An Expedient Mind: Layer-4, The Theory of Mind

AN EXPEDIENT MIND

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

Layer 4: The Theory of Mind

On The Recognition of Faces
We see faces, even when there is none - on a rocky cliff, on the moon, in a scattering of paint. It is not hard to understand why that is. For social animals, as we are, other people are the most significant source of danger, as well as being the most important source of help and support. For that reason it is important that we do not fail to recognise a face, and to read the expressions we find there. But if we overdo it - what if we recognise faces where there is noe to find? So long as that tendency does not become excessive, there is no harm in recognising a face on the surface of the moon. Erring on the safe side is tendency which the practical consequences of natural selection forces upon us.

And The Recognition of Minds
Not by faces alone do we anticipte another's behaviour. Body posture and tone of voice also provide important clues. But clues to what exactly? We are so familiar with the idea that the behaviour of every person is driven by what we call "a state of mind", that it is sometimes difficult to remember that no one has ever actually observed another person's mind. The notion is pure conjecture. But like the recognitionn of faces, the conjecture of a motivating mind, must err on the safe side if we are not to overlook a vital clue which will enable prediction.
Imagine what you would think if you stepped on to a strange planet and observed rocks moving about in ways which appear to defy gravity amd emitting noises which are unrelated to any observable event. I think you would be bound to suspect that these rocks were "alive" and had some motivating internal driving force which caused these things. That, at least, would be the initial reaction. The general, but not always accurate rule, is UNPREDICTABLE BEHAVIOUR => CAUSED BY A MIND.

Behaviourism
But what form should we assign to this assumed mind? Behaviourists avoid that question. They tabulate only the observable. LIst every stimulus and its associated response.
A hopeless task. We might manage that with simple organisms. Perhaps even with a pigeon. But the idea that we can capturre the entirre range of human behaviour in such a way is ridiculous. Every stimulus would need to be accompanied by so many caveats and qualifications, and detailed notes about the previous history of the system, that the required table would quickly outstrip the capacity of any possible information storage system. Not only that, but the list of possible histories woould need to include histories which had not actually happened yet.
But then again, perhaps it is not quite so ridiculous if we could separate that table into sub-tables. Then, if the selection of sub-table was determined by those qualifications and previous histories, then the listing might not be quite so preposterously large. If we could classify various types of experience, by the effect they have have on selecting the sub-table we might be able to chop the overwhelming profusion of possibilities down to a few standard types.
A change of terminology. Call that behaviourist's stimulus-response table a "MIND". Call each of these sub-tables an "EMOTIONAL CONDITION" or a "STATE OF MIND". Now we must try to identify the experiences which select a particular kind of table (or drive the mind into a particular emotional condition). Once we have identified the relevant emotiional condition, it is not so hard to list the potential behaviour we migght expect in response to a number of possible stimuli.

Table-1
Pain -> strike back
Ask a favour -> refuse
Help needed -> ignore
Food shortage -> refuse to share

Table-2
Pain -> seek help
Ask a favour -> comply
Need help -> offer help
Food shortage -> share

I think two tables are sufficient for the purposes of illustration.
We can see, that the sub-table is promising but not enough. We need also to list the experiences which might causse pain. Pricking (with a pin or any other sharp implement). Burning (with a match or any other very hot object). Striking (with a hammer or any heavy object).
That's an improvement. By classifying things like that we are able to avoid the combinatorial explosion which would occur if we tried to list everything individually. We can see also that the representation of a mind relies heavily on the way thinigs can be classified by being linked (by causal links) with their consequences. What is pain? What is anger? What is happiness? Pain is what makes a person do this ... Anger is what makes a person do that ... Happiness is what makes a person do the following ...
Now recall those bits of information which are donated to the system by the Layer-1 sub-system. I refer to the priorities assigned to each perception, and to the broader categories NICE and NASTY which indicate whether or not the system will try to repeat or avoid those circumstances. If we now add those categories into the tabulated lists, we have a way of representing the emotional drives for behaviour. We also have the basis for a representation of personality. A short tempered person is one who will lash out in response to low level NASTINESS where others might refrain until the level of provocation becomes much greater. We even have a means to represent the concept of "provocation" itself. It is a NASTY form of behaviour which causes a NASTY response to be more likely.

