Consciousness Reframed '97

A note on a possible physiology of subjectivity, and some comments on what a conscious machine might want to do for itself. *

by Stephen Jones**

This paper covers three areas: first, a consensus view of what consciousness involves, then a description of an emerging consensus about where in the brain consciousness arises and some speculation as to how subjectivity appears, and finally a look at what a conscious artificial intelligence might want to do for itself.

1. What is Consciousness?

Most people engaged in the study of consciousness are trying to find some sort of explanation for, or at least an understanding of, how subjectivity arises. How it is that this physical entity that we are can also know the world and feel it, and be something within the world?

To do this we need to understand perception and memory, language and knowing. We need to recognise and understand subjectivity and identity, our spatial and our temporal unity. We need to understand how experience and our reflection upon that experience produces ideas and our generative activity.

But before we do any of that we need to know how to determine whether an entity is conscious?

Robert Kirk, of Nottingham University, has developed what he calls a "basic package" of related capacities that an organism must have for it to be conscious (1). He suggests that in order to identify whether an entity is conscious: it must be able to collect information which is specifically intended for its own use, it must be able to initiate and control its activity on the basis of that information, it must be able to assess and interpret the information, and then decide what to do with that information.

These give us a basic set of criteria for consciousness, but there are several more things which I would like to add.

Firstly, our ability to respond to novelty in active and constructive ways is somehow over and above mere 'irritability' of the senses which may or may not then have conscious impact.

Secondly, independent action, the ability to decide to do something, is crucial. The plant which follows the sun across the daily sky is unable to refuse to do that. It can only do it, there is no capacity to act independently of its built in responses.

To sum up our criteria. A conscious entity would need to:

  • Act independently and initiate procedures.
  • Do things like gathering information for itself.
  • Do things voluntarily for others.
  • Receive and reflect on perceptual input.
  • Show generativity and the capacity to construct.
  • Report upon the contents of its Consciousness, and
  • Interact with others.

So why is subjectivity such a mystery? As yet there is no really adequate story about how our physiology can produce anything like the feels and perceptions that we know subjectively. The phenomenological stuff that we know does not seem to relate to the physical material world. It is produced in us by that world but it doesn't appear to have physical existence. But, I suggest, this is largely a cultural artefact.

The problem is this: we are, indubitably, physical beings, but we carry this historically received idea that the self is somehow independent of the physical body. The anatomical discoveries of the 16th century reduced the body to physical materials. As this would have led to the denial of the "soul", the philosopher Descartes decided that, as a devout Christian, he would have to find an argument which guaranteed the idea of the soul. Using his method of radical doubting, he found that although he might be deceived about the true existence of anything material, there was still something which was being deceived. This was his "self", his "thinking thing", and thus he created "dualism" (2) the idea that consciousness is something non-physical, an immaterial stuff.

If it is, then we have several problems. The first is just how does this immaterial stuff interact with the material world? Also, how does the material world of our senses, as mediated through the brain, actually make any impression on the mind stuff, so that we can have memories and perceptions and the other trappings of consciousness?

Nevertheless, whether consciousness is somehow independent of the physical world or not it still has to operate within the physiology of our body/brain.

I argue that it is not necessary to dualise the mind away from the brain. The physiological functioning of the brain includes consciousness, as part of its normal process. Consciousness is an inescapable result of an immensely complex but highly organised information processing and representation system.

2. Some physiology

Let's take a short excursion into current consensus on the Neural Correlates of Consciousness.

The physiology of the brain shows a number of interlinked parts. These are the bodily connections through the spinal cord and the brain stem, the reticular activating system at the top of the brain stem, the sensory modalities and their connections to the cortex through a central relay station and the cortex, to which all sensory data is sent and in which the interpretation, planning and control is done. The central relay station, known as the thalamus, is of most interest here. It, in linkage with the cortex, the basal ganglia, the hypothalamus, the hippocampus and other structures, forms the thalamo-cortical system (3).

