Distributed Meta-Management for Self-Protection and Self-Explanation

A non-hierarchical system in which distributed meta-levels mutually monitor and repair each other can act collectively as a self-protective system. Such a system can resist attacks against any part of itself, including its self-monitoring and self-repair processes. A distributed self-model is acquired by a process of bootstrapping where the different components learn by mutually ob-serving each others ’ internal processes. In this paper, we argue that such a decentralised architecture is also capable of high level reasoning and self-explanation. Using some examples, we show how objectively de-fined interactions between distributed software compo-nents can be translated into a subjective narrative about the agent’s own mental states. Furthermore, the global coordination needed to provide coherent actions and ex-planations is possible in a decentralised architecture, provided that the meta-level which is managing the co-ordination is also subjected to critical monitoring and modification. An important principle is that the dif-ferent meta-levels are not necessarily static components but are best understood as different roles, for which the same components may be re-used in multiple contexts