Talk:Sensorimotor theory of consciousness
- //author’s response to review preceded by //
This is a well-written and clearly-structured article providing a comprehensive but concise summary of the approach.
I have the following comments/suggestions for improvement:
The terms sensorimotor dependency, sensorimotor regularity and sensorimotor law seem to be used interchangeably. Is this correct, or is there be a stronger requirement for a law?
- //The word ‘law’ could indeed suggest stronger requirements, while sensorimotor dependency/regularity is a statistical notion that allows for some level of noise. We therefore now restrict the use of ‘law’ to strong cases, as in ‘the laws of occlusion.’
How does the theory deal with the use of similar descriptive terms across modalities, e.g. softness of a sponge, soft lighting, a soft sound. Is this a meta-regularity? Such an interpretation would seem to be consistent with the discussion of synesthesia.
- //We like the suggestion! For example in case of a soft sound the sensory consequences of moving the head may be less strong than in case of a loud sound, a bit like squeezing a soft object implies less change of pressure on the senses than squeezing a hard object. However, as such similarities have been relatively little explored in the present literature on sensorimotor theory, we decide to leave it for perhaps a later review. We would certainly be interested in further explorations of the cross-modal use of descriptive terms in different languages in relation to sensorimotor dependencies.
In the section on conscious experience, the claim that access has a phenomenal character does not rule out that there can be aspects of consciousness that are not directly to do with access. The discussion of cognition and language also seems to allow this. This discussion reminded me of theories of symbol grounding (e.g. Harnad), that is that language/cognition consciousness has to grounded in sensorimotor interactions. Would it be useful to make this link here?
- //We think that it makes little sense, in the everyday use of the notion of consciousness, to conceive of forms of consciousness involving no access. What would it mean to be conscious of something without having access to it? We’ve reformulated our view on access and phenomenal character. We further sharpened our presentation of the sensorimotor account by adding remarks about implications of the sensorimotor approach for artificial intelligence research. As concerns symbol grounding, we feel it would go too far to include an explicit discussion of this in the explanation of sensorimotor theory because that would compel us to get into the assumptions of the symbol grounding idea, taking us too far from our main topic. But here are some notes on the issue:
- //Given that sensorimotor theory appeals to cognitive capacities as well as to sensorimotor engagement with the environment, one might be reminded of the idea of symbol grounding [Harnad 2007 www.scholarpedia.org/article/Symbol_grounding_problem]. Symbol grounding is typically understood as a question about elements in a symbol system, which is “a set of symbols and syntactic rules for manipulating them on the basis of their shapes” (Harnad 2007). The computational-representational view of mind and cognition asserts that such ‘symbols’ are implemented in the brain, raising questions about how these ‘symbols’ acquire meaning and grounding in the world and sensorimotor engagement of the agent.
- //Sensorimotor theory is wary of this computational-representational view of mind and cognition, it does not think that it makes sense to say that there are symbols or ‘words’ in any sense “in” the brain, and it therefore does not accept this symbol grounding challenge in the form that it is usually stated. Still, whether or not there are symbols in the brain (understood as elements of symbol systems), questions may be asked about the meaning of our spoken or written words. Clearly, nothing makes sense except in the stream of life: it is only in the context of our active engagements with the world and each other that we can speak of meaning. But sensorimotor theory specifically offers an account of perceptual consciousness and it does not offer a general theory of meaning (i.e. the meaning of words). Sensorimotor theory defines perceptual consciousness as a subset of our sensorimotor engagement characterized by cognitive engagement with the environment: engagement involving the potential for (verbal) report, further thought, and/or deliberate planning of action.
- //Thus the claim is not so much that the meaning of the word ‘perceptual consciousness’ is grounded in sensorimotor engagement, but rather that a subset of sensorimotor engagements is the referent of the word ‘perceptual consciousness’. (Perhaps one way to view this may be as making links within a symbol system?)
Would it be appropriate to provide some reference to the philosophy of Merleau-Ponty as an important antecedent? (For instance, this could be mentioned in the section on enactivism).
- //We’ve added Husserl’s ‘thing and space’ as well, in the subsection on the relation with enactivism.
Does the sensorimotor research program have a synthetic component—e.g. to construct robots that perceive according to the principles of the sensorimotor account? If a robot were able to classify colors (for instance) according to sensorimotor contingencies would the robot necessarily have a phenomenological experience of color? How might building such a robot advance the paradigm?
- //We added a new subsection addressing these issues in the ‘sensorimotor research program’ section.
The discussion of Bayesian models could include a mention of the broader predictive brain hypothesis which has been extended to include action (e.g. Clark, BBS, 2013).
- //Reference added.
