An Expedient Mind: Outline of a Mind Mechanism

AN EXPEDIENT MIND

Tartan Hen Publications : Home | more books | Contact : feedback@tartanhen.co.uk

An outline of the System

Development by Pseudo-Evolution
Although my "consciousness mechanism" is not designed to resemble the brain of a biological organism, I will, nevertheless, adopt a design strategy which is similar to biological evolution. Starting with a very simple system, I will add extra features, one at a time. Each step will be small, and (in an equivalent biological context) each would bring with it, some marginal improvement in the system's ability to survive in a hostile environment.
I have a particular reason for adopting that approach. If my thesis is correct, then that must have been the way that real biological intelligence evolved. As the brush-strokes of a painting, bear witness to the way the artist applied the paint, so the products of an evolutionary process, carry evidence of their step-wise development. My claim is that intelligence and consciousness would not have the functional characteristics which they do have, unless they had developed in that way. So, if we want to design an artificial system with the same functional characteristics, then we must developed it in a similar fashion.
We often speak of the design of, for example, the motor car, or the aeroplane, as having "evolved". And the design of a human artefact does take place by a process of evolution. But not by Darwinian evolution. Adaptations to the design of a motor car can be made in a piecemeal way, as new ideas occur, and faults are recognised and eliminated. But it is also possible for the designer of a motor car, to go back to "square-one" and start all over again (avoiding some unfortunate previous choice).
That is not possible with Darwinian evolution. Each step in any process of Darwinian evolution must be based on what already exists. The traces of previous characteristics will always be visible, either in vestigial form, or incorporated into the new design, sometimes modified and happening upon some new purpose.
There are aspects of the conscious mind which are difficult to explain any other way. Why is it that were are consciously aware of only a small part of our own thinking and actions? If consciousness has only an observational role, and plays no mechanistic part in the process of thinking, why does its remit not cover all aspects of that thinking process?
Why do we sleep? Why do we dream? Why do we switch off our ability to detect approaching danger, for long periods of each day? Surely that invites extinction. If consciousness and dream consciousness, are not an integral part of the nuts, bolts and cogwheels of our thinking mechanism, and if they are not a modification which fortuitously corrects some previous defect, these human (and animal) characteristics, would seem to me to be inexplicable. If consciousness is not part of the physical mechanism of the brain, why is it that injury to the physical brain or some physical congenital conditions, can give rise to blindsight and synaesthesia?
The Darwinian evolution of thinking provides an explanation of these things and that explanation will emerge as I develop my consciousness machine.

The Five Layered System

The mechanism has five layers. Layer-1 is at the bottom, and is the first to be developed and discussed. But that layer itself must evolve from something simpler - something so simple that not even the most sceptical of my readers, will be able to deny that such a system could be constructed and would operate as described. That simple system can then grow in small steps into the complete form of layer-1.

