The editor kindly highlighted those sections of this reviewer's comments that he considered most important to address, therefore this response will focus specifically on those highlighted sections.
1: The author attempts to overthrow too many widely held beliefs as to how perception should be explained without arguing against those positions in detail (e.g. content-vehicle distinctions, traditional computational modeling.)
As discussed in the general response above, the message of this paper is not to overthrow those many beliefs individually, but to challenge the fundamental assumptions on which those beliefs are based. In particular I propose to challenge the belief that the properties of visual consciousness as experienced phenomenally are irrelevant to models of the neurophysiological mechanism that subserves that experience. The method I use is to quantify the dimensions of conscious experience in a quantitative model that is explicitly defined in subjective phenomenal variables, and thus it is a model that remains safely on the subjective side of the mind/brain barrier. This inoculates the model from the problems inherent in the "hard problem" of consciousness, because I do not address the issue of how the model is implemented neurophysiologically. This makes issues of content-vehicle distinctions, and explicit v.s. implicit representation irrelevant to this model, because those issues address the relation of the phenomenal to the physiological. The perceptual model by definition explicitly models the content of consciousness as a content, and a process of consciousness as a process. This issue is now elaborated in the new version of the paper.
Traditional computational modeling is generally expressed in neural network or connectionist terms, even models of subjective phenomena such as subjective contours and illusory figures, and therefore those models must address issues of content-vehicle distinction. But the perceptual model represents only the subjective component of the experience, and is therefore expressed in the same variables or dimensions as the subjective experience itself. This therefore represents a separate class of model with different objectives than the neural model, and is not subject to the same neurophysiological critiques. I do not claim that traditional neural network modeling is invalid as such, and therefore I do not argue against traditional computational modeling. I only take issue with neural network models whose informational content is less than that of the subjective phenomena that they supposedly explain, although that happens to include a great many of them.
These issues have now been elaborated and clarified in a whole new section in the new version of the paper, that clarifies the objectives and methods of perceptual modeling.
2: [the author] does not systematically or convincingly defeat the standard counterargument presented against homunculus- internal perception-, sense-data-, or Cartesian Theatre -theories of consciousness, although those arguments are the ones that will probably be immediately raised against a view such as the present one.
It is not sufficient for the reviewer to state that I do not convincingly defeat the homunculus objection, without explaining what it is in my argument that he finds unconvincing. For my argument is very clear and specific.
There is no need for an internal observer of the spatial model because the model is merely a data structure, like any other data encoded in the brain, with the only difference that this data is in explicit spatial form.
If a picture-in-the-head required a homunculus to view it, then the same objection would also apply to symbolic or verbal information in the brain, which would also require a homunculus to read or interpret that information.
In fact any information encoded in the brain needs only to be available to other internal processes rather than to a miniature copy of the brain as a whole.
If the reviewer finds this unsatisfactory, he must explain why. Does he propose that explicitly spatial information requires a homunculus to view it while symbolic information does not? If so, on what grounds? Or does he challenge the validity of any form of representation in the brain? If so, how does he account for the data represented in a computer? Even image data are represented in computers, with algorithms to process them, without any sort of homunculus involved.
The above argument is not my own, I have referenced Earle (1998) and Singh & Hoffman (1998) who made essentially the same argument in the open peer commentary in response to the Pessoa et al. paper, and I did not find a clear response from Pessoa et al. to this central point. A similar defense of internal representations has also been made by Pinker (1984, "Visual Cognition: An Introduction". Cognition 18, p. 38), Revonsuo (1994, "In Search Of the Science of Consciousness", in Revonsuo A. & Kamppinen (Eds) Consciousness in Philosophy and Cognitive Neuroscience, Hillsdale NJ: Erlbaum, p. 275), and Smythies (1994, The Walls of Plato's Cave. Aldershot: Avebury, p. 75), and I have yet to hear a credible rebuttal from anyone. The fact that this fallacious homunculus objection pops up so frequently in the debate highlights the importance of getting this issue settled once and for all.
3: Some of the evidence [the author] presents in support of his theory, e.g. the interpretation of neglect, is based on an insufficient review of the actual phenomenon (which is a much more complex disorder than the author seems to be aware of) and he does not mention the different cognitive theories that have been presented to explain it. It is implausible that all the different varieties and forms of neglect could be explained by referring to the disruption of a single spatial representation.
The phenomenon of neglect was introduced at the very end of the paper in the discussion section, i.e. this is not the central focus of the paper. I do not propose to account for all of the subtleties of the neglect syndrome with a single simple explanation. However there is an undercurrent in the debate on neglect as to whether the phenomenon is spatial at all, or merely "attentional", and the basis for this debate is ultimately neurophysiologically motivated, i.e. it is difficult to imagine how explicit volumetric images could possibly be encoded in the brain, especially images that can rotate and translate with respect to each other in a non-anchored manner. It would be very convenient for neural network theorists if the neglect syndrome could be wished away, which would conveniently dispose of its troublesome implications. Similar objections are often raised with regard to mental imagery.
