What has Mill to say about data mining?

Data mining is an application that needs a theory. It is a significant application to which AI is well suited and applicable, and over the years a number of projects have attempted to build such systems with varying degrees of success. At least two major issues present themselves when examining a theoretical basis for data mining. First one must ask how the agents perform in the presence of inaccurate or incomplete data, and second one must ask what is the coverage or the "sorts of things" that the agents can discover. A theoretical underpinning for the first question is provided by probability theory and statistics. We examine the second question using the theory behind methods of discovery and discuss in particular the relevance of Mill's Methods and their basis in Bacon's Novum Organon. Although science in general has moved away from a purely mechanistic view of the world, the insights in the methods are particularly relevant to automated processes of discovery and duly provide a theoretical basis for assessing the coverage or capabilities of intelligent agents.< >