Algorithmic simulation in system design and innovation

Abstract

This thesis explores the use of genetic programming as a tool in the system design and innovation process. Digital circuits are used as a proxy for complex technological designs. Circuit construction is simulated through a computer algorithm which assembles circuit designs in an attempt to reach specified design goals. Complex designs can be obtained by repeatedly combining simpler components, often called building blocks, which were created earlier in the algorithm's progression. This process is arguably a reflection of the traditional development path of systems engineering and technological innovation. The choice of algorithm used to guide this process is crucial. This thesis considers two general types of algorithms-a blind random search method, and a genetic programming search method-with variations applied to each. The research focused on comparing these algorithms in regard to: 1) the successful creation of multiple complex designs; 2) resources utilized in achieving a design of a given complexity; and 3) the inferred time dependence of technological improvement resulting from the process. Also of interest was whether these algorithms would exhibit exponential rates of improvement of the virtual technologies being created, as is seen in real-world innovation. The starting point was the hypothesis that the genetic programming approach might be superior to the random search method. The results found however that the genetic programming algorithm did not outperform the blind random search algorithm, and in fact failed to produce the desired circuit design goals. This unexpected outcome is believed to result from the structure of the circuit design process, and from certain shortcomings in the genetic programming algorithm used. This work also examines the relationship of issues and considerations (such as cost, complexity, performance, and efficiency) faced in these virtual design realms to managerial strategy and how insights from these experiments might be applied to real-world engineering and design challenges. Algorithmic simulation approaches, including genetic programming, are found to be powerful tools, having demonstrated impressive performance in bounded domains. However, their utility to systems engineering processes remains unproven. Therefore, use of these algorithmic tools and their integration into the human creative process is discussed as a challenge and an area needing further research.