Comment by photochemsyn
4 years ago
In the modern world, a good fraction of potential problems can be estimated using good computer models. Examples include building design parameters for withstanding earthquakes and hurricanes, river flow models for projecting potential flooding, aircraft design models that discover problems before planes start crashing, climate models that show the results of burning fossil fuels for decades, etc.
Yes, many people don't trust models, because they worry something might have been left out, but there's all kinds of ways of running real-world tests on models to get around those issues. For example, climate models accurately predicted the effects of the Pinatubo eruption in 1991, so there was no reason to distrust their predictions by the late 1990s.
The problem is that lawmakers are rarely statisticians and are sometimes even the people not trusting the models because they don't understand them. Someone needs to write up a bayesian form of government and then take over a country.
Or they are incentivized to not trust the models...
This is why command economies like China work today, when they did not in the past. With massive amounts of data and advances in theory, a centrally planned economy (and government) can be dramatically more efficient than a democracy.
The problem with command economies is that the incentives are broken, not that just that central planning is hard. The established elite uses political power to keep their position.
The USSR kept it running for half a century. Give them a while.
1 reply →
I want to be in charge of the priors.
On the other hand, models said that Japan could never experience an earthquake as bad as the 2011 earthquake right up until 2011 proved them wrong (https://physicstoday.scitation.org/doi/10.1063/PT.3.1361)
The other side of this is "whats the alternative?"
You can't prepare for an arbitrarily strong Earthquake, so the only lesson we take is that our models needed correcting and we'll work to that next time.
No one has been able to make a climate model that is able to predict present conditions from historical data.