Emergence out of interaction: Developing evolutionary technology for design innovation

https://doi.org/10.1016/j.aei.2006.01.006Get rights and content

Abstract

We consider Class III problems in emergent synthesis methodology. Our aim is to minimize human interaction in coping with problems of incompleteness. We introduce and discuss an agent-based simulation informed from biological evolution. We deal with the problem of persistent species evolution in an artificial evolutionary system and argue that a species evolution process can help addressing design problems, especially design innovation and changing function spaces. Our simulation is based on the theory of ‘fat’ phenotype applied to the dynamic generation of new evolutionary tasks. We present the model and its computational results showing how ‘fat’ phenotypes can yield changing interaction spaces to define new selection forces that recursively give rise to new ‘species’ that solve new selection tasks. We discuss prospects for radical evolutionary technology and for emergent synthesis.

Introduction

We deal with design as a synthesis problem. The question considered in the paper concerns the generation of novel design in the context of emergent synthesis as introduced by Ueda [1]. In this theory, the difficulties in synthesis are categorized into three classes [2]. Class I is characterized by systems with a complete description of the design specification and the environment; Class II means a complete design specification but incomplete information about the environment; finally, Class III denotes systems with both incomplete specification and an incomplete environment description. It is generally held that the most difficult challenge is posed by the Class III problems. Because of the underspecified nature of some of these problems, a continual human interaction is often held significant. For example, manufacturing systems have been treated as co-creative systems of this kind [3].

However, sometimes human interaction is infeasible, costly or impossible to achieve (such as in systems dispatched at distant locations). A minimization of human interaction may be required in such cases. A problem with this approach is that incomplete information does not make it possible to achieve a direct design method. We need to find another solution that allows for complementing the missing information in an automated way.

The situation is illustrated in Fig. 1. The figure is adopted from Ueda et al. [2] and refers to Yoshikawa’s General Design Theory [4], [5]. Ideally, design functions and the attributes realizing them are linked by mappings that reflect a complete knowledge. To handle incomplete knowledge without human intervention, we will consider an iterated process with changing function space and changing attribute space as in Fig. 2. Our aim is to study the possibility of minimal knowledge allocation by the autonomous generation and solution of a series of new design problems. This procedure may be especially relevant in unknown or unpredictable environments, like in the autonomous operation of populations of robots. At the same time it models certain features of design innovation where the introduction of new function spaces is critically difficult to achieve.

Our approach is very similar to that taken by Luh and Cheng [6] in an earlier paper of this Journal, using “reactive” instead of “deliberative” systems in robotics in a work informed from immunology. They write: “… robots must operate without the direct intervention of other agents (possibly human) and maintain interaction with their environment”. To support this, “… try to explore principles of the immune system focusing on its self-organization, adaptive capability, and … memory”. Our own work is motivated from evolutionary theory, where self-organization, adaptability and system memory offer a framework for coping with function space generation and hence design innovation as an example of Class III processes.

We will study how in an evolutionary system it is possible to achieve sustained evolution with a changing function space. Sustained evolution is known to be one of the most profound current challenges for Artificial Life modeling as argued repeatedly by different authors such as Kampis [7] and Holland [8]. In this paper we consider the challenge of sustained evolution to be equivalent with the problem of the production of new species (i.e. reproductively isolated sub-populations) with a different function space. We present both a theoretical model and a computer simulation to approach the question. In our model, the function space will be the evolutionary task space of selection and the attribute space equivalent with the evolutionary phenotype of artificial organisms.

Achieving open-ended evolution in design space is just a first step towards a minimal allocation design methodology, that is, an Evolutionary Technology [9]. Ideally, in Evolutionary Technology a system should be able to develop its own increasing task space by solving a hierarchy of real-world problems. In our model, to support the iterative process as in Fig. 2, an Evolution Engine is used that could easily be augmented in the above way with engineering tasks, like Genetic Algorithms and other adaptive search techniques. However, the novel step in our study is the testing of the generation and utilization of variability in the task space, which we consider the missing element to be solved. Therefore, we concentrate on the “free” evolution of a system with internal constraints only. In a next step, a complex and dynamically structured real world environment could be added to imply environmental problem solving. Here, again we follow the in-principle test of [6] where a purposeless abstract environment in the form of a set of discrete states was assumed in a similar minimal model.

