it is hard (sometimes impossible )to establish causal relationship purely from a statistical model. Science, especially mathematics and physics, build causal models.
Yes, that was very good. And it was also pretty much the only "useful" thing I got from the whole talk. I think even wise experts need constraints and structure to do and say meaningful things. We all have a tendency to bloviate, apparently even the intelligent ones among us.
No wonder, his knowledge of Bayesian stat could be traced to Francis Galton. Advisor tree: Francis Galton advised Karl Pearson, Karl Pearson advised John Wishart, John Wishart advised William Cochran, William Cochrane advised Donald Rubin, Donald Rubin advised Andrew Gelman.
Everything he's talking about are signs of a mature scholar, where he's realizing that the most fundamental issues are not what's "in the box," but what's "outside" of it. He's graduated from statistics to the philosophy of statistics.
Great video, although I don't understand how come a Foundations of Probability seminar has come to have half of its audience not knowing what sigma/sqrt(n) is.