Layer 1 - The Stimulus-Response Automaton
Layer-1 is a stimulus-response automaton. It is equipped with various kinds of sensory aparatus. These are analogous to the retina of the eye, touch-sensitive finger-tips, to the ear, and so on. Layer-1 will also have sensory equipment similar to the internal sensors with which we are able to sense the positions of our limbs, the tiredness of our muscles, the state of nutrition, or hydration. So it has a wide range of sensors which are able to send signals to the rest of its "brain" (if I may be allowed to use that term, in this context). There is no doubt that such things can be constructed using electronic components. The electronic retina of a digital camera is an obvious example. We also have robot devices with jointed limbs and sensors which monitor the positions of these joints. And we have sensor-electrodes which can read the chemical composition of blood. And so on.
Within layer-1 we will also find what I call "#perceptions". (Note the hash-prefix. That is a notation I will use to indicate that the word is being used as a label for a particular kind of physical structure) These are similar to "perceptrons" defined many years ago by Rosenblatt and now widely used in the field of neural networks. But my #perceptions are a little more general and more flexible. Each has the power of a mini-computer, with its own processor and its own memory store. The now defunct and superceded transputer device, would be a means to implement this component.
Each #perception is able to receive a set of input sensory signals, to analyse them, and then to emit another signal (or signals), depending upon whether or not it has detected a particular pattern within that group of signals. Could such a think have evolved in small steps. I think so. We could start with a thing which is capable of responding to a single sensor - like a thermostat. Is it it ON or is it OFF? Then two signals. Is X greater than Y? And so on in gradual steps to increasingly complex analysis.
The signals sent out by the #perceptions, are targeted at another component of the system. For generality I call it a "#responses" (note the hash-prefix again). We can think of them as being like the muscles which we use to control our limbs.
So layer-1 reads sensory signals (the stimuli) and then responds to the patterns of signals which those #perceptions are able to detect, in one of a fixed range of pre-set physical responses.
The kind of sensory perception to which it may respond could be the distinctive shape of and approaching predator, the scent of a morsel of food, intense sunlight, or any one of a host of possible environmental circumstances. Initially the stimuli will be very simple and very limited in range. Even the ability to respond to one single environmental variable is a survival advantage when compared with other system which do not have any such ability. New stimuli and new responses can then be added one at a time, by a system of trial and error (mostly error). But given sufficient time, and a sufficient number of systems to try their luck, improvements will slowly emerge.
There might be some dispute about the degree of sophistication we could build into a system like that, but I do not anticipate objections to the idea in principle. The system, indeed, could be regarded as a caricature of the robotic system in the imagination of most people. In the popular imagination, if it encounters something, which is not listed in its repertoire of perceptions and responses, it has a standard get-out response. It says “this does not compute”, or it shouts "exterminate!" in an electronic monotone. That limited mental model of a robot is perhaps the reason why most people are sceptical of the whole idea of machine intelligence. How could such a thing ever be regarded as “conscious”? It's a preposterous idea. Isn't it? We shall see.
We could, in principle, simulate the evolutionary development process. We could build an enormous number of similar automata, with random variations in the detail of their design. We would then reject those which failed to respond to events in an appropriate way, and then we would build a new generation based on the design of the successful ones. And so on and on for billions of years. As I said, it is possible in principle, but there might be a few practical difficulties - such as remaining alive long enough to witness the results.

Layer-2 - The Memorizor
Layer-2 is able to process the information provided by layer-1. That information includes more than just the pattern of input signals. It includes the whole of the repertoire of the #perceptions and the extent to which the incoming signals are identified as corresponding to one or other of the standard set. It also includes the #responses which are associated with those incoming signals and the priorities which layer-1 assigns to those stimuli and responses. In other words, layer-2 can read and analyse ALL of the information provided by layer-1. It can analyse it, and it can store all of that information for future reference (for a very short period of time). It can also store SOME of that information, for a longer period.
Certain advantages accrue from this ability to store information. If, for example, some dangerous condition is recognised, and the system triggers its escape response, it will then be able to keep on escaping, for a while at least, after the initial stimulus is not longer perceived. Also, by remembering some important episodes in its past experience, the system can recognise the possible consequences of its current perceptions. It could do that before any adverse conditions actually arise. In other words, it can anticipate. Even a limited facility of that kind, is better than none at all. Over time, it can be improved in gradual steps.

