Are You Hip Or Square?

Regular readers likely have noticed that oft times you'll see: "statistics isn't all that difficult, it's just squared differences". That's a bit dogmatic, but nearly always true of stat problems involving finding the central tendancy of a pile of numbers; regression being the most famous example. Even with non-linear models or wacky distributions, you'll see something like, "if we assume sufficiently large sample, we can use the normal approximation...", which then allows the pile of numbers to be used to calculate the parameters. Very convenient that.

Well, turns out, I'm not the only dogmatist in the world. And I'm not sure, but think his is a tad more so than mine:

All the impressive achievements of deep learning amount to just curve fitting.
-- Judea Pearl

The quote leads off the article, but is not from the mouth of the author; Smith sets out to show that Pearl is, basically, right. What's also interesting, from a teeVee point of view, is that the project being discussed is from Max Tegmark, who I see in numerous episodes of "How the Universe Works". My impression has been that he's mostly a naysayer when it comes to the more envelope pushing bits of cosmology.

So, no, AI isn't the next Newton or Einstein or Feynman. Interesting: AI is mostly being used to flog consumer widgets. Mammon always wins.