Models in the Brain

Dan Ryder (UNC Chapel Hill)


There are many different kinds of representations: pictures, maps, words, graphs, diagrams and more. I present evidence that the kind of representation that typifies the cognitive/perceptual part of the brain, the cerebral cortex, is model representation. A model M is a structure that represents another structure E because M is designed to be isomorphic to E. M has the function of being isomorphic to E, or mirroring E, and the structural nodes {m1, m2,} of M have the function of corresponding to the structural nodes {e1, e2, e3...en} of E. When this is the case, we say that {m1, m2,} stand in for and so represent {e1, e2, e3...en}.

Models can be produced by machines. A simple example would be a machine that takes an object as input, produces a mould from that object, then shrinks the mould and eventually produces a miniature model of the original object from that mould. In this case, there is no intentional agent who designed the model directly, so in deciding what the miniature represents, and what its parts have the function of corresponding to, we cannot consult the intentions of such a designer. Instead, we consult two things: 1) the general operating principles according to which the model building machine was designed, and 2) the object that acted as a template for the machine to produce the model in accordance with its operating principles. Thus if both the original object and its miniature model come to have a gouge in corresponding positions on their left sides, but only by accident, we do not say that the gouge on the model represents the gouge on the original object. Only features of the model that are produced by the model building machine from the original object in accordance with its general operating principles represent the corresponding features in the original object. This is not to say that a model building machine can produce only perfect models. On the contrary, the machine's operating principles may face limitations in certain situations, such that although the machine is functioning according to its operating principles, it may produce an imperfect model.

Recent evidence suggests that the cerebral cortex may well be a very general purpose model building machine, designed through natural selection to produce dynamical models under guidance from the environment. I outline the principles under which the cortical model building machine may operate, based upon the SINBAD theory of the cortex. These principles allow us to pick out what kinds and individuals served as "templates" for the cortex to produce a model of some portion of the environment, and thus to determine what particular cells or groups of cells in the cortex have the function of corresponding to, i.e. what they represent. (Unlike in the simple mould-plus-shrinking device described above, it is usually a determinate matter whether it is an individual or rather a kind that has guided the cortex to produce a particular model.) Since this mechanism would develop representations of just the sorts of things that we represent, there is reason to suppose that it reveals the nature of mental representation, at least in our kind of mind.

A dynamical model of the sort that the cortex constructs is a useful thing. Since it operates isomorphically to some dynamical structure in the environment, it can "fill in" missing information when that information is not directly available. For example if the thermometer in the Space Shuttle's hull is broken during atmospheric re-entry, a model of the Space Shuttle may be used to discover what the hull temperature is simply by plugging the known variables (velocity, atmospheric density, angle of descent etc.) into the model and reading off the temperature from its corresponding node in the model. This "filling-in" ability of models may also be used to discover the best action to take in a particular circumstance. For instance, one could discover the velocity below which one would need to travel in order to avoid exceeding a certain maximum hull temperature. One would simply plug in that hull temperature, and the atmospheric variables, and read off the maximum velocity from its corresponding node in the model. These uses of a model provide the basis for an account of both occurrent belief (i.e. judgement) and occurrent desire.