But the troublesome issue of neglect is not that half of space is missing, but that it highlights the fact that there is a spatial representation at all in the brain. This fact is easily overlooked in normal perception where the percept of the world is easily confused for the world it represents, but it can no longer be ignored when half of that world disappears. Once we recognize the world of experience around us for what it really is, it becomes immediately obvious that the brain is capable of generating vivid three-dimensional spatial percepts that can rotate, translate, and scale by perspective as they move about in the perceived world. In other words the brain contains a three- dimensional volumetric imaging system capable of generating volumetric spatial structures which are dimensionally identical to the ones you see around you now. Once we accept this capability of the brain, a great host of otherwise deeply mysterious phenomena suddenly seem to make more sense, i.e. they no longer require heroic efforts of denial to account for their manifest properties. Those phenomena include hallucinations, dreams, mental images, neglect syndrome, the Kanizsa and Necker cube illusions, apparent motion phenomena, neon color spreading, etc. etc. These phenomena are now quantifiable in a perceptual model exactly as they are observed, and that model in turn sets a lower limit on the information that must be encoded in the corresponding neurophysiological state.
The criticism that my discussion of neglect is based on an "insufficient review of the actual phenomenon" suggests that the reviewer does not understand the paradigmatic nature of what is being proposed. I am not offering a specific computational model to account for all the properties of neglect syndrome, for that would require a whole paper devoted to that specialized topic. Instead, I am proposing that if spatial perception and mental imagery (in neglect syndrome or elsewhere) appear phenomenally as volumetric spatial structures, then that is how they are encoded explicitly in the brain.
One of the most troublesome aspects of the neglect syndrome is that it is hard to know even what it means to say that half of phenomenal space is missing. Again, confusion on this issue stems from the naive confusion of the perceived world for the real world that it represents. It seems paradoxical to say that half of space appears to be missing, when that space must in fact also be present. The quantitative phenomenology of the Gestalt Bubble model offers at least a concrete description of this otherwise paradoxical phenomenon, for by quantifying the dimensions of spatial experience in normal vision, this now offers an objective way to quantify the properties of neglect syndrome, whatever its neurophysiological basis might be. For neglect is experienced exactly AS IF half of phenomenal space is missing, and therefore that is how it is modeled perceptually. This is not a neurophysiological theory, but a more precise method of quantifying the phenomenal experience of neglect, as it is described by patients. This eliminates the paradoxical nature of the syndrome, which now becomes an ordinary deficit like a scotoma, except in a spatial representation. The additional subtleties of the syndrome can now be expressed as variations on the basic Bubble model in a paper devoted to that more specialized topic.
Paradigm debates do not come around often in science, and when they do, they require a more general handling than the debates over details that characterize "normal science" as discussed by Kuhn. In the discussion section I touch on a great variety of different phenomena which have been deeply problematic for models of visual representation, but which can be addressed much more readily using the explicit spatial representation of the Gestalt Bubble model. My intent is not to address them individually here, but merely to suggest that they are ideal candidates for the perceptual modeling approach, for they are difficult to even describe in more abstracted terms. This is made clear in the concluding paragraph of the discussion section which states:
"It is perhaps too early to say definitively whether the model presented here can be formulated to address all of the phenomena outlined above. What is becoming increasingly clear however is the inadequacy of the conventional feed-forward abstraction approach to account for these phenomena, and that therefore novel and unconventional approaches to the problem should be given serious consideration."
4: The author seems to believe that there is no neuroscience research that would go beyond the single cell approach and try to understand Gestalt phenomena, but in fact the studies on neural synchronization ... and theories of possible neural mechanisms underlying holistic Gestalt perception ... try to do exactly that.
Studies of neural synchronization do indeed attempt to address Gestalt phenomena, although they are still in a very early stage of development, so it is hard to say exactly what they account for just yet. In any case they do not yet account for the properties of visual consciousness addressed by the Gestalt Bubble model, i.e. the world of solid volumes, bounded by colored surfaces, embedded in a spatial void, and therefore there remains a large explanatory gap between phenomenology and physiology. The new version of the paper now includes reference to the synchronization literature and how it relates to the present argument.
5: [the author] also does not mention people whose ideas apparently come quite close to his own. (e.g. John Smythies' defense of representative realism in walls of plato's cave, 1994; O'brien & Opie's 1999 paper in BBS where they defend a novel view of computa- tion that takes consciousness into account etc.) That is a pity, for connecting his model with other people's work more explicitly would make the paper much more eligible for peer commentary.
Smythies (1989, 1994) has now been referenced in a whole new section on the epistemological implications of the theory, along with Harrison (1989), and Hoffman (1998) who present similar ideas, and also Kant (1781), Köhler (1971), and Russell (1927). The neural network modeling literature has now been referenced with the work of O'Brien & Opie (1999), Zucker (1998), Grossberg et al. (1975, 1988), and Lesher (1995), and the literature on pure perceptual modeling has been fleshed out with references to Nagel (1974), Chalmers (1996), Price (1932) and Clark (1993). The reviewer will find that the paper has been substantially revised to connect with the relevant literature.