Acquiring a Self-Model to Enable Autonomous Recovery from Faults and Intrusions
C.M. Kennedy,, Aaron Sloman · Journal of Intelligent Systems · 2002
In this paper, we explore reflective architectures which enable an autonomous system to build a model of its own operation and use this model for the purpose of survival (including anomaly-detection and self-repair) in a hostile environment.The simplest way to introduce reflection is to add a reflective layer or meta-level which monitors and protects the system's capability to operate.If the meta-level is damaged, however, it cannot reliably be inspected or repaired by the system itself.Our solution to this problem is to distribute the reflection so that components mutually observe and protect each other.This involves two stages: first, multiple versions of the anomalydetection system acquire models of each other's 'normal' patterns of behavior by mutual observation in a protected environment; secondly, they compare their models against actual patterns in an environment allowing intrusions, in order to detect and repair anomalies in each other's behavior.