Collaboration with Thomas Ray

Thomas Ray’s goal is an independent evolution of computer programs with as few defaults from the outside as possible. In a technical environment, in which the only goal of programs is their own reproduction, and only limited system resources such as processor time and storage location are available, accord­ing to Ray’s view, programs will develop independently into complex problem solutions, which can be observed by humans and used for our purposes. An as­sumption with regard to the kind of the developing programs cannot be made, since evolution leads to new solutions that exceed our imaginative power. Thus, only pure principles of evolution are permissible, like self-reproduction and “natural” selection with a system-inherent fitness function.

genetic algorithm ^ His system “Tierra” is a virtual machine and an embedded operating system that assigns computing time to programs, and knows different virtual machine instructions that can be implemented by the programs. Tierra was started with programs that copied themselves into other memory areas, and then introduced them to the operating system as new programs. During the execution randomly erroneous operations are introduced. If this happens during a copying process, the program is changed randomly, and thus corresponds to a mutation in nature. In the competition for computing time, different patterns of programs devel­oped, such as those that use the computing time of other programs to multiply themselves (parasites), or self-optimizations of programs regarding the use of computing time, such as, for example, the linearization of loops.

Ray also worked at the Evolutionary System Department at the ATR Labora­tory, and describes in [170] a qualitative classification of evolutionary systems based on genetic languages, interaction and fitness functions. He differentiates the following:

■ Genetic Algorithms (GAs) Versus Genetic Programming (GP)

In genetic algorithms a vector – in the simplest case a bitstring – is changed by parameters and optimizes a fixed problem solution parametrically, while in genetic programming the solution strategy is changed in the form of an executable program within an evolutionary process. Here, principal new forms of solutions can develop (emergence).

■ External Versus Internal Fitness Functions

In a system with external fitness functions the fitness is given from the out-

side and independently of system parameters, and remains constant, thus Section 12.4 providing the problem. Internal fitness functions arise as a result of the in – Bill Viola, Tree of Knowledge

teraction of the system parameters, i. e., change during the interaction of in­ternally participating “individuals”, and thereby can create new problems.

A special form is an external self-changing fitness function, for example one operating through the aesthetic selection of a viewer interacting with the system, as found in Karl Sim’s “Genetic Images” and “Galapagos”.

■ Interaction of Entities of the System

Individuals can promote or impair their reproducibility and thereby mutu­ally affect the fitness function. The complexity of the system is not linear.

Using this classification the genetic codes of the individuals in A-Volve cor­respond to genetic algorithms. The developing forms can thus vary greatly; however, they always move within the genetic code of the interpreting algo­rithm. Thus, no qualitatively new forms or behaviors can develop. According to Ray, there is an interaction between the individuals of the system. The fit­ness function corresponds to the reproduction ability of the individuals, with influence from the inside through interaction with other individuals, and with outside through the human viewer.

In its function as an exhibit piece, A-Volve shows the principles specified above, and refers thereby to the creative potential of evolutionary processes that can be used in the artistic context within a technical medium. It is above all an open system that does not create in principle new solutions by itself, but rather from the concept, the technical expertise of the expert, and the develop­ment of new forms by the viewer.

In 1994 A-Volve won the golden Nica of the renowned Prix Ars Electronica. The first versions of A-Volve ran on an SGI Onyx; in the meantime a PC with a standard graphics board has become sufficient. The hand movements of the viewer in the water are determined by a self-developed camera recognition system.

Updated: October 12, 2015 — 12:11 am