Reviewed by
Itzhak Benenson
Department of
Geography and Human Environment, Tel Aviv University, Israel.
The editors of this volume define its goal as 'capturing the universality of life, to transplant it into a different medium and to study its general and logical characteristics'. One can argue that this is the objective of all scientific theory, and the medium could be mathematical equations, analogous natural or mechanical systems, computer programs or still other methods. The participants of the 'Artificial Life' conference dealt exclusively with computer simulations. An overview of the papers presented during the conference and previous meetings since 1987, when the series began, make it clear that most researchers simulate and investigate the dynamics of complex adaptive systems, in which the elements are complex self-organising systems. We tend to think of living systems in this way (Portugali 1999) and the truth is that simulating each specific phenomena at such a level of complexity demands the building of new models. As a result, more models, each simulating specific aspects of life, come into being. The goal of the sixth conference - 'Life and Computation: the Boundaries are Changing', was to go beyond the traditional domain of applications. The papers presented at the conference extend the boundaries of Artificial Life studies to incorporate new models in artificial chemistry, molecular and cell biology, and population genetics. Most interesting for JASSS readers, these models also describe organism development and behaviour, and simulations of human systems - social, economic and linguistic. It is in these areas that I will focus my comments.
The papers presented at the conference can be classified into three categories. The first - 'classic' - aim to study the evolutionary dynamics of complex systems, thus providing the background for specific applications. Several papers offered important improvements to Kauffman's classic NK model of 'ruggedness' in evolutionary landscapes and to Langton's model of self-replication. For instance, L. Barnett ('Ruggedness and Neutrality - the NKp Family of Fitness Landscapes') extends the classic model to the NKp one. The latter displays a form of neutrality, recently considered an important evolutionary mechanism. Neutrality entails a new type of evolutionary dynamics, namely neutral drift punctuated by drastic transitions rather than local hill climbing. H. Sayama ('Introduction of Structural Dissolution into Langton's Self-Reproducing Loop') introduces 'death' into the classical Langton model and assumes that the loop can dissolve its own structure when faced with a difficult situation, such as a shortage of space for self-reproduction. The dissolution mechanism qualitatively changes the model behaviour, making it dynamically stable and resulting in the evolution of a colony of self-reproducing loops.
Three other papers devoted to the general problems of artificial life examine 'socially grounded' simulations. A. K. Seth ('The Evolution of Complexity and Value of Variability') explores the classical hypothesis that environmental variability promotes the evolution of organism complexity in the context of social activity. The model deals with populations of individuals whose strategy in playing Iterated Prisoner's Dilemma is determined by genotypes of variable length. The results demonstrate that environmental variability is necessary for the evolution of complexity (genotype length). In a related paper ('Generic Behavior in the Lindgren Non-Spatial Model of Iterated Two Player Games'), T. Shannon demonstrates that punctuated equilibrium is a generic property of this model. Macroscopic measures of evolutionary activity and concentration within population of players are introduced, making it possible to characterise a model's evolutionary dynamics, including a description of the epochs of equilibrium and punctuation in genotype space. C. Adami, R. Seki and R. Yirdaw ('The Critical Exponent of Species-Size Distribution in Evolution') consider adapting populations of single-stranded self-replicated genomes, represented by the programs in Avida (Adami 1998) world. The evolution of the population results in a power law that emerges in complex landscapes and evolves towards separation of two fundamental time scales: time for the population to return to equilibrium after a perturbation and time between mutations that produce fitter genotypes. The theoretical results regarding the distribution of avalanche sizes are discussed in relation to fossil records.
The papers that belong to the second category simulate development, behaviour and locomotion of individual organisms. I will refer only to one, namely the paper by S. I. Nishimura and T. Ikegami: 'Emergence and Maintenance of Relationships among Agents'. The authors study the emergence and maintenance of relationships among individual agents in a model ecosystem. The model agents are not interacting directly, but behave according to the "follow the others" principle, that is, they react to the instantaneous vector of velocities for neighbouring agents. The collective dynamics of agents on the 2D plane is examined with regard to the emergence and breaking of spatial patterns, consisting of several individuals. It is demonstrated that qualitatively different sets of model parameters determine various dynamically sustainable motions of the groups of agents, and these motions are portrayed in detail.
