Artificial Life: An Overview.
Are computer viruses a form of artificial life? Some of us may have pondered over this question having suffered the vagaries of the former or possibly, having noticed the excitement in some circles about the latter. The question is addressed by Eugene Spafford in a paper that appears in Artificial Life: An Overview edited by Christopher Langton, a book that surveys the new and emerging field of artificial life (AL). Spafford concludes that computer viruses are not forms that can be classified as AL. They do exist as patterns in space-time, self-reproduce, store information about their self-representation, interact with their environment, exhibit growth, and have different species. However, they do not have a metabolism, do not have interdependence amongst their parts, and, most importantly, they do not evolve (not yet anyway!). Spafford's argument locates a very interesting question that forms the basis of many artificial forms, whether hardware, software, or wetware, that have been evolved or developed in laboratories across the world, namely, what indeed is AL's relation to biology or computer science, and in what way does it explain life?
Langton's editorial introduction to the book highlights the methodology of synthesis as the approach by which many hitherto unanswered questions of biology can be addressed. Synthesis implies `putting the constituent pieces of matter together in new and different ways', which enables the construction of chemicals, cellular organizations, or behavioral patterns. By this method artifacts can be constructed that can then be examined and tested in the laboratory. Thus, scientists have a means by which to examine the origins, constituents, and behavior of molecular cellular, organismic, or population patterns that is different from what is available in nature. Theories about life and life patterns can now be formed because, as Langton phrases it, 'to have a theory of the actual, it is necessary to understand the possible'.
The synthesis is done by an actual production of whole forms, which may be molecular, as in the construction of ribonucleic acid (RNA), or may be at the organism level, as in the construction of mobile robots. In either case the phenomena of interest to scientists are invariably not the properties that are defined by the laws of the constituent parts, but are the patterns that emerge from the interaction of the parts with the environment. The phenomenon of emergence is at the same time the most exciting aspect of AL research, as well as the most troublesome, when considered from the perspective of epistemology Let us first bear on the former, to consider the papers in the book that describe research in AL in various disciplines. I will return to the latter issue towards the end of this review.
AL research in biology ranges from construction of RNA to studying populations of organisms. In many cases the artificial life forms are software modules that follow specific rules to stimulate the behavior of actual life forms. So `ant' modules will follow rules corresponding to ant behavior. These rules will specify if the ant will move towards food locations or away from them, follow pheromone trails or drop pheromone, etc. The environment is simulated by specifying locations for food deposits, their quantities, etc. The software modules then simulate the behavior of the ants in this artificial colony and the emergent patterns of ant movement are studied. Dyer's paper entitled `Synthesizing artificial neural nets' describes a number of experiments that model animal behavior, including foraging and trail-laying by ants, food discrimination by aquatic creatures, etc. Dyer contends that the grand challenge to the discipline is that of explaining the diverse behaviors of animals across the globe. This view is shared by Taylor and Jefferson also who, in their paper `AL as a tool for biological inquiry', list some of the major outstanding problems in biology that are likely to be studied by AL. These problems mostly deal with evolutionary theory--origins of life, evolution of cultures, origin and maintenance of sex, amongst others.
In many cases the simulations are not made to follow any specific rules of animal or plant behavior, and instead compute mathematical forms such as differential equations, or calculate game-theoretic outcome patterns. The emergent phenomena here are either graphic depictions of actual life forms or structural depictions of specific ecologies. Prunsinkiewicz's paper on `Visual models of morphogenesis' describes the formation of graphic patterns from differential equations. These patterns accurately model the patterns on shells, leaves, branches, etc. Lindgren and Nordahl study `Cooperation and community structure' by using the Prisoners Dilemma problem. They use multiple iterations of the problem, as a game played by multiple software modules, to generate models of cooperation. These models are used to explain phenomena such as food webs in specific ecologies.
The computer science research in AL has its roots in theories of cellular automata first proposed by von Neumann (1966), and later in genetic algorithms, as described by John Holland (1975), both of which led to the field of evolutionary computing. The goals of this field are to `evolve' software of generate emergent patterns that are useful in an engineering sense. A paper by Mitchell and Forest, `Genetic algorithms and artificial life', lists the multiple areas in which genetic algorithms have been successfully applied, ranging from optimization problems to models of social systems.
Within the field of artificial intelligence (AI), the focus is on the organism level of detail, where the units are hardware or software structures that have sensing/acting capabilities. Robotic `creatures' are built that are made to function in a particular environment where fixed sense-act rules produce complex emergent behavior. In `Modeling adaptive autonomous agents' Maes delineates the field from traditional AI research, as being more focused on low-level behaviors, goal determination, etc. The open questions remain those of determining goals and learning from past experiences. In another paper Steels, in `AL roots of artificial intelligence', outlines the role of AL research in defining the AI field.
It is clear from the papers in the book that the issue of emergence has to do with the visible properties of the patterns obtained from the system. This is clearly at a phenomenological level. Foraging and trail-laying behavior of `ants' is simulated and observed as patterns on a screen. Bonabeau and Theraulaz argue, in a paper entitled `Why do we need artificial life', that even if the ants deposited, in the simulation, several tons of pheromone to mark their trails, which is impossible in a real ant colony, this would be acceptable model. When the subject of study is simply exploratory behavior of the `ants', the issue of quantity of pheromone is not relevant.
Many such disquieting features of AL modeling have been pointed out in the book. Papers by Dennett and Harnad point to the caution required in extrapolating the synthetic and reductionist models produced by AL, beyond their explicit boundaries. Although AL is a powerful tool for conducting very elaborate thought experiments, its power to explain is limited. In spite of such cautions, however, the authors barely contain their excitement about the field and its potential for scientific advancement.
The book is meant to be a comprehensive overview of AL. To this end it is highly successful. All the articles include extensive bibliographies and point to many avenues for further exploration. The book also includes a chapter on other books dealing with the subject and related areas. I feel that the book would be an excellent guide for both the uninitiated as well as the well-entrenched AL researcher.
Holland, J. H. (1975). Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control and Artificial Intelligence, University of Michigan Press, Ann Arbor, MI. (Reprinted in 1992 by MIT Press, Cambridge, MA)
von Neumann, J. (1966). The Theory of Self-reproducing Automata, University of Illinois Press, Urbana, IL.
|Printer friendly Cite/link Email Feedback|
|Publication:||Systems Research and Behavioral Science|
|Article Type:||Book Review|
|Date:||Jan 1, 1997|
|Previous Article:||Migrating to Windows 95: A Programmers Guide to What's New.|
|Next Article:||A systems approach to studies of creativity and consciousness.|