I. INTRODUCTION
Evolutionary systems are of interest to both biologists and
computer scientists. For the former, they represent
challenging objects of study. For computer scientists, they are
both a source of inspiration and increasingly a domain within
which computational techniques are being profitably applied.
Traditionally, biologists have tended to use mathematical
methods to model biological systems. In the field of
behavioural ecology, for example, game theory has been the
preferred tool since it was first championed by Maynard
Smith and Price in the early 1970s [1].
Computer scientists have used simulation models to
explore similar systems with some success [2]. However,
their uptake within the biological community has been
limited by the perceived lack of rigour associated with
simulation in comparison to equational modelling techniques
[3]. Formal models are expected to explicitly present all of
their assumptions and to provide an intelligible account of the
moves leading to the model's conclusions. Simulations, on
the other hand, are often
opaque,
even to their designers [4].
Unlike the interpretation of mathematical models,
understanding why a particular simulation model produces
the results that it does is often a significant undertaking. As a
result, even when the actual details of the implementation are
made clear (which they are often not) by providing the source
code, for example, simulation models are difficult to assess
and appreciate. This sometimes leads to artefactual claims
[5]-[6].
Despite the success of game theory modelling, it suffers
from the same tractability limitations as all mathematical
modelling techniques. While simple games can be analysed
fairly easily, including more detail often renders models
effectively insoluble. This has not been a severe problem in
behavioural ecology partly because theoreticians have been
interested in quite simple games with few equilibria. These
simple models have thrown light on many important
questions, but further exploration often requires addressing
more complicated scenarios.
We believe that this is best accomplished by using game-
theoretical models as the basis for evolutionary computer
simulation modelling. Wedding the two approaches may
serve to overcome the limitations of each-the perceived lack
of rigour in the simulation design, and the tractability
constraints on mathematical modelling.
One might consider a game-theoretic model to represent
the
pure selective force
driving an evolutionary system. This
is because game theory assumes evolution takes place in a
world where even the smallest of selective forces will
eventually overcome any limitations [7]-[8]. The only
constraints on evolutionary change imposed by a game-
theoretic model are that strategies must be drawn from the
pre-defined strategy set. There are no genetic constraints, no
developmental limitations, no noise, no fitness landscapes to
traverse, etc., there are only evolutionarily stable strategies
(An ESS is defined as a strategy, or mixtures of strategies,
that, when prevalent, cannot be invaded by any others).
In contrast, an evolutionary simulation model immediately
introduces evolutionary constraints, many of which might be
described as "logistic" factors. These factors would include
stochastic effects such as noisy fitness functions or sampling
errors, and genetic constraints, which govern how far one
strategy is from another across a fitness landscape. In
addition, population structure, direct vs. indirect costs, life
history strategies and other influences are also often brought
to bear upon the evolutionary process being modelled. It is,
of course, possible to include these kinds of constraints in a
game theory model, but only at the cost of greatly
complicating the mathematics required to obtain a solution.
Importantly, a simulation modelling approach also enables
a researcher to study any
non-ESS
, transitory phenomena.
This is particularly important with systems that have non-
stable or multiple equilibria, as well as potentially revealing
system behaviours that might give an indication of an
evolutionary trajectory toward an ESS. Just because a
behaviour is not an ESS does not necessarily mean that it will
rarely be observed. Commonly occurring transitory
phenomena also require explanation.
In the remainder of the paper we will apply both game
theory and evolutionary simulation modelling techniques to
an important problem in biology. Our primary aim is to
demonstrate that the synthesis of both techniques provides a
better understanding of the problem than game theory alone,
but we also aim to contribute to the biological literature on
the evolution of honest signalling systems.
II. THE EXAMPLE PROBLEM
One problem that has been studied extensively through
both game-theoretic [9]-[10] and evolutionary simulation
modelling [11] approaches is the discrete action-response
game, which models communication between individuals
who may or may not have a conflict of interest. In the
majority of cases, only one modelling technique was used.