Of Data Mining, Markets, and Second - Order Effects

A question for the chaos theorists: suppose we have a system whose behavior is a function of the independent actions of multiple agents, and that these actions are in turn occasioned by the agents' expectations regarding the future behavior of the system based on detected patterns in data.  Can the scope of pattern and the speed and efficiency with which it is detected have implications for the overall stability of the system, independent of the ontology of the data schema and the methodology that determines a pattern's salience?  And if scope, speed, and efficiency of detection do have implications for the stability of the system, can the choice of ontology and methodology mitigate or exacerbate these effects?

Case in point:

http://www.theatlantic.com/technology/archive/2010/12/wall-streets-latest-bubble-machines/68547/

The concerns this article raises are not small - though I wonder if the author fully appreciates the general importance of his claim:  self-regulating mechanisms for a market system like the NYSE, which are completely sufficient when the pattern detection capabilities of market agents remain within the speed and scope of human ability, may cease to be so if pattern recognition can be accomplished within a spatiotemporal scope that is some orders of magnitude larger than what humans are capable of, or with a speed that is some orders of magnitude smaller.  What had been a stable market may then become highly volatile in the event such pattern recognition capability is introduced. 

Whether or not this is the true is, to my mind, an empirically open issue, but one that is well-deserving of investigation for obvious reasons.