The research objective of this EArly-Concept Grant for Exploratory Research (EAGER) award will be to develop a proof of concept system that integrates value-driven agents into an agent-based information processing workflow simulation of design organizations to develop new capabilities for simulating design decision-making for complex systems. The investigators will develop an agent-based simulation of large design organizations in which agents are capable of making design decisions in response to requirements or various forms of incentives. This will provide a controlled environment in which to explore five research questions on Values, Incentives, and Hierarchy - all identified as being critical to successful design of complex systems. The investigators will also provide a simulation which can be validated against actual design organizations.
If successful, the research resulting from this EAGER award will lay the foundation for the future development of a robust agent-based simulation environment that can be used by the systems engineering research community to study fundamental sociological and cognitive phenomena in the detailed design phase of the development of large scale complex systems like aircraft and automobiles. Tens of billions of dollars per year are lost to cost overruns and schedule delays in the development of large scale complex systems. The investigators propose to reverse these losses through a research agenda to employ the social sciences to understand more fundamentally how large organizations design complex products. This exploratory research will enable a reformation of present processes for designing large scale complex systems.
Decision-Making Artificial Agents Are Used to Study Engineering Design Organizations Raymond Levitt at Stanford University, Daniel Shapiro at the Institute for the Study of Learning and Expertise, and Paul Collopy at the Value-Driven Design Institute have succeeded in embedding decision-making agents inside an agent-based simulation of an engineering design team. The NSF-funded research project modeled a team of engineers designing a jet engine compressor blade, and choosing the blade size and material. The team discovered that the hardest design decision is choosing whether a design is good enough, or whether the engineering team should spend more effort to improve it. Designing large complex systems such as Boeing airplanes or NASA launch rockets is a social activity where thousands of engineers may work together. Large engineering projects often suffer huge cost overruns and years of delays. It is important to understand what causes these problems, but very difficult to observe large organizations at a detailed level. So Levitt, Shapiro, and Collopy propose to study design teams in simulation. Levitt has already developed Virtual Design Team, a widely used agent-based simulation that models an organization as an information processing system. By putting agents that can make decisions inside Virtual Design Team, the researchers will be able to study the way design choices interact and change the results of large engineering teams. Agent-based simulations are now widely used to study complex phenomena, from economic markets to traffic jams. The advantage of using agent-based simulations is that scientists can look at very detailed interactions between the agents and relate these to complex behaviors of the whole system in a controlled environment. However, the agents used in these simulations are simple. They follow a set of rules, and the rules generate their behavior. Artificial intelligence experts call these "thin" agents. Daniel Shapiro designs "thick" agents, agents with cognitive skills. Thick agents have preferences and beliefs. Shapiro says his agents "might choose to work on the compressor blade, or they might choose to steal the office furniture." Agent-based simulations using thick agents open up a whole new horizon of possibilities for studying complex social phenomena in simulation. The Value-Driven Design Institute, with the support of NSF and the Department of Defense, is working to reduce the time and cost of designing and developing aircraft, spacecraft and other very complex systems. Today, these systems cost tens of billions of dollars more than they should, and development takes years longer than we expect. In a future where we are more proficient at large system development, travel to the moon and Mars will become more practical, vacations in space will not be unusual, and conveniences like air travel will be less expensive with less delays. To improve our engineering and development processes, we first need to understand how large engineering projects work. It is amazing that ten thousand people can cooperate to produce an aircraft which is so complex that no single person understands even a small fraction of its most intricate details. An agent-based simulation of a design team with cognitive agents making design decisions is a first step toward a more detailed understanding of large design teams, and eventually toward developing methods to perform very large design projects at less cost, without delays, and with better results.
