In 2000 the John Holland wrote concerning Learning Classifier Systems (LCS)
In recent years there has been a focus on classifier systems as performance systems or evolutionary incarnations of reinforcement systems . This has been fruitful but, I think, falls short of the potential of classifier systems. Smith makes the point that classifier systems, in their ability to innovate, offer broader vistas than reinforcement learning. In my opinion the single most important area for investigation vis-a-vis classifier systems is the formation of default hierarchies, in both time (``bridging classifiers'' that stay active over extended periods of time) and space (hierarchies of defaults and exceptions). Having been unable to convince either colleagues or students of the importance of such an investigation, I intend to pursue it myself next summer (year 2000).
Is anyone aware of whether Holland's ideas were taken up and developed esp. in the area of non-stationary learning?