How to Make a Robot that Feels

How to Make a Robot that Feels

This article is divided in two parts:

1. A summary of Kevin O’Regan keynote talk at CogSys 2010 by Raúl Arrabales.
2. An invited extended discussion about the self and the role of action in sensation by Kevin O’Regan.

How to Make a Robot that Feels

“How to make a robot that feels” was the title of the keynote talk given by Kevin O’Regan at the CogSys 2010 conference last week. During this talk O’Regan introduced the so-called hard problem of consciousness (as coined by Chalmers) and explained his sensorimotor approach to (phenomenal) consciousness [1]. This talk and related ideas are of special interest for Machine Consciousness researchers since O’Regan offers an account for sensory feel which virtually eliminates the hard problem, and therefore the explanatory gap itself. In the following I will try to summarize the key ideas that I got from both the talk and further discussions we had with O’Regan during the CogSys conference.

Using the redness of red, quite typical example in philosophy of mind, O’Regan addressed the problem of designing a robot that feels. Note that in this context the word feel is not used as in Damasio’s work, but to refer to the what-is-it-like or qualia associated to conscious contents, i.e. sensory feel.

In order to explain the redness of red and to find out how it could be generated in a robot, there are some key associated properties that we must understand: (1) associated cognitive states, (2) associated behavior, and (3) the raw feel or subjective – first person – experience. This last property is the one which is often considered something special and has led to dualistic views in the past. O’Regan remarked that the apparent impossibility to describe raw feel, as suggested by Chalmers, should not make us fall in the fallacy of the hard problem. Although other philosophers like Dennett have argued that the hard problem doesn’t really exist, they have failed to provide a convincing explanation of qualia. In fact, neurologists can’t explain experience either. What neurologists can explain is that a neuronal red-green channel is activated in the red direction when you see the red color. However, this doesn’t explain experience (the red sensation). In other words, we don’t have an obvious way to bridge the explanatory gap just by looking at neurobiological results. The presence or “feel like something” seems to be impossible to explain by physiology alone.

O’Regan argued that we can progress in understanding raw feel by considering exactly what are the main things that seem difficult to explain about it. He suggests that the main mysteries of feel are: (1) their ineffability (impossibility to completely and accurately communicate about them), (2) the fact that qualia have structure, and (3) the fact that they have presence (or “feel like something”). These facts about raw feel seem very hard to understand in terms of physico-chemical processes in the brain. On the other hand, according to O’Regan, these mysteries can be unveiled if instead of thinking of the brain as the generator of feel, feel is considered as a way of interacting with the world. Under this “sensorimotor” view, experienced qualities are equivalent to sensorimotor laws. “Redness” is not something generated by the brain, but the way a red thing changes the light. We have the redness feel when we attend to the fact that we are interacting in a skilful way with an object that changes the light following a determined pattern.

The three main mysteries of feel can then be explained in the following way:

   Ineffability: qualia are ineffable because we don’t have cognitive access to all the details of our own sensorimotor skills. For instance, we don’t have access to the lengths of particular muscle fibers.

   Structure of qualia: The similarities and differences between feels within and between different sensory modalities can be accounted for in terms of the similarities and differences between the objective laws that determine the sensorimotor interactions involved. For example, the quality associated with seeing is constituted by all the laws that are potentially obeyed when you are involved in exercising the skill involved in seeing. As a confirmation of the theory, sensory substitution [2] shows that vibration on the skin can be felt like vision, provided vision-like laws are obeyed when the observer moves.

   Presence: Accounting for presence or “what it’s like” requires explaining why sensory experiences have something very special that other mental activities lack. To explain this difference under the sensorimotor approach it suffices to look for objective differences in the sensorimotor laws governing perceptual acts, compared to those governing mental activities like thinking or remembering. It then appears that what particularly characterizes sensory feels, and distinguishes them from mental experiences, is that sensory interactions have bodiliness and grabbiness (see O’Regan’s further points below and [1] and [4] for details).

Unfortunately, there was not enough time during the keynote talk to discuss in detail about the former points, and specifically about the emergence of the self during cognitive development. Conscious-Robots.com hereby invites Dr. O’Regan to include in this summary additional considerations that could not be covered in the talk due to time constraints (see below).

O’Regan’s further points about the Sensorimotor Approach

Bodiliness is the fact that when you move your body, sensory input changes. Mental activities possess no bodiliness. Grabbiness is the fact that sensory systems are wired up so as to interrupt cognitive processing when there is a sudden change in sensory input. Mental activities also possess no grabbiness. Bodiliness and Grabbiness are objective facts about sensory activities that seem intuitively to capture the fact that we are subject to our sensory experiences — they impose themselves upon us. This would correspond well to the notion of “presence” or “what it’s like”. The fact that mental activities, and activities associated with autonomic processes in the nervous system, do not possess bodiliness and grabbiness thus provides an explanation for why only sensory states possess sensory “presence” or a “what it’s like”.

