In the typical genetic algorithm experiment, the fitness function is constructed to be independent of the contents of the population to provide a consistent objective measure. Such objectivity entails significant knowledge about the environment which suggests either the problem has previously been solved or other non-evolutionary techniques may be more efficient. Furthermore, for many complex tasks an independent fitness function is either impractical or impossible to provide. In this paper, we demonstrate that competitive fitness functions, i.e. fitness functions that are dependent on the constituents of the population, can provide a more robust training environment than independent fitness functions. We describe three differing methods for competitive fitness, and discuss their respective advantages.

review. Views or opinions expressed herein do not necessarily represent those of the Institute, its National Member Organizations, or other organizations supporting the work. IIASA STUDIES IN ADAPTIVE DYNAMICS NO. 27 The Adaptive Dynamics Network at IIASA fosters the development of new mathematical and conceptual techniques for understanding the evolution of complex adaptive systems. Focusing on these long-term implications of adaptive processes in systems of limited growth, the Adaptive Dynamics Network brings together scientists and institutions from around the world with IIASA acting as the central node. Scientific progress within the network

The paper surveys work on the computational modeling of the origins and evolution of language. The main approaches are clarified and some example experiments from the domains of the evolution of communication, phonetics, lexicon formation, and syntax are discussed. 1 Introduction The paper surveys research in which software simulations and experiments with robotic agents are used to explore the viewpoint that language is a complex dynamical system. The main goal of the paper is to outline the approaches and show example experiments. Much more work needs to be done to arrive at a full-fledged theory of the origins of language and even about the work already done much more can be said than is possible in a single paper. Nevertheless, I hope to show that a new exciting approach to the study of the origins and evolution of language is taking shape. The rest of the paper is in four parts. The next section clarifies the notion of a complex system and the multi-agent perspective. Section 3...

For Artificial Life applications it is useful to extend Genetic Algorithms from a finite search space with fixed-length genotypes to open-ended evolution with variable-length genotypes. A new theoretical analysis is required, as Holland's Schema Theorem only applies to fixed lengths. It will be argued, using concepts of epistasis and fitness landscapes drawn from theoretical biology, that in the long run a population must havegenotypes of nearly equal length, and this length can only increase slowly. As the length increases, the population will be nearly converged, and hence evolving as a species.

We present results from experiments in which a genetic algorithm (GA) was used to evolve cellular automata (CAs) to perform a particular computational task—one-dimensional density classification. We look in detail at the evolutionary mechanisms producing the GA’s behavior on this task and the impediments faced by the GA. In particular, we identify four “epochs of innovation ” in which new CA strategies for solving the problem are discovered by the GA, describe how these strategies are implemented in CA rule tables, and identify the GA mechanisms underlying their discovery. The epochs are characterized by a breaking of the task’s symmetries on the part of the GA. The symmetry breaking results in a short-term fitness gain but ultimately prevents the discovery of the most highly fit strategies. We discuss the extent to which symmetry breaking and other impediments are general phenomena in any GA search. 1.

Tag is a children's game based on symmetrical pursuit and evasion. In the experiments described here, control programs for mobile agents (simulated vehicles) are evolved based on their skill at the game of tag. A player's fitness is determined by how well it performs when placed in competition with several opponents chosen randomly from the coevolving population of players. In the beginning, the quality of play is very poor. Then slightly better strategies begin to exploit the weaknesses of others. Through evolution, guided by competitive fitness, increasingly better strategies emerge over time. 1. Introduction Many of us remember playing the game of tag as children. Tag is played by two or more, one of whom is designated as it. The it player chases the others, who all try to escape. Tag is a simple contest of pursuit and evasion. These activities are common in the natural world, most predatorprey interactions involve pursuit and evasion. Tag also includes an aspect of role-reversal, b...

One of the persistent themes in Artificial Life research is the use of co-evolutionary arms races in the development of specific and complex behaviors. However, other than Sims’s work on artificial robots, most of the work has attacked very simple games of prisoners dilemma or predator and prey. Following Tesauro’s work on TD-Gammon, we used a 4000 parameter feed-forward neural network to develop a competitive backgammon evaluation function. Play proceeds by a roll of the dice, application of the network to all legal moves, and choosing the move with the highest evaluation. However, no back-propagation, reinforcement

We present empirical evidence that long-term evolutionary dynamics fall into three distinct classes, depending on whether adaptive evolutionary activity isabsent (class 1), bounded (class 2), or unbounded (class 3). These classes are de ned using three statistics: diversity, new evolutionary activity (Bedau & Packard 1992), and mean cumulative evolutionary activity (Bedau et al. 1997). The three classes partition all the longterm evolutionary dynamics observed in Holland's Echo model (Holland 1992), in a random-selection adaptivelyneutral "shadow" of Echo, and in the biosphere as reected in the Phanerozoic fossil record. This classi-cation provides quantitative evidence that Echo lacks the unbounded growth in adaptive evolutionary activity observed in the fossil record.

Metastability is a common phenomenon. Many evolutionary processes, both natural and artificial, alternate between periods of stasis and brief periods of rapid change in their behavior. In this paper an analytical model for the dynamics of a mutationonly genetic algorithm (GA) is introduced that identifies a new and general mechanism causing metastability in evolutionary dynamics. The GA’s population dynamics is described in terms of flows in the space of fitness distributions. The trajectories through fitness distribution space are derived in closed form in the limit of infinite populations. We then show how finite populations induce metastability, even in regions where fitness does not exhibit a local optimum. In particular, the model predicts the occurrence of “fitness epochs”—periods of stasis in population fitness distributions—at finite population size and identifies the locations of these fitness epochs with the flow’s hyperbolic fixed points. This enables exact predictions of the metastable fitness distributions during the fitness epochs, as well as giving insight into the nature of the periods of stasis and the innovations between them. All these results are obtained as closed-form expressions in terms of the GA’s parameters.

In many problems of interest, performance can be evaluated using tests, such as examples in concept learning, test points in function approximation, and opponents in game-playing. Evaluation on all tests is often infeasible. Identification of an accurate evaluation or fitness function is a difficult problem in itself, and approximations are likely to introduce human biases into the search process. Coevolution evolves the set of tests used for evaluation, but has so far often led to inaccurate evaluation. We show that for any set of learners, a Complete Evaluation Set can be determined that provides ideal evaluation as specified by Evolutionary Multi-Objective Optimization. This provides a principled approach to evaluation in coevolution, and thereby brings automatic ideal evaluation within reach. The Complete Evaluation Set is of manageable size, and progress towards it can be accurately measured. Based on this observation, an algorithm named DELPHI is developed. The algorithm is tested on problems likely to permit progress on only a subset of the underlying objectives. Where all comparison methods result in overspecialization, the proposed method and a variant achieve sustained progress in all underlying objectives. These findings demonstrate that ideal evaluation may be approximated by practical algorithms, and that accurate evaluation for test-based problems is possible even when the underlying objectives of a problem are unknown.