Rational and resilient protocols

Abstract

Cryptography and distributed computation have been very successful in advancing the study of interaction of distinct computing agents. Moreover, both fields have been very successful in conversing with each other, sharing models and techniques. Notably, they both model agents as being either 'good' (i.e., following their prescribed programs) or 'bad' (i.e., deviating from their prescribed program, by stopping, by acting maliciously, or even by coordinating their malicious strategies). I believe however, that we have been neglecting a fundamental ingredient, UTILITY, which has long been recognized and studied by another scientific field, game theory, and in particular by a beautiful subfield of it, mechanism design. Mechanism design aims at obtaining a desired outcome by engineering a game that, rationally played, yields a desired outcome. In such games, multiple players interact very much as in a cryptographic/distributed protocol. But here players are not good or malicious. Rather, every player is RATIONAL, that is, always acts so as to maximize HIS OWN utility. I believe that properly incorporating utility/rationality in our models will dramatically increase our range of action. Viceversa, mechanism design stands to gain a lot by properly incorporating cryptographic/distributed notions and techniques. In particular, rational players may, by colluding (and making side-payments to one another), increase their utilities. And they too value privacy, which may indeed represent their strategic interests in unforeseen and not yet modeled interactions. Thus, privacy and collusion can disrupt the intended course of an engineered game, and ultimately prevent a desired outcome from being achieved. Mechanism design has been only moderately successful in protecting against collusion, has largely ignored privacy, and might gain precious resiliency by taking into consideration our notions and techniques. In sum, there is an opportunity for cryptography, distributed computation, and mechanism design to join forces to study more general and accurate models of interaction, and to design more realistic and resilient protocols that simultaneously take into account utility, collusion, and privacy. No sufficiently complex and sufficiently large system, no organism can successfully work or merely sustain its existence without recognizing and harmonizing these basic forces. To be successful, this designing effort will require a good deal of modeling and the development of new conceptual frameworks. It will require open minds and open hearts, so as to leverage past and successful scientific experiences, without being trapped or confined by them. There is the promise of a great deal of fun, challenge, and excitement, and we must recruit as much talent as possible to this effort. As a concrete, simple, and hopefully provocative example, I will describe a (quite) resilient mechanism, designed by me and Jing Chen, for achieving a (quite) alternative revenue benchmark in unrestricted combinatorial auctions. In such auctions there are multiple distinct goods for sale, each player privately attributes an arbitrary value to any possible subset of the goods, and the seller has no information about the players' valuations.