Reviewed by
Colin Johnson
Computing
Laboratory, University of Kent.
This substantial volume consists of the papers presented at the fifth European Conference on Artificial Life, held in Lausanne in 1999.
Artificial life is at something of a transition point in its development as a subject. Following an initial burst of enthusiasm and a number of interesting early developments, it has reached the stage at which researchers are beginning to look back over its achievements with a critical eye and ask what its role might be as a mature discipline. A review of early artificial life work can be found in Langton (1995) with a more popular account in Levy (1993).
For those unfamiliar with the subject, artificial life is the study of life through computer programs, robots and similar technologies. However it is not primarily concerned with "life as it is", that is, the simulation of existing living systems. Instead artificial life is concerned with "life as it could be"; that is, using simulation to understand what conditions can give rise to life, in what ways information can move through a living system, what sorts of reproductive and evolutionary systems are effective and so on. The perspective is typically one of "sufficiency": what are the minimal models required for the creation of a particular phenomenon? What is essential to the production of a particular observed phenomenon, and what is specific to the unique "implementation" of life found here on earth?
One of the main motivations for this type of study is the limiting presence of only one "instantiation" of life that life scientists can study; it could be argued that if a wide variety of life forms had arisen independently and were still available to study then research in the artificial life mode would not be necessary. However I am not convinced by this argument; it seems that the kind of questions asked by this stance towards life provide a refreshing contrast to the types of questions asked by other approaches to bioscience.
A natural question in the context of this journal is whether a similar stance can be adopted towards the social sciences. Clearly the immediate need for such a stance is less pressing; many different societies have developed largely in isolation from each other until recent centuries. Nonetheless the stance of examining societies from the perspective of "society as it could be (or might have been)" seems to be an interesting one. One of the powerful things which artificial life type "simulations" have given us is the ability to find minimal models producing behaviour that has been observed in the real biological world. Often it has been surprising to observe how a small set of basic rules can give rise to complex phenomena biological systems. Applying the same methods in social simulation could be a valuable application of these ideas.
The first section of the volume deals with epistemological issues relating to artificial life. Of particular interest is the paper by Ronald, Sipper and Capacarrère which picks apart the often-used concept of emergence in artificial life. They suggest that the idea of emergence results from a dissonance between the language in which it is natural to describe the design of the system and the language used to describe the observed phenomena. This makes it clear that emergence is not a property of particular phenomena but a property of their descriptions. It also implies that as the usual language used to describe these behaviours changes, our sense of what is and is not emergent is likely to change with it.
Evolutionary dynamics is the subject of the next section of the book. Many of the papers in this section are concerned with the use of simulation to tease out details of evolutionary theory and to distinguish between various hypotheses about the functioning of evolution. Another concern is the use of simulation to test measures of biological complexity and bio-diversity. One of the best papers in this section is the paper by Ochoa et al., which provides an elegant connection between the theory of error thresholds in evolutionary theory and the idea of optimal mutation rates in evolutionary algorithms. This kind of work, in which ideas from biology and computer science support each other, is of great value to both subjects.
The third part of the book is entitled evolutionary cybernetics and is largely concerned with the evolution of robot control programs of various kinds, using both real and simulated robots. These papers typically cover two main themes. Either they are concerned with models inspired by particular life forms or they attempt to reproduce certain aspects of those life forms (in particular their neural circuitry). Perhaps the most interesting paper in this section is Ijspeert's paper which uses an evolutionary algorithm to evolve simulated neuro-circuitry for the motion of creatures such as salamanders and lampreys.
The following section continues the emphasis on practical robotics and investigates various topics in the area of bio-inspired robotics and autonomous agents. The majority of the work in this section is concerned with implementation of neural networks to control various kinds of robots. A number of novel robot architectures are described, including an interesting design by Hasagawa et al. that moves by brachiation, the kind of movement used by apes as they swing through trees. In terms of novel control architectures the paper by Adamatazky et al. is the most innovative. Their paper describes the use of a computational system not built around conventional circuitry but by interactions between chemicals on a two-dimensional surface. A very literal interpretation of "autonomous" is given in the paper by Kelly et al. Their paper discusses the design and construction of a robot that maintains itself in the world by catching and processing slugs in order to generate electricity! A final paper that deserves mention in this section is that of Hyvärinen and Honkela, which discusses the issue of emotion in agent models. Whilst their work is at an early stage, they highlight an area which is likely to increase in importance in future, particularly as models of social behaviour take on board the effects that emotions have on social interactions.
