Comment by asr | Hacker News Reader

Comment by asr | Hacker News Reader

Parent Full thread

I agree with your point, but strongly disagree with your example. You think DC should be running the risk of accidents which kill riders because of trains that lurch into stations? Loss of life is a massive cost, even if only economic factors are accounted for. Please show me a cost-benefit analysis where that is outweighed by lurching trains (even many millions of lurching trains).

Also, it's incorrect to say that we could have used decades of problem-free operation to extrapolate and predict a low risk of similar accidents in the coming decades. The 2009 crash was caused by an aging system, and unless your solution involved replacing all the switches/sensors, we would have seen new failures at a more frequent rate.

Finally, the 2009 accident highlighted an inherent problem with the old system: putting a human operator in a train who is only responsible for acting in an emergency will not work. Just like the recent self-driving car fatality. It wasn't a case where we knew and should have accepted the risk, it was a case where the crash caused us to understand that the prior risk assessment was wrong (because it assumed that a human operator could prevent these kinds of accidents, when that's just not going to happen). So it was appropriate to change operations.

> You think DC should be running the risk of accidents which kill riders because of trains that lurch into stations?

DOT estimates the cost of a human life at about $6 million. Times 9 is $54 million. Let's say such an event happens once every 10 years, so $5.4 million per year. The D.C. metro has 180 million rides per year. That works out to 3 cents per ride. Do you think people would pay 3 cents for a smoother, ATO-controlled Metro ride? I bet they would. (That's not even counting all the delays caused be forgoing automated control.)

As to the switches and sensors--Metro already replaced the track circuits (per the NTSB's recommendation): https://wamu.org/story/17/04/06/metro-shelves-plans-bring-ba....

It's worth elaborating on a central point; the idea that 'preventable deaths are unacceptable' only makes sense at ordinary human scales. In someone's day-to-day life, they aren't reasonably going to be in a situation where letting someone die will be the best alternative.

The situation changes completely at the level of the decision makers of countries and large cities work at. /Everything/ is a variant of the trolley problem and all resources could in theory be used to save a life. Not accepting a risk of death can do more harm than good and realistically infrastructure; it just obscures the cause and effect.

Eg, we'll never know if someone died because their doctor was late because a train operator made a mistake operating a train. The cause-effect chain is too convoluted. But statistically speaking, something like that will happen every so often. Deaths from inefficiency are just as bad as deaths from direct accident; and if the scale is large enough that they can reasonably be detected then they need to be considered too.

Obviously most of the time a high standard of safety is going to be best, but trains are a high-risk environment moving 100s people at speeds they were never meant to move. True perfection over 30 years is probably not a required standard. Great goal though.

This has been interesting to chew on. Let me say that if there are actually delays, obviously those impose other costs (as a sister post explains here) and so the cost-benefit calculation may come out differently. But your original post was only about whether it's worth risking lives to avoid lurching trains, which is why I thought it was interesting.

I think the problem with your cost-benefit analysis is that you are using different methodologies to value the cost and the benefit. If you value the benefit of non-lurching trains by asking what people would pay, then you need to use a value for human life that asks what people would agree to receive to be killed--which I believe is much more than $6 million.

Conversely, if you are going to value a human life based on lost earnings/support to family/etc., or some other figure which might produce results in the ballpark of $6 million, then on the other side you need to consider only the lost earnings/hard economic costs of an annoying train ride, which are $0 -- there is no economic loss caused by lurching into a station.

I think the devil's advocate argument here is: would deciding not the run trains on auto cause an increase of car riding which would lead to greater than the number of deaths generated by the aging automatic driving system?

Reply on news.ycombinator.com