Artificial Life Evolutionary Models

Artificial Life (ALife), as an area of investigations, took its form in the late 1980s [1,2]. The main motivation of ALife is to model and understand the formal rules of life. As C.G. Langton said, “...the principle assumption made in Artificial Life is that the ‘logical form’ of an organism can be separated from its material basis of construction” [1]. ALife “organisms” are man-made, imaginary entities, living in computer-program worlds. Evolution, ecology, and the emergence of new features of life-like creatures are under special attention of the ALife researches.

The ALife evolutionary models include:

• “PolyWorld” by L. Yaeger: a computer model of artificial “organisms”, which have structured neural networks, possess a color vision, can move and increase their energy resources by eating foods, can mate and fight with each others. Some kinds of non-trivial ecological strategies were evolutionary emerging during PolyWorld simulations.
• “Tierra”by T. Ray: a model of the evolution of self-reproductive programs. The Tierra’s “organisms” include genome strings, which determine the executive program instructions. The organisms are able to exchange the program code segments. The interactions between the organisms result in an evolutionary emergence of complex “biodiversity” of the self-reproductive programs.
• “Avida” by C. Adami et al. is an Tierra-inspired model. Basing on Tierra and Avida, C. Adami et al. constructed a mathematical model describing the distribution of species in evolving populations. This model quantitatively supports the point of view that the evolution has a punctuated character rather than gradual one.
• The analysis of interactions between learning and evolution by D. Ackley and M. Littman. This work expressively demonstrated “that learning and evolution together were more successful than either alone in producing adaptive populations” [2, pp.487-509].
• “ECHO” by J.H. Holland. This model describes the evolution of simple agents, which can interact by mating, fighting, and trading. The interaction between agents results in complex ecologies: “arms race”, symbiotic relations, etc.
• The model of co-evolution of host and parasite populations by D. Hillis [2, pp.313-324]. The individuals of the host population in this model are algorithms, which are intended to solve a certain practical problem (e.g. the sorting problem), whereas the parasite population is a set of tasks to be solved. The host population evolves to find a good solution of the problem, while the parasite population evolves to make the problem more difficult. The competitive host-parasite co-evolution ensures finding the significantly better solution, than the host population could find alone.
• Models of evolving cellular automata, e.g. the models by M. Mitchell et al., describing the evolutionary search of cellular automata, which can perform simple computations [3].
• “AntFarm” by R.J. Collins and D.R. Jefferson is a Connection-Machine-based model for simulation of foraging behavior in huge evolving populations of artificial ants [2, pp. 579-601].

• Classifier system by J.H. Holland et al is a model of evolving cognitive process [4]. The classifier system models the inductive inference scheme, which is based on a set of logic rules. Each rule has the following form: “if <condition> then <action>”. The rule system is optimized by both learning and evolutionary search. Learning implies that the priorities of the use of the particular rules (the rule strengths) are modified by means of so-called backed-brigade algorithm. The evolutionary search is the discovery of new rules by means of the genetic algorithm.

ALife evolutionary modeling is currently developing field of evolutionary investigations. Mainly, the models are clever computer experiments. The serious mathematical description of ALife evolution features is still at the very beginning. A good example of significant mathematical investigations is Adami’s model of species-size distribution in evolving populations [5]. This model is based on the theory of “Self-Organizing-Criticality” [6] and provides a reasonable interpretation of both ALife computer experiments (on Tierra and Avida) and real biological data.

ALife evolutionary modeling is developing in close relations with life origin models, Kauffman’s NK-automata investigations, and researches of evolutionary algorithms. Adaptive behavior of ALife “organisms” is often based on operation of artificial neural networks; the evolution is mainly modeled by means of the genetic algorithms. It should be noted that – in contrast to the original version of the applied genetic algorithm – the majority of ALife evolutionary models don’t include explicit fitness function. The fitness is usually endogenous: the organisms are naturally born (when their parents are ready to give birth to the children) and die (e.g. through starving or being killed by predators).