Layer-3 - The Conceptualizer
Layer-3 is able to process ALL of the information provided by layers 1 and 2. The ability of layer-2 to store information about past experiences, is constrained by the physical limitations of storage space. It is impossible for any system to go on storing more and more information without eventually running out of space. Layer-3 introduces a new strategy to reduce the amount of space required. It introduces procedures for compressing the stored information.
Anyone who has ever taken a digital photograph will be familiar with the idea of compression. Although there are many different algorithms for squeezing data into a smaller and smaller space, the basic principle is common to all and it is also easy to understand. Find the common elements, the patterns of data which are repeated. Extract the chunk of data which is repeated and store just one copy of it somewhere. Then, in the original store of data, replace all those multiple occurrences of the repeating chunk, with a bookmark. A bookmark takes up less space than the original chunk. If you then want to reconstruct the original image, read what it says on the bookmark, find the chunk of data to which it refers, and copy that original chunk back into the slot which is marked by the bookmark. The penalty is that the reconstruction process takes time. The advantage is that much more data can be stored in the same amount of space. In some circumstances, the advantage will outweigh the disadvantage and that will be the survival advantage provided by this layer.
But there is more to layer-3 than just a saving of space. Sometimes, evolution takes an unexpected turn. Sometimes an evolutionary development, which provides a genuine advantage at one point in time, turns out later to be a disaster. The development of sabre teeth in tigers allowed those tigers to pierce the tough skins of their prey. But when that kind of prey was killed off, those large cumbersome teeth turned out to be worse than useless. And sometimes the reverse happens. The passage of time and changing circumstances may reveal a particular development to be even more advantageous than was initially apparent.
And that, I suggest, is what happens to the compression process of layer-3. If we examine the repeating chunks of data, which have been derived from sensory perception, we can recognise that these chunks represent more than just data. They correspond to objects, and events which are encountered on many occasions. That secondary store of repeating chunks, is actually a store of information which can be turned into what we call “concepts” and concepts are the organising principle of all intelligent analysis.
A concept contains more than just a fleeting experience. It is a compendium of experience, built up over a period of time, from many encounters. When you look at a sheep, you don’t actually see the whole sheep, not at that instant, not in terms of instantaneous perceptions. But you do ‘see’ the whole sheep in another sense. The concept in the mind, which corresponds to a ‘sheep’, contains information about what that sheep would look like from every direction. It also provides information about what the sheep should sound like, feel like (if touched), and taste like (if cooked and eaten). The concept is the whole thing. It also tells us how sheep behave, so that we can anticipate how they are likely to respond in certain circumstances. Concepts enable us to predict. Layer-3 provides my system with that facility.
The concepts which are developed by the layer-3 mechanism, can be classified in several ways. One is the classification of physical objects. Each encounter with a physical object happens a single moment in time. A second classification is of events, each of which is spread over a period of time.
But there is a third classification which though less obvious, is of fundamental importance. This is the experience of two momentary and distinct circumstances, which habitually occupy chronologically adjacent moments in time. Two conditions - X and Y, such that X always, or nearly always, precedes Y. We can then say that X is a reliable predictor of the occurrence of Y. This is the fundamental basis for prediction. It is also the concept, which we call “a causal link”. When we say that X “causes” Y that is simply shorthand way of saying that when we experience X we expect to experience Y immediately afterwards. Those familiar with the philosophy of David Hume will recognise that that view corresponds exactly to his analysis of causation. Causation, is not something inherent in the universe which connects events and makes the occurrence of some events inevitable. There may indeed be such a connectivity in the universe, but in so far as we can observe what happens on the universe, we have only that observation of temporal association to guilde us. We call it a causal link, meaning that we can rely upon it and that we can use past experience to predict future events. Causation is a human creation. It is a mental construct which we use to render predictable an otherwise bewildering universe.

