One specific and relatively common version of this are people who believe that women have a lower standard deviation on measures of IQ than men. This belief is not incompatible with believing that any particular woman might be astonishingly intelligent, but these people all seem to have a great deal of trouble applying the latter to any particular woman.
I don’t think that this requires a utility-function-changing superbias. Alternatively: We think sloppily about groups, flattening fine distinctions into blanket generalizations. This bias takes the fact “women have a lower standard deviation on measures of IQ than men” as input and spits out the false fact “chicks can’t be as smart as guys”. If a person updates on this nonfact, and he tends to value less-intelligent individuals less and treat them differently, his valuation of all women will shift downward, fully in accordance with his existing utility function.
Placing “a high value on not discriminating against sentient beings on the basis of artifacts of the birth lottery” is not a common position. Most people discriminate freely on an individual basis. They also aren’t aware of cognitive biases or how to combat them. Perhaps it’s safer not to learn about between-group differences under those circumstances.
One argument you could give a Less Wrong audience is that the information about intelligence you could learn by learning someone’s gender is almost completely screened off by the information content gained by examining the person directly (e.g. through conversation, or through reading research papers).
That is exactly what should happen, but I suspect that in real life it doesn’t, largely because of anchoring and adjustment.
Suppose I know the average intelligence of a member of Group A is 115, and the average intelligence of a member of Group B is 85. After meeting and having a long, involved conversation with a specific member of either group, I should probably toss out my knowledge of the average intelligence of their group and evaluate them based on the (much more pertinent) information I have gained from the conversation. But if I behave like most people do, I won’t do that. Instead, I’ll adjust my estimate from the original estimate supplied by the group average. Thus, my estimate of the intelligence of a particular individual from Group A will still be very different than my estimate of the intelligence of a particular individual from Group B with the same actual intelligence even after I have had a conversation (or two, or three) with both of them. How many conversations does it take for my estimates to converge? Do my estimates ever converge?
After meeting and having a long, involved conversation with a specific member of either group, I should probably toss out my knowledge of the average intelligence of their group and evaluate them based on the (much more pertinent) information I have gained from the conversation.
If your goal is to accurately judge intelligence this may not be a good approach. Universities moved away from basing admissions decisions primarily on interviews and towards emphasizing test scores and grades because ‘long, involved conversation’ tends to result in more unconscious bias than simpler, more objective measures when it comes to judging intelligence (at least as it correlates with academic achievement).
Unless you have strong reason to believe that all the unconscious biases that come into play in face to face conversation are likely to be just about right to balance out any biases based on preconceptions of particular groups you are just replacing one source of bias (preconceived stereotypes based on group membership) with another (responses to biasing factors in face to face conversation such as physical attractiveness, accent, shared interests, body language, etc.)
I don’t think that this requires a utility-function-changing superbias. Alternatively: We think sloppily about groups, flattening fine distinctions into blanket generalizations. This bias takes the fact “women have a lower standard deviation on measures of IQ than men” as input and spits out the false fact “chicks can’t be as smart as guys”. If a person updates on this nonfact, and he tends to value less-intelligent individuals less and treat them differently, his valuation of all women will shift downward, fully in accordance with his existing utility function.
Placing “a high value on not discriminating against sentient beings on the basis of artifacts of the birth lottery” is not a common position. Most people discriminate freely on an individual basis. They also aren’t aware of cognitive biases or how to combat them. Perhaps it’s safer not to learn about between-group differences under those circumstances.
Strange advice for Less Wrong, though.
One argument you could give a Less Wrong audience is that the information about intelligence you could learn by learning someone’s gender is almost completely screened off by the information content gained by examining the person directly (e.g. through conversation, or through reading research papers).
That is exactly what should happen, but I suspect that in real life it doesn’t, largely because of anchoring and adjustment.
Suppose I know the average intelligence of a member of Group A is 115, and the average intelligence of a member of Group B is 85. After meeting and having a long, involved conversation with a specific member of either group, I should probably toss out my knowledge of the average intelligence of their group and evaluate them based on the (much more pertinent) information I have gained from the conversation. But if I behave like most people do, I won’t do that. Instead, I’ll adjust my estimate from the original estimate supplied by the group average. Thus, my estimate of the intelligence of a particular individual from Group A will still be very different than my estimate of the intelligence of a particular individual from Group B with the same actual intelligence even after I have had a conversation (or two, or three) with both of them. How many conversations does it take for my estimates to converge? Do my estimates ever converge?
If your goal is to accurately judge intelligence this may not be a good approach. Universities moved away from basing admissions decisions primarily on interviews and towards emphasizing test scores and grades because ‘long, involved conversation’ tends to result in more unconscious bias than simpler, more objective measures when it comes to judging intelligence (at least as it correlates with academic achievement).
Unless you have strong reason to believe that all the unconscious biases that come into play in face to face conversation are likely to be just about right to balance out any biases based on preconceptions of particular groups you are just replacing one source of bias (preconceived stereotypes based on group membership) with another (responses to biasing factors in face to face conversation such as physical attractiveness, accent, shared interests, body language, etc.)