The applications of the third category, devoted to social dynamics, financial markets and the evolution of language, were presented in several plenary lectures and papers. At the population level, F. Klugel, F. Puppe, U. Raub, J. Tautz ('Simulating Multiple Emergent Phenomena Exemplified in an Ant Colony') attempt to build an explicit simulation model of a real-world ant community. The model is inherently complicated and unifies the main ant activities during the complete life cycle of a colony - foraging, recruiting, storing energy, breeding and individual development. The paper presents the concept of a model and concentrates on a computer environment, called 'Shell for Simulated Agent Systems', which aims to be used for model implementation. It demonstrates how the shell enables direct interpretation and modelling of important population phenomena, such as mass recruitment for defence of territory, development outside of the nest, and so on. However, like many others, this environment does not go beyond the conceptual stage. C. K. Hemelrijk ('Spatial Centrality of Dominants without Positional Preference') verifies the proposition that, for gregarious animals, the centrality of dominants is aimed at optimal protection from predators. The model confirms this approach, demonstrating that the evolutionary basis for this form of behavior can be the result of individual level dominance interactions among agents, supposing that winning and losing are self-enforcing, and the attack rate is negatively dependent on rank differences and distance between the individual agents. The paper by T. Terano, S. Kurahashi and U. Minami, ('TRURL: Artificial World for Social Interaction Studies') describes a simulation multi-agent environment, which utilises genetic algorithms to evolve societies. They use this environment to investigate features of so-called Face-to-Face, E-Mail, Net-News and Mass-Communication oriented societies, each one presenting different emerging social structures.
Several papers within the social category dedicate attention to economic systems. M de la Maza, A. Ogus and D. Yuret ('How Do Firms Transition Between Monopoly and Competitive Behavior? An Agent-Based Economic Model') consider the fundamental economic problem of transitions from monopoly to competitive behavior as firms enter a market. They study the collective behaviour of agents existing in the framework of a single product market and two input markets for capital and labour. The agents maximise their profit by locally setting the price of the product and purchase capital and labour to produce it. In the long term, the resulting global economic behaviour is an equilibrium distribution of prices provided by the known analytical models. However, the agent-based model additionally presents the path of transition from monopoly to equilibrium, which could not be obtained in the standard equilibrium model. J. O. Kephart, J. E. Hanson and J. Sairamesh ('Price-War Dynamics in a Free-Market Economy of Software Agents') examine Internet-based economies, where software agents provide trading and use a variety of goods and services in an open free market. The paper focuses on brokering (retailing) information, gathering it from the correct producers and distributing it to the appropriate consumers. In addition, the paper investigates price-setting mechanisms, based solely on the system's current state, where brokers tend to optimise their short-term profit. Analytical and numerical results show that the dynamic behaviour of the system is generically an unending cycle of competition wars in price/product space and the existence of multiple peaks in the broker's profitability landscape. The authors contend that friction effects preventing price wars in human economies are considerably weaker in economies of simple agents. The latter market lacks consumer inertia and the ability to predict indirect effects of action. K. Izumi and K. Ueda ('Emergent Phenomena in a Foreign Exchange Market: Analysis Based on an Artificial Market Approach') integrate fieldwork and a multi-agent model, where agents interpret data, manipulate dollar and yen assets and make market predictions based on a weighting of seventeen economic factors. The simulation results show the phase transitions in the structure of the agent population regarding the forecast of the rise/drop of a dollar and several other macro-level emerging phenomena. It is one of a few papers in the proceedings which operationally compare the model to a real world situation.
Regarding linguistic systems, the paper by L. Steels and F. Kaplan ('Stochasticity as a Source of Innovation in Language Games') demonstrates that coherent sets of linguistic conventions and their transmission from one generation to the next can be considered as a result of self-organisation. The paper explores the role of stochasticity in various aspects of lexical communication by means of the multi-agent model of communicating agents. It is demonstrated that there exists an upper bound of the amount of stochasticity which can be tolerated, while below this bound, stochasticity causes and maintains language variations.
Considering the proceedings as a whole, it is clear that Artificial Life studies are still in an initial stage. No commonly accepted list of problems exist and it seems that researchers still prefer to widen their horizons. The next step ought to involve a departure from surface studies, towards deeper details of specific artificial life sciences, where JASSS readers are engaged in the development of the 'social side'. The papers from this volume will contribute to this endeavour and an index for the volume, which is unusual for conference proceedings, will help to locate topics of interest.
References
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2001