Layer-4 - The Mind-Reader
Layer-3 has cracked the problem of predicting the behaviour of the inanimate universe. We know (and my system will know) from experience how stones will fall or fly through the air if thrown, how water pours, how wind blows. We know what will hurt us and what will be good to eat. We can predict how to construct a comfortable shelter.
But there remains the problem of predicting the behaviour of the animate (or apparently animate) universe. How can we predict the behaviour of other human beings, or of animals? How, for that matter, how can we predict the behaviour of a volcano, a lightning bolt or a tsunami? For a social species like humans, it is a regrettable fact, that the behaviour of other human beings is the source of our greatest danger and of succour. So the ability to predict that behaviour would bring with it a very great survival advantage.
To solve this problem, my system will fall back on the concept of cause. That inate facility has been successful in enabling all kinds of predictions, so why not in this case too? If Y is a special kind of behaviour, the prediction of which is important for our survival, then we should look for some X which will be a reliable predictor of Y. So use the apparatus of perception. Look. Feel. Touch. Smell. Seek X in every circumstance possible. Seek a repeating item of some kind. Seek it in nature. In the stars. In the entrails of beasts. And seek it in the faces of the people we live among.
Soon we will find there certain facial expressions, certain tones of voice, certain body movements which are the tell-tale signs of aggression, friendship, selfishness and so on.
A facial expression does not cause someone to be aggressive. It is not the X which we seek. But a facial expression can be the clue which tells us that X is present within the person whose behaviour we want to predict.
Note please what I am NOT saying. I am not saying that the system does a bit of intellectual analysis and works out that it needs to form such a concept of mind for other people. It is an error (and one which often traps the unwary) to think that a developing brain can ‘work out’ what it needs to make some improvement in its own thinking processes. That would be to suppose that the developing brain has available to it, the kind of intellectual abilities which it has not yet evolved. That's a bit like stone age man trying to make himself a stone tool, and doing that with modern stone grinding tools. It doesn't work that way.
Any improvement, of that kind, has to happen spontaneously and in gradual faltering stages. It is a process of trial and error. The random mutation of genes proposes, and the brutal axe of natural selection disposes (or, alternatively), allows to survive).
The ability to predict future events is crucial for survival. The development of the concept of “causal link” is crucial to the ability to predict. That is a natural ability which could arise by chance. It is also an ability which can be promoted or inhibited by the way the system operates. We can, as it were, turn up the knobs, to enhance the propensity to form the concept of causal-link. So important, is it, that potential causal links are not overlooked, that it would be advantageous to have the brain hardwired, and perhaps a little over-wired, to perceive causal connections. A bias like that, would make it easy for a brain to develop the concept of a mind, which seems to drive the behaviour of other people. That mental trait, is a very important survival advantage, so it is easy to see why it would be produced by evolution, without any need to “work out” what is needed.
While a mental bias towards the forming the concept of mind as a predictor of behaviour, would ensure no causal link was overlooked, it would also result in a few spurious causal links being formed, where there might be none. The behaviour of a volcano is unpredictable, so let us hypothesise a mind which controls it. The seasons operate in a cycle, which is in phase with the Sun’s behaviour. So why not give the Sun a mind of its own, and imagine that it is causing the seasons to happen in some mysterious way. A woman’s body is roughly in phase with the Moon. So let’s have another causal connection. So too, between the Moon and the tides. Note that some of these connections are spurious while others are completely accurate (Moon and tides for example). Superstition is nothing more than an early example of scientific enquiry. It ceases to be science and becomes superstition, only when better explanations are available and someone goes on clinging to that older, less adequate theory.
Layer-4 introduces what psychologists call “The Theory of Mind” which helps the system to predict the behaviour of animate objects. It also helps the system to predict its own behaviour. There is not much survival advantage in being able to predict the likely course of events, if the system is unable to predict its own reaction to these likely future events. It has to be able to characterise its own behaviour in ways that facilitate its own reactions. Minds, including its own, can be characterised as brave, cowardly, kind, cruel, reliable, unreliable, honest, dishonest, intelligent, stupid and so on. All of these are characteristics which, to a degree, allow future behaviour, in a variety of circumstances, to be predicted.
To enable all this to happen Layer-4 requires to have a representation of itself. I have already explained the problem, which that entails. I have also indicated how that problem might be solved. The representation contains a memory record of the events experienced. This record, however, unlike the memory stores of the lower layers, does not hold the raw data of experience. It holds a record of events in conceptual form. That record will therefore not contain most of the incidental aspects of experience. It will have only those aspects which are considered important (and by important I mean those aspects which were given a high level of priority by layer-1). Thus the record held in conceptual memory will be events which were dangerous, or delightful, stressful or stress-relieving.
By a continued process of compression, many of these conceptual memories will also be converted into compendia of similar experiences. Subtle differences will be eliminated and what remains will be the commonality of those experiences. Crucial to this process of self-representation, however, will be the way the record is limited. It will omit any record of how the relevant information was acquired by the self-representation entity. If it did not abide by that limitation, it would immediately get involved in the endless recursion I mentioned previously. As a result, the self-representation will appear to acquire information by magic (i.e. by an unexplained process). The self-representation will appear to be “observing” the rest of the system in operation. Hence the dualism of Descartes. The aspect of all this on-going processing, which we call “consciousness”, is linked to the focus of attention. The other layers of the system continue to process data in their accustomed fashion. Every now and then a particular bit of the on-going experience, generates an alarm mechanism. This is triggered by the priority levels assigned to certain perceptions by layer-1.
When this happens, Layer-4 is directed to this aspect and processes the data using its store of remembered concepts and their associated predictions. Other aspects of on-going experience continue to be processed by the lower layers, as before. Layer-4 can (in effect) pick and choose, on which bits of experience it will focus its attention. I say “in effect” because it does not actually have much of a choice. Try a little experiment. Try concentrating on something of your own choice, and invite a friend to do something unexpected - like exploding a paper bag or setting fire to your hair, and see if you can avoid switching your focus of attention to this new event. If you know in advance what is going to happen, you might just be able to succeed, but I will lay good odds that you cannot do it, any other way. Layer-4 is the layer of consciousness. So why do I want to introduce a fifth layer?