A minor stylistic comment. In the opening paragraph: “(what explains that some things are consciously experienced while others are not). […] (what explains that some things are consciously experienced while others are not)” the “what explains that…” here reads a bit awkwardly, you perhaps write “(how we explain why …”
- //Changed by the more precise phrasing mentioning the examples of the larger questions: (e.g. why some things are consciously experienced while others are not’); in the main text other examples are explicated (e.g. that some systems are conscious while others are not).
- //author’s response to review preceded by //
Here my comments!
Overall this is an excellent summary of the sensorimotor approach, as expected given the author is the founder and primary exponent of these influential ideas. In addition to the specific comments below I have a general suggestion that the authors include a section explicitly addressing the limitations of the SMC approach, as the authors see them. These might include, for instance, a bit more on the challenging cases of synaesthesia and dreaming – which are often raised as counterexamples to the theory. (I’m not saying these counterexamples are valid – just that the authors might do well to pre-empt critiques on their basis). Another limitation which is perhaps more self-evident is how SMCT connects with neural evidence generally. While SMCT nicely challenges strongly naïve representational (sense-think-act) views, there is not much offered on the positive side that predicts or interprets neural data. In short there is a rather large open question about what SMCT can say about the brain! (And what the brain can say about SMCT). I think explicitly acknowledging the scope, limitations, and open questions will overall make the theory more compelling to readers who have not previously encountered it.
Another general comment: what does SMCT say about affective and visceral experiences like hunger, thirst, sadness, disgust, pain etc.? Are these dependent on ‘intrasomatic’ SMCs?
- //We fully agree that limitations of the present sensorimotor account deserve explicit attention, especially concerning the relation to neural processes, and we’ve added a section on scope and challenges, addressing the brain as well as affective experiences. (There we’ve further added perceptual development to the list of issues that need to be addressed in more depth in relation to a sensorimotor approach.)
- //Concerning dreaming (and synaesthesia), we made some minor clarifications, we think that together with our present discussion of neural processes this should sufficiently tackle the issue.
In particular, regarding perceptual consciousness, a distinction can be made between two groups of explanatory gaps, the intermodal gaps and the intramodal gaps (Hurley & Noë 2003).
This seems to be non-standard use of the term ‘explanatory gap’ when compared to the sense introduced by Levine and as used as the start of this article. I’d suggest rephrasing this as just the problem of explaining how and why the character of perceptual (or sensorimotor) experience varies within and between modalities.
- //In the new version we’ve explicated how the use of the term is appropriate (and sufficiently standard): the challenge is to avoid a gap between physical/biological descriptions of differences between the modalities and phenomenal descriptions of the differences between the modalities.
Sensory experiences differ from non-sensory experiences like thinking or unfelt visceral goings-on in these respects: the “sensory presence” of a world is dependent on the bodiliness, insubordinateness, and grabbiness – objective properties of sensorimotor engagement which are lacking or less pronounced in non-sensory experiences and unfelt visceral processes.
I like the concepts of bodiliness, insubordinateness, and grabbiness (though to me they call out for neural-mechanistic explanatations). But I’m wondering how they apply to emotions and ‘felt’ visceral experiences, which are typically thought to be grounded in perception of bodily responses. What about pain, hunger, fear, etc? Maybe these issues will be treated later on … ah, they aren’t. Perhaps add a short paragraph here? (See general comment above)
- //The main contrast here is between sensory experiences and cases where sensory presence is lacking. Emotions and other experiences such as pain and huger are set aside in an added section (Section 3) as something in need of further exploration.
The notion of implicit grasp needs careful unpacking. Some interpretations of the notion of implicit grasp or ‘implicit knowledge’ of sensorimotor dependencies suggest or even imply commitment to representationalism, as criticized by Hutto (2005). It should be clear that such an interpretation is not intended (see also Noë 2001; Myin & O’Regan 2002; Myin & Degenaar 2014). When one literally grasps on object with one’s hand, the hand gets a grip on the object without representing it (in the sense of forming a model of it). Similarly, it is claimed, when one ‘grasps’ the obtaining sensorimotor dependencies, one gets a grip on the environment without having to represent it. Thus, having grasp of sensorimotor dependencies is no more (and no less) than being attuned to these sensorimotor regularities. When one is attuned to the environment, this means that one can get under the influence of the obtaining sensorimotor dependencies, which in turn implies that one potentially can act accordingly, e.g. by picking out red objects or soft objects.
This is all put very nicely. Some people however construe sensorimotor theory as a sort of ‘higher order’ theory of consciousness, in virtue of the importance of ‘mastery’ of SMCs. Do you condone this interpretation, perhaps in a non-representational sense, wherein the embodied agent in virtue of being ‘attuned’ to SMCs therefore instantiates a non-representational higher-order relation to the SMCs, and that this is what underpins conscious experiences? (Also, I wonder if you could say anything more about what it means to ‘grasp’ SMCs. I mainly understand this in terms of what is not meant – i.e. a representational relationship).
