Publication details
Evolving Interaction in Artificial Systems: An historical overview and future directions
Tim Taylor
2003
Abstract
Over the last 50 years, a growing number of attempts have been made to engineer systems which can support the open-ended evolution of self-replicating components. This work includes not just software systems, but also those employing robotic, electronic, and chemical media. Much of this work has been (explicitly or implicitly) based upon the neo-Darwinist premise that the existence of living organisms can fundamentally be explained in terms of the processes of reproduction, variation and natural selection. However, the stark fact is that none of these attempts to recreate such processes in an artificial system has succeeded in producing an ongoing evolutionary process. Rather than producing ever more complex organisms/societies/ecologies, evolution in these systems generally leads to a stable and simple end point in which only the smallest and fastest reproducers survive. These results suggest that the neo-Darwinist position is, at best, only a partial explanation of the evolution of complexity. Commenting on one such attempt (Orgel’s experiments with the in vitro evolution of RNA sequences), John Maynard Smith remarks: "This raises the following simple, and I think unanswered, question: What features must be present in a system if it is to lead to indefinitely continuing evolutionary change?" [Maynard Smith 88]. Even going back to the earliest attempts to engineer such systems, the importance of interactions between organisms for providing co-evolutionary selection pressure for increased complexity has been recognised as a key issue [Barricelli 62; Conrad & Pattee 70]. From a biological perspective, Waddington too recognised the need for elaborating the necessary and sufficient conditions under which open-ended evolution might arise [Waddington 69]. His proposal included the requirement for "an indefinite number of environments, and this is assured by the fact that the evolving phenotypes are components of environments for their own or other species." One of the major challanges facing designers of artificial evolutionary systems – and one which remains largely unsolved – is to understand how to design systems in which indefinitely new types of inter-organism interaction can evolve. From Waddington’s perspective, this can also be stated as the need to understand how to design systems in which evolving organisms are fully part of the environment experienced by other organisms. These matters concern the fundamental design issue of the relationship between organisms and their environment. The nature of the interface between organism and environment determines the organism’s potential for sensing, acting, and communicating with its environment (which may include other organisms). Furthermore, the capacity for this interface to evolve determines the potential for new forms of inter-organism interaction to evolve. In this talk, an overview of the work described above will be presented, together with a discussion of the (often very limited) capacity of various well-known artificial evolutionary systems (both software- and hardware-based) to evolve new forms of interaction. Drawing from the experiences of these existing studies, the talk will conclude with some suggestions for improving the design of future artificial evolutionary systems in order to increase their evolutionary potential.
Full text
- Presentation slides: pdf
Reference
Taylor, T. (2003). Evolving Interaction in Artificial Systems: An historical overview and future directions. In P. McOwan, K. Dautenhahn, & C. L. Nehaniv (Eds.), Abstracts from the Evolvability and Interaction Symposium, held at Queen Mary, University of London, UK, in October 2003. University of Hertfordshire Computer Science Technical Report No. 393.
BibTeX
@incollection{taylor2003evolving, author = {Taylor, Tim}, title = {Evolving Interaction in Artificial Systems: An historical overview and future directions}, booktitle = {Abstracts from the Evolvability and Interaction Symposium, held at Queen Mary, University of London, UK, in October 2003}, publisher = {University of Hertfordshire Computer Science Technical Report No.~393}, year = {2003}, month = oct, editor = {McOwan, Peter and Dautenhahn, Kerstin and Nehaniv, Chrystopher L.}, category = {workshop}, keywords = {evoca, history, oee} }