Layer-5 - The Conversationalist.
The answer is simple, but, I fear, will elicit some incredulity on the part of some readers. These will probably be the same readers who expressed incredulity about the idea of a machine being able to think and be intelligent. In a by-gone age, these would also be the people who expressed incredulity about the idea of building a ship using metal instead of wood. Or of building a machine which can fly - “Do you seriously think it would be possible to build a machine - made of metal for goodness sake, which would be able to rise off the ground and transport people halfway round the world. You’ve got to be mad!”
Yes. I do. Actually. And what is more, I am prepared to back up that assertion with an explanation of how it can be done. Read on sceptic.
And the answer to the other question - why do I want this fifth layer - is this - I want to be able to talk to the system and ask it what it feels like to be a mechanism with this kind of consciousness procedure running in its brain. It would be more accurate to say that I want to be able to imagine talking to the system. Before I even start to imagine, however, I must first describe how such a conversational ability would be constructed (in principle).
It is at this point that I run into a bit of serious opposition from the serried ranks of the academic establishment. It has, for many years, been the received wisdom of researchers in the field of automatic language processing, that syntax is the bedrock of linguistic ability. This is a view first mooted, and since sustained, by the redoubtable Noame Chomsky. In a best selling book, Steven Pinker expounded the idea and argued enthusiastically for the existence of a mechanism of syntax analysis, an inherited characteristic of the human brain - and a characteristic which has made possible the human use of language.
But there is a snag to the Chomskian thesis. No one has been able to explain how a mechanism of syntax analysis could have developed by Darwinian evolution prior to the development of language. Pinker made a brave effort to do so, but even Chomsky himself has admitted that the idea seems problematic, to put it mildly. There does not seem to be any survival advantage to be gained by having some kind of partial syntax machine, operating in one’s brain. But that is what the gradualism of evolution demands. An ability to see, or hunt, or run fast, or to predict the likely outcome of events, are all immediately and individually advantageous, even if only partyially developed. So they can start small and grow big. It is difficult to see, however, how the acquisition of a facility for syntax, by one individual, would give that individual a survival edge over his neighbours or any other species. Language, unlike those other abilities, is a means of communication. It provides no advantage to an individual, unless there is a community of more individuals who are similarly equipped and with which one can communicate.
So I reject the Chomskian theory. In my system, syntax will evolve later, after an ability to communicate, using language, has already been established. Syntax, in my system, is a consequence of language, and a refinement, not its precursor.
It is acquired by a continuation of exactly the same evolutionary processes, driven by exactly the same step by step acquisition of marginal survival advantages. The development of concepts has already taken place at the level of layers 3 and 4. In addition to this, there is also an ability to utter a wide variety of sounds and to hear, discriminate, and recognise, these sounds. The mechanism of concept formation, which is a development of the compression algorithm, is able to recognise and build into a concept structure, any number of perceptions which occur simultaneously and do so repeatedly. That process must then build into the concept of any physical object, the sound of a word being uttered whenever that physical object comes into view. So the association between word and object is established. The sound of its word-label becomes a property of the concept.
The “meaning” of a word, is the concept for which it is the label. Once this association has been established, words can be used as a means to communicate concepts from one individual to another. In a primitive form of language all that is needed for a small marginal survival advantage is the means to communicate single concepts - “Wolf, Wolf!” “Mummy”, “tree”, “rock”. If consciousness is a means of applying a complex process of reasoning and prediction to selected circumstances, then the utterance of single words can be the means to direct the focus of attention (and therefore of consciousness) to important circumstances. Human society, becomes safer if individuals can draw the attention of others to important objects and events. Of course, once the link between words and concepts has been established, words can be used by one individual to introduce others to new concepts. Communication serves to establish a standard set of concepts, and that in turn, serves to promote better communication.
After the word-labels for physical objects, come the word-labels for actions - “run”, “sit”, “cook”, “hunt”. Next comes the ability to communicate multiple concepts - “Mummy run”, “Wolf hunt” and so on. This linguistic ability can grow in small steps with a marginal advantage associated with each. The idea that words are just the labels for concepts has often been suggested, and just as often denounced. The major obstacle to acceptance of the idea has been the existence of words like “how”, “at”, “why” and so on. But this objection can be overcome without undue difficulty.
First we have to establish, not just single object-words, and action-words, but we have to classify them. Exactly the same process of compression (and concept formation) will identify the similarity of structure between a number of concepts - note that that is a similarity between the concepts which these words label, not a similarity between the words themselves. So “dog” and “cat” will be grouped together, not because the word “dog” has any similarity with the word “cat” but because the concept “dog” and the concept “cat” both have the same properties of being physical object which is animate, has four legs, a tail and a head. And much more. In the same kind of way, words like “skip” and “run” will be associated. Both are active energetic things which people do. In this way concepts are grouped into a hierarchical structure.
In this way too, words can be organised in to groups to which we can now, with hindsight, attach the names “noun” and “verb”. Thus does syntactical classification emerge as side-effect from a primitive linguistic ability. Every word is a sensory experience. We hear the utterance of words and we hear them in a temporal sequence.
As an utterance is being heard it is also being processed. Each word, as it is encountered is used as a label to extract from concept store, the appropriate concept structure. Just as the word-labels for physical objects are established by repeated occurrence in association with a particular physical object (object-concept) so these little words like “how” “at” and “why” will be associated by constant repetition, with a pattern of concepts.
These concepts will have already been classified and so we can see how it is that a word like “to”, in the utterance “Run to Mummy”, can be recognised as the label for a concept in its own right. This is a concept which identifies a relationship between two other existing concepts. It is, in my classification scheme a “meta-concept” label (a concept which refers to other concepts). Words like “How” and “Why” express a relationship between other concepts as they are within the mind structure of a person. Specifically they indicate a lack of information and a desire to acquire that missing information.
This then is the contribution made by Layer-5. It extends the mechanism of concept formation to the experience of the spoken word and makes possible the construction of a representation of the world, not from that world as it is experienced by direct sensory perceptions, but by analysing spoken utterances, and constructing the representation of a hypothetical world based on those utterances.

