From: Rolf Nelson (rolf.h.d.nelson@gmail.com)
Date: Fri Jan 04 2008 - 17:54:46 MST
> Here's what I mean. Say the world had never known about the drug
> "alcohol". Then some people at a pharmaceutical company discovered it and
> showed that it was safe in controlled clinical trials, and could be used for
> some medical purposes (say as a low-grade anaesthetic). They also showed
> that it produced a state called "drunkenness", associated with mild euphoria
> and disinhibition, which let's for the sake of argument assume is a good
> thing.
>
>
> Now, if the discovery of alcohol happened in modern times, people would be
> very suspicious. They would worry that drinking a lot has bad side effects
> (they'd be right). They'd worry that it kills brain cells (check). They'd
> note that it is habit forming (yup). In the face of all these potential
> hazards, most people would avoid the drug, and so humanity would fail to
> discover that alcohol consumed moderately is basically a good thing. Of
> course, it might still be used to a limited degree for its original purpose.
> You could do a similar thought experiment for coffee.
>
Is your point just that some people fail to be rational Bayesians, and use a
"first, do no harm principal" rather than weigh expected costs versus
expected benefits? If so, I guess I agree, with the caveat that the "priors"
for any *random* drug are *hugely* in favor of it harming rather than
helping. The priors are so large that it would be wise to wait for a
controlled study to conclude that the drug helps rather than harms, and even
then maybe wait to see if the study is successfully replicated. (In other
words, the priors are so large that the probability that the first
experiment was botched or was a statistical fluke may be *greater* than the
probability that the drug is actually more helpful than it is harmful.)
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:01 MDT