In summary, since the sensorimotor approach accounts for the ineffability, the structure and the presence of feel, it provides a way of bridging the explanatory gap as concerns the quality of experience.

There remains a point concerning what then constitutes phenomenal consciousness. To answer this question, the approach suggests that being conscious of a sensory feel amounts to, first, an agent having the notion of self, and second, the self cognitively accessing the fact that it is currently engaged in exercising the sensorimotor dependencies corresponding to the feel.

Since cognitive access and the notion of self are not part of the “hard” problem of consciousness, but can be accounted for in terms of computational processes available to AI and robotics, it follows that if robots have sufficient cognitive capacities to have a self and to access their own interactions, they should be able to have phenomenal consciousness.

NB: A written version of the talk given by K. O’Regan at CogSys 2010 will (shortly) be available on the web site http://nivea.psycho.univ-paris5.fr

An article which conveniently summarizes the approach is

O’Regan, J.K. (2010). Explaining what people say about sensory qualia. In N. Gangopadhyay, M. Madary, F. Spicer (Eds) Perception, Action, and Consciousness: Sensorimotor Dynamics and Two Visual Systems. Oxford: OUP. (http://lpp.psycho.univ-paris5.fr/pdf/2630.pdf)

Action and Cognition

Raúl Arrabales:

In light of recent evidence of willful modulation of brain activity in patients with disorders of consciousness [3], there is an ongoing discussion about the requirement of motor action for cognition and consciousness. According to the results presented in [3] it seems that patients can be conscious even in the absence of any motor activity. How can this be explained in terms of the sensorimotor account?

Invited comment by Kevin O’Regan:

It is important to note that the sensorimotor approach does not claim that for sensory experience to occur, action is necessary at a given moment. What the theory says is that for sensory experience to occur, an observer must be in a state where they are aware that if they were to move in certain ways, certain sensory changes would occur. But they do not actually have to move!

As an analogy, take the example of my feeling that I’m at home sitting in my living room. I have the feel even though I’m not moving at all. The feel comes from me implicitly knowing that if I were to stand up and go this way, I would come into the kitchen, and if I were to go that way I would come into the bedroom. But actually I need do nothing at all.

Thus the role of action in sensation is a potential role not an actual role. To perceive now the perceiver must b in a state where they assume (although perhaps incorrectly — as in the case of the vegetative state patients, or, in dreams, or in tachistoscopic presentations!) that similar sensorimotor dependencies apply now as when some time previously they were in the same sensory state.

For these reasons, the finding in [3] that there can be cognition (and perhaps sensation) without action is compatible with the sensorimotor approach.

First of all, as regards cognition, the sensorimotor approach says nothing about the need for action in this case. On the contrary, just as there is symbol processing in computers without motor action, the approach admits cognitive processing in humans without action.

As regards sensation, there is also no incompatibility. The patient must merely have cognitive access to the fact that if they were to move, then certain sensory changes would usually occur. The fact that they cannot actually move is unfortunate, but provided that previously in their life the patient has undergone movement-related sensory changes, and was currently aware (or convinced, even incorrectly!) that the current potential changes are the same ones, this will be sufficient for him or her to have the same sensory experience that they previously had.

Acknowledgements

I wish to thank Dr. O’Regan for his thoughtful comments, additional commentary, and the review of my original summary, which made it much clearer. I also wish to thank Ricardo Sanz and the rest of Retecog members for the references and discussion about the role of motor action in cognition.

References

[1] O’Regan, J.K. (2009). Sensorimotor approach to (phenomenal) consciousness. In Baynes, T., Cleeremans, A. & Wilken, P. (Eds) Oxford Companion to Consciousness. (pp. 588-593). Oxford: Oxford University Press. (http://lpp.psycho.univ-paris5.fr/pdf/2532.pdf)

[2] Bach-y-Rita P, Collins CC, Saunders F, White B, Scadden L.(1969). “Vision substitution by tactile image projection.”. Nature, 221:963–964 (http://www.nature.com/nature/journal/v221/n5184/abs/221963a0.html)

[3] Monti, M.M. et al. (2010) Willful Modulation of Brain Activity in Disorders of Consciousness. The New England Journal of Medicine. Feb. 3, 2010. (http://content.nejm.org/cgi/content/full/NEJMoa0905370)

[4] O’Regan, J.K. (2010). Explaining what people say about sensory qualia. In N. Gangopadhyay, M. Madary, F. Spicer (Eds) Perception, Action, and Consciousness: Sensorimotor Dynamics and Two Visual Systems. Oxford: OUP. (http://lpp.psycho.univ-paris5.fr/pdf/2630.pdf).