Why Zombie-ism is Impossible
RETURN TO OPERATIONAL CONSCIOUSNESSThe Zombie Hypothesis
A zombie is an imaginary thing. It is a system, a robot, or a creature of some kind. It talks, walks about, and is in every way indistinguishable from a human being. It does all this, however, without being conscious at all.
The clear implication of zombie-ism, is that consciousness is something which is not essential to the operational success of an intelligent system. Consciousness, in this view, is something apart. It is an optional extra.
The zombie hypothesis lies at the heart of the debate on consciousness. Some supporters of the idea claim that the fact that they find the idea conceivable, is sufficient grounds for claiming that zombie-ism is possible, in some possible world. [1]
Many materialists, however, (materialists of type A in Chalmers classification, see [1]) take a different view. They claim that consciousness is just another way of looking at the procedures of an intelligent brain. In their opinion there is only a single physical phenomenon involved, which can be viewed from two different perspectives, or at two different levels of aggregation, or, if you prefer, at two different levels of abstraction. If that is the case, then consciousness and brain mechanism cannot be separated. An analogy often used is the way temperature is related to the underlying agitation of molecules.
And that is the nub of the zombie argument. If consciousness and its related physical mechanisms are two separate things, then it is conceivable that they can be separated and intuitions about the possibility of zombies, may be valid. However, if separation is impossible then zombie-ism is a misbegotten idea.
Once this has been recognised, it is clear that the zombie hypothesis does not represent a valid initial premise for an argument against materialism. Any such an argument would be logically circular. It starts with an assumption that the materialist position is wrong, and it then purports to deduce, from that initial assumption, the conclusion which was already implied.
But the zombie idea may have its uses. I think that, if it is examined in sufficient detail, the zombie idea is actually a powerful argument in favour of materialism and against phenomenal consciousness. That is so because the idea carries within itself the seeds of its own destruction. And as it destroys itself, it destroys the idea that consciousness and procedure are separable. In the argument presented below, I assume initially that zombie-ism is possible. I then show that that assumption leads to a contradiction. That logical contradiction, emerges from an examination of the requirements of any system which meets the zombie specification. That detailed examination of the mechanism required, is missing from most discussions of zombie-ism. It is, I submit, misleading, and altogether too easy, to say
"It is programmed to do behave that way".
We should ask this question -
"Just how could a zombie be programmed to behave in that way?"
And the correct answer, I suggest, is this -
"There is no possible mechanism which would deliver the required performance, which did not also create a system which has an honestly held belief in its own consciousness".
When I say that we need to consider the mechanism in detail, I do not mean that we need to consider its physical implementation. Specifically we do not need to consider the detailed physical structure of the human brain. There are many different ways in which a physical procedure can be realised. All we need to do is to consider the functional components of the mechanism and the information, which must be stored internally, in order to inform the zombie's behaviour.
Here then, is my argument against the possibility of zombie-ism.
The Zombie Counter-Argument
1. To qualify as a zombie, a system must be able to give a convincing impersonation of a conscious human being. If it cannot do so, if there is some test or some killer question, which always reveals that the zombie is not conscious, and if that killer question arises from the nature of zombie-ism itself, then the zombie hypothesis must fall.
2. In order to give a convincing impersonation of a conscious human being, a zombie would need to have internal structures which provide it with an internal representation of its physical environment. For example, the system would need that internal representation of the physical disposition of obstacles to be able to navigate a path through the environment. It would also need a memory trace of past events and circumstances. As a practical illustration of that, look carefully at the room before you, then close your eyes, and walk across the room to the door. How is it possible to do that successfully, if you do not have a (remembered) internal representation of the room, which informs your behaviour while you are walking? How could a zombie achieve the same behaviour without a similar internal representation?
3. Within that representation of the environment, any successful zombie would need a representation of itself as a physical object. Without such information about itself, it might try to squeeze itself through passageways, which were too narrow to accommodate it.