In Conversation with a Robot
That is an outline of the five layer brain mechanism which I claim would possess the property we call “consciousness”. Having described the whole thing again, in much more detail, I will then ask the reader to consider what would happen if we were able to have a conversation with the system.
The basic principle involved here, is that the system cannot answer questions correctly, unless he - may I use the word 'he' in this context? - unless he has available some internally stored information which will inform his response.
For example if I ask him what colour my eyes are, he cannot answer that question in a way that will convince us he is conscious, unless he can look into my eyes and record the colour of those eyes in some internal register. I reject the idea that the system can be convincing by means of verbal tricks, like claiming the light is not strong enough for him to be able to see my eyes, or claiming temporary blindness. We are not going to accept a game of pretend. So the system has to convince, and do it handsomely. But since we have studied the system architecture in detail, we will know what information is stored within, and what is available to the system.
And now comes the acid test. We guide the conversation round to the issue of his own feelings about his being conscious. Note that, to be able to take part in such a conversation, and to convince us that he knows what he is talking about, the robot must have a stored concept, which is labelled by the word "consciousness". And if we ask him if he thinks he is conscious, he will fetch out that concept structure, examine his own mental processes and his own representation of self. And having done that, he is bound to say that he does indeed believe that he has this thing called consciousness. Why would he answer in any other way? He (the mechanism) has the requisite concept structure. He knows (or thinks he knows) what it means to be conscious. And he will find that he is able to do exactly the procedure defined by that concept. Some may argue that when he responds that he is indeed conscious, that he is not really conscious of what he is saying. He is just "going through the motions". That idea raises interesting issues which i want to defer until later.

