4 ms·
My argument doesn't rely on dentists being experts on toothpaste. It relies on them knowing more than I do about toothpaste, on average. I can think of plausib
by philh 12y ago
My argument doesn't rely on dentists being experts on toothpaste. It relies on them knowing more than I do about toothpaste, on average.
I can think of plausible mechanisms that would discriminatingly push an unusually good toothpaste to the top of dentists' recommendations. Offhand, I can't think of any that would discriminatingly push an unusually bad toothpaste up there.
By "discriminatingly", I mean that e.g. advertising would work just as well for a good toothpaste as a bad one. But "the evidence suggests that this toothpaste is actually really good" is more likely to show up for a good toothpaste than a bad one. And I can't think of anything which is more likely to recommend a bad toothpaste than a good one.
So if I think there's a 90% chance that toothident is only recommended because of advertising, then I have no reason to think that toothident is particularly good or bad; but if there's a 10% chance that dentists do actually have a reason to recommend toothident, then I should use toothident, because I have no reason to favour anything else.
Certainly there are cases where this goes wrong, but I never claimed it was infallible.
- gress 12y agoIt sounds as though you are assuming that the signal that dentists receive from advertising will be equal for all toothpastes and so the signal from some evaluation of knowledge of toothpaste quality will still be enough to make the better product be the one they recommend.
- philh 12y agoI'm not assuming that. Advertising is noise, and knowledge is signal. I can't cancel out the noise, I can only observe noise+signal. But I expect the noise to be uncorrelated with the signal, which means that wherever noise+signal is highest is my best bet for where noise is highest. The more noise there is relative to signal, the less effective this method gets, but it's still better than any other strategy (in this toy model).
- gress 12y agoWhat makes you think the signal is stronger than the noise?
- philh 12y agoIt doesn't need to be. My argument never assumed that. I don't know where you get the impression that it did. Take nine 0s and a single 1, then generate ten random numbers, identically distributed however you choose, and add one of them to each of your starting numbers. Ask someone to pick out where the original 1 was. However you generated the random numbers, their best strategy is to choose the largest resulting number. Probably tapping out here.
- gress 12y agoAdvertising isn't random or independent of the signal - it's systematic bias.