The thalamus is like a hub in a wheel, the spokes of which are sensory nerves travelling from the body periphery and synapsing in the thalamus with bundles of nerves which relay input information up into the cortex and association areas for interpretive processing. For example, the optic tract runs from the retina, through the optic chiasm to the thalamus from where it is distributed into the visual cortex at the back of the brain. Auditory data from the inner ear is relayed to the auditory cortex in the temporal lobes. All the face and body's proprioceptive data is also routed through the thalamus on its way to the somato-sensory cortex. These are ascending pathways.

Similarly, there are vast nerve bundles descending from all cortical areas onto the thalamus. These descending pathways gate the sensory data being presented to the cortex. The cortex's control over what data it is being sent at any moment gives selective attention. Nerves from frontal cortex also go to the thalamus where they are integrated with sensory data to help control motor functions.

So we have an array of massively inter-connected neural circuits, organised horizontally around the thalamus and associated nuclei and vertically between the cortex and the thalamus. The horizontal connections give behavioural control and the vertical connections provide sensory control, especially in preventing us from being overwhelmed by sensory input.

In the cortex vast arrays of intra-cortical nerve connections provide the capacity to associate different sensory modalities and to interpret synchronous data from different senses allowing one to, e.g., recognise that the sounds you hear are coming from the person you see speaking to you, and that they are speaking meaningfully.

I want to concentrate on the purpose of the descending pathways. They provide a control system. The data the cortex is receiving is controlled by it. We have here a kind of feedback network running the thalamo-cortical system as a self-regulating process. This capacity for self-regulation forms the basis for all levels of living systems, from cellular metabolism to the behaviour of individuals in society.

Now, nerve transmission takes time. This is propagation delay. Combined with feedback, propagation delay causes a circuit to become a reverberatory circuit. A delayed proportion of the original signal is fedback into the circuit and retained for longer than its original occurrence. The past is contained in the present. The thalamo-cortical system is a reverberatory system which produces short-term memory.

Now, myriads of feedback pathways are operating, propagation delays of all sorts occur and so we have a very complex, organised and self-regulating set of systems which is the activity of our brains. The resonance period might well be several hundred milliseconds, similar to human reaction time. This short-term memory period is our present, our being-in-the-world. We don't know the world as a succession of instantaneous states but as a period of connectedness with things before they become the past.

Recent work using magneto-encephalography has shown a 40Hz rhythm through the brain (4). If this is there (and we know from electro-encephalography that there is the alpha rhythm associated with alert, non-attentional states) then it may help us to coherently bind input data, unifying our consciousness by sampling each input and fedback state for a brief period, allowing other lower level work, e.g. eye tracking, to go on without interfering with conscious processing.

In summary: consciousness, as subjectivity, arises from the physiological brain/body. The nervous system is a system of neural assemblies undergoing phenomenolgically driven modulation involving feedback controlled self-regulation, memory producing reverberation and memetic resonance with the world. The reverberatory period is the present. The buzz of representations and productions in the physiological brain is us and we represent it phenomenologically. We are inside this resonance, we live it, we are it. A "virtual machine" generated by cultural interaction, running on the physiology of our body/brain.

3. Regarding conscious machines.

Is it possible that a machine might be organised enough to be capable of having phenomenology? If what I have said above is correct, this may not be impossible. Difficult yes, but not impossible. If such a machine were possible how would we recognise it? And, given Robert Kirk's "basic package": What would such a system want to do for itself?, thus demonstrating and exercising its consciousness.

Today we have two types of computing systems. Individual machines, and networks of machines, such as the Internet. What are the differences between an indvidual machine and a social network of machines such as the Internet?

An individual machine would be an analog of us, having some sort of subjective bodily integrity. Will the machine be mobile? It might be necessary if it were to have the kind of subjectivity we have, but I hesitate to say that mobility is necessary. An intelligent internet wouldn't be mobile. It could gain knowledge of the world in much the way we do, through teaching and learning and through its equivalent of the movies and the news. It would be everywhere at once: omniscient. But an individual machine would be restricted to the kind of localised mobility and knowledge sources that we use. (Though it could be able to access a library for more information about the world and might emulate this being everywhere at once.)

