that relevant information in a percept comes into a dendritic tree. But I argue that it is unreasonable to suggest that all the information in a percept comes into a single dendritic tree. Suppose you perceive a red Jaguar speeding from your left to your right on a road running along a fence between you and the road. Various colors, shapes, occlusions, and distances, as well as motion and the direction of motion, have to be properly registered (bound) in your egocentric spatiotemporal representation of your current subjective world. I have struggled with this problem for many years and found no credible solution within the constraint of what we know about any single neuron. The only way that I have been able to solve the problem is by positing the neuronal structure and dynamics of the mechanisms that together constitute our putative retinoid system. And many diverse experiments and natural observations lend support to the retinoid model of consciousness.
Masataka: "Can you point me to your paper because I don't have academic internet address during the weekend."
For more about SMTT see A. Trehub. *The Cognitive Brain* (MIT Press 1991), pp. 227 - 242.
Arnold,
Thanks! I've read the SMTT and also about the other illusions included in the part I found on the internet. It is interesting that you can explain many illusions with your single framework.
My hypothesis claims more of a generic mechanism that may "wrap around" any proposed neural mechanism relying on distributed neural codes (so almost any, besides Jonathan's!). In relation, although I didn't mention it in the video, I only consider my "chaotic spatiotemporal fluctuation" as a necessary condition for subjective experience, and to determine what exactly enters subjective vision, I believe that we need an additional sufficient condition.
For example my fMRI experiment (Watanabe et al. Science, 2011) suggests that contents of V1 does not enter conscious vision, but I don't necessary claim that the neuronal fluctuation in V1 should be different from higher areas which provide its contents to subjective vision, although it will be interesting if there is a difference.
To answer your question, it is not a kind of a mechanism that can be tested by explaining or predicting visual illusions (although I do have some sufficient conditions in mind and things might change in the future).
So in my position, your retinoid system is very interesting, but it lacks mechanisms for holistic access. For example, if you combine my mechanism with yours, it would completely satisfy me, (but probably not Jonathan!).
Jonathan,
I see your logic and since I am on the same side arguing against "barebone neuronal population vectors" as consciousness, I will try to suppress my "common-sense" criticisms (e.g. criticisms against "Grandmother Cells" that applies to your theory).
As a starter, let me ask some questions, to clarify my thoughts.
1) At a given time, is only one dendritic tree of a neuron responsible for all visual consciousness?
2) What about all other modalities (sensory, motor, proprioception, thoughts, emotions, etc.)? Do they also converge onto the single dendritic tree?
3) Is this "conscious" neuron fixed as long as its intact, or does it switch around? If so in what circumstances does it switch from one neuron to another?
4) Does this neuron need to collect its "conscious contents" monosynaptically from the right neurons at the right level of representation? This is related to my "transparency of neuronal information transfer" argument which I use to build my logic (linked video: 24:50)
Thanks in advance!
Jonathan, thanks!
'I think the big problem with your proposed experiment is the practicality. I agree that if you get behavior that indicates that inferences have been drawn without the necessary cross-hemisphere passage of information you have proven that experience, and indeed physical causality can violate locality. I don't think that will happen but you can try.'
Its totally fair if you don't think its a doable experiment. Just that, I was wondering if you would still have problems in the logic/framework itself, even with positive results. If you do, it will be a great opportunity for me to think and further shape my ideas.
BTW, by travelling and seeing all the great unpublished things that people are up to, I am quite optimistic about my experiment up to a certain scale.
'I am also a bit worried that if you cut the callosum then your fluctuations will not go global either so any surprising behavioral abilities would not be explained by your theory? Maybe I have misconstrued that.'
Yes, cutting the CC, AC etc, would block the fluctuation too. The fluctuation also needs to travel to the other hemisphere with rewired connectivity.
Dear Masataka,
Arnold and I have been arguing amicably for about ten years but I have always acknowledged that I think Arnold's model is by far the best for solving specific representational problems. Like you I am interested in what I see as the complementing problem of hositic access. I get that from Arnold's model just by joining up some convergent axon branches but you are bold enough to want to test a more radical dynamic solution. I don't doubt that you may find some important answers along the way. One thing I learnt in science is not to tell someone not to do an experiment. In the meantime we may remain as three blind men inspecting the elephant of consciousness, but we do have the internet these days, and discussion always sharpens ideas - I have enjoyed this one.
Masataka,
Our posts have crossed. Back to some practicalities.
Grandmther cells are not a problem. Horace Barlow voted against them in 1972 and eveyone followed him but more recently he has voted for them. There will not be just one grandmother cell, there will be partial redundancy, but the basic idea of Jerry Letvin's gnostic cell seems pretty well proven by Rodrigo Quian Quiroga to me (and Horace). What is crucial is to recognise that the grandmother cell is the site of recognition of grandmother, but that in no way implies that this is the only site of experience of grandmother, or indeed that this is where the sort of experience of her we report occurs at all. I suspect not.
In response to the specific questions:
1. No, this theory is quite different from either WIlliam James's (rejected) proposal of a pontifical cell or Sherrington's proposal of a 'dog-pontifical cell'. Since most cortical neurons send their outputs to about 10,000 places we can expect any cell receiving data about granny or a motorbike to be one of 10,000 also receiving those data. The difference will be that the granny cell is tuned to respond. The theory Steve Sevush and I advocate says that at any one time probably millions of neurons (via relays) are getting inputs which they are each experiencing as the visual scene of the moment, plus the noises and smells. People worry that this should give a sense of multiplicity but it should not (as James points out) since the input carries no information about there being 'any other cells online'. When I read a newspaper I do not see a headline saying that 4 million other people are reading it at the same time.
2. The presumption is that the input is in Aristotle's 'common sense' - meaning that there are no modalities at this level of processing. Everything is in a common langauge of 'things in space'. Our ideas of 'five sense modalities' have pretty much collapsed in the face of cross-modal binding studies as I see it.
3. There is clearly no switching, although I guess it is possible that neurons have some down time - like when you cannot remember a name.
4. The neuron will receive its conscious contents encoded in terms of the relationship betewen the spatial array of the synapses that are giving EPSPs (or indeed IPSPs) and the dynamic response structure of the tree as a whole. I cannot get the video to run today so I not quite sure what your transparency argument is. I do not think individual synapses deliver 'pixels'. At this level of representation I think integration will be much more like words in a sentence - as both David Marr and Horace Barlow have suggested and most neuropsychologists tend to favour. So a scene of five red roses will not have inputs for the colour for every petal or even every rose. There will be encoding of a colour and an encoding of 'all five roses the same'. The encoding is likely to shift violently with introspection or change in salience without us being aware of this. So if an artist is painting the roses and is interested in how dark to do the shadows his percept will override 'all the same' and give data on relative shadow depth. He will not experience this as the rose changing but he will be aware he is thinking in a different way. I think this is relevant to the symmetry perception issue - there may be a synapse that signals 'and its all completely symmetrical', although I think the code is actually going to be much more counterintuitive than that still.
