812 karma · joined February 4, 2020
There's also Bit's About Money describing this effect in banks https://www.bitsaboutmoney.com/archive/seeing-like-a-bank/
As someone on an insulin pump they do. Iirc they have reps showing up at hacker conferences looking for red teams.
Definitely agree with your worries generally though.
Also, most of the worst parts of American development have happened since 1945.
When you say noticeable, do you mean in like, traffic statistics? Or in what the model knows that it clearly shouldn't if it wasn't trained in legally dubious ways?
Most disasters follow power laws and other fat tail which don't have the same effects in the tail as a Gaussian. If you shift 1/x^a by c, you "only" get a polynomial increase.
But also, if you shift the mean of a Gaussian, the increase isn't exponential, it's super exponential (e^(x^2) to be specific).
> Insurance is based on statistics. The math they use assumes stationary distributions. Insurance companies can't deal with shifting distributions well so they take the losses and then exit markets.
Sure they can, that's why they hire statisticians. They routinely deal with insurance of much rarer events where we have much worse models than climate change. They're just banned from charging the actual rates, because it's politically unacceptable.