Friday, February 22, 2013

Anti-Fragility

As I read and listen to Nicholas Nassim Taleb on the topic of his coined word, anti-fragility, I struggle to keep an open mind but I increasingly begin to feel he's jumped the shark.  He claims to be defining something that is not resilience, because an anti-fragile system is one that actually gains from disorder/stressors/events. But when he gives examples, each one makes me immediately think "But..."

In his conversation with Daniel Kahneman, he offers a typical example. He talks about how his glass of water is not anti-fragile, because if there is an earthquake it will likely shatter, it will certainly not benefit from that unexpected event. Hard to argue against the idea that a glass is not anti-fragile, though beware of the subtle psychological bias that introduces towards making you believe the concept actually exists, simply because we can point to its opposite (much as the existence of God is sometimes argued to imply the existence of Satan or vice versa).

But then, he claims that we can actually measure fragility. His example again is the earthquake: which is likely to break first, the glass or the table? Obviously, the table is "more" fragile than the glass. Well, maybe, but...

First of all, which will break first in an earthquake? The real answer (as someone who wrote Fooled by Randomness and The Black Swan surely knows) is: We Don't Know.  Like anyone who grew up in Kansas knows, tornadoes invariably leave in their wake little miracles of chance. A house explodes into chaos, but there on the front lawn is a carton of eggs intact, or a glass vase unscratched. A straw might be driven 6 inches into a tree by the same tornado that picks up an infant and drops it many yards away unscathed.

Of course, statistically, the eggs/glass/infant are more fragile than a tree, but Taleb's writing career has been all about not being deceived by incomplete statistics and the Flaw of Averages. Well, OK, this is not the strongest critique. Let's try another.

What if the "unexpected" (as soon as we envision it, it's no longer unexpected!) event was a fire instead of an earthquake? Now which is likely to be destroyed first, the glass or the table? Since glass melts at about 2700F while wood ignites at perhaps 500F, the tables have turned for the little glass, so to speak, and I begin to like its odds much better than the table. What "measure" of fragility can account for this reversal of fortune?

One of Taleb's go-to examples for anti-fragility is that the "stressor" of gravity causes your bones to get stronger. So your body actually "benefits from stress". Here, I can no longer find anything to agree with. If you become obese, your bones will also get stronger. Does that mean your body benefits from the stress of obesity?

And in what sense is the skeletal system in your body actually able to benefit from unexpected events? None, I think. It seems much more accurate to me to say that the skeletal system responds to various signals that evolution has taught it about, said instruction residing in DNA. Your skeletal system isn't anti-fragile, it doesn't gain from stressors, it doesn't benefit from the unexpected -- it just adapts to incoming events as best it can given its limited model of possible events and responses.

Taleb frames the weightlessness of space (in which astronauts lose bone density stunningly quickly if they don't relentlessly exercise) as a lack of stressors. But it can be equally well framed as sending the skeletal system a truly unexpected signal (zero gravity). How does the skeletal system handle the truly unexpected? Disastrously, by decreasing bone density to the point that, even in weightlessness, the organism becomes highly likely to suffer life-threatening injury.

Organisms are not anti-fragile, they don't benefit from the truly unexpected. Instead, organisms have models of reality that they use to try to adapt when (possibly rare but envisioned) expected events happen. We already have a perfectly fine word for this, resilience, which Taleb insists is inferior to the (hypothetical) anti-fragility.

I speculate that anti-fragility is Taleb's own nemesis (the unpredictability of the future) come back to haunt him. It's ironic that in his conversation with Kahneman, immediately after Kahneman asserts that they both believe it is arrogant to predict the future and we shouldn't try, Taleb responds with the example of the glass and the table, asserting that "we can measure the fragility" and "I can't predict the event that will break the [glass], but I know if there's an event what will break first, the table will break after the [glass]". Inherent in his claimed ability to measure fragility is a claim to predict the future. Since he didn't anticipate the "event" might turn out to be a fire, he feels assured that he knows the glass is more fragile than the table.

I suspect this example of how judging fragility implicitly requires predicting the future undermines the entire idea of "anti-fragility". It is very hard (perhaps impossible) to give up our desire to predict the future.