My own experience, again, is that only about 1 in 5 medical students know what Bayes’ Theorem is.
Just like how 1 in 5 medical students knows what the Hindu-Arabic number system is. So what? The whole problem is that medical students excel at rote memorization at the expense of genuine, model-based understanding.
What matters is whether med students can perform the kind of statistical inference codified by Bayes’s Theorem—whatever they call it—not whether they can spit out the right password in response to the question, “What is Bayes’s Theorem?” (and then go on to unjustifiably tell their patients they have breast cancer).
Just like how 1 in 5 medical students knows what the Hindu-Arabic number system is.
Over here the Arabic and Roman numeral systems are compared and contrasted in primary school with a short word on both their development. This is done again with a bit more history in High School. I’m quite confident that if you said “Arabic numerals”, 90%+ of Medical students would know what you are talking about. And in context 80%+ would infer or remember what the Hindu-Arabic numeral system is.
But me pointing out that your own password guessing example is country dependent or perhaps even false is utterly irrelevant because it makes a vivid example.
Its highly probable that Henry Harpending has a good handle on the difference between password guessing and knowledge it is reasonable to assume he meant that less than 1 in 5. can perform the kind of statistical inference codified by Bayes’s theorem. Which I think you would agree with considering you, in my opinion accurately, stated that :
The whole problem is that medical students excel at rote memorization at the expense of genuine, model-based understanding.
Remember the sentence you are criticizing was a quote from a note he sent to the FDA not a quote from a rationalist discussion. Saying that:
that only about 1 in 5 medical students know what Bayes’ Theorem is.
Is much more expedient than explaining what one really means. Short notes are more likley to get read. It also has the advantage of sounding bad to someone who has no inkling of what Bayes’ Theorem or the knowledge that password is associated with is.
When one is sending a letter aimed at convincing someone of your point it is optimized only to convince.
So if I understand you correctly, HH’s phrasing of his complaint about an irrelevant deficiency (whether the students know “Bayes’s Theorem”) is understandable because there’s no similarly-expedient way to convey the same problem. I disagree. Watch:
“only about 1 in 5 medical students … … can perform Bayesian inference.” [1] … understand the impact of false positives in medical tests.” … reason consistently with Bayes’s Theorem.” … can correctly estimate false positive rates.”
I don’t think those rephrasings require any kind of brilliance, just a willingness to convey what you actually mean, and looking for a way to do so.
When one is sending a letter aimed at convincing someone of your point it is optimized only to convince.
I hope not, because that’s all the more reason for someone to ignore it! (If someone is willing to say anything to make a point, the listener should not expect a helpful relationship between the arguments and reality.) It should be optimized for persuasiveness under the constraint that one only say stuff that is actually relevant and true. No such constraint → no reason to care what they have to say.
If it seems like I’m belaboring a minor point, it’s because HH’s phrasing implies that the important thing is for students to know this or that password—the exact mentality we want to get rid of.
[1] Note: there is a huge difference between whether one can “perform” Bayesian inference—which happened long before Bayes walked the earth—and whether someone “knows what’s Bayes’s Theorem is”.
I hope not, because that’s all the more reason for someone to ignore it! (If someone is willing to say anything to make a point, the listener should not expect a helpful relationship between the arguments and reality.) It should be optimized for persuasiveness under the constraint that one only say stuff that is actually relevant and true. No such constraint → no reason to care what they have to say.
I think this goes without saying and all falls under optimizing to convince someone.
Its a queer scenario where you could get away with with not saying anything relevant and actually have this be optimal. But saying that the most persuasive argument possible is one where everything said is relevant and true is a bit hopelessly naive.
What really made your point is this:
So if I understand you correctly, HH’s phrasing of his complaint about an irrelevant deficiency (whether the students know “Bayes’s Theorem”) is understandable because there’s no similarly-expedient way to convey the same problem. I disagree.
And the examples you give.
In any case on second thought Harpending probably wasn’t optimizing only for persuasiveness (though one might argue he should). I think he most likley was pretty cost efficient with regards to the amount of thought and effort he put into the note and its expected effect. So while I now concede your point I don’t think this is particularly damning criticism. LW standards on this are, relatively speaking, pretty high.
Just like how 1 in 5 medical students knows what the Hindu-Arabic number system is. So what? The whole problem is that medical students excel at rote memorization at the expense of genuine, model-based understanding.
What matters is whether med students can perform the kind of statistical inference codified by Bayes’s Theorem—whatever they call it—not whether they can spit out the right password in response to the question, “What is Bayes’s Theorem?” (and then go on to unjustifiably tell their patients they have breast cancer).
Over here the Arabic and Roman numeral systems are compared and contrasted in primary school with a short word on both their development. This is done again with a bit more history in High School. I’m quite confident that if you said “Arabic numerals”, 90%+ of Medical students would know what you are talking about. And in context 80%+ would infer or remember what the Hindu-Arabic numeral system is.
But me pointing out that your own password guessing example is country dependent or perhaps even false is utterly irrelevant because it makes a vivid example.
Its highly probable that Henry Harpending has a good handle on the difference between password guessing and knowledge it is reasonable to assume he meant that less than 1 in 5. can perform the kind of statistical inference codified by Bayes’s theorem. Which I think you would agree with considering you, in my opinion accurately, stated that :
Remember the sentence you are criticizing was a quote from a note he sent to the FDA not a quote from a rationalist discussion. Saying that:
Is much more expedient than explaining what one really means. Short notes are more likley to get read. It also has the advantage of sounding bad to someone who has no inkling of what Bayes’ Theorem or the knowledge that password is associated with is.
When one is sending a letter aimed at convincing someone of your point it is optimized only to convince.
So if I understand you correctly, HH’s phrasing of his complaint about an irrelevant deficiency (whether the students know “Bayes’s Theorem”) is understandable because there’s no similarly-expedient way to convey the same problem. I disagree. Watch:
“only about 1 in 5 medical students …
… can perform Bayesian inference.” [1]
… understand the impact of false positives in medical tests.”
… reason consistently with Bayes’s Theorem.”
… can correctly estimate false positive rates.”
I don’t think those rephrasings require any kind of brilliance, just a willingness to convey what you actually mean, and looking for a way to do so.
I hope not, because that’s all the more reason for someone to ignore it! (If someone is willing to say anything to make a point, the listener should not expect a helpful relationship between the arguments and reality.) It should be optimized for persuasiveness under the constraint that one only say stuff that is actually relevant and true. No such constraint → no reason to care what they have to say.
If it seems like I’m belaboring a minor point, it’s because HH’s phrasing implies that the important thing is for students to know this or that password—the exact mentality we want to get rid of.
[1] Note: there is a huge difference between whether one can “perform” Bayesian inference—which happened long before Bayes walked the earth—and whether someone “knows what’s Bayes’s Theorem is”.
I think this goes without saying and all falls under optimizing to convince someone.
Its a queer scenario where you could get away with with not saying anything relevant and actually have this be optimal. But saying that the most persuasive argument possible is one where everything said is relevant and true is a bit hopelessly naive.
What really made your point is this:
And the examples you give.
In any case on second thought Harpending probably wasn’t optimizing only for persuasiveness (though one might argue he should). I think he most likley was pretty cost efficient with regards to the amount of thought and effort he put into the note and its expected effect. So while I now concede your point I don’t think this is particularly damning criticism. LW standards on this are, relatively speaking, pretty high.