The Representation of Motivation
In an attempt to clarify what is, admittedly, a complex situation, I have drawn a schematic diagram of the way the system would represent the way a person is motivated to perform some action. Note please the very heavy use of causal-links.

FIG: The diagram is in two parts.
(1) is a symbolic representation of some arbitrary circumstance. It could, for example, be the representation of a person eating chocolates. The representational structure would contain two entities (the person and the chocolates) and then in chronological sequence the person causing the chocolate to enter his mouth and that causing pleasure (NICE) to the person concerned.
(2) Is a representation of a person and that person's MIND (just another structure). The representation shown in (1) is then placed inside the MIND structure and its presence there. The full representation also shows the physical person CAUSING some action to be carried out (presumably the same action of eating chocolates). The crucial link, between the mental representation inside the MIND structure and the external physical action, is yet another CAUSAL LINK. In words, it is having the idea in his head, (and the way that idea predicts a NICE condition resulting from the action), which causes the person to cause the action to take place.

As I said at the outset - the individual components of the system are simple. It is the way they are put together which is a bit complicated. It should be clear from that example, the way various modifications could be made to the representational structure to show the difference between deliberate and accidental actions, or how a person might plan to do one thing, and end up doing something very different. Since the MIND structure is just another data structure like the rest, there is no limit to the arbitrary structures which can be placed there. We shall also see later (in the Layer-5 description) how the ability to place complicated structures inside the MIND of a person (and one's own MIND structure) can solve a number of problems related to the meaning of words like "how" and "why" and "certainly" and so on.
For the present we can note the extra scope whhich this method provides for the separation of what might be called a "fictional" truth from one that is considered "really" true. Did Hamlet kill the King with a poisoned sword? Yes. That is true. Is it "really" true? No. Not really. Well ... it might be. There was a historical character who was the Prince of Denmark, and it is possible that the story on which Shakespeare based his play, had some basis in fact. But when we describe that event as "true", we mean that that is indeed part of the plot of the play called "Hamlet". It is part of a structure in the MIND of Shakespeare (or was when he was alive) and is now part of the MIND structure of anyone who knows the play. The story is "true" in the context of that play. It is "not true" in the context of the so-called "real world". Relative truth. That is an issue to which I will return in the last sections of this book.

The Theory of Self Mind
That reference to one's own MIND structure introduces the topic of this section. There is little advantage to be gained from an ability to predict the likely course of events if one cannot at the same time predict one's own likely reponse to those predicted events. Specifically, one has to know whether these future circumstances will be NICE or NASTY in order to know of they should be sought or avoided. And that means that one has to make a prediction about the MIND representation of these future events, when they occur.
In the previous section (on layer-3) I described the interpretation procedure and how, in general terms, it builds a structure which i call the #interpretation. The #interpretation structure is dynamic. By that I mean that its construction is continuous and takes place while new sensory experience is being gathered. In that respect it is rather like the #trace structure which I described for the simpler system consisting only of layers 1 and 2. And like the #trace, there has to be some way of gettinig rid of old material before it swamps the storage capacity of the brain, and to do so without losing vital information. So like the simple, short-term #trace memory, there has to be a way of processing the continuously growing #interpretation. That means more off-line processing, more compression, more memory store, and more consolidation and re-organisation of that memory store. This time, however, thememory store does not consist of raw sensory data. this time it consists of an interpreation of events in terms of concepts which have been modified to fit the observed circumstances.
Take, for example, a visit to the supermarket. There has to be a short-term interpretation memory store which will record the interpreted events as the interpretation was formed at the time. This will then be processed to extract the chunks of conceptual structure which are shared by this particular set of circumstances, and previous memories of various visits to a supermarket. Quite soon after that type of compression has been performed, the original information will be deleted. After that it will not be possible for the robot to remember (for example) where he parked his car on that particular visit or which particular items he purchased. If something unusual happened, like meeting an old friend or finding something unexpected on the shelves, it will stull be possible to recover detailed information. But slowly, that too will be merged with the general and more highly compressed form.

