Agency-GP: Agent-Based Genetic Programming for Design

Agency-GP is a prototype for a system using genetic programming and software agents for architectural design exploration. Its software structure is noteworthy for its integration into a high-end three dimensional modeling environment, its allowance for direct user interruption of evolution and reintegration of modified individuals, and its agent-based evaluation of fitness. 1 BACKGROUND AND MOTIVATION The Emergent Design Group at Massachusetts Institute of Technology (web.mit.edu/arch/edg) conducts research into architectural morphology and the emergent and adaptive properties of architectural form. Our in-house developed software tools exploit models of "natural computation". Natural processes can be viewed both as pragmatic, successful problem solving activities and, as generators of interesting and aesthetic physical or dynamic outcomes. For this reason, they are compelling to abstract and use for inspiration in the implementation of computer software tools for Architecture. In the context of our work, our natural computation design tools variously employ generative algorithms that respond to environmental properties (Hemberg, 2001, or evolutionary computation. Evolutionary computation makes sense for architectural design both in the role of optimization (not to be discussed herein) and in the role of form design. A key reason is that it can create novel, adapted (and often unanticipated) forms without requiring explicit input that directly specifies how to formulate or reformulate a form. In evolutionary computation the architect instead supplies operative criteria (e.g. the state of a form's environment, how a form should fulfill architectural program) or directive information that guides or influences the adaptive process. This provides a means of "saying what is wanted'' rather than how it must be designed. It furnishes an architect with a less constrained approach. The inheritance and variation mechanisms of evolutionary computation contribute to apparent form "revisions'' that are neither totally random nor totally deterministic. This spontaneity combined with the coherent genealogical process intuitively coordinates with the human design process. Furthermore, evolutionary computation can use specifications of complex and divergent form properties to judge a solution. This implies it has the potential to explore solutions that resolve partially conflicting requirements. In the sense that evolutionary computation is used to pursue a populationbased adaptive process rather than optimization, it suits architecture because architects are interested in families of forms rather than finding one absolute best. Also, evolutionary computation allows architects to creatively explore a vastly expanded form space that is too large to investigate by hand. Figure 1. An evolved form from Agency-GP. Shadings denote architectural program. Agents were used to operationalize neighbor constraints, program constraints and height and size constraints. Agency-GP is the software arm of a larger project called AGENCY (web.mit.edu/arch/edg/agency) currently under way at the Emergent Design group. The project seeks architectural responses to the radical transformation that business organizations are currently undergoing. The pace of organizational change is being driven by the rapid development of commercial technology, global markets and reengineered institutions. This constant need to change gives rise to organizational structures that are no longer stable, but continuously adapting to their shifting environments. Such structures can be said to be "emergent" and describe many of today's commercial and governmental organizations. Vertically structured office buildings no longer provide the model for most businesses. With the advent of widespread use of telecommunications, information technology, and corporate reorganization in the 1990's, new forces are actively reshaping the architecture of office buildings. There is a shift in the United States toward research and development, management and finance, consultancy, and the culture industry, productive activities less prone to standardization and bureaucratization. Driven by the demand to improve office productivity, businesses and organizations have begun to experiment with a variety of alternative officing methods. However, there exist no working models of an intelligent adaptive architecture. The AGENCY project focuses on application-oriented basic research to develop new design software that generatively models the complex interactions of physical space and information technology within emergent