Great article on how media responded to experts re covid. Prediction becomes weird in a context where you have partial control, because you simultaneously want to be wrong while having people take you seriously. slatestarcodex.com/2020/04/14/a-f…
11 tweets · April 2020 · 11 likes · 3 retweets · read on Twitter
Great article on how media responded to experts re covid. Prediction becomes weird in a context where you have partial control, because you simultaneously want to be wrong while having people take you seriously. slatestarcodex.com/2020/04/14/a-f…
One approach might be to specify conditionals: "conditional on [these decisions/behaviors by govt & citizens], I predict X hundred thousand deaths; conditional on [alternative], I predict X hundred deaths" +"former scenario looks more likely, but I obviously ADVISE the latter"
The point, as @slatestarcodex makes very clearly in that article, is that when there's huge uncertainty & meta-uncertainty... focusing on predicting what will happen is a diversion from the real question: ❓What Does It Make Sense To Do While We Have This Uncertainty?
❓What Does It Make Sense To Do While We Have This Uncertainty? And *sometimes* the answer is "what we were doing before the question was even asked". But it's very risky to assume so! And if saying "10% odds of pandemic are too low to take action", then what's NOT too low?
Also, the answer (including the threshold at which it makes sense to change behavior at all) is different depending on who "we" is!
US govt could've reasonably started developing drive-thru testing infrastructure on Dec 31st (even if estimating 1% chance of pandemic) on the basis that that would come in handy eventually even if it wasn't needed in 2020. Citizens didn't need to change behv then or even know.
Another example, if experts figured on a 15% chance of a particular volcano erupting this month: - people living nearby 😷🛫 - governments nearby 🤔🧐🤝🧑🚒 - people intending to vacation there 😬👀🛬 - entrepreneurs 💡🤔🚌 - and most people far away, probably nothing 🤷♀️
if it were 95%, and a huge volcano, all the people nearby would certainly gtfo if they at all could, and the government might pay to safeguard these citizens. That probably doesn't make sense at 15%, but *something* makes sense. And how do we discern what that something is?
One thing that seems to help stimulate this kind of thinking is Scenario Planning—an adaptation of some military methods. Thinking in advance what it might make sense to do under certain conditions can generate better actions to take in the present. en.wikipedia.org/wiki/Scenario_…
Better actions now could be: 🟢 convergently instrumental (in any scenario, having invested in X helps) 🟠 preparation (in scenario C, we're fucked *unless* we did Y 3wks earlier) 🔴 fat-tail disaster-avoidant (scenario D would cost >$10T, currently P=10%, $10B NOW → P=5%)
(This thread 👆 ended up being mostly about how to cope with the weirdness of predictions in these situations, rather than "wtf is weird here and why?", so I wrote a second thread in which I elaborate on the more fundamental question 👇)