Introduction to Artificial Life

Citation

Andrew, A.M. (1999), "Introduction to Artificial Life", Kybernetes, Vol. 28 No. 2, pp. 239-242. https://doi.org/10.1108/k.1999.28.2.239.1

The Preface begins with the question: “What makes living systems alive?” and goes on to point out that our level of understanding of biological systems is such that it is difficult to answer even this fundamental question. In contrast, our understanding of non‐living systems has advanced dramatically. The “Artificial Life” approach is meant to clarify the basic principles of the biological sciences and to bring them more nearly into line with the physical sciences.

To clarify the essential characteristics of life, its origins are considered, probably as an RNA‐world before the emergence of cells. In the RNA‐world, molecules replicated directly, presumably in a primeval soup. It has been shown that the spontaneous formation of amino acids is reasonably likely under conditions that can be assumed to have existed on the earth. The significant step is the appearance of molecules capable of replication. Research is mentioned that may reconstruct a molecule, RNA replicase, from which all life is descended, a molecule that has been extinct for three billion years.

Although the RNA replicase molecule is extinct in its original form, an important result of replication is the preservation of molecular structure, and its evolutionary enhancement from a survival point of view, over periods enormously greater than the time an individual molecule would survive. The latter may be a matter of hours or days, but the main features of the RNA replicase molecule have survived over these three billion years, and we trust they are not about to expire now. In a (slightly tentative) definition on page six of the book, this feature of persistence of information in the face of disruptive forces is postulated as the essential characteristic of life.

There is a review of computer programs aimed at simulating biological evolution, including one that produces designs for physical “bodies” capable of swimming, etc. The possibility of replication was studied by von Neumann in the context of cellular automata, and Conway’s game of “Life” is an interesting example allowing a form of replication.

The main focus of the book, however, is on populations of computer programs capable of replication, and subject to some kind of mutation as well as to overall culling by a “reaper” so that there is effective competition. Computer viruses are an example of self‐replicating programs, which have evolved to more powerful forms during their period of existence. Their evolutionary development, however, is not a valid model of biological evolution. It is not attributable to random mutation and selection but rather to the purposeful intervention of highly ingenious malicious human programmers.

Another example of a life‐like situation in a machine is the game of “Core Wars” in which the players write programs that are aimed at taking over space in computer memory, the various programs competing with each other for space. To run on an existing computer, the programs are written in a specially‐devised language and are restricted to a specified section of storage.

A system that convincingly simulates aspects of biological evolution is given the name tiera. In this, as in Core Wars, programs compete for space but are now subject to mutation. The instruction set differs from that used in Core Wars and is designed to be less “brittle”. A “brittle” program is one whose behaviour is likely to be drastically affected by a mutation, compared to one which moves to a new mode that is different but not drastically so. A “reaper” operates to limit the overall population.

Given the simple instruction set allowed in tiera, it was possible to find a minimum length of program allowing replication. It was found in operation, however, that shorter programs emerged, apparently with the capability of replication. Closer examination showed that these were effectively parasites whose replication depended on plugging‐in to instruction sequences in other programs. Short programs, and hence the parasites, have a reproductive advantage, but only so long as there is also a population of susceptible “host” programs. At a later stage in the simulation the numbers of these parasite programs declined, and it was found that the longer programs had evolved a means of preventing parasitic use of their replication facility. The correspondence to biological interaction is compelling.

The book introduces a comprehensive facility for this sort of modelling, with the name of avida. The necessary software is provided on the CD‐ROM included. It differs from tiera in a number of respects, one of them being that a larger instruction set is provided. The set provided in tiera is insufficient to achieve universality of computational function, and it is felt that a set allowing this is preferable. The user of the avida software can specify the size of the instruction set used in any experimental run, but the minimum set allowing universality is already considerably larger than that of tiera.

The exciting results obtained from tiera were greatly facilitated by its small instruction set, which meant that the number of possible sequences of instructions, of sufficient length to be interesting as possible self‐replicators, is large but not astronomical. For avida the space of possibilities to be sampled is much larger and experimental runs definitely have to be seeded with one or more “mother” replicators. The instruction set can be referred to as the chemistry of the system, analogous to the amino acids to be incorporated in gene sequences.

The avida system differs from tiera in its general arrangement, in that all of the putative programs in tiera share one linear address space, but in avida they are like distinct organisms, and can have a spatial distribution. This allows spatial phenomena in avida, such as diffusion and waves of change across the population. The avida software allows the user to set up experiments with control of the instruction set, or “chemistry”, and of such things as the “mother” replicators installed, the mutation rate and the type of mutations allowed. The viability of a program type is determined partly by its rate of reproduction, but is also influenced by a user‐specified “fitness landscape”. Needless to say, the package includes comprehensive facilities for collecting and processing data from the runs.

Besides introducing the avida system, and providing an Appendix that is a users’ manual, the book deals with theoretical approaches that allow analysis of the results and comparison with those on true biological systems. One of these is Shannon’s Information Theory, where it is emphasised that the optimal communication between programs is not perfect uncorrupted transmission but transmission introducing the right amount of entropy.

Other connections with thermodynamics are developed in another chapter, and yet another discusses the spatial concept of percolation through a regular array of elements of which a subset is tagged in some way. The problem is to assess the probability of finding a path through adjacent tagged elements that links opposite faces of the array. In one dimension the problem is trivial as the path is only formed if all elements are tagged, but there is a challenging problem when two or more dimensions are considered. The theory is shown to be relevant to analysis of evolutionary processes.

A type of behaviour that can be observed in evolving systems, living or artificial, is described as “self‐organisation to criticality”, in which a succession of epochs of self‐organisation reach a kind of stagnation and crumble away. In a later chapter it is argued that particularly interesting adaptation occurs when the population is close to an error catastrophe, which happens when a code string is likely to be mutated before it generates a copy of itself. These phenomena are related to phase transitions in physics and to analysis associated with the “hypercycle” theories of Eigen and Schuster.

The software on the CD‐ROM allows avida to be installed under Windows 95 or NT, or on a UNIX system. The software can be reproduced and shared freely, provided that it is not for profit, and it can also be obtained (even without buying the book) by downloading from a Web page whose address is given on page 322. The version available from the Web page will be updated.

The reader is invited to register his interest by returning a postcard to the TELOS organisation, so as to receive updating information, particularly since the field is described as fast‐moving. The Website from which the software can be downloaded is also part of TELOS, which is an imprint of Springer‐Verlag specialising in publications that “wed the traditional print medium with the emerging new electronic media”.

It is easy to feel swamped by the deluge of persuasive theorising in the book, but it gives a good and readable introduction to this important topic, with the invitation to run the software and so to get hands‐on experience and possibly contribute fresh findings.