I hope that is reasonably clear. There is a lot more on this in one of my web essays (the 300 page one on Reality Meaning and Knowledge) but I think that was what you were asking for.
Sorry, it seems long posts are banned now. To continue:
4. The neuron will receive its conscious contents encoded in terms of the relationship betewen the spatial array of the synapses that are giving EPSPs (or indeed IPSPs) and the dynamic response structure of the tree as a whole. I cannot get the video to run today so I not quite sure what your transparency argument is. I do not think individual synapses deliver 'pixels'. At this level of representation I think integration will be much more like words in a sentence - as both David Marr and Horace Barlow have suggested and most neuropsychologists tend to favour. So a scene of five red roses will not have inputs for the colour for every petal or even every rose. There will be encoding of a colour and an encoding of 'all five roses the same'. The encoding is likely to shift violently with introspection or change in salience without us being aware of this. So if an artist is painting the roses and is interested in how dark to do the shadows his percept will override 'all the same' and give data on relative shadow depth. He will not experience this as the rose changing but he will be aware he is thinking in a different way. I think this is relevant to the symmetry perception issue - there may be a synapse that signals 'and its all completely symmetrical', although I think the code is actually going to be much more counterintuitive than that still.
I hope that is reasonably clear. There is a lot more on this in one of my web essays (the 300 page one on Reality Meaning and Knowledge) but I think that was what you were asking for.
Jonathan, Thanks for the detailed answers!
Let me clarify one more thing before I comment, because I am a bit confused.
5) At a given moment, does our full subjective experience (including all contents of vision, and all contents of all other modalities) originate from a single dendritic tree? In a sense, is it little like the parallel universe interpretation of quantum mechanics, where there are multiples of you, but “you” only experience this one?
I tried to rephrase the questions no.1 and no.2 slightly, because you surprisingly answered "no" to them, and noticed that theses questions may have multiple meanings due to my ambiguity.
I hope its clear this time, and I am guessing that the answer has to be "yes", otherwise you are still stuck with the very problem you are trying to address, non-locality of representation by multiple neurons.
Masataka: "... your retinoid system is very interesting, but it lacks mechanisms for holistic access."
This is interesting. Maybe we have different concepts of "holistic access". Can you explain what you mean by "holistic access"? Can you explain what mechanisms for holistic access are lacking in the retinoid system? What would such mechanisms require?
Arnold:
In terms of holistic access, Jonathan and I share views that, yes, your system needs something more for holistic access, and it that terms, I totally agree with Jonathan's earlier arguments and his suggestion of connecting your retinoid system to his single dendritic tree.
My take would be "Chaotic Spatiotemporal Fluctuation". I am sorry to force you, but as a starter, may I ask you to watch a part of my video? Diagrams help a lot, and the problem is that we don't have one here. The part I discuss holistic access is at 42:00-45:00 of the linked video. https://archive.org/details/Redwood_Center_2014_04_30_Masataka_Watanabe
Dear Masataka,
The difficulty is now with the words 'our' and 'you' in your question 5. A full subjective experience is in a dendritic tree. I am not sure if it originates, because it doesn't go anywhere, but maybe it does. If it is in one of my neurons it would in the past have included, on introspection, that this experience 'belonged to Jonathan Edwards'. More recenly I have got used to the idea that it belongs to one of the cells in the body named Jonathan Edwards and it still enjoys the sense of being that person, but from a slightly shifted angle. The play may go on with suspension of disbelief or not, according to mood.
So I think the Everett QM many worlds metaphor is distantly applicable but not too literally. There are multiple 'mes' qua subjects. Each me-qua subject is only a subject once over for its own input. Me in the other sense of me as a total functioning multicellular organism is not a me-qua-subject so it does not have to choose which me-qua-subject to be. We all accept that the brain is massively parallel and that there is 'no one place where eveything comes together' but rather lots of places where lots of things come together. I am simply saying that experience matches up with that as it should do - it is massively parallel and multiple.
So I think you would agree that I have locality of EACH representation even though there is distributed representation taken as a whole. The common mistake is to confuse unity of consciousness in the sense of unified - all elements of a percept being together - and unity in the sense of unitariness, or there only being one copy.
Put another way, there are no persons in biology except in the sense of stories, or dramas, that come with an audience of many listeners. The idea is very unfamiliar, but if we were all Mr Spock in Star Trek, with no emotional prejudice, it would be obviously the answer.
Jonathan, Thanks!
I believe we are getting closer to the critical differences between your theory and mine, but before we go on, I need to clarify one matter.
I think your answer to my previous question,
What would be the solid building block that bridges "holistic access by quantum mechanisms" and "subjective experience"?
was,
'The reason for tying subjective experience to the interaction between a notionally quantized mode and the field potentials in its domain is that we have to have an intrinsic definition of a 'single causal relation' or for James 'physical fact' to underpin an experience, otherwise we have a regress in space and time.'
I understand if you are only claiming the above as a constraint that a neural mechanism of consciousness needs to satisfy.
But I was asking for something that fills in the "Explanatory Gap". How any physical process can lead to subjective experience.
Most scientists agree that there are no theories worth listening to as of today. Are you not one of them?
No, Masataka, I am not one of them. Let me start with a brief background post. The explanatory gap is a new gap created by what CP Snow called The Two Cultures. The people studying practical aspects of our knowledge of the world, or 'scientists', and the people studying the foundational issues, or 'philosophers' no longer learn each other's disciplines. So those being practical do not understand the foundations and those dealing with foundations no longer know what they are founding. Bertrand Russell, in my grandfather's generation, was already very worried by this, although he understood both. By my father's generation people like Ryle, Quine and Armstrong had made the gap more or less unbridgeable. Both scientists and philosophers have been left with an overview of things a bit like the man in the street's naive 'materialism' - 'there's stuff out there, I know it'. Dan Dennett has championed this but it has nothing whatever to do with real science or its foundations. Chalmers got sucked in to it, despite the fact that he actually rather likes Russell's analysis. The explanatory gap is a mirage, for reasons I will come to. There are some seriously difficult issues in describing how experience relates to dynamics proximally, almost as difficult as Wittgenstein said, but not for the reasons he gave. There is a whole new discipline of proximal phenomenal/dynamic correlations that will open up in a decade or so once we have thrashed out this issue of where the experience is happening. Working in this area will be at least as difficult as the tensor algebra that Einstein had to get help with from Noethe, so brace yourself, but the future of your subject couldn't be more fascinating.
