No Title

Selecting Naturally for Differentiation:
preliminary evolutionary results
Thomas S. Ray
ATR Human Information Processing Research Laboratories
2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto, 619-02, Japan
Tel: 81-774-95-1063 Fax: 81-774-95-1008
ray@isd.atr.co.jp, ray@santafe.edu
http://www.isd.atr.co.jp/ray/

Abstract:

A progress report on an effort to create conditions under which natural selection will favor cell differentiation in multi-threaded (multi-cellular) self-replicating machine code programs. ``Cell differentiation'' means that different threads of the process execute different code (express different genes). The machine code algorithms exist in a networked environment in which they are able to move between machines. They have access to sensory data on network conditions, which can be used to support strategies for foraging for CPU cycles. The network ancestor is differentiated into reproductive and sensory tissues.

Evolution in the medium of organic chemistry on Earth has produced a rich diversity of life forms which are the subject of study of the science of Biology. However, our Biology is based on a sample size of one, in that there is only a single evolutionary tree on Earth, with all living organisms sharing a common ancestor.

If we may consider evolution itself to be the signature of life, then all genuine examples of Evolution should fall within the domain of Biology. In addition to evolution in the organic medium on other planets (and Earth), Biology should also consider evolution in any medium that can support it, such as in the digital medium on Earth (or elsewhere).

Unfortunately, evolution on other planets remains beyond the reach of our observations. At the same time, digital evolution on Earth, while within range of our observations, remains so primitive that many biologists do not consider it worthy of their attention.

In fact, the problem goes deeper. Because everything we know about life is based on a single example, our every conception of life is highly parochial. For most people, it is impossible to separate our concept of life from the material in which our one example is embedded. Organic, wet, material, etc., are integral parts of our conception of what life is. It is difficult for us to admit an example of life which lacks most of the familiar forms.

This author considers life to be a process which can potentially inhabit various media, not only organic chemistry. And evolution is considered to be both the creative process and the defining property of life. Thus it is presumed that it is valuable and interesting to study the process of evolution regardless of the medium in which it is embedded.

This work began with an experiment known as ``Tierra'', in which self-replicating machine code programs evolved into a diverse ecological community, and exhibited a variety of adaptations and optimizations [6, 7, 8]. Yet the replicators in Tierra exhibited only modest evolutionary increases in complexity, whereas on Earth, evolution in the organic medium has caused replicators to undergo a large increase in complexity.

While it is not surprising that digital evolution has not produced a richness comparable to that of organic evolution, the reasons for the difference are not certain. Certainly digital evolution has not yet operated on adequate scales of time and space. But perhaps the digital medium itself is a poorer medium which will never be able to support such a rich evolution. More work is needed to explore the limits of digital evolution, and the conditions under which those limits can be reached.

We may be able to look to our one example of rich evolution, life on Earth, for clues to how we may enhance the richness of digital evolution. The work presented in this paper is based on a fairly elaborate metaphor between organic and digital life. The objective is not to create a digital model of organic life, but rather to use organic life as a model on which to base our better design of digital evolution.