Game-theoretic Risk Assessment for Distributed Systems (GRADS)
Report Number: SERC-2019-TR-011
Publication Date: 2019-07-02
Project: Game-theoretic Risk Assessment for Distributed Systems (GRADS)
Dr. Paul Grogan
Distributed or collaborative system architectures present a fundamental strategic tradeoff in the design of engineering systems in large-scale domains such as defense and critical infrastructure. Effective collaboration among multiple design actors can lead to improved emergent capabilities and opportunities for cost sharing to field systems impossible to achieve through independent action. However, collaboration also requires a non-reversible commitment of effort that presents a fundamental source of uncertainty and can lead to coordination failures with significant cost penalties. Thus, interdependencies among design actors contributes both an upside potential and a downside risk associated with collaboration.
This project introduces a game-theoretic perspective to assess and evaluate the risk of collaboration in distributed architectures. Design actors are modeled as playing a binary coordination game with two alternatives: either to pursue an independent architecture or a collaborative architecture. Economic literature of equilibrium selection in game theory helps to measure the strategic dynamics present among design actors using a property called "risk dominance." Risk dominance, like payoff dominance for an alternative that is most preferred for all actors, is a desirable property that identifies favorable strategic dynamics and can help to guide architecture decisions under essential uncertainty in collaboration.
Starting from the foundational economic theory, this project proposes a new design method that uses risk dominance to assess the relative stability of collaborative architectures. The analysis is oriented towards conceptual design phases before significant cost commitments have been made and while the architecture is still fluid. A validation study using multi-agent simulation shows how the risk dominance metric helps to differentiate between dominant strategies (collaborate versus remain independent) under varying contextual conditions. Results show that each context is somewhat unique to interpret point values of risk dominance; however, generalizable patterns among relative values can guide strategic decision-making.
Finally, this project demonstrates how such a risk dominance measure can be applied to inform a realistic architecture decision using an example of a joint polar-orbiting satellite program similar to the National Polar-orbiting Environmental Operational Satellite System (NPOESS). NPOESS was envisioned to be a collaborative architecture between multiple government agencies to replace existing independent missions; however, it was ultimately canceled after more than a decade of turbulent history due to excessive cost overruns. Using a game-theoretic assessment method similar to that proposed in this project could help to evaluate the relative stability of a proposed joint architecture relative to other proposed joint architectures to help give collaborative projects the greatest opportunity for success.