Herzig, Andreas and Maffre, Faustine (2017) How to share knowledge by gossiping. AI Communications, vol. 30 (n° 1). pp. 1-17.

Preview

Text Download (174kB) | Preview

Abstract

We provide a logical investigation of a simple case of communication in a network of agents called the gossip problem. Its classical version is: given n agents each of which has a secret - a fact not known to anybody else -, how many calls does it take to achieve shared knowledge of all secrets, i.e., to reach a state where every agent knows every secret? Several protocols achieving shared knowledge in 2(né2) calls exist and were proved to be optimal: no shorter sequence of calls exists. We generalize that problem and focus on higher-order shared knowledge: how many calls does it take to obtain that everybody knows that everybody knows all secrets? More generally, how many calls does it take to obtain shared knowledge of order k? This cannot be achieved simply by communicating facts: the agents also have to communicate higher-order knowledge of facts. We give an algorithm that works in (k+1)(né2) calls. We analyse its properties in a logic that we have investigated in previous work and that is based on the concept of observability of propositional variables by agents: Dynamic Epistemic Logic of Propositional Assignment and Observation DEL-PAO. This enables us in particular to give a formal proof of correctness of the algorithm.