Page 12 -

1 Introduction 5

1947, 1976; Sent 1997). With the advent of computational simulation, it is now
fairly common to represent the cognition of agents in a model with a series of rules
or procedures. This is partly because implementing substantive rationality is often
infeasible due to the computational expense of doing do, but more importantly it
seems to produce results with a greater “surface validity” (i.e. it looks right). It turns
out that adding some adaptive or learning ability to individuals and allowing the
individuals to interact can often lead to effective “solutions” for collective problems
(e.g. the entities in Chap. 23). It is not necessary to postulate complex problem-
solving and planning by individuals for this to occur.
Herbert Simon observed further that people tend to change their procedure only
if it becomes unsatisfactory; they have some criteria of sufficient satisfaction for
judging a procedure, and if the results meet this, they do not usually change what
they do. Later Simon (1956) and others (e.g. Sargent 1993) focused on the contrast
between optimisers and satisficers, since the prevailing idea of decision-making was
that many possible actions are considered and compared (using the expected utility
of the respective outcomes) and the optimal action was the one that was chosen.
Unfortunately it is this later distinction that many remember from Simon, and not the
more important distinction between procedural and substantive rationality. Simon’s
point was that he observed that people use a procedural approach to tasks; the
introduction of satisficing was merely a way of modelling this. However, the idea
of thresholds, which people only respond to a stimulus when it becomes sufficiently
intense, is often credible and is seen in many simulations (for some examples of
this, see Chaps. 24 and 27).
Along with Alan Newell, Simon made a contribution of a different kind to the
modelling of humans. He produced a computational model of problem-solving
in the form of a computer program, which would take complex goals and split
them into sub-goals until the sub-goals were achievable (Newell and Simon 1972).
The importance of this, from the point of view of this book, is that it was a
computational model of an aspect of cognition, rather than one expressed in
numerical and analytic form. Not being restricted to models that can be expressed
in tractable analytic forms allows a much greater range of possibilities for the
representation of human individual and social behaviour. Computational models
of aspects of cognition are now often introduced to capture behaviours that are
difficult to represent in more traditional analytic models. Computational power is
now sufficiently available to enable each represented individual to effectively have
its own computational process, allowing a model to be distributed in a similar
way to that of the social systems we observe. Thus, the move to a distributed and
computational approach to modelling social phenomena can be seen as part of a
move away from abstract models divorced from what they model towards a more
descriptive type of representation.
This shift towards a more straightforward (even “natural”) approach to modelling
also allows for more evidence to be applied. In the past, anecdotal evidence, in the
form of narrative accounts by those being modelled, was deemed as “unscientific”.
One of the reasons that such evidence was rejected is that it could not be used to