Memory and the Self-Mind
And where will these short and long-term conceptual memories be stored? There is one obvious location - in the SELF-MIND structure. These memory stores become part of the history of that self-person (or self-robot). They will in fact be part of his personality. To a considerable extent, a person and that persons memories of his/her own experience is synonymous with that person's self-image. In short, consciousness is this memory store (of current and past experience) and the process by which it is formed.

Internal Models
I want now to sum up how the system works. In order to function as it should, the system must construct a mental representation of its environment. Much of what has gone before has been a description of how that internal representation can be formed. The final stage is that interpretation procedure which builds a dynamic model of the environment using concept structure as the constructional units.
The system refers to this model constantly. We could say therefore that this internal model "informs" all of its behaviour. I will be dealing with language in the next section, but if I can anticipate that for a moment and assume that it is able to talk freely, I can make my next point more easily.
When it speaks, it does not produce an utterance from a pre-stored list of standard utterances. It constructs each utterance afresh. And if it is asked a question, it must consult the internal model to ensure that its answer is correct.
I ask it, "What colour are my eyes?" It must examine the model, identify the part of it that corresponds to me, and my eyes, recognise the colour, and then reply with the correct verbal description of that colour.
I say, "Fetch me that book please." It must consult the internal model, find within it the bit which is the model of the book, note its location relative to other items in the environmental space, then plot a course, from the robot's own current location, to the location of the book, and return.
"Where is Australia?" The robot must consult his mental model, not of his current local environment, but the wider environment as defined by the concept "Australia". Here he will find a representation of the world and the location of Australia on that world-model, and then frame an answer in these terms.

Self Representation
I ask it, "Are you happy?" The system must now consult the model of his/its immediate environment, and that part of it which corresponds to himself. He must look into that bit of the model and identify there whether he is in a NICE or a NASTY condition. But it is not quite as simple as that. He has to identify which things are giving him a NICE feeling, and which are NASTY. And he has to predict his own future condition. He has to read the degree NICEness or NASTiness associated with each condition. Does he anticipate continuing to be ini a NICE condition (or NASTY) and why (what is causing these continuing conditions). Only then can he answer honestly.

NOTE: From time to time, critics of artificial intelligence have suggested that it would be possible to construct a talking robot, using only pre-stored text or based only on grammatical analysis. By assuming that a system can operate successfully, in that way, those critics are then able to show that the system lacks some important human characteristic, such as "understanding". I will be dealing with these ideas later, in the discussion, and in the section called "arguments". I will limit my comments here, to this - these are extremely silly ideas which have no validity whatever. Systems based on such simplistic mechanisms have no hope whatever or performing as their inventors claim. Any conclusions based on these systems are, therefore, also invalid. Briefly - If you start by making silly assumptions, it is not surprising if you reach silly conclusions. My system, although there would undoubtedly be technical difficulties to be overcome, would at least be furnished with the kind of information which would be needed to perform as required.

Self-reference and the Problem of Infinite Recursion
Self-reference is the name I will give to those processes by which the system consults or refers to the information stored in its representation of itself. There is, however, at least one aspect of its own internal mechanisms, which it can never represent inside that self-representation. That, of course, is that same self-reference mechanism. If it tried to represent that, it would need to represent itself representing itself representing itself ... and so on indefinitely.
There will be many other aspects of the system which will not appear in the self-representation. These will include most of the mechanisms of the lower layers. These sub-systems are not needed in the self-representation, for successful operation. To include them would waste space.
But the exclusion of the self-referencing system is an absolute requirement. There is no way it could ever be included without producing that endless loop, and with it, total system failure.
Consider what that implies. The system can refer to own internal circumstances. It can observe that the information in that self-MIND is being consulted. But it can never consult, or observe the mechanism of the self-referenciing system in action. The process of observation cannot be observed. The system can answer a question such as "Why are doing this?" where "this" is some action. It can answer such a question by identifying the causal precursors of the action concerned. But it cannot answer the question - "Why are you observing your own feelings?" The causal precursors for that, can never be centre stage, always they must hover in the unobserved wings.

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