How can we learn useful things from big data? Data mining from the perspective of Meno's problem

In modern physics, many data sets arise not because there is theoretical motivation to study a set of variables, but rather because new instruments allow for the speedy accumulation of huge sets of measurements. An important challenge then is to make scientific use of the data. As L. Floridi puts it, the challenge is to find small patterns in big data. The aim of this talk is to understand how methods of data mining may meet this challenge.

I approach this topic from the perspective of a puzzle presented in Plato's "Meno". There, it is argued that we cannot search for something yet unknown (nor investigate something yet unknown). For to claim success, we would have to have a criterion of success, and such a criterion may only be given if we knew what we are searching for, which we do not. Whereas the paradox can be resolved in a rather trivial way for many searches, it has more plausibility in the context of big data, because scientists are there looking for something they don't have any clue about.

My philosophical project thus is to explain how data mining may produce new knowledge despite the paradoxical conclusion from "Meno". I do so by presenting a case study from astrophysics and by analyzing representative examples of methods of data mining. My focus is on the aims of, assumptions behind, the methods.