4 ms·
> In 1960, the computer at... NORAD... warned, with 99.9-per-cent certainty, that the Soviets had just launched a full-scale missile attack against North Americ
by chriswarbo 12y ago
> In 1960, the computer at... NORAD... warned, with 99.9-per-cent certainty, that the Soviets had just launched a full-scale missile attack against North America... They later discovered that the Ballistic Missile Early Warning System at Thule Airbase, in Greenland, had interpreted the moon rising over Norway as a missile attack from Siberia.
This is why numbers shouldn't be reported with too many significant figures. I doubt the engineers thought the reports could be %99.9 certain, given that they're extrapolated from noisy sensors, networked across huge distances and that the opponent is adversarial (actively trying to avoid detection).
Percentages can cause a false sense of certainty too; as Leonard Mlodinow points out in The Drunkard's Walk about changing wine ratings from out-of-10 to percentages. We might be quite sure that an 8/10 compares favourably to a 7/10 but not to a 9/10; we can't say the same about an %82 with an %83 and an %81.
If you're forced to report percentages, or tenths of a percent, write the estimated error alongside. Raw numbers like "%99.9" have an implied "+/- %0.05" afterwards.
And that, ladies and gentlement, is why Die Hard 4.0 contains too many many significant figures.
- fela 12y agoIn which case can you define what an error interval on a probability estimate means? :) I think probabilities can either be well calibrated or not. (Which then of course begs the question of how you define being well calibrated for probabilities of events that happen very rarely). An error interval does not seem to make too much sense.
- DennisP 12y agoPut it that way and it sounds like 99.9999% is just an excessively precise version of 99%. Really it's the difference between .01 and .000001. It's just an expression of the order of magnitude: one in a hundred, a thousand, a million.
- aragot 12y ago...and we did hit that one chance in a million that it was wrong.
- jessaustin 12y agoWhat are the chances?!
- pdpi 12y ago“Million-to-one chances,” she said, “crop up nine times out of ten.” - Equal Rites
- vadman 12y agoThe actual points range when rating wines/spirits is only 20 or 30, depending on who does it: http://en.wikipedia.org/wiki/Spirits_ratings http://en.wikipedia.org/wiki/Spirits_ratings Does not make it any less ridiculous.
- Strilanc 12y agoThe flaw there wasn't the number of significant digits, it was not taking into account the probability of false positives and bugs.