Comment by LeonB | Hacker News Reader

Comment by LeonB | Hacker News Reader

In safety we try to prevent rare events (such as death) and these are rare enough that “past deaths” are not a useful way to predict the likelihood of a death.

A crucial metric that occurs more more often and does help predict deaths is “HPI” - high potential incident. A worker eight metres in the air dropped a spanner. It thudded on the ground, harmlessly. No one was hurt. No problem, right? That’s a HPI. A workplace that has a death will have had many HPIs. The dice-roll is not “it’s a new day, we roll the dice”. When a HPI occurs, that’s when we roll the dice.

Or in nuclear war estimates: a HPNWI occurs (high potential nuclear war incident). The Cuban missile crisis springs to mind as a particularly big example. But many many smaller HPNWI’s have occurred. Border skirmishes between nuclear powers. Sabre rattling by nuclear powers. Elect a president who uses threats and brinkmanship. Do this year in, year out. The dice is rolled not because time passes but because risky decisions are made and risky actions are taken, over and over and over and over and over and over, year in, year out.

https://en.m.wikipedia.org/wiki/List_of_nuclear_close_calls

Ah so that Lemmino video is just a rundown of this wikipedia page hah.

This runs counter to the black swan concept popularized by Taleb. In that they are by nature unpredictable, what one can predict is that there will be a next one.

I like this framework but I'd appreciate hearing more about it.

Ok I was waiting for someone to say this.

Taleb’s ideas on this would be very interesting. But is a nuclear war a black swan? Think about it. (Hint: have we ever observed a swan of this color before? Have we ever observed countries engaged in total war? Have we ever observed countries deploying nuclear weapons? Yes and yes)

Now a stronger form of what I think you’re saying is that, even if modelled all of the ways we think a nuclear war could break out — there are ways we would never think of. And the consequences of failing to account for those would be catastrophic, unlimited downside.

I agree with that whole heartedly. To that I would say — having nuclear weapons, allowing them to exist, exposes us to an unlimited downside. No matter how remote the chance of nuclear war seems to be, the only course we can take is complete nuclear disarmament. Global moratorium on nuclear weapons. End of story. We can’t foresee why/how a massive war would unfold. But we can say that it has unlimited downside. Immediate international cooperation on total banning of nuclear weapons is the only way forward.

Amazing way to think about HPIs, thanks for the introduction of this concept.

Thanks randomjoe2. I do want to be very clear that it’s not a concept I’ve created, and also I have probably mis-stated parts of it - it’s something that I’ve heard about many many times from working for years in the oil and gas industry.

I am not sure if I understand what the point is you’re trying to make

I took it to be: the original author's point was that nuclear attacks are so rare and our sample size so small that we can't use history to predict their occurrence in the future. This commenter is saying the typical way to handle situations like this is to look for more-common events that are proxies for the less-common ones in some predictable way.

To give a different example of my own: on an intersection-by-intersection basis, pedestrians dying from getting hit by a car is relatively rare, so if it's perceived that a particular intersection is unsafe and the design of the intersection is modified with a goal of reducing the chance of death, it can be difficult to know if that intervention worked. Supposing last year there was one death there and this year there weren't any; was that because of the design change, or just chance? But we do know that, in aggregate, some relatively constant percentage of collisions will result in deaths, and collisions overall are much more common than deadly ones, so even if the policy goal is to reduce deaths, measuring all collisions is a very useful, less-noisy proxy for determining if an intersection has been made less deadly.

So: original author is arguing that we can't meaningfully reason about these things because they're rare, and this commenter is saying "yes we can, if we find a more-common, less-noisy thing that's a good proxy for the rarer thing," and proposes that nuclear close calls are such a proxy.

Thank you, yes. As my buddy Jeeves would say “rem acu tetigisti”

Thank you, that was helpful

The author was saying we don’t have enough data to make a prediction. OC is saying we do have enough data to make reasonable assumptions if you look at high risk situations (which can reasonably be assumed to be a prerequisite to an actual disaster like death or nuclear war).

For example, take two people who have never been in a car accident. Driver A gets a speeding ticket a year and a DUI every 3 years. Driver B has never had a violation. It’s reasonable to assume that Driver A is at higher risk of a future accident. We don’t have to only look at the 0 accidents data point, we can look at factors that are typically common in accidents like speed and alcohol.

Do you think it’s possible to model a mad man’s mind with this?

Reply on news.ycombinator.com