Sorry for what seems like nitpicking, but this kind of quiet background categorization is necessary to human cognition. I’m not trying to say “Don’t categorize” but rather, “Since you have no choice but to categorize, do it right.” You can just visualize someone saying, “Oh, I have no choice to assume that Anne’s a female” and then assuming that you, I don’t know, own 20 pairs of shoes, when this is not so much forbidden categorization as bad categorization—if you say “Anne is a member of the ‘likes spaceship Lego’ class”, that’s also categorization, but it’s more detailed categorization, and it screens off any default (stereotypical?) inferences one might make from the now superseded, higher-level ‘female human’ category. But I’m still licensed to assume you’ve got red blood, because that aspect of the ‘female human’ category hasn’t been overridden.
Exactly. One thing that I’ve found helps, is to remember to pick up and put down categories based on what particular decision I’m trying to make.
For example, let’s say I’m going to plan a group outing to see a cool sci-fi movie, and I need to decide whether to invite Anne along. (Let’s say I only have 8 tickets, and I want to maximize the chances that the other 7 tickets go to the seven of my friends who will most enjoy the movie, because I’m that kind of maximizer. To further constrain, let’s say the outing’s going to be a surprise, so I can’t just call up Anne and ask her; I have to go on facts I know about her.)
If I know that Anne is female, but don’t know anything about whether she likes spaceship Legos or not, then that’s actually relevant information, and indicates that she might need to go lower down on my list. (This isn’t a chauvanism thing; it’s just a bare fact that females in our culture tend to not like cool sci-fi movies as much as guys. If I don’t like that, I can do something about it, but the moment of deciding how to allocate movie tickets is not the optimal time to do something, given the kind of optimizer I am.)
Now, if I know that Anne likes spaceship Legos, but not that Anne is female, that indicates that they need to go higher on my list. “Liking spaceship Legos” and “liking cool sci-fi movies” tend to correlate pretty strongly.
Now, if I know that Anne likes spaceship Legos, AND I know that Anne is female, that actually places them higher on my list than merely knowing that they like spaceship Legos, even though knowing that Anne is female by itself would place them lower on my list than not knowing. Because my stereotype of “female AND likes spaceship Legos”, as a sub-class, happens to contain cached information about how the “likes spaceship Legos” and “likes sci-fi movies” data happen to clump together inside the “female” super-class.
One of the things that Bayesian analysis has been helping me with, is learning how to back-propagate new information about a particular sub-class into its containing super-class, and then how to forward-propagate the update to the super-class into its remaining sub-classes.
Exactly. One thing that I’ve found helps, is to remember to pick up and put down categories based on what particular decision I’m trying to make.
For example, let’s say I’m going to plan a group outing to see a cool sci-fi movie, and I need to decide whether to invite Anne along. (Let’s say I only have 8 tickets, and I want to maximize the chances that the other 7 tickets go to the seven of my friends who will most enjoy the movie, because I’m that kind of maximizer. To further constrain, let’s say the outing’s going to be a surprise, so I can’t just call up Anne and ask her; I have to go on facts I know about her.)
If I know that Anne is female, but don’t know anything about whether she likes spaceship Legos or not, then that’s actually relevant information, and indicates that she might need to go lower down on my list. (This isn’t a chauvanism thing; it’s just a bare fact that females in our culture tend to not like cool sci-fi movies as much as guys. If I don’t like that, I can do something about it, but the moment of deciding how to allocate movie tickets is not the optimal time to do something, given the kind of optimizer I am.)
Now, if I know that Anne likes spaceship Legos, but not that Anne is female, that indicates that they need to go higher on my list. “Liking spaceship Legos” and “liking cool sci-fi movies” tend to correlate pretty strongly.
Now, if I know that Anne likes spaceship Legos, AND I know that Anne is female, that actually places them higher on my list than merely knowing that they like spaceship Legos, even though knowing that Anne is female by itself would place them lower on my list than not knowing. Because my stereotype of “female AND likes spaceship Legos”, as a sub-class, happens to contain cached information about how the “likes spaceship Legos” and “likes sci-fi movies” data happen to clump together inside the “female” super-class.
One of the things that Bayesian analysis has been helping me with, is learning how to back-propagate new information about a particular sub-class into its containing super-class, and then how to forward-propagate the update to the super-class into its remaining sub-classes.