Section snippets

The theoretical framework

In the model, we utilize the consequences of the changing dynamics of a phenotype-to-phenotype interaction system in a population of sexually reproducing agents.

The approach is based on the evolutionary application of “fat” interactions of natural causal processes as introduced by Kampis [9]. The causal interactional framework provides a natural tool for discussing the problem of the production of species understood as sub-populations with different function spaces. Such stable species’

Basics of the model

In the rest of the paper, we study the phenotype-based evolutionary dynamics of a simple sexual selection system in order to bring forth new species in the sense as discussed above. Our work was motivated by the hypothesis, corroborated by the results to be presented, that in a model of sexual selection, evolution can transform and finally split the population when genetic mutations (or other factors) produce individuals with new phenotype traits. This is because the new phenotype feeds back on

Specification of the model

Our experiments were conducted using the basic evolution engine that maintained a stable population. The engine defines a usual evolutionary setting and provides a sympatric environment without a spatial component. This engine simulates a partial artificial ecology with a single resource, energy. Each organism has an equal chance to ‘eat’ in every time step. (This fact introduces an implicit competition for energy, which leads to density-dependent effects. Genotypes represented with a higher

Computational results

We summarize our main results by first pointing out the observed dramatic difference between the ‘flatline’ behavior of the basic engine and the divergent, rich species production behavior of the same system when interaction change is allowed. Interaction change introduced, a formerly stable convergent species becomes more extended in property space, and finally it splits, giving rise to two or more new stable sub-populations, or species. The process can repeat itself several times, ultimately

Discussion

Using a simple sexual selection example our model demonstrates the validity of the in-principle claim that a changing interaction field can lead to emergent effects producing sustained evolution of populations or artificial organisms with a changing function space. A detailed sensitivity analysis is published elsewhere [14].

The presented model clearly lacks biological feasibility (dimensions are only added and never dropped; new properties are assigned too radically in the whole population,

Acknowledgments

Part of the work reported here was carried out during the first author’s stays in the School of Knowledge Science, JAIST, Japan, and in Kalamazoo College, MI, USA. The hospitality of these institutions, as well as the personal support of Professor S. Kunifuji (JAIST) and Professor Peter Erdi (Kalamazoo) is gratefully acknowledged. Computer simulations were done on the BeoWulf cluster of the Center for Complex Systems Studies, Physics Department, Kalamazoo College, as well as on the SUN E10K,

References (18)

  • K. Ueda et al.

    Emergent synthesis methodologies for manufacturing

    Annals of the CIRP

    (2001)
  • G.-C. Luh et al.

    Behavior-based intelligent mobile robot using an immunized reinforcement adaptive learning mechanism

    Advanced Engineering Informatics

    (2002)
  • K. Ueda, Introduction of the project ‘methodology of emergent synthesis’, in: Proceedings, Workshop on the Methodology...
  • T. Kito, N. Fujii, K. Ueda, Co-creative decision making in artifactual systems in consideration of bounded rationality,...
  • H. Yoshikawa

    General design theory and a CAD system

  • Y. Reich

    A critical review of general design theory

    Research in Engineering Design

    (1995)
  • G. Kampis, Complexity and causality, Complex Systems 7, JAIST report, Japan Advanced Institute for Science and...
  • J.H. Holland, Challenges for Agent-based Modeling, 2003, Available from:...
  • G. Kampis, A causal model of evolution, in: Proceedings of SEAL 02 (4th Asia-Pacific Conference on Simulated Evolution...
There are more references available in the full text version of this article.

Cited by (7)

  • Soft computing in engineering design - A review

    2008, Advanced Engineering Informatics
  • Cellphone evolution - Applying evolution theory to an info-communication system

    2012, 3rd IEEE International Conference on Cognitive Infocommunications, CogInfoCom 2012 - Proceedings
  • The species problem in artificial life

    2007, Proceedings of the 2007 IEEE Symposium on Artificial Life, CI-ALife 2007
View all citing articles on Scopus
View full text