So, how does a physical process lead to subjective experience? First look at the dictionary. A 'physical' object is one we can access through our senses of sight or touch. ('Mental' things seem to pop up inside without coming through senses.) So a 'physical' thing like a flower pot is a pattern of causal relation that is defined in terms of causing, quite distally, an experience of 'seeing a flower pot'. Although we tend to think that physics describes 'stuff' it can only describe patterns of causal relation, becasue only those can cause us to know anything. So 'mass' and 'charge' are patterns of causal relation, and in abstract terms are 'dispositional properties' - tendencies to resist force or attract and repel. Dispositions to what, though? To change in space and time? But what are space and time? Wherever you look the gold standard is experience. They are our metrics of how we describe the contents of our experiences (not the experiences themselves note) in terms of the causal relations we infer to be behind them (distally). The naive realist will say, 'nonsense, there's actually a real flower pot on the wall there - look you can see it there!', not realising the irony of his claim.
It is true that much of the time we define physical dispositional properties in terms of relations between things in the outside world, but all of these can only be defined in the gold standard of experience - Russell says this in the Analysis of Matter. Otherwise we are just calibrating against calibrating against - experience.
So what is the problem with saying that there might be an experience in a flower pot? We define flower pots in terms of causal patterns that give rise to experience, so why not? What is discordant is the idea that 'physical' processes give rise to experience proximally. We are used to the idea that they give rise to experience distally in somebody's head but we are not used to the idea that they give rise to experience proximally. But then we should be a bit surprised that brains give rise to experiences proximally, should we? Well, that would be the hard problem or explanatory gap. But, hang on, we are quite happy with flower pots giving rise through causal relations to experiences in our own brains; that is what we mean by a flower pot, something that does that. So we are quite happy with experiences arising proximally, as long as they are hidden away in 'minds' that we pretend are somehow different from 'things'. If we start saying that the experience is in a causal relation in a dendritic tree people start saying 'Whoooah!!!, that's not a mind, that's a thing, that's crazy'. But what was the difference between phsyical and mental in the dictionary? - well just the route of access to what causes the experience - whether it is distal or proximal. Any other duality belongs to the religious mysticism of the masses.
I could say more, but I have come to see that you are a man of considerable perspicacy, so I suspect this is enough. There cannot be an explanatory gap, can there?
The real problem is that Newton believed matter was a continuum so physics grew without any concept of 'a single event'. But the black body radiation spectrum showed that there must be single events - and modern physics was born. Leibniz had told us there must be dynamic indivisibles. He got some things wrong, but now we have the tools to get it right.
Jonathan, Thanks again for the very elaborated explanation, but unfortunately, I don't think I fully understood. I still feel like you are only talking about constraints.
Let me question again from another viewpoint, using Newtonian physics as an example.
The law of universal gravitation is something irreducible. In the classical sense, it is a given, and we may not question it any further.
But from this law, and another irreducible law, Newton's second law, we may explain how a point of mass would behave in the non-extreme terrestrial environment without any 'gap' whatsoever (well, besides air resistence, etc.). So once we take the two laws of nature for granted, the loop is closed.
How would you place your 'quantum mechanism of holistic access' and 'subjective experience' in the above example?
My current understanding would be that, you need to introduce yet another irreducible law of nature, such as,
‘Subjective experience arises when the quantum wave function collapses ’ (but don't ask me why, that's how the universe is)
to close the loop. Can you do without such?
And btw, this is what I meant by 'we need to introduce something beyond physics that we know of today' in my previous comments.
p.s. I recently read the book 'Biocentrism'. Apart from the fact that there are a lot of problems with this book not referreing to past work etc., do you basically agree with their claims?
Masataka,
This is my working definition of consciousness:
*Consciousness is a transparent brain representation of the world from a privileged egocentric perspective.*
Would you say that, according to this definition, consciousness is "beyond physics that we know today"?
Arnold, Thanks! My discussion above is also very much related to your model too.
And yes, from my point of view,
*Consciousness is a transparent brain representation of the world from a privileged egocentric perspective.*
would be a new irreducible law of nature, or at least built upon some new irreducible law of nature, and therefore, it goes beyond physics that we know of today.
But this is exactly where I was trying to get to. Once you introduce such a law as your working hypothesis, you are no longer obliged to explain things further. So in your case, you can just claim the above, and tell us that, consciousness mechanisms can do without whatever ‘holistic access neural mechanisms’.
I my case, I wanted to go as far as I can with established physical mechanisms and delay the introduction of such new laws of nature until the very end. I placed the responsibility of explaining 'holistic access' on the 'known science' side, and the link between that and subjective experience on my new irreducible law of nature.
But still, my theory does not make sense at all without such a new law. And from my point of view, this applies to all proposed theories of subjective experience.
In this sense, dualism can be placed at the extreme end of the spectrum. Although the important point is that we are ultimately in the same category as long as we are relying on such new laws of nature. And hence, we cannot claim which is 'true' or ‘better’ based only on pure logical reasoning. Every player has a ‘joker’ in their hand, which they can choose to play at any stage they please to.
This is the reason why I put focus on experimental testability. And it is also the reason why I need to ask Jonathan to come down to our 'playing grounds', before I dive into the details!
Masataka: "This is the reason why I put focus on experimental testability."
I agree with your approach. I would be very much interested in your thoughts about dual-aspect monism, and the points I raise in my book chapter "A Foundation for the Scientific Study of Consciousness", on my RG page.
Arnold: Thanks! Good that we have an agreement. And thanks, I will read your article.
Dear Masataka,
I am not quite sure what more you are needing.
We agree there is subjective experience and that it seems to involve 'holistic access'. We may have different ideas of what holistic acces is but I go with Leibniz and James - some A is informed/influenced by or aware of some complex pattern i,j,k,l,m,n in a way that is somehow co-dependent. A computer gate that gets a 1 signal, then a 0 then a 1 does not seem any good firstly because each signal seems to be unconnected to the others and secondly because there seems to be an infinite regress - when does a series stop and a new one start?