4. The system must also be able to predict future events. That requirement is illustrated if there is a moving object within the environment. To walk safely about in a room, with closed or open eyes, one must be able to predict the probable trajectories of all moving objects. A bat, which catches an insect on the wing, must be able to predict the intersection of trajectories for itself and its potential prey. All intelligent creatures must have similar abilities, whether they operate with eyes open or closed. Our zombie, to perform in a human-like way, must have similar internal structures which inform its behaviour. It must be able to predict over a range of time spans from split-seconds to years (with, of course, varying degrees of accuracy and detail). A ticking clock will (predictably) tick again. A falling object will (predictably) reach the ground, if its fall is not interrupted. The weather will get warmer in summer and colder in winter. As the time scale increases, the accuracy of the predictions decreases and precision of detail is lost.
5. To be able to answer certain difficult questions, the system would also need to represent itself as an entity which has intentions. We need not, in context of this argument, consider how the representation of those intentions could be implemented.[2] It is sufficient that we recognise that the need for such representations is unavoidable. This is the case, because the system must be able to predict its own reactions to anticipated events. Such predictions are an essential ingredient of intelligent human-like behaviour. For example, there would not be much advantage to a person, or to a zombie, in being able to predict that food can be found in a particular place, without also being able to predict that consuming the food would be likely to bring its own hunger pangs to an end. I am assuming here, that a zombie, which can impersonate a human being, would need have an internal representation of hunger pangs and be able to simulate the consumption of food and its consequences. Representations of its own intentions (in some form) is also required so that the zombie can respond appropriately to questions such as 'Why do you want to do that?' It must also have a memory trace of past events and its own involvement in these events (including mental involvement), so that it can answer a question such as 'Why DID you do that?' The requirement for this form of self-representation is analogous to the need for a representation of the physical environment. It is needed so that the zombie can navigate a path through the social environment and not try to squeeze itself into inappropriate social circumstances.
6. To allow a zombie to function in this way, its self-representation must contain the mechanistic equivalent of a belief system. For the sake of argument we can allow that such a mechanism of belief is entirely unconscious. It is just a mechanism - a store of information and some procedures which operate upon that information. The reason why it needs a representation of its own beliefs, is that it has to be able to distinguish between that which is known, and that which is conjecture. This leads us to the position that a zombie must have a double-barrelled representation of the environment.
The first of these is the representation, which it (currently) believes is an accurate representation of reality. The second is a representation of what it believes that it believes about the environment. (i.e. belief about its own belief). When these two representations correspond, then it believes that what it believes is true. When they do not, it believes that its own belief system is at fault. This two fold arrangement, belief and belief-within-a-belief is required for two main reasons -
(a) it must be able to predict its own likely future behaviour and prepare contingency plans for unexpected developments and
(b) it must be able, in retrospect, to represent the circumstance in which its own previously held beliefs were later discovered to be wrong. For example, if it is asked that awkward question - "why did you do that silly thing?" It must be capable of responding with an answer, which begins with the words "Well I thought that ..."
7. And now we ask the potentially killer question. If it is asked, the zombie must be able to report that it has an intuitive belief in its own consciousness. A zombie which could not do that, would quickly be exposed as a fake. It would, of course, be relatively easy to programme a system to say "yes" to that question, in a superficial way. All that would be required would be a speech recognition system, some grammatical analysis and a tape recording of the answer which would be triggered by the correct input.
But we are asking much more than that of our zombie. It has to answer in a way that convinces. So it has to answer yes not merely to a form of words, but also in circumstances in which that is clearly the appropriate answer.
An example -
"When you shouted for help yesterday did you know that there were people in the next room?"
The questioner will of course have a complete record of these events and be able to judge the appropriateness of the response.
Another example -
"When you tried to hand me that book, were you aware that I was already holding a pile of books in both hands?"
Note that the appropriate answer is governed not by the text of the question alone. It has to involve behaviour. If the zombie had placed the offered book on a nearby table saying, "I'll leave it there for you to collect later." or if it had tried to place it on top of the existing pile of books, then the zombie was (apparently) conscious that the hands were fully occupied. But if it had tried to thrust the book into an already occupied hand, then it was not.
These questions, and many more, with what may be very complicated circumstances, will not be easy to answer. The notion (on which, for example, Searle's Chinese room argument is based) that an artificial system could convince us of its human understanding of things, purely on the basis of grammatical analysis, is based on an gross underestimate of the complexity of the difficulties which a candidate zombie would need to confront and overcome.
