Genetic Algorithms and Artificial Life

A genetic algorithm (GA) is a technique which mimics natural evolution, made famous by Holland's seminal work Adaption in Natural and Artificial Systems [16]. GAs mimic evolution by populating a gene pool with potential solutions to a problem--these are known as `chromosomes.' These chromosomes are traditionally encoded as a bit string, however more complex representations are often used. Selection of pairs of solutions--the parents--is usually based on some fitness function, or rating of how `good' the parent solutions are. This is followed by crossover and mutation, resulting in a `child' chromosome. In the case of a bit string representation this is usually performed by choosing a crossover point in each string and concatenating the head of one chromosome with the tail of the other. Finally, mutation of the resultant chromosome is performed. For bit strings, this consists of stochastic flipping of the bit values (see [11, Section 20.8]).

A variation of the above technique is known as a steady-state genetic algorithm. In this case, the `solutions' are imbued with the ability to choose when to reproduce: there is no explicit fitness function, selection is analogous to Darwinian [6] natural selection is performed. The concept known as `artificial life' is an extension of this. Artificial life involves autonomous agents who have a variety of perceptors and effectors with which they manipulate their environment. These agents can then, through evolution via GAs and/or individual learning, discover how to survive in their given environment. Artificial life has been used to evolve intelligent behaviour (eg. [17]). Also, Mascaro et al. [18] used an artificial life environment to successfully evolve altruistic suicide in autonomous agents. These uses suggest that artificial life may be of use in studying emotion.