Artificial Evolution and Lifelike Creativity

Abstract

This paper discusses the aims and goals of artificial evolution in relation to two of the founding features of A-Life: how to characterize the domain of the possible and the criterion of lifelikeness. It is argued that artificial evolution should aim to understand the evolution of organizations and that this will bring about a better understanding of possible evolutions that could have taken place on Earth.

Artificial Life (A-Life) emerged as a research program for studying living phenomena in artificial media so as to exempt the notion of life from its dependence on a single example. Three main features characterize the goals of this science as conceived by Christopher Langton [1]. First, its object is to explore the domain between the existent life-as-we-know-it and the possible life-as-it-could-be. In relation to biology, which studies life on Earth, the scope of A-Life includes universal living phenomena. Second, this universality will be achieved by studying artificially generated invariants of form or organization, and not of matter. The systems under study are synthesized, instead of (or, ideally, before) being analyzed, and it is assumed that the relevant phenomena depend more on the organization of the parts than on the nature of the components. Third, the extension of the systems under inquiry is open, but characterized by the notion of lifelikeness; thus, the question of defining life is left aside and replaced by a criterion based on experience and intuition. Research may emulate complex organization; stability in a given medium; capacity to absorb, transform and use energy; self-reproduction; evolution; development and growth; adaptation; etc., and those aspects will be interesting according to how lifelike they are. All three-the domain of the possible, universal organizing principles independent of matter, and lifelikeness-are interesting starting points that deserve careful discussion.

In previous work several authors have pointed out the problems of analyzing universality by means of organizations that do not emerge from a realistic material dynamics [2-5]. In this paper I intend to discuss artificial evolution in relation to Langton's other two features: how to characterize the domain of the possible and the criterion of lifelikeness.

The distance between existent life and possible life is manifest in the idea that different possible evolutions could have taken place on Earth starting from the same or similar initial conditions, because the history of evolution contains both contingent, or fortuitous, events and necessary ones, determined by the properties of evolving matter. An interesting line of research for a discipline focusing on the universal aspects of possible life is then to distinguish the necessary and the contingent in the history of evolution. In this sense, and presuming that laws are universal, the domain of the possible has at least two orders of magnitude: the scope of non-realized possibilities of life on Earth (new evolutions starting with the same initial conditions), and the scope of other possible lives (other evolutions with different initial conditions). Both are counterfactual approaches to evolution.

However, another way of considering the domains of possibility that artificial evolution may reveal concerns evolutionary theory itself and the way it conceives creativity or production of novelty. The theory of evolution is epistemologically challenging because it introduces creativity into the realm of science. Unlike other fields, such as physics, that try to discover natural laws underlying the behavior of all systems, evolutionary theory may be read as positing that almost anything is possible; it describes a procedure by which novelty appears and develops in nature.

It is in this sense significant that the criterion of lifelikeness is a foundation for research in artificial evolution. The creativity and open-ended nature of life has been characterized as "supple adaptation" [6], the hypothesis that life can be defined in terms of a system that exhibits "lifelike" evolution. This is an interesting idea, but it has somehow developed at the cost of considering lifelikeness at the level of individual living beings and instead focusing on the whole of life as a process.

I am interested in the problem of how to conceive the evolution of embodied agents, a problem...