The biggest practical advantage of MI is that it applies to categorical data as well as numerical data, whereas correlation only applies to numerical data (or at least, in the case of ordinal correlations, data that can be sorted from “smallest” to “largest” even if it's not strictly numerical).
