PostScript
Adobe PDF
|
PostScript |
Adobe PDF |
To exhibit some autonomy and to be able to treat unexpected situations, a system must have the possiblity to monitor its own behavior: it must be able to assess, analyze and possibly fix its own behavior. A natural extension to this approach is to give the monitoring system the possibility to apply upon itself. Some researchers have claimed that this is impossible because it would lead to infinite regress. The claim of this paper is that a necessary condition for a self-monitoring system to avoid infinite regress is that its knowledge is sufficient for tackling the monitoring system possible misbehaviors. We call this condition "reflective self-sufficiency". On one hand, this approach eliminates possible infinite regress. On the other hand, this has the advantage of reducing the amount of knowledge required by the system. We support our claim by describing the experiments conducted with the SADE system. SADE has been designed as a testbench for meta-level architectures. In its present version, the main concern of SADE's monitoring is to prevent both looping and stopping, which is a minimum necessary condition for survival. We show from various examples how to implement reflective self-sufficiency in this case.
