Efficient Mechanisms for Socio-technical Systems
| Modern society is based on large-scale engineered systems, often at the service of human end-users, e.g., transportation and power networks. While the control of such systems is typically grounded on purely engineering principles, their performance greatly depends on how human users interact with them. A common is- sue arising in these settings is the performance degradation often incurred when the users’ interests are not aligned to the “greater good” (e.g., traffic routing). In this context, a natural question arises: how can we design behavior-influencing mechanisms to incentivize efficient use of the existing infrastructure? In this presentation, I answer this question in relation to the well-studied class of congestion games, of- ten used to model traffic assignment problems. More precisely, I show how to design mechanisms that utilize only local information, and robustly maximize the system efficiency. Surprisingly, optimal mechanisms designed using only local information perform closely to those designed using full information (1% difference for affine latency functions). Additionally, I show how the proposed approach recovers and generalizes a number of well-known results in the literature. Finally, I discuss how the marginal cost mechanism, known to be optimal in the continuous-flow approximation, results in a lower efficiency than that encountered if no mechanism was used.