Reading 1.5: Defining modelling
A model is always made for a purpose, and whatever that purpose is decides what aspects of the real entity are represented in the model. The aspects of the car which are modelled in a toy car and the aspects of the human modelled in a doll will vary depending on its purpose.
A model is a model because its form, the structure and organisation of its parts match in some way the form, structure and organisation of the parts of the entity that is modelled. The model is a simplification – that is some, but not all, aspects of the entity represented are present in the model.
The idea of a mental model is that it is just such a representation of an external entity stored within the brain or nervous system of a human. The notion of mental models as a way of modelling the thinking process was first proposed by Kenneth Craik (Craik, 1943) in his book, The Nature of Explanation, and is now a well established set of ideas (Johnson-Laird, 1983). The significance of Craik's proposal is explained as follows. Imagine a small robot, perhaps a tricycle, with a steerable front wheel, able to move around on the surface of a table.
There are two ways of building the robot so that it does not fall off the table. I will call them Mode 1 and Mode 2, as illustrated in Figure 3. The first and the simplest strategy is to build a detector on the front of the robot, which can detect the edge of the table when the robot reaches it. Then as the robot moves forward, this device continuously checks for the edge of the table. On detection the front wheel is caused to turn so that the robot moves away from the edge. This is the mechanism that most of us would come up with if asked to design such a robot, but there is an alternative, Mode 2. The second way is to build a more elaborate robot. On the tricycle we construct a model table, and on the model table a model robot. The model robot is geared to move across the model table at a proportionate rate, and so the robot 'knows' when it nears the edge of the table and can steer accordingly. Perhaps it is not an easy task to design the mechanical linkages required as I have described it, but with today's electronics and computing capability this could certainly be done. As long as no one knocks the robot, and all remains aligned, the robot will stay on the table as we require. Of course in time, with inherent errors accumulating (e.g. someone knocks the robot or the table) eventually the table and model table will become misaligned. So it is necessary to incorporate into our design something from Mode 1 to check occasionally that the table and model table remain aligned.
The information processing requirements of these two techniques are quite different. Mode 1 requires a large ability to process information from the outside world and minimal internal processing. Mode 2 requires a minimal ability to process information from the outside, but a sophisticated model building apparatus internally; all the evidence points to nature adopting a heavy reliance on the second strategy in humans. The information carrying-capacity of our external sensors is much less than that required to continuously-monitor our outside world. Incoming signals collected from the outside world using the five senses are used to build models over time, and the models are then used to process the incoming signals from the outside world. This system brings advantages, the most important being the ability to use relatively crude outside sensors, but then to be able to elaborate the information collected internally using accumulated experience, and therefore to be able to react quickly to complex outside threats.
As always, nature has developed a brilliant solution to a difficult problem, but the solution does have a downside. Sometimes our brains pick the 'wrong' model – an extreme example could be a police marksman seeing a criminal carrying a gun, rather than a law abiding citizen carrying a table leg. Your preconceptions, your expectations framed in the model you are using, determine what you perceive, and what you perceive determines how you respond. Acting too quickly driven by the wrong model can have dire consequences. Psychologists are coming to understand that this effect plays a crucial role in some driving accidents.
So far we have been thinking only in terms of humans creating models and modelling their environment in order to deal with it and survive. As humans, we have taken control of our environment to extraordinary lengths; we can now build air-conditioned or centrally heated buildings, and elaborate distribution systems to supply everything from food to entertainment.
But this way of thinking has much wider implications. Modelling to regulate and control our environment is an example of the more general problem of regulation and control. In 1970 the notion that modelling was at the heart of regulation and control was formalised. The Conant-Ashby Theorum (Conant and Ashby, 1970) fixed this idea of modelling in the sense that has been described, as essential to any understanding of regulation and control, and therefore essential to any organism in regulating its environment in order to survive.
Therefore, from the Craikian robot's point of view the model of the table provided is enabling it to survive in its table environment. You can see that from this point of view, within any living organism there are models of the relationship of that individual organism to its environment. This is the way in which an organism seeks to regulate this relationship, to survive and to make existence comfortable. In the simplest of cases, an animal having the ability to move from one place to another means it can change its environment. When threatened, a frog simply needs to hop away.
But even in the simple situation of the frog, it must be able to perceive and recognise a potential threat. For example, if something approaches it must be able to discriminate between a potential mate and a potential threat. It needs to model the difference. It used to be that the autopilot built into all airliners kept the airliner flying straight and level without any attention from the pilot, but it had no model of the aircraft's relationship to terrains such as mountains. To avoid this danger the pilot, who did have the model, had to intervene. This is now built in and autopilots have models of the land they're flying over. Together with GPS this is really useful for avoiding mountains. So being able to survive in a particular situation means having an internal model of that situation – if there is no internal model then the situation cannot be perceived and appropriate action cannot be taken.
So now you see that in order to interpret signals that are significant to it, an organism must contain internal models that enable it to perceive, interpret, and act upon those signals. Therefore, you can extend the definition of an internal model to encompass the internal organisation of a system which enables it to recognise and respond to a particular pattern of signals from the environment falling upon its receptors.
This internal organisation may involve the whole system or just a small part of it, and may be simple if the pattern to be recognised is simple, or complex if the pattern is complex.