Scan barcode
A review by ferris_mx
Noise: A Flaw in Human Judgment by Cass R. Sunstein, Daniel Kahneman, Olivier Sibony
2.0
This book wasn't nearly as good as "Thinking Fast and Slow", and was really diluted for a mainstream audience. At some points I was frustrated and put off by the oversimplifications. Nevertheless, some of the ideas were still relevant for me. Let me boil it down to a TL;DR.
There is noise any time there is a human "judge".
Error can be decomposed into the following orthogonal dimensions: bias, level (some judges are more critical than others), pattern (even adjusted for level, judges are not uniform in what they rank highly), occasion (representation of judge inconsistency across time/conditions). We pay lots of attention to bias, but it is harder to pay attention to the other sources of error.
What can be done about it?
* "Noise Audit"
* Picking good experts
* Aggregating results (mean, prediction market, etc.)
* If your process has multiple experts, they must be independent.
* Avoid setting a preliminary decision based on partial information.
* Keep updating your personal prediction model.
* Try to decide components before coming up with an overall recommendation.
There is noise any time there is a human "judge".
Error can be decomposed into the following orthogonal dimensions: bias, level (some judges are more critical than others), pattern (even adjusted for level, judges are not uniform in what they rank highly), occasion (representation of judge inconsistency across time/conditions). We pay lots of attention to bias, but it is harder to pay attention to the other sources of error.
What can be done about it?
* "Noise Audit"
* Picking good experts
* Aggregating results (mean, prediction market, etc.)
* If your process has multiple experts, they must be independent.
* Avoid setting a preliminary decision based on partial information.
* Keep updating your personal prediction model.
* Try to decide components before coming up with an overall recommendation.