I listened to your Superforecasting discussion and I also read your review of Radical Uncertainty. I am equally baffled how can you not understand Mervyn's line of argumentation. However, it is not entirely inexplicable, given you are a big fan of Bayesianism and given your (mis)understanding of Popper and his view of knowledge which I figured out from a previous podcast.
Let me try to explain - please let me know what I am getting wrong:
Your confusion appears to stem from the fact that you do not seem to differentiate between Mediocristan (normal distri) and Extremistan (fat tailed distri). Bayes & the concept of expected value is very useful but only works in Mediocristan. In Extremistan, expected value is rather useless, downside protection is key. A lot of the finance types act as if they are in Mediocristan when they trade in financial markets. A combination of incentives, hubris, lack of intellectual rigour, power etc - a longer discussion.
The other confusion stems from your reliance of probabilities expressed in %s to define your confidence in propositions & hence grow your knowledge VS conjecture/ criticism (see Popper) and good explanations to grow your knowledge (including "I dont know" to reflect "I do not have a good explanation, need to think about it some more & if the event is consequential, need to buy insurance in the meantime").
Lets take the Osama example. The range of probabilities of 30-90% of Osama being in the house is at best misleading, most likely confused.
-Stating confidence levels on some small scale like 1-5 is useful.
-Using a 1-100 scale is splitting hairs - is there a real difference between 68 and 83 in ones mind? also my 68 might reflect a higher level of confidence than your 83.
-Importantly, using %s such as "63% likely" is confusing. And thats b/c we don't have any historical sample of events in which Osama was/wasn't in the house. "63% likely" gives the illusion we do as people intuitively understand such statements as parallel to coin toss or rolling dice.
-I would be much more interested into WHY different people had different views and only then have then put some non-probabilistic confidence level on their views.
If you are interested in understanding how other people may think differently than you & Bayes fans, I suggest you read a combination of The Chaos Kings (Scott patterson), Fluke (Brian Klaas) and definitely The beginning of Infinity (David Deutsch). If pressed for time, try The Precautionary Principle article by Nassim Taleb (assume you read Black Swan et all and are not convinced?). Brian Klaas also has a great article on his substack criticising Nate Silver's election forecasting.
Back in the 1970s when I did A level Physics , first question on the exam paper was a Fermi estimate. Something like ‘how many molecules of Caesars last breath are in every breath you take’.
Tom,
I listened to your Superforecasting discussion and I also read your review of Radical Uncertainty. I am equally baffled how can you not understand Mervyn's line of argumentation. However, it is not entirely inexplicable, given you are a big fan of Bayesianism and given your (mis)understanding of Popper and his view of knowledge which I figured out from a previous podcast.
Let me try to explain - please let me know what I am getting wrong:
Your confusion appears to stem from the fact that you do not seem to differentiate between Mediocristan (normal distri) and Extremistan (fat tailed distri). Bayes & the concept of expected value is very useful but only works in Mediocristan. In Extremistan, expected value is rather useless, downside protection is key. A lot of the finance types act as if they are in Mediocristan when they trade in financial markets. A combination of incentives, hubris, lack of intellectual rigour, power etc - a longer discussion.
The other confusion stems from your reliance of probabilities expressed in %s to define your confidence in propositions & hence grow your knowledge VS conjecture/ criticism (see Popper) and good explanations to grow your knowledge (including "I dont know" to reflect "I do not have a good explanation, need to think about it some more & if the event is consequential, need to buy insurance in the meantime").
Lets take the Osama example. The range of probabilities of 30-90% of Osama being in the house is at best misleading, most likely confused.
-Stating confidence levels on some small scale like 1-5 is useful.
-Using a 1-100 scale is splitting hairs - is there a real difference between 68 and 83 in ones mind? also my 68 might reflect a higher level of confidence than your 83.
-Importantly, using %s such as "63% likely" is confusing. And thats b/c we don't have any historical sample of events in which Osama was/wasn't in the house. "63% likely" gives the illusion we do as people intuitively understand such statements as parallel to coin toss or rolling dice.
-I would be much more interested into WHY different people had different views and only then have then put some non-probabilistic confidence level on their views.
If you are interested in understanding how other people may think differently than you & Bayes fans, I suggest you read a combination of The Chaos Kings (Scott patterson), Fluke (Brian Klaas) and definitely The beginning of Infinity (David Deutsch). If pressed for time, try The Precautionary Principle article by Nassim Taleb (assume you read Black Swan et all and are not convinced?). Brian Klaas also has a great article on his substack criticising Nate Silver's election forecasting.
Back in the 1970s when I did A level Physics , first question on the exam paper was a Fermi estimate. Something like ‘how many molecules of Caesars last breath are in every breath you take’.