A basic premise of modern physics, which underlies the 'double slit phenomenon' that Feynman described as the core of all QM is that the relation between a quantised mode and the pattern of potentials in its domain is totally co-dependent, in the sense that no relation between mode and potential exists independently of all the relations to all the potentials. It is exactly what we are looking for. However, once we consider relations that span more than one mode this ceases to be the case and we have the traditional classical situation that every causal relation can be demonstrated to exist separately from the others.
So we are agreed that holistic access is a feature of subjective experience and science tells us that holistic access is a defining feature of the direct relations of quantised modes to their environment, but not a feature of any relations beyond that. So there is no need to postulate any new principle. Subjective experience must be a feature of a causal relation, in order for it to cause us to talk about it. To comply with modern physics it must be a direct relation between a mode and a field of potentials. As William James said, nothing more classical can work.
'Wave function collapse' is an idea associated with rather naive interpretations of QM that I would consider obsolete. There is no point in time when a 'wave function collapses'. A wave function is a descriptor of an ensemble or type of event, not an individual event and it is certainly not the event itself. All this stuff from people like Penrose I see as pretty crazy. The holistic access of the mode to its potentials is what QM is actually about and independent of any 'interpretation'. For experiences to form a sequence I think we have to say that there is a sequence of modes, each with a fleeting experience. That implies that the modes arise and are annihilated. That gets around any talk of WFC. The experience is the temporal domain of the mode in toto. A sequence of acoustic modes in dendrites would be expected since an action potential is likely to 'wipe out' any existing mode, with a new one arising with repolarisation. I may be wrong about the detialed nature of the modes but at least there is a plausible option and one that can be tested for - perhaps using LASER interference.
Jonathan, Thanks again for the detailed description!
So I guess we should move on. Let me interpret for the time being that
'The experience is the temporal domain of the mode in toto. '
is your 'new irreducible law of nature', although you might not agree.
The main concern I have with your hypothesis is how you deal with neuronal invariance.
You claimed in your previous comment that, your single dendritic tree collects spike from quite abstract levels of representation. (e.g. Jennifer Aniston neurons in the Quiroga paper).
These neurons themselves cannot tell apart which way the face is facing, whether its a photograph or a cartoon, or even whether its simply the word "Jennifer Aniston"presented on the screen.
So if you are to collect information from these highly invariant neurons, how does it lead to our vivid 'photorealistic' vision? The necessary information for our visual experience will not be there in that single dendritic tree.
And if you are to instead collect spikes from more raw levels of representation, the number of synapses in a single neuron will not be sufficient. Moreover, there are no known areas in the brain that collect raw information from all modalities (which leads to the binding problem!)
A second point in relation would be, if all necessary information to aid full multi-modal subjective experience converges onto a single dendritic tree, how does it happen that some spikes are experienced as the color 'red' and some other are experienced as the sound of a 'trumpet'? In the world of a single dendritic tree, they will be all qualitatively non-discriminable changes of voltage etc.
The above are the two main reasons I prefer my own theory as an alternative mechanism for holistic access of distributed neural codes that accesses it on the spot and also may take advantage of the whole network structure. I guess there are always hacks to get around problems, but then a theory becomes like what "String Theory" have become in the last two decades.
Masataka: "Moreover, there are no known areas in the brain that collects raw information from all modalities (which leads to the binding problem)."
In my view, thinking in terms of brain *areas* instead of brain *mechanisms* is an impediment to understanding how the brain works.
Arnold: Thanks, I completely agree, and luckily we are allowed to rephrase it as,
'Moreover, there are no known neurons in the brain that collects raw information from all modalities'
in this special case, or to be safe,
'It is very less likely according to what we know about cortical areas that there exists neurons that collect raw information from all modalities'
Jonathan,
Maybe a more fundamental problem related to the previous two questions is, how a single dendritic tree tell which spikes are representing faces, which are representing flowers and which are representing notes of a trumpet.
This problem is somewhat related to the 'Markov Blanket'.
http://en.wikipedia.org/wiki/Markov_blanket
If a single dendritic tree is all you have, there is a severe limit in tagging 'meanings' to incoming spikes, whereas if you have the whole polysynaptic network to work with, you can trace things back to its origin as in the generative model.
BTW, a mechanism which relates to the generative model is what I am thinking as the 'sufficient condition' that I mentioned quite earlier. For the mechanism of holistic access to take into account such network morphology, I believe that we need to go side by side with the network and have access to the distributed code on the spot, as I attempt to do with my Chaotic Spatiotemporal Fluctuation Hypothesis.
Jonathan:
And the other important issue, in regard to our previous discussion, is how you can possibly test the "single dendritic tree" aspect of your theory with near future doable experiments.
Dear Masataka,
Thanks for the further questions. I am glad these are the ones you come up with since they are the ones I have spent the last few years addressing. Some of that is in the essays on my site but buried a bit deep. The questions are right but there are answers, I believe. I am just about to go out to sea in my boat for a few days so a blank from me is not a cessation of interest but a lack of internet access.
Briefly, to start, as I hinted above, the JA cell is probably not where the experience of her face we talk about is hosted, nor for the granny cell. This is partly for the invariance reason but also because when we experience 'the face of JA' the experience has the recognition already stamped on it. The experiences we talk about would have to be in cells that are downstream of the recognising cell bank. The dynamics of integration may be very different in recognising and experiencing cells - ranging from something like linear integrate and fire to something much more pattern dependent maybe.
You say there are no cells known to receive raw data from all modalities. I would be interested to know how one can tell that? The connections do not have to be direct and my naive understanding is that if you take into account thalamocortical loops you can trace connections between pretty well all cortical cells via a few relays. I am also cagey about the idea of 'raw' data. I agree that for the experience we want to ratchet back to lower levels of invariance or abstraction but I would deny that we are somehow going back to something like pixels. You also say that no dendrite would accommodate enough raw data but I think that is generally recognised to be unfounded. Colin Blakemore, who I see fairly regularly at the Institute of Philosophy Censes seminars (he now has a chair in philosophy!) reckons that an experience only needs 50 bits. I think it needs 1,000-10,000 but I don't know of any neuropsychologist who suggest more than that except maybe Victor Lamme. For sure, there are lots more bits held in buffer in somewhere like V1 but the number let through to experience at any one time looks from the change blindness and recall studies to be pretty small. You don't need, or indeed want pixel type raw data. When I see a spray of yellow azalea flowers I don't need data for colour for each of forty flowers, my experience is of frty odd flowers of 'the same yellow'. So I don't see a problem here.
More later, maybe after the boating.