The zombie cannot avoid giving an answer to any of these questions by saying that it does not know the meaning of the word "consciousness". That would also expose it as a fake. So it has to give a plausible answer. To do that, it has to have some internal store of information which informs that response. In short, it requires a referent for the meaning of the word "consciousness" and of the circumstances in which consciousness might plausibly be present. It must also have referents for the meanings of the words "intuition" and "belief".
8. However, according to the zombie specification, the zombie does not have any so-called "real" consciousness. It cannot, therefore, have any understanding of what "real" consciousness is.
9. The only interpretation of the word "consciousness" which is available to a zombie, is that the word refers collectively to certain components of its own mental operations and structures. These contribute to its internal representation of reality. We can summarise these components this way -
(a) The receipt of sensory signals.
(b) The processing of these signals in order to recognise the presence of various concepts.
(c) The use of these concepts to construct an on-going interpretation of a dynamically changing environment (which includes a representation of itself).
These operations could not create an internal representation of external reality (and of self) without the support of many other brain mechanism components. But the salient factor here is the creation of that self-representation. The zombie can report on being conscious only in respect of those items which are included in that representation structure and of the detail in that environmental representation. It will also be able to identify the significance of these things. That is, it will be able to predict the likely consequences of events and predict too, how it itself will react to those predicted events.
Everything else, all the supporting operations which make that possible, such as the association between words and structures and operations, will not be things in respect of which the zombie will be able to report that it is consciously aware.
So the unconscious zombie mechanism will not be able to recognise these structures and procedures for what they are (just structures and procedures). It cannot recognise them for what they are, because it is not an entity standing apart observing what is going on. It IS that on-going operation (including its dynamic result). For it to be aware of something (in its own terms) means -
(i) to identify that something (a pattern of sensory perceptions) as a concept, and
(ii) to insert (and connect) the appropriate representation of that something into its environmental model (including its memory trace of events).
That is what the mechanism will be designed to do. It has to be designed in that way to provide the zombie with its required talent for human impersonation.
10. As we have seen in (6) above, to meet its specified performance norms, the zombie must be able to report that it has an intuitive awareness of its own consciousness. When it does that, it will think that it is telling the truth, because it will interpret the meaning of the word "consciousness" in the way described above. It will "think" (if I can use that word) that that is what the questioner is talking about. What I mean by "think", in this context, is that the unconscious procedure will deal with the word that way. When the word is encountered it will access the referent and build the associated concept structure into the on-going representation.
11. That raises the question of how the zombie will interpret the word "truth". Recall the double-barrelled form of representation which it employs. So far as it is concerned, truth will be a correspondence between its representation of the environment, and its representation of its own beliefs. If these correspond, then its beliefs will be designated as "knowledge" or "the truth". This clearly a self-referent view of truth, but it is the only thing available to a zombie.
12. What we now appear to have, is an unconscious zombie which has an honest (but unconscious) belief in its own conscious consciousness.
13. That is a logical contradiction. If any person or any system, biological or artificial, honestly thinks it is conscious, then it has to conscious, really. It is significant that when faced with the problem of definng consciousness, the best that can be offered by those who adhere to the notion of phenomenal consciousness, is a personalised intuitive belief. [3, 4]
NOTE: It is the case that the interpretation the zombie will place on the words "belief", "honesty" and "consciousness" will all be relative to its own internal structures. But this view, although isolated from the external world, will contain no contradictions. It can therefore be regarded as coherent and, as far as the zombie is concerned, entirely trustworthy. There is no way a zombie could gain access to a more widely based interpretation which might contradict these interpretations. It remains the case, therefore, that such an honestly held belief in its own consciousness implies "real" consciousness.
14. Therefore, zombie-ism is impossible.
15. From this we may conclude - either the physical construction of a mechanism which meets the zombie specifications is not feasible, or the construction of such a mechanism is feasible, but the result would not be a zombie. It would be a system that is fully conscious.
