Introduction: Genetic Algorithms in Visual Art and Music

Leonardo has a long tradition of publishing work at the mutual frontier of art and science. We are pleased to present in the pages of Leonardo a special project that demonstrates the application of ideas from science (evolutionary biology), technology (computing) and art (both visual art and music). The topic of this special section is the application of genetic algorithms and related heuristics to visual art and music. These texts were first presented at the Genetic Algorithms in Visual Art and Music workshop at the 2000 Genetic and Evolutionary Computation Conference (GECCO) in Las Vegas [1]. The texts will be published in two installments in Leonardo-the first installment here and the second set of papers in Vol. 35, No. 4 (August 2002).

Genetic algorithms, invented by John Holland in the 1970s [2], are part of a heuristic method that abstracts the processes found in biological evolution and simulates them on a computer. However, instead of being used to simulate real biology, genetic algorithms are used to solve problems in many non-biological domains.

A typical use of genetic algorithms is in optimization, where we want to search some virtual space for the individual (in this case, either an image or a piece of sound/music) that scores highest in some selected measure. The genetic algorithm first generates a random set of individuals drawn from the set of possible individuals that could exist (this set is known as the search space). In the context of an artistic system this could be the generation of random sounds or images. Typically, these individuals are represented as binary strings, which facilitates

Fig. 1.

Genetic algorithms-a summary. The genetic algorithm consists of three main stages. Members of the population represent solutions to the problem at hand. In the first stage the quality of these solutions is measured, and the best ones chosen to form the basis for the next generation. Members of this chosen group are then crossed over, exchanging information. Finally a local mutation process is carried out, which makes minor changes to the individuals. (© Colin Johnson)

the application of the crossover and mutation operators described below. After the best individuals in the population are selected (either by a human, a computer or a combination of the two), they exchange information with other individuals (a process known as crossover) and small changes are then made in the individuals (i.e. they undergo mutation), resulting in a new population. This process is repeated (see Fig. 1) either until a satisfactory result is found or until the population reaches a state where the individuals are all very similar to one another. In this latter state there is no variation left for evolution to work on-it is said that that population has converged.

Procedures such as this have proven to be a powerful way to search many different kinds of search space, with many real-world applications, such as timetabling, scheduling, operations research problems, design of control systems, computer-aided design of architectural structures, parameter setting in neural networks, and data mining [3]. A number of other techniques, such as cellular automata [4], artificial life [5] and autonomous agents [6], have similar "flavors" and have also been used in artistic and musical areas.

A number of researchers have investigated the use of genetic algorithms in artistic domains. The Genetic Algorithms in Visual Art and Music workshop was held to review the state of the art in this area. This workshop included presentations of papers, general discussions of the topics presented and demonstrations of art and music created using these systems, including a live performance by John Biles and his GenJam system. The articles presented here are extended versions of the papers presented at the workshop.

In this introduction, we begin by providing some general background to this area and a review of prior work on these topics. One of the aims of the workshop was to consider whether there are any general principles or major open questions that underlie this field of research. Some of...