Masataka,
A quick word about 'Biocentrism'. I have not read the book but I have been trawling through the review material. As far as I can see this is a very low level analysis of the problem - what I call a re-invented wheel that isn't quite round anyway. Heraclitus probably did a better job 2.5K years ago and Leibniz certainly did better. It looks as if they start with some useful insights - of the sort that any decent neuropsychologist should be fully aware of - and then go off in the wrong direction. It reminds me of the discussion we had as teenagers at school 50 years ago to be honest.
Maybe I am suggesting a new irreducible law but, as indicated on my thread about this, all I think I am saying is that experience obeys the locality rules of physics. I see that as more a clarification or endorsement of laws already existing.
For the issue of the red and the sound of the trumpet I think the boot is actually on the other foot. If these are encoded in action potentials in distributed brain areas we have no reason at all to think they would be experienced differently (apart from there being nothing specified to do the experiencing). The spatial arrangement of nerve cells has nothing to do with their informational capacities - as I said earlier, it is probably just embryological convenience. I don't know what you mean by tracing back to origins and I am not familiar with a generative model. What would do the tracing back -there are no railway track maintenance men to do that for the brain. And the redness of a tomato is not 'back', it is a high level inference that might be based on 95% monochromatic sodium yellow light in a dark street. This sounds to me like violating cauality and information theory again.
In contrast, if encoding of qualia is at synapses we have the possibility of a very rich range of qualia because these would be encoded in **local dynamic co-relation**. This is essentially the basis of Tononi's idea of integrated information - you get rich options when you consider combinatorial relations between signals. In my model the relations between PSPs are REAL relations because they are causal relations within a single event. In contrast the relations between spikes in separated cells in a brain have no real relation - except of course when those spikes come to converge on a downstream dendritic tree! That is the only sort of integration the brain actually has. So Tononi's maths is right but it only has a real implementation in one dendritic tree at a time - phi has to be a dendritic tree.
When we talk of testing hypotheses we are not talking of proving an idea. We are talking of either disproving our idea or finding that the idea remains plausible but not proven. So testing is not for the faint hearted. A useful experiment is one that destroys an idea. However, there is of course the possibility of doing an experiment that destroys the possibility of all the alternatives to an idea that are consistent with ones general theoretical framework. My argument is that the data in a simple neuroanatomy textbook destroy all explanations for experience compatible with physics other than one based in an individual dendritic tree, simply because that is the only structure where there is integration, or causal co-relation, of signals. Popper tells us in one of his essays on Conjectures and Refutations that before doing further experiments one should look to see whether an idea can be refuted on grounds of inconsistency with the theoretical framework.
But as a bonus it is nice to do corroborative experiments as well. My model requires that there are indivisible dynamic modes occupying at least large portions of dendritic trees. If someone could show that there are no such modes my model would be refuted. The trouble is that I doubt this could be shown. On the other hand if I predict that there ought to be an acoustic mode with a frequency of around 1-10MHz and this is found (I came to this figure some years back quite independently of what has been reported in isolated microtubules recently) then at least I would have a corroboration.
Experiments neither prove nor disprove a theoretical model. Experiments serve to provide evidence for or against a particular theoretical proposal. As far as I can see, the weight of available evidence supports the retinoid theory of conciousness.
Jonathan: "So a scene of five red roses will not have inputs for the colour for every petal or even every rose. There will be encoding of a colour and an encoding of 'all five roses the same'."
It seems to me that you have fallen into a very common conceptual trap. You confuse a description of an immediate conscious presence/experience with the conscious experience as such.
Suppose you are transported to an unknown garden in full bloom while your eyes are closed. Immediately upon opening your eyes you will only see an egocentrically organized spatiotemporal array of colors and shapes. It is only after your non-conscious cognitive mechanisms decompose and analyse parts of the global experience that you would recognize the scene as a garden. And only then you might think of "all five roses" and their color. The thought itself might be expressed as inner speech, which would then become an integral part of your current conscious experience.
Masataka,
In your "metaphysical definition" of consciousness, you make a sharp distinction between consciousness and the content of consciousness. What would consciousness be like without any content?
Jonathan,
Thanks! Let me respond by the time you are back. Have a great trip!
Meanwhile, I will answer your following question,
'I don't know what you mean by tracing back to origins and I am not familiar with a generative model. What would do the tracing back -there are no railway track maintenance men to do that for the brain. And the redness of a tomato is not 'back', it is a high level inference'.
The key point of the 'generative model' or 'predictive coding' (Kawato, Mumford) is that, with topdown projection, the brain simulates what causally occurs in the environment. Taking vision as an example, let's say there is a house. The house has surfaces with reflective properties. The light from the sun shines on them and some of them reflect back. Some reflection directly enters our retina, some other will bounce repeatedly in the environment and it may or may not enter our retina. This causal process is coined 'forward optics' and the generative model claims that the brain simulates this process using topdown projection. It starts from a high-level symbolic representation (e.g. there exists a house) and creates a low-level retinotopic representation just as in the rendering process of computer graphics. Then the difference between the simulated low-level representation and the sensory driven low-level representation, so-called the prediction error, is calculated and fed-back by bottom-up projection to update high-level representation, leading to minimization of prediction error.
Through this process of forward optics, high-level symbolic representation is dynamically routed towards its real-time low-level counterparts (or traced back!), depending on other high-level representation such as position, size, facing angle, etc. of the house, which is also updated by prediction error.
Why I favor this model in the context of sensory qualia is because the simulated forward optics/acoustics/etc. reflects the natural statistics of the given modality (e.g. vision: high probability of continuation of features in space, audition: high probability of continuation of features in time). In a Japanese book chapter I wrote few years back (渡辺正峰.意識.イラストレクチャー認知神経科学), I have claimed that this might be the key to explaining qualia, why color is color and sound is sound, etc. In fact, I bravely proposed that, parts of the topdown forward process, as a neural algorithm, leads to subjective experience.
I still believe it to be a possible solution to the very fundamental problem of consciousness, 'how some neuronal firing leads to the color red and how others lead to the sound of a trumpet'.
And when I was trying to come up with a neural mechanism of holistic access, one of the constraints was that it goes nicely with the above proposal. That, 'it accesses the distributed neural code on the spot' so that it may 'wrap around neural algorithms'
Arnold, Thanks for the great feedback!
'What would consciousness be like without any content?'
In case of an innately blind person, they are conscious, but they don't have any contents of vision. Another example would be, we have visual consciousness, but there is no content in the back of our head, leading to 'nothingness'.
