Intelligent Data Analysis of Intelligent Systems

We consider the value of structured priors in the analysis of data sampled from complex adaptive systems. We propose that adaptive dynamics entails basic constraints (memory, information processing) and features (optimization and evolutionary history) that serve to significantly narrow search spaces and candidate parameter values. We suggest that the property of “adaptive self-awareness”, when applicable, further constrains model selection, such that predictive statistical models converge on a systems own internal representation of regularities. Principled model building should therefore begin by identifying a hierarchy of increasingly constrained models based on the adaptive properties of the study system.