The Black Swan

The Impact of the Highly Improbable

by Nassim Nicholas Taleb


Simultaneously better and worse than I expected. (2nd edition is much better than the 1st, includes a section re: his argumentative errors).

The book is tripartite: Primo, Umberto Eco’s Antilibrary deals with “how we humans deal with knowledge” with sometimes-clever anecdotes and philosophical arguments; Secundo, We Just Can’t Predict rails against the ability of humans to predict complex systems (with mentions of Phil Tetlock); Terso, Those Grey Swans of Extremistan attempts to add mathematical rigor to his arguments against the Gaussian.

(I previously had some stylistic criticism here, but I don’t think it was good enough to be worth anyone’s time.)


Humans turn things into stories. Taleb claims we do this because it condenses the inherently complex world around us into easily digestible information, and as a result we constantly run the risk of running afoul of the randomness of the world (when the information we use to construct our models implicitly excludes possibly relevant data, our models will be more simplistic than they should be).

Post-hoc rationalization is simply a corollary of this tendency - we view the past as a story as well (it is the job of the historian to make history make sense with stories, to ignore the role of complexity and random chance). Sometimes the narratives constructed based on “principles” and “thought” fail (the Ludic Fallacy, or why the layman can predict things as well as or better than an expert). Taleb claims narrativization (among other things) has subsumed the social sciences, specifically economics, and blinds academia to unknown unknowns, randomness, and the incompetence of their own economic theory.

Which of these two statements seems more likely? Joey seemed happily married. He killed his wife. Joey seemed happily married. He killed his wife to get her inheritance.

The second makes more sense, while the first is more likely (the second is a strict subset of the first).

Mediocristan vs. Extremistan

I view the heart of NNT’s argument against the Gaussian as 1) the Gaussian cannot exist in Extremistan and 2) Extremistan is where many (if not most) important things lie.

To understand Mediocristan and Extremistan, we must first understand the dichotomy of Type 1 and Type 2 randomness: Type 1 obeys the law of large numbers, such that with a large sample an individual outlier will not change the aggregate that much. Type 2 does not follow this rule. (Stereotypical example of Type 1 is the Gaussian, that of Type 2 is a Power Law distribution). Mediocristan is where Type 1 randomness exists, Extremistan is that where Type 2 exists.

Extremistan is characterized by “scalability” - where it is possible for a few to monopolize most of the value in a market (he uses the example of Horowitz’s classical piano records vs. the local live performance - music is now scalable, therefore lives in Extremistan). He claims that luck plays a large role in who succeeds and who fails in Extremistan, and that these worlds are extremely Black Swan prone. The failures in current modeling stem from modeling Extremistani environments as Mediocristan (Gaussian) ones.

Later, in Part Three, NNT introduces the idea of “fractal randomness”: randomness which can appear superficially Gaussian, but is not; randomness which has similar variation at all levels; randomness that can appear to be in Mediocristan but actually exists in Extremistan. While I think his association to fracticality is because of his adoration of Mandelbrot (not underserved, for what it’s worth), it is true that assuming scale-invariant variation in data gives you good models of things (Pareto models, 80/20 law, the distribution of wealth in the top 1% is similar to the overall distribution of wealth). He claims this turns Black Swans into Grey Swans, I somewhat agree.

Assorted Philosophizing

Taleb clearly has Heroes: Poincaré, Popper, Mandelbrot, etc. (he even says this himself):
Half the people I know call me irreverent (you have read my comments about your local Platonified professors), half call me fawning (you have seen my slavish devotion to Huet, Bayle, Popper, Poincaré, Montaigne, Hayek, and others).

His heroes are very interesting. Popper has shown up as the fundamental critic of historicity, Hayek as one of the only sane economists, Poincaré as a luminary polymath, and the anecdotes he drops make for good starting points for reading material (apparently Yevegenia was fictional, so that doesn’t count here).