The fifth section is largely concerned with artificial life as a tool for exploring various details of biological systems such as self-replication, self-maintenance and gene expression. Many of the papers in this section show how simulation can be used to understand particular biological phenomena, in particular via the reconstruction of minimal models that demonstrate the phenomenon in question. Work of this kind seems to be where the artificial life literature interacts most closely with the literature on simulation in biosciences. Perhaps the most interesting paper in this section is that of Bersini, which begins to tie together various notions from object-oriented software engineering with ideas from chemistry. There is much work to be done in this area, both in using software engineering ideas as a way of understanding complex systems in science and in the production of new software engineering ideas which are more attuned to the kinds of complexity found in science rather than that generated by interactions between people.
The papers on societies and collective behaviour are likely to be the most interesting to readers of this journal. Many of these papers refine existing arguments which have previously been the subject of simulation or mathematical modelling studies, such as the evolution of co-operative behaviour, the role of imitation as a learning mechanism in animal societies and strategies for parental investment in offspring. A number of papers also tackle questions concerning ant colony behaviour, both from the perspective of understanding real ants and from the perspective of using the ant colony as a metaphor for computation; this kind of "swarm intelligence" is now maturing as a field of study in its own right. Two papers are of particular interest in the context of understanding social behaviour. Hemelrijk's paper uses simulation to understand the development of dominance relationships in primate societies, whilst the paper by Ugolini and Parisi discusses a model of the evolution of technological artefacts in a society. This latter paper opens out a large area of study for future work; in particular it is pleasant to see an appreciation of the fact that the artefacts and technologies in a society develop (co-evolve?) alongside the society itself, and a move away from simplified models of the gathering of some amorphous "resource" from the environment.
The final section of the book consists of a number of papers on communication and language, and these are again likely to be of interest in the social simulation world. The papers in this section consider the evolution and development of these ideas at a number of levels, from pheromone-style communication in ants to the simulation of changing vocabulary in a society of agents which already possess a language capability.
Several of the papers in this section deal with the ways in which communication systems can arise "from scratch" in animal societies, and how evolution and learning shapes such systems. The paper by Kawamura et al. shows how a society of ant-like agents driven by neural networks can create a communication system based on pheromone trails without the meaning of the pheromone being pre-specified. Noble's paper makes use of simulations to study the conditions under which the handicap principle in communication (Zahavi and Zahavi 1999) is likely to be observed. This principle suggests that signals about fitness between potential mates in an animal population become more trustworthy if there is a large fitness cost to making the signal; the canonical example of this is the tail-feather display in peacocks. Again this illustrates the power of artificial life models in refining biological notions and providing explanations for the sufficient conditions needed to explain a particular observed phenomenon.
The papers on the evolution of language provide one of the most interesting sections of the book. Several papers use simulation to test hypotheses about evolution of certain language features, such as the development of a basic word-order grammar from a set of individual words and the way in which the basic phonetic components of words in languages are shaped by learning and evolution. Another group of papers are concerned with how a vocabulary evolves in a society and is transmitted between agents acting in that society. There are a number of ways in which this could be extended, for example in studying the ways in which new words arrive in the language and the conditions which cause them to be adopted into the language, or in understanding how some words and phrases remain the property of particular subgroups within society whilst others spread throughout society or die away completely.
One feature which is of particular note in several papers in this chapter is the use of diagrams which convey the internal dynamics of the process described. This would seem to be an important area of research for this whole field; we would like to be able to visualise the internal dynamics of these processes as well as observing the output. Indeed such visualisations might enhance our understanding of the slippery concept of emergence; what appears to be a step-change in behaviour between initial rules and observed behaviour may turn out to be a much smoother transition when we can observe intermediate steps.
Overall this book provides a broad snapshot of the artificial life field and related areas. As might be expected of a field at this level of development, there is lots of emphasis on consolidating and refining ideas and providing more examples of the basic types of mechanisms studied by the subject. In particular, one impression I was left with was that many of the core concepts in the field (such as emergence and embodiment) need to be more deeply understood and better defined before they can be applied in powerful new ways. In contrast to some such volumes, there was little sense of a single "next big thing" in the artificial life domain. This may be a good or bad thing; clear and well explored new ideas can be inspiring, but they can also lead to a subject lurching from one big new idea to the next without a clear direction for the subject as a whole.
The format of the book, consisting of a large number (90) of short papers collected into a number of broad categories will be familiar to readers coming from a background in the computational sciences but perhaps less comfortable for social scientists who may be more familiar with a more focussed and coherent presentation. One problem with this approach is that papers which present a new, self-contained idea leap out of the book more than papers which attempt to build on existing work; this perhaps biases the reader's attention more towards work which describes easy-to-grasp "one-shot" ideas rather than those work which will make a longer-term contribution to the consolidation of the subject.
Overall the book provides a good snapshot of the diverse topics which fall under the "artificial life" heading. Many researchers will find something of interest here, though this is a book largely for "dipping into" rather than reading from cover-to-cover. Readers who are curious about this topic may also be interested in the forthcoming 2003 European Conference on Artificial Life: details at http://www.ecal2003.org.
References
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2002