I will begin by providing some motivation for caring about the directed
information between two processes, beyond feedback communication. I will then
describe several approaches to estimating the directed information rate,
building on universal sequential probability assignments. The estimators inherit
much of the algorithmic simplicity and universality properties of the latter, in
senses I will briefly describe. I'll conclude by showcasing the results of a few
experiments, demonstrating the potential of the proposed schemes in practice and
the efficacy of directed information estimation as a tool for detecting and
measuring both causality and delay. 
Based on a pleasurable collaboration with Jiantao Jiao, Haim Permuter, Lei
Zhao, and Young-Han Kim.