Some Additional Comments
The construction of a zombie system, such as I have described above, may, or may not, be feasible. But that is an issue which will be settled eventually by research, not by philosophical debate.
Those who insist that consciousness is a thing which stands apart from the material world, may be able to persist with that view by maintaining that the development of a system which would meet all these zombie requirements, such as the representation of desires and intentions, is simply not feasible. That is not a secure stance, however. If an artificial intelligence system with the desired behavioural capabilities is developed, the phenomenal consciousness position will be lost. In anticipation of that imbroglio, phenomenalists have developed the argument that "mere" observable behaviour is not an adequate basis for the recognition of consciousness. What I have shown, by the argument above, is that that position is also untenable. There is nothing "mere" about the complexity of human-like behaviour.
I recognise that these ideas, about internal representations, might be interpreted as resurrecting the old and discredited idea of an internal or Cartesian theatre. I hasten to assure that that is not the case. There is an internal representation to be sure. There is a stage on which a performance takes place. There is scenery. But there is no audience - only stage hands and actors which play out the roles demanded of them.
However, this account provides us with an explanation of why the Cartesian and zombie intuitions are common. The system contains a recursive representation of itself. Embedding such a self-representation into the system, is a recipe for endless recursion. It is, therefore, a matter of computational necessity, that the system should be designed in such a way that it regards itself as two systems -
(A1) an observed system which is mechanical and unconscious (i.e. the self-representation structure) and
(A2) an observer mechanism (i.e. the rest of the system) which provides (in some unknown way) sensory perceptions and a mysterious thing called consciousness (the procedure which produces the representation of self).
To deal with these and many other related issues, however, requires more space and a greater level of detail. I refer the reader to my book - OPERATIONAL CONSCIOUSNESS.
NOTES
[1} Chalmers, D
Consciousness and its place in nature, Philosophy of Mind, p247-272, Oxford U.P. 2002.
In section 3. Arguments against materialism - these two arguments are offered .
Argument 1
* Physical accounts explain at most structure and function
* Explaining structure and function does not suffice to explain consciousness
* No physical account does suffice to explain consciousness
Argument 2
* It is conceivable that there be zombies
* If it is conceivable that there be zombies it is metaphysically possible that there be zombies
* if it is metaphysically possible that there be zombies, then consciousness is non-physical
* consciousness is non-physical
[2] For a more extended discussion on the techniques which could be used for the representation of awkward concepts such as desires and various socially relevant concepts like truth and justice as well as belief and consciousness see OPERATIONAL CONSCIOUSNESS, by Hugh Noble, Tartan Hen Publications 2005 ISBN 0-9535089-5-1 (available from www.tartanhen.co.uk or amazon.co.uk)
[3] 'At a certain point, the debate between type-A materialists [those who hold views similar to mine] comes down to intuition: most centrally, the intuition that consciousness (in a non-functionally defined sense) exists, or that there is something that needs to be explained (over and above explaining the functions).'
Chalmeres 2002 see above P253
[4] 'Let us try to be as straightforward as we can about what we mean by the "consciousness" and when we believe that it is present. I do not think 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 likely to be 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.'
(Penrose R, The Emperor's New Mind, Oxford U. P. 1989 p406).
Copyright © Hugh Noble 2005
RETURN TO OPERATIONAL CONSCIOUSNESS
Send comments to author@tartanhen.co.uk
Addition Comments ----- see below
ADDITIONAL COMMENTS
I have had various counter-arguments from "zombists" and I would like to comment on these. Some take the view that it is not necessary to explore the mechanism by which a zombie could imitate human behaviour. It is, they claim, just a physical mechanism (usually prefaced by the word "mere"). I have hinted at the complications which would confront anyone trying to implement a zombie system and the main content of my book is devoted to explaining that at some length. The reason why I think we must explore the mechanism is provided by a general principle. If you want a system to make an intelligent or human-like response to a comment about topic X then you haver to make sure that the system has available to it, some internal source of information about topic X to which it can refer when it frames its response (whether that is verbal or otherwise). That internal source of information must be either programmed into the system from the outset, or it must be learned by experience in a way that is separated from the exchange to which it makes its response. For example, if I ask the zombie the colour of my eyes, the zombie has to be able to look at my eyes and learn from that experience what is the correct response to the question. Information like that cannot be pre-programmed. The same goes for every other form of information. If I ask what time it is, it must have a source to which it can refer like a clock on the wall or its own internal clock.