But yes, the way I use 'consciousness' and 'contents of consciousness' might be quite confusing. 'Chaotic spatiotemporal fluctuation' can be explained simply as the 'neural mechanism of holistic access to distributed codes' instead of claiming that it is 'consciousness' itself. This combined with the 'contents of consciousness (= neuronal population vector)', becomes subjective experience.
Jonathan,
Since you admitted that
'Maybe I am suggesting a new irreducible law',
you are relying on a beyond physics hypothesis to close the loop of ‘neural mechanism -> subjective experience’, meaning that the loop is not closed in terms of established science. Then you are also in the game of playing with a 'joker (=a new irreducible law)' in our hands ,where we may choose to play it any stage we want (e.g. dualism).
Hence, basically, you cannot make such strong claims as,
'My argument is that the data in a simple neuroanatomy textbook destroy all explanations for experience compatible with physics other than one based in an individual dendritic tree, simply because that is the only structure where there is integration, or causal co-relation, of signals',
because we should admit that other jokers for holistic access like mine, might just be true. Well, I admit that we are totally in equal grounds, and even admit that we cannot logically negate neural mechanisms of consciousness without holistic access. This is regardless of my personal position that 'holistic access' should be taken seriously as a contraint on the neural mechanism of consciousness.
Therefore, I take the position that experimental testability is crucial.
But still, it is very interesting to debate over which hypothesis make more sense, and it surely helps to further shape our ideas, at least for me. So thanks for the great discussion!
Masataka: "In case of an innately blind person, they are conscious, but they don't have any contents of vision. Another example would be, we have visual consciousness, but there is no content in the back of our head, leading to 'nothingness'."
1. I want to stress that in the retinoid model all exteroceptive and interoceptive sensory modalities have recurrent (feedforward-feedback loops) projections to and from retinoid space, not just vision (see Fig. 8 in "Space, self, and the theater of consciousness").
2. If we "have visual consciousness" but "no content in the back of our head", why should this lead to "nothingness"? This proposal seems incoherent.
3. I argue on the basis of my own experience, and the testimony of others, that we are conscious if and only if we have at least a primitive sense of something somewhere in relation to our self. So there must be this minimal subjective content for consciousness to exist. This would corespond to C1 in my article "Space, self, and the theater of consciousness".
4. What principled reason leads to the conclusion that 'Chaotic spatiotemporal fluctuation' together with a "neuronal population vector" becomes subjective experience? I ask this because, in my view, subjective experience must depend on some kind of spatiotemporal representation having a fixed locus of perspectival origin.
Dear Masataka,
I am online for a while again. I am very familiar with predictive coding. In fact I think one can deduce from very simple introspection that the brain must work like that. My understanding is that Bishop Berkeley worked out something similar! But I do not see how it relates to subjective experience and access to signals. The brain generates effent copies, forward models, or whatever you like to call them, subtracts from sensory input and that results in the set of signals that encode what is experienced. Somethng still has to experience a set of signals that it has as informational input. And that set of signals does not bring look up tables about past learning routines with it - it is just a set of signals.
VS Ramachandran invented a thought experiment that I think he got wrong and I think is relevant. The idea is that we have a congenitally blind man A and a normal man B who can see red. We get to the stage of knowing which action potential(s) in B's brain 'encode red'. We then connect the axon via a synthetic axon substitute to A's brain and he will see red for the first time. This must be wrong. The nature of the quale A gets will surely depend entirely on where the signal arrives, not where it came from. The qualia associated with signals in brains must be a matter of where the signal arrives, as much as where it comes from. For A we have to ask if the synthetic axon is grafted on to one he uses for hearing or smell, or is a new synapse built ofr it, or indeed is it plugged in to an unused synapse designed emberyologicaly to indicate red (or blue maybe).
This is why I can see no possibility in qualia being determined by some aspect of a cell body firing off a signal as part of a distributed set of cells sending off such signals. These are indeed only 'signals', functionally completed, if they arrive somewhere and I see no other possible meaning to 'signals being sensed together' than them arriving together - it would make no computational sense.
So I come back to a key puzzle for me about your model. You say that my suggestion of a dendrite would not provide access to enough signals (say the input to 10,000 synapses). But it is unclear that there is anything in you rmodel that has access to signals as input at all, let alone in a bound or holistic fashion and more than 10,000 options. For my example of A having an input pattern of signals i,j,k,l,m,n aboe, what in your model is A? If there is no need for an A then I am not sure what your difficulty with Arnold's explanation is. I am afraid I am a bit lost.
Jonathan, Some comments in relation to your previous comments.
I basically agree with your following comments on validation of theories by experiments,
'We are talking of either disproving our idea or finding that the idea remains plausible but not proven. ..., useful experiment is one that destroys an idea'.
But most established theories like Relativity Theory had good 'nearly only the winner survives' experiments (e.g. light bending around the sun observed during a solar eclipse), but others like String Theory don't, which remains in limbo after all that hype.
I am curious if you have such an experiment in mind regarding the ‘all conscious contents are converged onto a single dendritic tree’ aspect of your hypothesis. I don’t see a ‘nearly only the winner survives’ experiment that is doable in the near future.
And I am afraid, Masataka, that I do not buy the argument in your last post. As indicated, my new proposal is simply to stick to the rules of physics for experience proximally as we do distally for its dynamic precursors. This is no joker, it is just being consistent. It is exactly the same argument that Descartes used and I think he was much closer to consistent science than most neuroscientists are today.
And I detect a certain sleight of hand in your 'beyond physics', because your beyond physics does actually violate the laws of physics as we know them. You are not just beyond in a metaphysical sense, you are suggesting an explanation of a form that physics normally forbids. (Or at least so far I can't see what has any 'access' of the input sort we need.) That seems to me not only more of a joker, but possibly real cheating! I sympathise to an extent because as far as I can see more or less everyone in neuroscience is cheating just the same - Tononi, Edelman, Dehaene, Freeman, etc. including all the forward model people. The person who had an honest causal explanation was Descartes. He just got the place wrong.
It is indeed good to discuss these things in this detail. it clarifies one's ideas a lot. I am sticking to my story so far - unless you can tell me what A is?
Dear Arnold,
I am not confusing description with the real thing. I am espousing a view taken by David Marr and my friend Horace B. The idea is that the conscious image is constructed not homotopically but in the way 'a sentence is bult from adjectives'. I do not think that I ever see colours without them belonging to things when I open my eyes. The first experience I have is of coloured things. And there is all that evidence in Purves and Lotto that the colours you see are totally dependent on what you think they belong to.
