Page 13 -

6 B. Edmonds and R. Meyer

help specify or evaluate formal models; such narrative evidence could only be used
within the sphere of rich human understanding and not at the level of a precise
model. Computational simulation allows some aspects of individual’s narratives to
be used to specify or check the behaviour of agents in a model, as well as the results
being more readily interpretable by non-experts. This has let such computational
simulations to be used in conjunction with stakeholders in a far more direct way
than was previously possible. Chapter 12 looks at this approach.
Herbert Simon did not himself firmly connect the two broad strands of his
work: the observation of people’s procedures in their social context and their
algorithmic modelling in computer models. This is not very surprising as the
computational power to run distributed AI models (which are essentially what
agent-based simulations are) was not available to him. Indeed these two strands
of his work are somewhat in opposition to each other, the one attempting to
construct a general model of an aspect of cognition (e.g. problem-solving) and
the other identifying quite specific and limited cognitive procedures. I think it
is fair to say that whereas Simon did reject the general economic model of
rationality, he did not lose hope of a general model of cognitive processes, which
he hoped would be achieved starting from good observation of people. There
are still many in the social simulation community who hope for (or assume) the
existence of an “off-the-shelf” model of the individuals’ cognition which could
be plugged into a wider simulation model and get reasonable results. Against any
evidence, it is often simply hoped that the details of the individuals’ cognitive
model will not matter once embedded within a network of interaction. This
is an understandable hope, since having to deal with both individual cognitive
complexity and social complexity makes the job of modelling social complexity
much harder—it is far easier to assume that one or the other does not matter
much. Granovetter (1985) addressed precisely this question arguing against both
the under-socialised model of behaviour (that it is the individual cognition that
matters and the social effects can be ignored) and the over-socialised model
(that it is the society that determines behaviour regardless of the individual
cognition).
Herbert Simon did not have at his disposal the techniques of individual-
and agent-based simulation discussed in this handbook. These allow the formal
modelling of socially complex phenomena without requiring the strong assumptions
necessary to make an equation-based approach (which is the alternative formal
technique) analytically tractable. Without such simulation techniques, modellers
are faced with a dilemma: either to “shoehorn” their model into an analytically
tractable form, which usually requires them to make some drastic simplifications
of what they are representing, or to abandon any direct formal modelling of what
they observe. In the latter case, without agent-based techniques, they then would
have two further choices: to simply not do any formal modelling at all remaining in
the world of natural language or to ignore evidence of the phenomena and instead
model their idea concerning the phenomena. In other words, to produce an abstract
but strictly analogical model—a way of thinking about the phenomena expressed