Virtue ethics: “there are no stupid questions, only stupid people.”
Seriously though, I wish LW was less attached to single parameter models of human competence (IQ tests, g, and the like). This undercurrent of ordering people from best to worst based on single parameter models from the early 1900s psychometrics is pretty toxic. These models are severely silly of course, like any vast dimension reduction strategy has to be, but its the ordering that’s bad.
As far as I’m aware, there aren’t any parameters that correlate with success quite as well as g does, though, apart from maybe economic status, which itself correlates with g. I admit to being largely uninformed on this topic, so of course I could be wrong—indeed, I’d go so far as to call it likely. Could anyone expand on this point?
g ← [a high dimensional space representing your brain] → success in life.
You may well be right that it may be hard to design another single variable that is a child of [...] that correlates as well with success in life as g does. And if you are in the business of prediction, g is certainly a nice dimension reduction strategy. But why one number? Why not [some other integer] numbers? Aren’t you losing information? Success is super complicated. There are huge components of success not measured by tests that measure IQ (morale, ability to navigate social settings well, etc. etc. etc.)
Where it also gets iffy is where people forget that g is a parameter reduction strategy for [...], not a status marker that admits a total ordering. People talk about IQ a lot here, not sure if there is a more charitable reading of this than status talk.
Or where people start recommending policy based on a g/success correlation (recommending policy based on correlations is always a tricky business).
“g correlates with everything” is a special case of a more general thing people noticed where “everything correlates with everything.”
Interesting. This of course raises the question of whether it’s possible to write a hash function mapping various brain configurations to a relatively small space of hash keys. If so, then there really could be an integer parameter that correlates extremely well with everything you can do in life. Given the complexity of the brain, I somewhat doubt the feasibility of this, but it’s at least interesting to think about.
Virtue ethics: “there are no stupid questions, only stupid people.”
Seriously though, I wish LW was less attached to single parameter models of human competence (IQ tests, g, and the like). This undercurrent of ordering people from best to worst based on single parameter models from the early 1900s psychometrics is pretty toxic. These models are severely silly of course, like any vast dimension reduction strategy has to be, but its the ordering that’s bad.
As far as I’m aware, there aren’t any parameters that correlate with success quite as well as g does, though, apart from maybe economic status, which itself correlates with g. I admit to being largely uninformed on this topic, so of course I could be wrong—indeed, I’d go so far as to call it likely. Could anyone expand on this point?
Here is what the graph looks like:
g ← [a high dimensional space representing your brain] → success in life.
You may well be right that it may be hard to design another single variable that is a child of [...] that correlates as well with success in life as g does. And if you are in the business of prediction, g is certainly a nice dimension reduction strategy. But why one number? Why not [some other integer] numbers? Aren’t you losing information? Success is super complicated. There are huge components of success not measured by tests that measure IQ (morale, ability to navigate social settings well, etc. etc. etc.)
Where it also gets iffy is where people forget that g is a parameter reduction strategy for [...], not a status marker that admits a total ordering. People talk about IQ a lot here, not sure if there is a more charitable reading of this than status talk.
Or where people start recommending policy based on a g/success correlation (recommending policy based on correlations is always a tricky business).
“g correlates with everything” is a special case of a more general thing people noticed where “everything correlates with everything.”
Also, it’s a high-dimensional space with rich topological structure. The space of possible human brains is emphatically not a plain metric space.
Interesting. This of course raises the question of whether it’s possible to write a hash function mapping various brain configurations to a relatively small space of hash keys. If so, then there really could be an integer parameter that correlates extremely well with everything you can do in life. Given the complexity of the brain, I somewhat doubt the feasibility of this, but it’s at least interesting to think about.