Not sure about the idea that all great theories have experiments to test them, Masataka. Thinking at random, it seems to me that a very high proportion of innovative theories just explained a large body of data already available: Copernicus, Newton on gravity and laws of motion, Faraday and Maxwell on electricity, Darwin, Einstein on the photon and special relativity (which explained the photoelectric effect and Michelson Morley), Bohr on quantisation of electron orbitals etc etc.
The problem I have in thinking of a new experimental test for the dendritic tree hypothesis is that I cannot think of another explanation that is consistent with physics and with known neuroanatomy. So I wouldn't be able to say what different predictions would be made by another theory. If the only places in brains to have access to lots of signals all together are dendritic trees I am stuck for saying that dendritic trees are more plausible than - what exactly? What other A can I argue is less good?
Jonathan,
I agree, it does not have to be new experiments. But I guess you do have to agree that these established theories had 'something' that made it establish itself, 'something' that killed alternative competing theories. But I don't see that in your current arguments, because you are not closing the loop of causal explanation as in the established examples. I mean you admitted the use of a 'new irreducible law of nature: quantum mechanism that transforms dendritic code to subjective experience', which leaves the loop open.
A related problem is, the abundance of circumstantial evidence that makes your theory hard to believe to the common-sense neuroscientists. Yes, its a simple solution to the problem, but it just does not feel right with all mounted evidence against it. I am pretty sure you faced many of these responses in the past.
But for now, let's forget common-sense.
I believe that the weakest points of your theory are still the ones that I pointed out previously. Namely the problem of information coded in a single spike. After ripping off all the network structure and leaving only a very local portion of it, a monosynaptic converging network onto a single dendritic tree, this tree cannot tag any meanings to spikes.
You say,
'The qualia associated with signals in brains must be a matter of where the signal arrives, as much as where it comes from.'
This means that somehow a single dendritic tree needs to 'learn' that a spike arriving at certain positions of the tree is the color red. How could this possibly happen when your single neuron has only single axonal output. It cannot really 'learn' by interacting with the environment. (BTW, do you support that subjective experience is epiphenomena, having no causal links to behavior?).
Basically it boils down to whether you have a real convincing explanation to
'The connections do not have to be direct'
Why so?
Another way to phrase it is, since the single dendritic tree has no access to the morphology of the network structure, you don't have the 'code book' of that incoming spike. There is no way a single dendritic tree can tell 'this spike is a face', 'that spike is a tree'.
On the other hand, if you let the whole network do the job, it has access to the network structure, or in other words, it is the network itself. It is similar to Bayesian networks, where the meaning is embedded in the topology of the network.
Arnold,
'1. I want to stress that in the retinoid model all exteroceptive and interoceptive sensory modalities have recurrent (feedforward-feedback loops) projections to and from retinoid space, not just vision (see Fig. 8 in "Space, self, and the theater of consciousness").'
Are you claiming that due to recurrent activity, the distributed code is in a sense, holistically accessed? I feel that this is a very important divide.
To my understanding this is how many theories of consciousness implicitly address the problem of holistic access.
Jonathan,
There is a clear analogy between your theory and mine.
'But it is unclear that there is anything in you rmodel that has access to signals as input at all, let alone in a bound or holistic fashion and more than 10,000 options. '
My mechanism is not critically different from how you gather your information onto a single dendritic tree and let some quantum 'irreducible law of nature' do the magic, the transformation from 'a yet distributed code within a dendrite' to 'subjective experience'.
In my case, spatiotemporal chaotic fluctuation which is spread-out all over the network, but still a single entity due to its connective nature, gathers information right on the spot in the sense that it is causally influenced by all small local changes in the distributed code. This effect on the chaotic fluctuation is magically transformed into subjective experience with my version of the 'irreducible law of nature'.
The analogy goes,
yours mine
where a single dendritic tree the neural network itself
what reads a quantum process network-spread chaotic fluctuation (or magic!)
The difference, on the other hand, is that my 'holistic reader' has access to all neurons in the brain, although like I said earlier, I assume a sufficient condition that filters what enters conscious vision. And another is that I have a harsh test that, if the result turns out to be positive, it non-ambiguously proves that at least 'fluctuation is the key to subjective vision'. (video:51:30)
Jonathan: "I do not think that I ever see colours without them belonging to things when I open my eyes."
You deceive yourself. Notice that this is different than knowing that colors belong to things. There is a natural sequence of brain events underlying the seeing of particular things in a novel environment.
1. In order to see/perceive a particular thing in the world in front of you immediately after you open your eyes, you must first have a global egocentric brain representation of some colors and shapes in the current visual scene. This is an integral part of your current conscious content. It is a precondition for the perception of particular things in the scene.
2. Next, you must parse a particular object/thing out of the global conscious representation.
3. Then you can say what color belongs to the thing.
Masataka: "Are you claiming that due to recurrent activity, the distributed code is in a sense, holistically accessed?"
Not at all. I make no appeal to a distributed code. The global spatiotemporal pattern of autaptic-cell activity in the short-term memory of retinoid space is a holistic event that can be accessed for subsequent cognitive processing (e.g., learning, perception, etc.) by non-conscious mechanisms (e.g., synaptic matrices). The recurrent parallel exchanges between preconscious mechanisms and the patterns of activity in retinoid space contribute to the phenomenal content of consciousness. For example, see "Analysis and Representation of Object Relations" and "Self-Directed Learning in a Complex Environment", on my RG page.
Arnold
I am a bit confused now. If you have answered yes to my question, it would be equivalent to my backup 2nd favorite hypothesis. I was already starting to think, what exactly is the critical difference, but I can talk about that later.
Can you clarify what you mean by,
'The recurrent parallel exchanges between preconscious mechanisms and the patterns of activity in retinoid space contribute to the phenomenal content of consciousness.'.
Maybe we have different vocabularies. Because to me, the activity pattern of your retinoid system is clearly a distributed code. What I mean by a distributed code is simply 'on-off or quantitative spatial activity patterns of a neural network'
Thanks!
Masataka "What I mean by a distributed code is simply 'on-off or quantitative spatial activity patterns of a neural network'"
If this is what you mean, then I agree that retinoid space is a distributed code.
Jonathan,
I can rephrase the problem using your concept of
These are indeed only 'signals', functionally completed, if they arrive somewhere'.
I agree, but arriving somewhere is only important if it has further dissociative consequences on neural dynamics.
In case of the neural network, there is a meaning to 'arriving to a certain neuron', because this neuron has causal and distinct influences to the network via axonal output. It works because, as a neuronal population, a vector goes in and a vector comes out.
