Chapter 4 . Applications of Evolutionary Computation
63
This was an extension of 'Tierra', with spatial structure based on a grid. The
results of this work demonstrated that evolving individuals with spatial structure
showed some of the observed properties from real systems, such as power laws
and self-organised criticality (Bak 1996). However, it should be noted that
although a model shows self-organized criticality, or some other property that can
be interpreted as ecological based, it is not necessarily useful as a model of an
ecosystem. Few works based on these complex system approaches have been able
to produce theoretical predictions that have been testable with real ecosystems
(however, see Seetion 4.4.6 for one example).
Extending this concept to communities, GP has been used to demonstrate the
evolution of cooperative behavior (Koza 1994). This work showed that high-level
cooperative behavior of a community of ants, operating in parallel and with only
local sensing, emerges by evolving each individual. Other work has used the
behavior of real ants to construct systems that solve optimization problems
(Dorigo and Caro 1999). These works relate to community ecology, where the
structure and function of a community can be shown to evolve based on a simple
task.
Simulated evolution has also been used to study the interactions of coevolving
individuals within a population (for example (Kaufmann and Johnson 1992;
Angeline and Pollack 1993; Fogel 1993; Ashlock et al. 1996». Allowing the
fitness function to depend on the constituents of the population, rather then being
a fixed measure against a problem, causes the population to coevolve. Many of
these simulations are based on an idealized model of interaction referred to as the
interated prisoner's dilemma (!PD). !PD describes the interactions between
individuals in a competitive environment, where there are varying payoffs based
on whether two individuals cooperate, defect or some combination. !PD has been
successful in modelling the evolution of cooperative strategies between
individuals in a community, and is a useful theoretical model of social interaction.
For example, Ashlock et al. (1996) studied partner selection as a process in social
interactions. The models showed that, based on the degree to which individuals
are intolerant of defections and social isolation, various ecologies dominated.
However, like many theoretical studies, this work has not been extended to
predictions of real ecosystem structure, although there is clearly an opportunity to
use this form of modelling for predictions of population structure, cooperative
behavior and more complex interactions such as language development and
symbiosis.
Extending this concept to community assemblages has been achieved using the
Echo model (Forrest and Jones 1994; Hraber and Milne 1997), where a set of
agents coevolves under the press ure of invasion and agent interaction. This work
allowed a study of species abundance patterns, community assembly rules, species
richness and ecosystem stability. Other work using a genetic approach has allowed
a model of plant-herbivore interactions (Hartviggsen and Starmer 1995). In this
work the plants have simulated genes that infer a certain resistance to grazing, and
the herbivores have simulated genes that produce conditions that overcome these
63
This was an extension of 'Tierra', with spatial structure based on a grid. The
results of this work demonstrated that evolving individuals with spatial structure
showed some of the observed properties from real systems, such as power laws
and self-organised criticality (Bak 1996). However, it should be noted that
although a model shows self-organized criticality, or some other property that can
be interpreted as ecological based, it is not necessarily useful as a model of an
ecosystem. Few works based on these complex system approaches have been able
to produce theoretical predictions that have been testable with real ecosystems
(however, see Seetion 4.4.6 for one example).
Extending this concept to communities, GP has been used to demonstrate the
evolution of cooperative behavior (Koza 1994). This work showed that high-level
cooperative behavior of a community of ants, operating in parallel and with only
local sensing, emerges by evolving each individual. Other work has used the
behavior of real ants to construct systems that solve optimization problems
(Dorigo and Caro 1999). These works relate to community ecology, where the
structure and function of a community can be shown to evolve based on a simple
task.
Simulated evolution has also been used to study the interactions of coevolving
individuals within a population (for example (Kaufmann and Johnson 1992;
Angeline and Pollack 1993; Fogel 1993; Ashlock et al. 1996». Allowing the
fitness function to depend on the constituents of the population, rather then being
a fixed measure against a problem, causes the population to coevolve. Many of
these simulations are based on an idealized model of interaction referred to as the
interated prisoner's dilemma (!PD). !PD describes the interactions between
individuals in a competitive environment, where there are varying payoffs based
on whether two individuals cooperate, defect or some combination. !PD has been
successful in modelling the evolution of cooperative strategies between
individuals in a community, and is a useful theoretical model of social interaction.
For example, Ashlock et al. (1996) studied partner selection as a process in social
interactions. The models showed that, based on the degree to which individuals
are intolerant of defections and social isolation, various ecologies dominated.
However, like many theoretical studies, this work has not been extended to
predictions of real ecosystem structure, although there is clearly an opportunity to
use this form of modelling for predictions of population structure, cooperative
behavior and more complex interactions such as language development and
symbiosis.
Extending this concept to community assemblages has been achieved using the
Echo model (Forrest and Jones 1994; Hraber and Milne 1997), where a set of
agents coevolves under the press ure of invasion and agent interaction. This work
allowed a study of species abundance patterns, community assembly rules, species
richness and ecosystem stability. Other work using a genetic approach has allowed
a model of plant-herbivore interactions (Hartviggsen and Starmer 1995). In this
work the plants have simulated genes that infer a certain resistance to grazing, and
the herbivores have simulated genes that produce conditions that overcome these