Présentation virtuelle de Aron Lindbergh, Assistant Professor of IS, Stevens Institute of Technology
Using Traditional vs. Autonomous Design Tools: Design Problems and Performance Differentials (by Aron Lindberg, Stefan Seidel, Corinne Coen, and Michael Gau). Autonomous design tools are increasingly used across different design fields, including semiconductor chip design, video game design, and generative design of architecture and engineering products. These tools differ from traditional design tools—-such as for drawing or computer-aided design—-in that they carry out design tasks and make decisions on behalf of the designer in a largely independent fashion. Using algorithms, ranging from applying heuristics to machine learning, these tools are difficult to understand and are often used with low degrees of intermittent human intervention. Hence, they are often used in an experimental fashion, allowing designers to produce multiple designs at high speeds, which can then be evaluated post hoc. To understand the key characteristics of autonomous tools and how these are related to their effectiveness in different design situations, we conducted a series of agent-based modeling simulations. Our results indicate that designers using autonomous design tools are more effective in simple design situations (situations with fewer possible solutions), while human designers using traditional tools are more effective in complex design situations where a large number of possible solutions exists. We suggest that performance differences can be explained by variables pertaining to both the fitness landscape and the properties of the design system (i.e., designers using traditional or autonomous tools) in terms of range of vision and number of restarts. Our findings hold implications for when traditional or autonomous tools are best used, and how they can be combined across various design processes.