An Expedient Mind: Building Bricks for a Mind Mechanism

AN EXPEDIENT MIND

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

Building Bricks

Components
In this section I will define a set of components with which the mechanism can be constructed. The specification of these is not sacrosanct. Any one of a wide range of alternatives could be used. Also, if an attempt was ever made to put the mechanism together with the units I have described, then I expect that some modifications would be found to be necessary. That does not concern me unduly. My purpose is to show that the system is feasible in principle. I do not claim to be able to dot every "i" and cross every "t".

The development of a mind mechanism

FIG: The three main components used in the construction of Layer-1.

Although the precise nature of the components is irrelevant, there is one requirement which they must all meet. It must be possible to see how we could build the components themselves. What we do not want, in any of them, is a bit of unspecified magic juice. So I will rely on computer technology to supply us with the components we need. We all know that computers can input, store, process and output information. We know that the messages generated by computers, can be used to control all kinds of physical apparatus - like games consoles, robot limbs and even aeroplanes. All I need to do, therefore, to stay within the bounds of what is obviiously feasible, is to specify the nature of the components in terms of what information they process and what functions they carry out. I do not need to specify anything at the level of cicuit diagrams.

The Hash-prefix Notation
One of the most basic of these components will be a unit which carries information between the other units of the system. To make it quite clear what this component does, I will called it "a signal". The use of ordinary words helps the reader (and myself) to remember what each component does. But there is a small snag associated with the use of ordinary words. The reader would not find it easy to tell when the word was being used to indicate a special component, and when it was being used with its ordinary connotation. To get round that problem, I will introduce a special symbol "#", which I will call the "hash-prefix". Henceforth, the word "signal" has exactly the same meaning as it always has had for the reader, When I want to talk about the component of that name, I will call it "a #signal".

The #Signal
The #signal is the first of our constructional units. Physically, it could be a train of pulses on a wire. But we could think of it as though it was a piece of paper with items of information written on it, or a shoe-box with things inside. What is important, are the items of information which it carries. Every #signal has a special unique identification number number. It has a source and a destination. It also has a magnitude (the strength of the signal). Those who know about the physiology of communication along nerve fibres, may protest that a pulse which is transmitted along a nerve fibre, is constrained always to have the same amplitude. Quite so. But the signals which are transmitted along natural nerve fibres, consist not of a single pulse, but of a whole train of pulses, the magnitude of that composite signal is indicated by the number and frequency of these pulses. I warn the reader, not to expect too strong a parallelism between my mechanism and any real biological mechanism. I intend that there should be a functional correspondence, but not a physical resemblance.
There are a few more items of information in each #signal, which I will list here, but which will be explained later. The structure of a #signal can be summarised this way -

#signal =
ID
SOURCE
DESTINATION
MAGNITUDE
CHANNEL
ASPECT

#Source and #Array
The next construction unit is the #source. A #source can be a single unit in isolation. But several will normally be grouped into an #array. It is easy to understand what kind of thing that is. The retina of the human eye, or the photosensitive array of a digital camera is a simple example of an #array. Each photosensitive cell within a retina is an example of a #source. Every #source has a unique ID and a unique location. If it is part of an #array, then that location refers to its location within the #array. Note please that a location is just another unique, but arbitrary identifier. There is no need to start thinking in terms of cartesian co-ordinates, or anything like that at this stage. We will, in due course, introduce a partial ordering of locations - for example a list like this ....

L1 > L2 > l3 > L4 > L5 > .....

.... where the symbols "L1", "L2" etc. are location indicators. A partial ordering like that, will gives us a handle on the relationships between these locations. We will know, or at least we will know that the system can tell, which locations are adjacent, and that one is "above" another, and so on.

#Perception
In 1957 Frank Rosenblatt, a research worker in the field of neural networks, invented what he called the "perceptron". This was an electronic component which processed several input signals, calculated the weighted sum of these, and then output another digital signal, the value of which depended upon whether or not that weighted sum of the inputs exceeded or fell below some pre-defined threshold value. In effect it was a large (weighted) "AND" gate which could be adjusted so that it was able to recognised certain complex features, like edges, lines and corners (in a visual field). Later, in the 1970s the idea fell somewhat from grace when it was shown that the unit, as specified by Rosenblatt, could not recognise certain unusual patterns of input signals. More recently however, the idea has been reinstated. It was realised that the specification could be modified to include a form of feedback and that would nullify these objections.
My #perception unit is a simple adaptation of that basic idea. I have not specifed the way it works in any detail, but have given it additional power. I replace the simple weighted sum recognition process, with a generalised but unspecified computer program. So the #perception is really a mini-computer. It has inpout signals, and output signals. It can store some information internally, and it can refer to that when it is processing the input signals. It also has that internally stored computer program which reads the input signals, compares them to the stored information, and then outputs one or more signals. It can assign to those output signals certain properties (like a magnitude) as appropriate. In this form, the #perception will do the task which I require of it. We know that computer programs can be written to do all kinds of clever tasks, so the fact that I do not bother to specify the program in detail, should not rob the idea of practicality.

The specification of the #perception is

#perception =
ID
INPUTS
RELATIONSHIPS
MATCHER-PROGRAM
OUTPUTS
PRIORITY
COUNTER

In the simplest case, the list of input #signals will be a list of unique ID numbers. A #perception can also specify a set of ID numbers of its own choosing and then test various groups of the input #signals to see if they are suitable candidates for the roles played by these ID numbers as specified by the relationship number. A simple example may help to clarify that. Say we have three input #signals (S1,S2,S3). Say also that a particular #perception is designed to recognise two #signals (designated A and B) which are adjacent, and with A being a much stronger #signal than B. This #perception can now be programmed to select the input #signals in pairs and put each pair to the test specified by the relationship data. It could be that two pairings meet the requirements (S1/S2 and S2/S3). The #perception would then output a pair of #signals, each of which would identify one of those pairings. The #signals output by each #perception can then be processed by more #perceptions. They can also be sent straight to a #response (described below). Or it can do both.
One departure from conventional systems of a similar nature, is the idea that a #perception can output a negative #signal, indicating that the required pattern of input #signals has NOT been recognised.
The item labelled PRIORITY, allows the system as a whole, to descriminate when two sets of input #signals, compete for attention. The system can respond preferentially to the #signal with the highest priority. This is equivalent to a simple creature being confronted by an approaching preditor, and a juicy morsel of food. In most circumstances, though not necessarily in all circumstances, the creature would chose to flee and not to feed. The assignment of these priorities, is yet again determiined by natural selection.
The item labelled COUNTER keeps a record of the number of times a #perception is able to recognise its designated pattern of input #signals. This is a simple device which allows the system, after a period of learning from experience, to purge itself of #perceptions which are never or very seldom used. This arrangement allows a given individual system to adapt rapidly to the prevailing environmental circumstances instead of waiting for the very slow processes of evolution over successive generations. It also avoids the system being overwhelmed by negative #signals.

#Response and Response #Array
The #response component is included to complete the system. I will not provide detail. While it is possible to have a single #response as the target for output #signals a typical response to a stimulus, will normally be a pettern of muscular actions. I envisage the #perceptions sending a group of #signals, which in effect, choreograph a sequence of #responses. It is for that reason that I have grouped the #responses into an #array (see the diagram at the top of this section).

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