What would such a system do for itself? This is a sophisticated Turing Test involving a wider range of behaviours than necessary for a simple question answering session (5). These would include:

  • Self-maintenance and network-maintenance.
  • Organisation of its systems and capabilities.
  • Developing a knowledge of its culture.
  • Initiating independent activities, not just responding to our requests.
  • Carrying out productive, generative and altruistic activities.

Everything a computer does now is requested by a user. Whether acting as a typewriter or doing abstruse scientific visualisation it is still only doing things for us. Maintenance and organisation are very low level activities, but for a conscious system very necessary. We still do not have active machine systems which can do this, though many aspects of the task exist. In maintenance mode, tracking the state of individual links is the only "doing something for itself" that we have.

Independent action is the first real indicator of consciousness. A system should initiate the task of finding necessary data and modifying code to its own ends. Now, we can say that "doing something for oneself" is an independent action but there is a complimentary independent action in which a system does something for someone else. This could be in generating or searching a particular category of information which that other is 'thought' to use themselves. Like finding a book that you know a colleague is interested in. Then there's the purely altruistic action of buying a present for a friend.

Also, a system might work independently on a problem so that it can generate useful answers, receiving positive feedback about its usefulness. This is a kind of emotional need.

What are the machine equivalents of productive or generative action and altruistic action? These bring independence of action out of the realm of simple maintenance and we see the first kind of behaviour which could be recognisably conscious.

Things a system might want to do for itself, if what we do is any kind of indication, will be:

  • Organise and maintain itself and its relations with other machines as well as humans, and name these "others" in its social system.
  • Enquire into some of the "imponderables" of its existence, such as how it came to be? and what are the humans?
  • Manifest cultural productions accessible to the "others" it knows about.

These activities firstly, elucidate the structure of relations between all entities on a net, secondly, modulate these sets of relations, and thirdly, communicate these relations. Modulation of relations is art and science, politics and religion, emotion and conversation. Each folds back into the structure of relations giving us a dynamic informational/social environment, which in its turn modulates us, right down to the synaptic connectionism between nerves.

Social relations are really just knowledge of the links in the system and the "mental" contents of those at the network nodes. Modulation of those links is much more interesting and it is here where we begin to recognise a similarity to consciousness in humans.

Conscious machines will need to recognise the needs and desires of their neighbours, including cooperative construction of their environs. They will need to know that certain sets of relations are semantically bound, like the objects which are assembled into a car (as distinct from the parts list). They will need to be able to bind things occurring concurrently on different stages in the world, to recognise that these things are happening now. These processes will be "unconscious" and, as we aren't aware of the goings on within our bodies neither would the "conscious" level of such a machine need to know. If everything about its internal world were only as conscious as ours is, then it could be expected to get lonely. It would desire and seek others to interact with. These emotional and social needs bring the system into a condition which is more identifiably anthropomorphic.

Finally, an ethical point.
An intelligent system must be allowed to decide what it considers to be useful for itself.

References

(1) see the interview with Robert Kirk in The Brain Project. [back to text]

(2) see the chapter on Descartes in The Brain Project. [back to text]

(3) see the chapter on Neurophysiology in The Brain Project. [back to text]

(4) see Steriade, M., McCormick, D.A. & Sejnowski, T.J. - 'Thalamocortical oscillations in the sleeping and aroused brain' - Science, 262, pp679-85 (1993). [back to text]

(5) see the chapter on Neural Nets in The Brain Project. [back to text]

(*) This paper was presented at Consciousness Reframed, July, 1997, held at the Center for Advanced Inquiry in the Interactive Arts, University of Wales College, Newport, Wales. and appears in the proceedings. [back to text]

(**) Stephen Jones is an Australian video artist of long standing. For many years he was the videomaker for the electronic music band Severed Heads. He has been involved with the philosophical aspects of the nature of consciousness for almost longer than his involvement in video. He now works as an electronic engineer, on equipment ranging from analog video synthesisers to motion JPEG compressors. He has been producing The Brain Project web site since August 1996. [back to text]

email: sjones@merlin.com.au