- //(SMCs can still be there when the brain is removed, with some convulsions, but in that case the perceiver will no longer display behavior that is under the influence of the obtaining SMCs – in other words: there’s no more attunement to/grasp of SMCs; we see no reason to appeal to representation or to higher-order states to account for this difference.)
Sensorimotor theory insists that being conscious of a stimulus implies that the stimulus can have an impact on further action (including speech behavior) and on thought and rational reflection (O’Regan & Noë 2001).
This seems completely compatible with ‘global workspace theory’ type explanations. Would you agree?
- //Absolutely. (Although we are of course no big fan of its formulation in terms of the ‘broadcasting of content’ to subsystems of the brain.) In the added section 3 we briefly address workspace theory.
According to the sensorimotor theory, the absolute gap is a matter of what is sometimes called ‘access consciousness’ or ‘awareness’, in the sense that it concerns functional capacities for accessing the environment (or the way we are interacting with it) – sometimes considered to be an ‘easy problem’ (Chalmers 1996). But rather than allowing for a dissociation of (‘easy’) access consciousness and (‘hard’) phenomenal consciousness (Block 1995; Chalmers 1996), as if there could be ‘zombies’ having full-blown access consciousness but lacking experiences with a phenomenal character, the sensorimotor theory offers an account of the phenomenal character of the process of accessing the environment.
I am not sure it is right to equate easy/access and hard/phenomenal. Certainly, access consciousness is easier to study than phenomenal consciousness, because the contents of the former are defined operationally by reportability. But the hard problem (sensu Chalmers) applies to both – for any sort of functional or neural process there lies the ‘hard’ problem of why it should be associated with conscious experience at all. I don’t think the appeal to ‘zombies’ helps because access consciousness is still consciousness! It seems to me you are arguing that the character of phenomenal consciousness is determined by the kind of access an agent has, thereby undermining the distinction between the two. Is this right? If so this is a nice point, perhaps worth amplifying.
- //‘Phenomenal feel’ is what allegedly makes the hard problem hard, and we’ve simplified our presentation to avoid the response that ‘access consciousness’ already is consciousness and thus a hard case. We now speak more neutrally of functional capacities of accessing the environment. Whether or not questions about phenomenal character can be raised depends on functional matters of access. The questions are not the same, so there is a useful conceptual distinction, but the distinction does not refer to independent phenomena in the world. In other words, there are criteria for distinguishing conscious from non-conscious (making use in further thought, rational action etc.), and for the case of conscious experience we give an account of its phenomenal character, which we do without postulating additional ingredients in the world.
To some extent, explicit thought has its own phenomenology, so that thinking changes the overall experience of the perceiver
It might be worth citing here this recent collection on cognitive phenomenology http://ukcatalogue.oup.com/product/9780199579938.do. On the relationship between language, thought, and perceptual experience I also wonder what SMCT would have to say about the finding that Russian speakers, who have more words for different shades of blue, are better at perceptual decisions among blue stimuli (http://www.nature.com/news/2007/070430/full/news070430-2.html). Does this challenge your claim that “thought does not fundamentally distort the perceptual phenomenology as such”?
- //Our aim here is more modest – more trivial perhaps – and we’ve reformulated accordingly. It may go too far to link to discussions concerning a possibly distinctively cognitive phenomenology. (For example, the experience of thinking ‘in English’ we refer to here could simply be cashed out in terms of imagining spoken language.) Concerning the color naming finding, we acknowledge the possibility of subtle interactions (perhaps working through the development of perceptual capacities); the point here is merely to emphasize that the difference between thinking “red” and thinking “rouge” is not what we should focus on when accounting for the perceptual quality of experience. And if there is a perceptual difference between the Russian perceiver and the English speaker, this should be reflected at the level of their different engagement with sensorimotor patterns, not just at the level of the different thoughts. Again, it may help that we’ve added a section explicating the restricted scope of today’s sensorimotor theory!
The point is not to deny that there are perceptual completion phenomena involving spreading neural activity – these phenomena exist (Pessoa, Thompson & Noë 1998). But in the sensorimotor theory such activity is not interpreted as part of a process of constructing a model of the world. Both in sensorimotor accounts, as well as in internal model based accounts, neural activity is required for engagement with the environment; the sensorimotor account is more straightforward or simpler because it does not require that the neural activity has the additional function of building up a model of the environment inside the head.
You very nicely make the point that one should not assume a ‘filling in’ role for neural activity with respect to detailed internal models. But can you say anything here about what, on SMCT, phenomena like ‘spreading activation’ actually do? Is is part of the implementation of an SMC? Or something else? It is a bit unconvincing to say that the sensorimotor account is ‘simpler’ because it doesn’t – as far as I can see – offer an account of the neural activity at all. See the general comment about the role of the brain.
