Comment by konstruktor | Hacker News Reader

Comment by konstruktor | Hacker News Reader

She obviously fails to grasp what Taleb calls a Black Swan, or ignores his definition for some cheap publicity: "The attacks of 9/11 were not black swans, she said. The FBI knew that questionable people were taking ﬂying lessons on large aircraft."

They are the perfect example of a Black Swan. Quote Taleb (Black Swan, p.xxii): "... in spite of its outlier status, human nature makes us concoct explanations for its occurence after the fact, making it explainable and predictable", which is exectly what she is doing. And, one page later on 9/11: "had the risk been easily conceivable on September 10, it would not have happened".

He is not advocating abandoning risk management, he is in favour of risk management that doesn't need us to predict the future, as it is harder to reliably estimate the likeliness of very unlikely events.

Concerning the second example of earthquake risk and nuclear power plants: That, again, is a post-hoc rationalization. Everybody knew that earthquakes were a risk factor for NPP and was planning accordingly. They were just not planning for a Tsunami this size, as it was very, very unlikely. So including higher error margins for earthquakes next time is nice but not enough. Taleb on this: http://www.valuewalk.com/2011/03/nassim-taleb-black-swans/

Consider the aviation industry: After an accident, they find the root cause and eliminate it. It is now something expected, and can be directly dealt with. But also try to improve the system (e.g. via training) to be more robust towards all the root causes the didn't anticipate.

> They are the perfect example of a Black Swan.

I disagree.

The black swan is something that is completely unforeseeable and for which there are no previous partial or complete examples, either of the final outcome or the contributing causes.

Which is her point: 9/11 was foreseeable from the information (the failure was in connecting it) and the fact that a previous, similar plot had been tried in France. The clues were there and there was a previous partial example.

> The black swan is something that is completely unforeseeable and for which there are no previous partial or complete examples, either of the final outcome or the contributing causes.

There is no such thing as something completely, entirely 100% unforeseen or unforeseeable. Someone, somewhere in the world will be seeing it coming... just as some intelligence analysts saw 9-11 coming, some economists saw the financial crisis, etc. etc. What we're considering is what the prevailing and strongly established consensus well outside the fringe areas says.

This is where we start to wander into problems with classical set logic.

No, I'm not joking. Fuzzy sets are pretty much required for any meaningful discussion of "failed", "foreseeable" etc, if they are to be at all useful concepts: http://chester.id.au/2012/04/09/review-drift-into-failure/

Reduce foreseeability and failure to a binary toggle and you destroy enormous amounts of information with high utility so that some syllogisms still work. Wasteful.

Taleb is a very intelligent man, but AFAICT he does often reinvent existing concepts with much cooler names. "Antifragile" sounds awesome. "Robust" sounds boring.

Take "black swan", for example. Given the technology of the day, black swans simply didn't exist. Iain Banks called these "Outside Context Problems", one might also call them "paradigm-busters".

Anyhow. I should have padded out my original definition with the usual legalese about "reasonably foreseeable".

Relabelling is the marketing game; to get your businessy-idea-book in the best-sellers list you need everyone to replace "practice makes perfect" with "10000 hours".

Yep. And in fact marketers didn't invent relabelling (Tallyrand had a witty remark that one of the functions of the State is to rename odious institutions and I would not be surprised if Confucius had this phenomenon in mind when he got grumpy about the Rectification of the Names).

"reasonably forseeable" is hard to define. Let us, for a moment, use being able to implement countermeasures without seeming like a crazy lunatic as a definition. As a politician in a pre-9/11 world, if you had done what would have been necessary to prevent a 9/11 (not go and arrest the very guys who did it, that would have required perfect foresight), i.e. add bulletproof doors to planes, implement the very invasive searches the TSA does etc., maybe ground commercial aviation until those measures would have been in place, it would be the end of your career. The remote possibility of a terrorist attack has only become a justification for about everything after 9/11. Also, 9/11 was just one of many terrorist threats at that time, it's just in hindsight that we consider this particular one inevitable. Edit: Spelling

Hijackings had happened many times. Including with weapons. There was plenty of knowledge that could have easily justified adding secure, lockable cockpit doors. And that would have been a good, low impact way of mitigating against known risks while also taking a whole host of unknown risks off the table or substantially reducing them, including reducing the chance of 9/11 having the impact it did or even being tried.

I think that is one of the changes that more focus on mitigating risk vs. avoiding it might have produced. E.g. when faced with heightened risk of terrorist attacks, a rational response would be to not focus so much on identifying potential attackers and stopping a specific plot, but spending more resources at looking into low impact ways to mitigate the consequences of at various broad modes of attack actually getting underway.

The specifics of 9/11 was probably near impossible to predict, but another eventual hijacking was a near certainty, and an eventual building collapse for whatever reason should have been considered a near certainty, as many high-rises have failed over the years too. The failure to reduce the potential impact of those known risks are the real failures of 9/11, not the failure to prevent the specific plot, as these were broad, known risks and the actual causes leading up to them would be largely irrelevant for a lot of mitigating actions.

> "reasonably forseeable" is hard to define.

It certainly is. The concept of negligence was reinvented or reintroduced into the Common Law starting in 1932 with Donoghue v Stevenson and the lawyers have been wrestling with it ever since. I hated Torts as a student, it's a damn fiddly area of law. Give me Trusts any day of the week.

Still: if you want to study how intelligent people have mapped out the concept of "reasonably foreseeable", then lawyers -- particularly Scots lawyers who also look to Roman law -- are the people to talk to for inspiration.

