ABSTRACT
But, models are not reality. We began this text by discussing different types of models, and these models can be used to help us understand reality. Yet, at every step, when we build a model, we make a simplification of the thing being modelled. Each model is an approximation. In the words of George Box: “All models are wrong, some models are useful.” It is not that the models lie, merely that they are economical with the truth, and the art in appropriate application of models is in finding the best economies. The model should be simple enough to be useful, but not so simple that it no longer reflects the interesting parts of reality. In modelling neuronal dynamics (Chapter 6), for example, we seek models
that are capable of expressing dynamical behaviour consistent with a single neuron. We do not (yet) have the knowledge or technical tools necessary to build a model of human consciousness. Yet if we wish to learn about the function of individual isolated neurons, then that level of detail is unnecessary. Conversely, the models of disease transmission introduced in Chapter 7 are
perfectly adequate for some diseases, but not others. They model influenza epidemics very well, but are unable to explain the data observed for SARS in Hong Kong. More complex models, using complex networks, are able to explain many of the features of transmission of SARS (Chapter 10). But these models too, are economical with the truth. In 2009 a new strain of swine flu (H1N1 influenza) spread from Mexico and raised global concern. In June of 2009, the World Health Organisation declared it a pandemic. Fear, confusion and panic spread as health experts and computational models predicted a dangerous and severely infectious global wave of infection. The models were wrong. While the H1N1 strain of influenza did go on to infect millions of people globally, and led to deaths in excess of 15,000, this should be compared to a background of several hundred thousand deaths from all other forms of influenza. But, in this case, the model is only wrong because the parameters (that is, the data which drive the model) was not accurate. For the parameter values which were employed, the behaviour of the model was entirely accurate, and, in the future a similar infectious outbreak could still illicit the response predicted by the model.