- //When you’re gradually starting to see things as you adapt to the dark, you start to engage with more and more aspects of the environment (thus your behavior can get influenced by more and more aspects of the environment). This requires increasing activity of your perceptual systems. Spreading activation in perceptual completion does just that: you start to be more and more as if you’re engaged with a physically complete stimulus. This is something that both internal model-based and non-representational accounts agree on; on our view an advantage of a non-representational account is that it does not need to find a convincing implementation of a non-trivial internal model.
This is what happens in the phenomenon of ‘change blindness’.
Your work on change blindess is of course absolutely fundamental. It might be nice here to have an animation demo of e.g. a mudsplash image? I am sure the assistant editor could help sort this out if you think it would add to the article (I do!)
- //Yes, we’ll do that!
The plasticity of experience: sensory substitution and visual inversions In this section it might be worth briefly discussing Peter Konig’s work on the FeelSpace belt, which seems to support the main principles of SMCT. Here the interesting thing is that it is an entirely new perceptual modality that is being developed, rather than substitution or inversion. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4154453/
- //Great study, we refer to it but leave discussion to a minimum to keep the text reasonably short.
The sensorimotor theory, building on the ideas of Poincaré and Jean Nicod, suggests that the only way the notion of space could emerge would be by discovering invariants in the sensorimotor laws that govern the relations between an agent’s motor commands and its sensory inputs. The experience of space, from this point of view, efficiently captures these sensorimotor regularities (Philipona, O’Regan & Nadal 2003; Terekhov & O’Regan forthcoming).
On one hand this seems very compatible with (representational) predictive processing views, which are all about extracting and encoding sensorimotor invariants through active inference. The idea that these invariants somehow underpin the experience of space itself is rather interesting.
… we should not interpret the difference between perception and imagery as consisting in a difference in the veridicality of alleged internal representations, but rather as a difference between real and apparent engagement.
Could you add a sentence on how SMCT deals with vivid dreaming? Dream experiences are typically considered as equally veridical to normal waking experiences (though they differ on many other dimensions). How does this relate to ‘real vs apparent engagement’?
- //Reformulated and sentence added on vivid dreams etc.
Higher order theories In this section I got a bit confused. Firstly, I was not convinced by the description of HOT theories. E.g.
Higher order theories would claim that making use of experience in deliberate planning necessitates a higher order state accessing the experience, if the planning is to count as evidence for conscious experience. The sensorimotor theory is not committed to such a model of the difference between conscious and not conscious: we are typically conscious of the environment rather than of our experiences.
HOT says that we are conscious of the environment in virtue of having a higher-order ‘representation’ directed at the first order state – it does not say we are conscious ‘of our experiences’ in virtue of the higher order state. This would require a third-order state – possibly underpinning metacognition.
- //Description changed accordingly.
It still seems to me that SMCT is a kind of non-representational HOT, where SMCs become conscious in virtue of a non-representational higher-order relation of ‘mastery’ or ‘attunement’ or something. Is this right?
- //Being tuned to SMC’s (i.e. having mastery of them) does not make them conscious. Whether or not behavior can be dependent on a sensorimotor invariant (that is, mastery versus no mastery) has nothing to do with the presence or absence of higher-order states, representational or otherwise. What makes a person conscious is the fact that (mastered) SMC’s are being made use of in further thought, rational behaviour, decisions, planning, etc.
One way to further clarify this would be to detail the relation between SMCT and global workspace theory (which is similar but not the same as multiple drafts – the appeal to ‘fame in the brain’ is about as close as it gets). Many times you say things like
It stipulates that for us to be conscious of something in the environment (or of our way of interacting with the environment), the thing must play a role in subsequent thought or behaviour
… which is entirely consonant with global workspace theory. In other words, I would like perhaps to see a short extra section about global workspace theory and a clearer account of how SMCT relates to and differs from both.
- //We like the suggestion; see the new section 3 on the brain.
Helmholtz and Bayesian models of perception All this is nicely said. However to my (biased) view, Bayesian approaches when grounded in active inference provide a very close mapping to SMCT in all but the fundamental theoretical commitments of representational in-the-headedness vs non-representational. I do not expect you to endorse this view of course, but it might be useful for readers to cite a recent attempt to integrate SMCT and Bayesian approaches – and in doing so provide some response to the important question of ‘what is going on in the brain during the exercise of mastery of SMCs”. http://www.ncbi.nlm.nih.gov/pubmed/24446823
- //We’ve modified the paragraph and included reference to the paper. In fact, we are not against the notion of internal models as such – we think a classical cognitivist/representationalist robotic system might be perceptually conscious, but if so not because there would be contentfull states inside the system, but because of its capacities for engagement with the environment.
- //We like to thank the reviewers, as well as Lucia Foglia, David Silverman and Christoph Witzl, for their very helpful comments!