> Taleb is a very intelligent man, but AFAICT he does often reinvent existing concepts with much cooler names. "Antifragile" sounds awesome. "Robust" sounds boring.

Antifragile != robust: "The antifragile is beyond the resilient or robust. The resilient resists shocks and stays the same; the antifragile gets better and better." (http://www.randomhouse.com/book/176227/antifragile-things-th...)

I don't think this is a meaningful addition. Resilient systems are not static, they evolve; that's how they become resilient in the first place.

"some economists saw the financial crisis"

At least one Head of Risk at a large bank not only saw the financial crisis coming but was sacked for trying to point this out:

http://www.telegraph.co.uk/finance/4582535/Senior-HBOS-execu...

Taleb's definition is subjective. From wikipedia:

1. The event is a surprise (to the observer). 2. The event has a major effect. 3. After the first recorded instance of the event, it is rationalized by hindsight, as if it could have been expected; that is, the relevant data were available but unaccounted for in risk mitigation programs. The same is true for the personal perception by individuals.

Thanksgiving is a black swan for the turkey but not for the butcher. Taleb is trying to solve the problem of how not to be a turkey, and prediction is insufficient for that problem, because errors of prediction will let the worst events through anyway. Rather one should evaluate how much worse things will be as an event grows. If problems grow faster than linearly, trouble will strike eventually.

I thought I replied to this. I guess I didn't.

Taleb's model seems to miss the distinction between two properties of predictions: discrimination and calibration.

Discrimination is the correctness of the forecast of a single event. Did X happen?

Calibration is the closeness of fit between the predictions made and the distribution of outcomes. Given predictions X1, X2 ... Xn, how closely do the probabilities fit outcomes Y1, Y2 ... Yn?

Even if your calibration is very good, there are always outliers which will upset your model. You didn't discriminate them.

> The black swan is something that is completely unforeseeable and for which there are no previous > partial or complete examples, either of the final outcome or the contributing causes.

Not according to Taleb, who should know as he developed and named the concept. According to him a black swan only has to be unpredictable to the observer.

But the point is that ignoring the possibility of unpredictable (to you) events is a fallacy. So a black swan event is any event an observer assumes is impossible, or so unlikely as to be ignorable, but then actually happens. Whether an event is a black swan, being an observational fallacy, is subjective.

So taking the 9/11 attacks, for some professionals in the intelligence agencies these were not black swan events because they knew they could happen, but for the population at large and to politicians in particular, they very much were black swan events.

> So a black swan event is any event an observer assumes is impossible, or so unlikely as to be ignorable, but then actually happens. Whether an event is a black swan, being an observational fallacy, is subjective.

I see.

I'm not really sure how that helps in any discussion of risk management at all, except to remind risk assessors that a) they should keep looking and b) they'll fail anyhow, just going on the numbers (the set of potential outcomes being unbounded, nobody will ever foresee all eventualities).

The point of drawing attention to black swans is to make it clear that you can't plan your way into a situation where you've avoided all risks. So, you're right. It doesn't help much past that.

This is obvious in many areas - nobody believes we can avoid all fires, for example (or we'd defund fire departments and cancel our fire insurance).

Yet in other areas there tends to be a view that we should be able to perfectly avoid risks.

This is dangerous, because a lot of black swan events that might be impossible to avoid, can still be substantially reduced in magnitude if you apply the other main forms of risk management to a greater degree to supplement avoidance.

Even when we don't mean to, it is easy to get thrown into this pattern of reacting to a specific risk that is in our face right now rather than to address broader methods of mitigating classes of risks.

Consider if prior to 9/11, past experience with dozens of hijackings over decades had been used to consider how to reduce the potential impact of a hijacking, and there'd been locks on the cockpit doors.

It seems far more feasible that effort into investigating risk mitigation by considering possible wide catching responses to known risks (past hijackings, mentally unstable passengers, illness) could have stopped 9/11 this way, than that more effort into risk avoidance by pouring money into intelligence agencies would have led to forever preventing someone from pulling off a 9/11 type plot, for example.

This is why this idea is important. But you're right: It is just a reminder of what ought to be obvious, that we can't avoid all risks because we can't possibly predict them all.

> nobody will ever foresee all eventualities

Not quite. It means that just because you can't imagine how your nuclear reactor might go into catastrophic failure, you still need to have a bullet proof, well rehearsed, properly funded response plan in place.

It means that just because you don't think there can be a simultaneous collapse of all your major banks, you still need to have robust capitalisation requirements, diversification of your financial institutions risk profiles, a plan on how to deal with the failure of institutions that are 'too big to fail', etc.

It means that heavily optimising for current stable conditions is a mistake, no matter how sure you are they will remain stable indefinitely.

But there's an infinite regress. How will your "bullet proof, well rehearsed, properly funded response plan" "go into catastrophic failure"?

What is the point where additional planning is counter-productive?

> The black swan is something that is completely unforeseeable and for which there are no previous partial or complete examples, either of the final outcome or the contributing causes.

As far as someone is concerned. You misunderstand what a Black Swan event is. A Black Swan is at the root an opacity problem, and not one of forecasting. 9/11 certainly wasn't a Black Swan event for the hijackers or the masterminds, but it was for everyone else.

In hindsight you can of course always identify indicators for the coming catastrophe. Even a storm needs some build up. The problem is we can not see the indicators in time.

Otherwise they would not be black swans, they would more be like the whale falling out of the sky in the Hitchhikers Guide To the Galaxy, that has formed spontaneously out of random particles.

The point of engineering approach to risk management the author makes while suggesting Black Swan phenomenon is something Taleb is actually suggesting by making systems anti-fragile in his new book.

Reply on news.ycombinator.com