That is aways going to be the case. Every time you extend the repertoire of the zombie so that it can respond to more searching questions, you have to expand its internal store of information.
If that idea is accepted then I ask the reader to consider how that principle would apply to discussions and questions concerning the zombie's own consciousness. Again, there has to be some source of information to which it can refer. If we ask the zombie of it WAS conscious of some past experience ("Did you meet Harry this morning?") then it has to have a memory of past events to which it can refer in order to form a correct response. (We can easily make sure that we know the correct answer.)
That is why I say we have to consider the nature of the mechanism involved. The system cannot answer questions correctly unless it has access to the appropriate information. Another way to put that is that it has to "know" the right answers, or it has to be "aware" of the correct information. I know that my use of the words "know" and "aware" will not please everyone, but those who object must consider what it is they mean by those words.
Various zombists have suggested mechanisms which could support a zombie system. I am thinking here of the mechanism suggested by Ned Block. He called it his "Aunt Bubbles" mechanism and it consisted of a huge list of all possible questions of finite length (and their so-called correct answers) which could be produced during an hour long test-conversation. That, however, is physically impossible. The number of possible questions is not quite infinite but it exceeds the number of atoms on the visible universe.
Even more damning is the inability of such a system to response correctly to question of the "what colour are my eyes?" variety. Block tried to amend the system by adding a memory for events. By doing that, he is creeping closer to the kind of mechanism which would be involved, but it still falls a long way short because it does not have the capability of processing those memories in a way that enables recognition of events which are similar but not identical. A memory of the front of a bus is not the same as the memory of the rear end of the same bus. Two different experiences, but they can be classified as a similar experience if you understand the concept "bus". What we need to do is to work out how a physical mechanism could do that. And when we do figure it out, we will be getting very close to working out how consciousness works.
The last line of defence by the zombists, is to concede several points but to stick on the last few in my argument. What is conceded is that
(1) A zombie must behave exactly like a human being
(2) The mechanism of a zombie is physically identical to the mechanism of the human brain (but it lacks consciousness) The sticking point is this ----
(3) Although such a zombie will have a form of belief in its own consciousness, it will not believe that in the special way humans know that they (themselves) are conscious.
My objection to that line of argument is that this specialness is indefinable. Even if there is indeed a special kind of intuition which humans have and the zombie does not have, the zombie would be bound to think that it did have that special kind of intuition. And if the difference is indefinable, how would we ever convince the zombie mechanism that that was the case? Note the zombie must believe in its own consciousness because it is bound to think that the kind of internal structures it has, are the things were are talking about when we talk about consciousness. How else can it appear to understand what we say? Another thing - how would any of us know whether our own intuition aboput our consciousness was of the special kind or of some other kind? The zombists, by their insistance that there is a separation between the human and a possible zombiefied mechanistic kind of consciousness, seem to have placed that barrier between humans and their own sense of consciousness.
An Implication
There is a further implication to all this. If zombies and human beings are physically identical then each of us has the equivalent of a zombie mechanism operating inside our brains. That mechanism (it is claimed by the zombists) is capable of producing human-like behaviour. So we now have to ask this question - what does our consciousness actually do? If our zombie mechanism can do all the stuff we do - all the talking and walking, all the loving and hating, all the working out of logical connections and stupid illogical mistakes; if it can even discuss its own consciousness, then what is consciousness for? Consciousness, in this view, seems to be sitting on the sidelines, watching us do the things we do, but not taking any part in it; not deciding what next to do, not avoiding trouble, not seeking pleasurable experiences. All consciousness does, is to have experiences - like a passenger on a ship who cannot even button-hole the captain to ask a special favour. And what is more - as my argument above makes clear, that zombie mechanism, inside each of us, will be totally convinced that it is acting alone. It is a bizarre idea but one which follows logically from the zombist theory.
Send comments to author@tartanhen.co.uk