ALife modeling throws a new light upon evolutionary phenomena. An excellent example is investigations of the Baldwin effect [7]. According to the Baldwin effect, the learned features of organisms could be indirectly inhered in subsequent generations. The Baldwin effect works in two steps. At the first step evolving organisms obtain (through appropriate mutations) an ability to learn a certain advantageous trait. The fitness of such organisms is increased; hence they are spread throughout the population. But learning is typically costly for an individual, because it requires energy and time. Therefore the second step (which is called the genetic assimilation) is possible: the advantageous trait can be “reinvented” by the genetic evolution and become directly genetically encoded. The second step takes a number of generations; a stable environment and a high correlation between genotype and phenotype facilitate this step. Thus, the advantageous trait that has been originally acquired can become inherited, though the evolution is of Darwinian type.

A number of researchers (G.E. Hinton & S.J. Nowlan, 1987, D. Ackley & M. Littman, 1992, G. Maylay, 1996) modeled the Baldwin effect. They demonstrated that the Baldwin effect can play the important role during the evolution of ALife organisms. See collections of papers [8,9] for details.

There is an obvious tendency towards modeling of evolution of cognition abilities. Whereas the evolutionary theories developed in the first part of 20-th century described evolution processes in terms of distributions of gene frequencies (see the node General Models of Evolution for details), the current ALife evolutionary models are actively incorporating such notions as learning, neural networks, adaptive behavior. This tendency is certainly supported by investigations in computer sciences, cognitive sciences, artificial intelligence; so we can conclude that the mention tendency isn’t accidental and the evolution of cognition features will be intensively investigated in the future.

The Internet resources of ALife evolutionary models can be found at Santa Fe Institute Artificial Life Site.

Conclusion. ALife evolutionary modeling is a rather new, very interesting area of evolutionary investigations. It is trying to find the formal rules, the formal laws of life and evolution by means of computer modeling of life-like entities. The ALife models are able to throw a new light upon old evolutionary problems (the Baldwin effect, the punctuated character of the evolution process). ALife evolutionary investigations have good perspectives, especially in the field of analysis of the evolution of cognition abilities in natural and artificial systems.

References:

1. Langton, C. G. (Ed.) (1989). Artificial Life. Reading, MA: Addison-Wesley.

2. Langton, C. G., Taylor, C., Farmer, J. D., and Rasmussen, S. (Eds.) (1992). Artificial Life II. Reading, MA: Addison-Wesley.

3. Mitchell, M., Crutchfield, J.P., Das, R. Evolving Cellular Automata with Genetic Algorithms: A Review of Recent Work // In Proc. of the First International Conference on Evolutionary Computation and Its Applications (EvCA'96). Moscow, Russia: Russian Academy of Sciences, 1996.

4. Holland, J.H., Holyoak, K.J., Nisbett, R.E., Thagard, P. (1986). Induction: Processes of Inference, Learning, and Discovery. Cambridge, MA: MIT Press.

5. Adami, C., Seki, R., Yirdaw, R. Critical Exponent of Species-Size Distribution in Evolution // In Adami, C., Belew, R., Kitano, H., Taylor, C., (Eds.) (1998). Artificial Life VI. MIT Press, pp. 221-227.

6. Bak, P. (1996). How Nature Works: The Science of Self-Organized Criticality, Springer, Berlin.

7. Baldwin, J.M. A new factor in evolution // American Naturalist, 1896. V.30, pp. 441-451.

8. Belew, R.K. and Mitchell, M. (Eds.) (1996). Adaptive Individuals in Evolving Populations: Models and Algorithms, Massachusetts: Addison-Wesley.

9. Turney, P., Whitley, D., Anderson, R. (Eds.). Evolution, Learning, and Instinct: 100 Years of the Baldwin Effect // Special Issue of Evolutionary Computation on the Baldwin Effect, V.4, N.3, 1996.

See also: Links on Complexity, Self-Organization and Artifical Life