Consciousness and Intuition
But by way of a preamble to that discussion, consider the words of Professor Roger Penrose, Rouse Ball professor of Mathematics at the University of Oxford, and author of "The Emperor's New Mind"

"Let us try to be as straightforward as we can about what we mean by 'consciousness' and when we believe that it is present. I do not think that it is wise, at this stage of understanding, to attempt to propose a precise definition of consciousness, but we can rely to good measure, on our subjective impressions and intuitive common sense as to what the term means and when this property of consciousness is present. I more or less know when I am conscious myself, and I take it that other people experience something corresponding to what I experience." [Emperor's New Mind Ch10 p406]

According to David Chalmers (another well-known advocate of the phenomenalist position), a "Type-A materialist" is one who holds that there is no distinction which can be made between the physical description of consciousness and the phenomenal description (the subjective experience of consciousness. I reckon that means I am, in his classification scheme, a type-A materialist. Chalmers then says the following -

"At a certain point, the debate between type-A materialists and their opponents usually comes down to intuition: most centrally, the intuition that consciousness (in a nonfunctionally defined sense) exists, or that there is something that needs to be explained (over and above explainng the functions). This claim does not gain its support from argument, but from a sort of observation, along with rebuttal of counterarguments." [Consciousness and it place in nature, by David Chalmers, originally published by Blackwells Guide to the Philosophy of Mind ed Stich and Warfield, and reproduced in Philosophy of Mind, classical and contemporary readings, ed David Chalmers, Oxford University press, 2002]

The Intuitive Robot
And now we have a problem. On one hand we have two well known advocates of the phenomenalist view of consciousness, admitting that the only evidence they have for the existence of phenomenal consciousness, is subjective intuition.
On the other hand, we have a robot, which obviously has only a physical mechanism of consciousness and could not possibly be in possession of any kind of phenomenal consciousness. Yet, that same robot is claiming to have exactly the same phenomenal experience of consciousness which some humans claim to have. Moreover, he is doing so honestly. We can tell that he is being honest because we understand his definition of honesty and the nature of the information on which he bases his claim. The conclusion is logically inescapable. The robot is conscious in every sense of the word.
In so far as he honestly believes that he has some kind of phenomenal consciousness, over and above the physical mechanism of consciousness, he is course, deluded. That does not prove that humans who also believe their own intuitions about their experience of consciousness, are similarly deluded. But unless they can produce some evidence, which is more substantial than intuition, that conclusion (that they are deluded) remains the simplest and most convincing explanation. Any other kind of explanation is a gratuitous infringement of the principle called "Occam's Razor".

NOTE: I have discussed this point about a robot believing in his own consciousness, with a well known proponent of the phenomenalist viewpoint. I would not be fair to name him, because I did not caution him in advance that his words may be taken down and used in evidence. But his argument was interesting and illustrates the tenacity with which phenomenalists cling to the validity of their own intuitive judgement even when it becomes logically self-contradictory. The claim was this - even if a robot was able to profess a belief in his own consciousness that would not be a valid claim because he was not actually conscious of believing in his own consciousness. His belief in his own consciousness was not the same as our belief in our own consciousness. And how are we able to tell whether our own intuitive belief is of the valid or the invalid kind? Our intuition tells us it is valid.

In the concluding chapters of the book, I will develop these arguments in more detail, tackle a number of well known counterarguments, and discuss some of the philosophical implications of my thesis.

Tartan Hen Publications : Home | more books | Contact : feedback@tartanhen.co.uk