Game-theoretic Risk Assessment for Distributed Systems (GRADS)
Dr. Paul Grogan
Dr. Paul Grogan
Interest in distributed system architectures presents an important tradeoff in conceptual design activities. Distributed systems often have superior performance compared to monolithic alternatives by increasing flexibility in phased deployment or operations, robustness to individual component failures, and resource efficiency to match available resources to localized demands. However, distributed architectures also introduce new interdependencies between components which can lead to poor performance if not understood or anticipated due to cascading failures and loss of critical functions. This paradoxical relationship has been described as “robust yet fragile” as efforts to increase robustness through greater coupling expose innate fragility from corresponding internal interdependencies.
Interdependencies are not limited to physical or logical connections between components. Multi-actor or federated systems pose engineering design problems where interacting decision-makers pursue individual objectives. Each actor follows a strategic decision to either work independently or cooperatively. A cooperative strategy, analogous to a distributed architecture, pursues mutual benefits at the cost of additional interdependencies between actors. Misunderstanding or inability to anticipate these interdependencies can yield similar “robust yet fragile” behavior, where the resulting joint pursuit may become inferior to an independent solution. Acquisition programs requiring significant inter-agency collaboration occasionally have substantial overruns, illustrating the potential magnitude of this problem. This project proposes using game-theoretic concepts and assessment metrics to help understand the strategic dynamics underlying a collaborative system architecture and to identify, mitigate, or augment undesirable dynamics before significant resources have been invested.
The research objectives of this project are to:
- Develop a body of knowledge and methodology for game-theoretic measures based on the equilibrium selection theory to assess risk dominance of collective design problems,
- Demonstrate the proposed method on a representative case in the space domain, and
- Validate results of the application case by selectively forming and dissolving joint programs to demonstrate the utility of such a metric in systems engineering problems.
From a theoretical perspective, this project investigates how applications of multi-actor value models and, specifically, the weighted average log measure (WALM) of risk dominance proposed by Selten (1995) contribute to a new design methodology to assess strategic dynamics in collaborative system design problems and evaluate stability of collective decisions. This thread of research interprets and evaluates WALM's underlying assumptions and evaluates its applicability to inform design decision-making in joint engineering projects. As the WALM of risk dominance is defined in a highly simplified context of a single-shot decision, a validation study uses multi-agent simulation to understand the relationship between the numerical outputs and dominant strategies over longer evolutionary periods within an actor population.
From a more practical perspective, this project applies the WALM of risk dominance to an application case based on the National Polar-orbiting Operational Environment Satellite System (NPOESS) as a joint program between NASA, NOAA, and the DoD. This application case establishes a design scenario, formulates a multi-actor value model, identifies a proposed joint architecture, computes risk dominance measures, and validates the stability of resulting joint programs. This thread of research provides a concrete example to evaluate the usefulness of the proposed methodology to gain insight about whether the NPOESS architecture carried unfavorable or unstable risk dynamics susceptible to dissolution of the coalition of partners.
The overall proposed design methodology aims to transition fundamental theory from the field of economics and game theory closer to practice in systems engineering by representing and modeling value among multiple actors and computing metrics to assess the stability of strategic dynamics. It advances systems engineering methods, processes, and tools by evaluating anticipated outcomes of joint projects earlier in the conceptual design phase. The results of this work are expected to improve system architecting to identify favorable strategic dynamics and increase the stability of joint programs.