> This is why VC and management is an art, not a science. :-)
Granted, but that said, Tetlock showed that in at least one very messy problem domain (international geopolitics), embarrassingly simplistic linear models crush expert judgement in terms of predictive power.
The problem is that the world is still sufficiently unpredictable that you can't use this insight to turn a buck, since events get "better" or "worse" with about the same frequency.
embarrassingly simplistic linear models crush expert judgement in terms of predictive power.
This is true in the vas majority of cases (andecdoctaly). But the inflection points are the subject of this essay. The analogue in internatinaly geopolitics, would be 9/11 and the ruthless pursuit of anti-civillian warfare. Which nobody predicted, and which <gamed> the theory of linearity[1] to maximize its damage.
[1] eg, the "expert" advice: 'go along with the hijackers', this has been empirically shown to be the safest/smartest course of action, etc
You'd find that book interesting. One of Tetlock's (many) findings is that almost everyone overpredicts outlier events. That is, we expect large deviations to occur far more often than they usually do. This is especially true of the "Hedgehogs", but "Foxes" do it also.
This is interesting, but their are two possible ways to read this: (1) Overdiagonoses; (2) Actual Bias. They are quite different in character.
An actual bias would be (for example) poor people taking unfair bets. We know they will dis-regard odds and buy lottery tickets that are negative NPV.
An overdiagnoses migh be something else altogether. If i know dis-proportionate rewards accrue to being "first" or early, I will overdiagnose. This is a variant of strategic precedence (to infer seniority, status, or special priveledge). This goes buy other monikers, like offensive "land grab" or the more defensive "lick the cookie" tactic.
In any event, the con is just a trick to manipulate someone who has linearly locked on a path (biased or not). However you feed them the bait, your goal is to get them to believe they do not need to revise their priors.
Yes - expertise is highly overrated too. Ask experts to give 90% confidence intervals on topics in their area of interest, and see how many they get wrong. This spans industries.
Granted, but that said, Tetlock showed that in at least one very messy problem domain (international geopolitics), embarrassingly simplistic linear models crush expert judgement in terms of predictive power.
The problem is that the world is still sufficiently unpredictable that you can't use this insight to turn a buck, since events get "better" or "worse" with about the same frequency.
I reviewed Tetlock's book here: http://chester.id.au/2012/07/29/review-expert-political-judg...
And I heartily recommend it to anyone.