In the case of a single dendritic tree, it is basically a dead-end. There is only one axonal output that can influence the rest of the network. So a vector goes in, but only a single scalar comes out. I cannot see how feedback from such rich information like our subjective vision can be reduced down to a single scalar value. Therefore position within a single dendritic tree does not have sufficient dissociative effects, and the 'arriving somewhere' has very little meaning in your case, if any.
BTW, I guess we are basically looking at the problem of 'tagging meanings to spikes', from the other side of the coin.
Jonathan,
So let's say you micro-stimulate a neuron. Like in Shadlen, Newsome's now classic experiment, you can bias the percept in dissociative ways depending on which neuron you stimulate. This is because the downstream effects are distinct for each and every neuron.
What if you micro-stimulate specific positions within a dendritic tree. Would the downstream effects be different for each dendritic position? No.
You may see differences in the magnitude of axonal spike output, but its degenerate in the mathematical sense.
This would be my current main argument against your theory, so I would appreciate if you can focus on this and the one above for the time being. Likewise, if you feel I am not answering any of your important questions, please tell me so.
Arnold,
Good!
Can you still explain what you mean by,
'The recurrent parallel exchanges between preconscious mechanisms and the patterns of activity in retinoid space contribute to the phenomenal content of consciousness.'.
Thanks!
Masataka,
First, let me say that I am reluctant to describe the pattern of activity in retinoid space as a distributed code because, in my theoretical model of the cognitive brain, global autaptic-cell activity immediately constitutes one's conscious experience of the world. No decoding of this activity is required for conscious experience. If we needed to decode our retinoid representation of the world in order to have a conscious experience, we would be caught in an infinite regress.
When I refer to recurrent parallel exchanges between preconscious mechanisms and the patterns of activity in retinoid space as contributing to the phenomenal content of consciousness, I am talking about the ongoing enrichment of conscious experience (retinoid space) by axonal projections from the imaging matrices of our various preconscious sensory modalities, and the parallel feedback from attention-parsed retinoid patterns into the contributing preconscious synaptic matrices. If you read the selected chapters from The Cognitive Brain that I suggested above, I think you will get a better idea of how this works.
This is from another thread devoted to discussion of the retinoid model of consciousness:
.................................................................................................
My bridging principle (see “A Foundation for the Scientific Study of Consciousness” on my RG Page): For any instance of conscious content, there is a corresponding analog in the biophysical state of the brain.
According to the bridging principle, two landmark experiments can be assessed quite differently as they relate to the problem of consciousness: (a) Sheinberg and Logothetis (1997), in single-cell recordings, showed that during binocular rivalry the behavioral response of a monkey could be predicted with a high degree of accuracy from the difference in firing rate of individual cells in specific brain regions; (b) Larsen et al (2006) , employing fMRI imaging, showed that when subjects experienced illusory motion between two locations of sequentially flashed lights, there was a path of brain activation in the visual system corresponding to the path of illusory motion. Notice that according to the bridging principle, the Sheinberg and Logothetis results (only correlates of phenomenal content) would not be selectively indicative of conscious content, whereas the Larsen et al results (analogs of phenomenal content) would be selectively indicative of conscious content because they bear a similarity relationship to the illusory event that is consciously experienced. So the scientific problem is to formulate a competent brain mechanism that can generate the kind of analog activation demonstrated by Larsen et al. My candidate is the retinoid system. What are the counter arguments?
.......................................................................................................
Arnold,
The above seems very interesting, but I don't think I get it. Can you define 'selectively indicative of conscious content', 'correlates of phenomenal content' and 'analogs of phenomenal content'?
Dear Arnold,
You say:
"1. In order to see/perceive a particular thing in the world in front of you immediately after you open your eyes, you must first have a global egocentric brain representation of some colors and shapes in the current visual scene. This is an integral part of your current conscious content. It is a precondition for the perception of particular things in the scene."
I don't think that's right. I am not aware of seeing disembodied colours before I see them as colours of things. Moreover, the sort of data Purves and Lotto present is strongly suggestive that it does not happen. It seems that we only 'see' colours consciously at the point where we have inferred what they belong to. At this stage our early sensory pathways have already factored in relative intensities of 'red', 'green' and 'blue' sensitive cone and rod stimulation and computed the expected distribution of shadows for objects and suchlike - using the sort of predictive coding Masataka has referred to. The fact that with some of these illusions you cannot 'see' the raw colour on a print that looks yellow in one picture and blue in another however hard you try, seems to indicate that this raw data is simply not available to conscious perception. To be honest I have never heard mention of what you are suggesting by anyone else - is there empirical evidence for this? I don't think there is any theoretical necessity.
Dear Masataka,
I am finding this discussion most enjoyable, since it is perhaps the first time I have met an empirical researcher in the field who is actually interested in discussing these critical issues in a constructive way. We still have a wide gap to bridge but at least we are trying to get there. I have not actually come across any common sense objections to my model from neuroscientists - really just misconceived objections about grandmother cells, mostly from the standpoint of not quite seeing the implications of what I am proposing. As far as I can see the main obstacle is an emotional reaction to the idea of having multiple mes and an obsession with thinking Descartes was wrong (for the wrong reason). Still I need to reply to your last set of posts. My internet access is a bit patchy but I will come back online in a while.
Masataka: "Can you define 'selectively indicative of conscious content', 'correlates of phenomenal content' and 'analogs of phenomenal content'?"
I can give examples.
1. In the binocular rivalry case, a particular neuron (N1) fires at a relatively higher rate when one rival figure is seen, and a different neuron (N2) fires at a relatively higher rate when the other rival figure is seen. Because neither N1 nor N2 is an analog of its correlated figure, neither is selectively indicative of conscious/phenomenal content (according to the bridging principle of corresponding analogs).
2. In the illusory motion case, fMRI displays a continuous path of BOLD activity between the location of stimulus-1 (S1) and the different location of the later stimulus-2 (S2), even though S1 does not move to the location of S2. The fMRI image is an analog of the illusory conscious experience of S1 moving and tracing a continuous path to the location of S2, even though S1 does not move. So the fMRI image, in this case, is selectively indicative of conscious/phenomenal content (according to the bridging principle of corresponding analogs).
Jonathan: " It seems that we only 'see' colours consciously at the point where we have inferred what they belong to."
Absolutely not. The individual colors in this heliokinetic collage are consciously experienced globally before we identify the shapes that they belong to. Click on the image to enlarge.