DRMacIver's Notebook

Homophily and the tyranny of false positives

Homophily and the tyranny of false positives

A while ago a friend of mine told me about what he called his Terribly Goth Theory of Friendship: You can't trust people who don't have problems.

The reason is of course that we have problems, and that it's hard to trust people without them to know how to handle that well.

I think it goes deeper than that: Actually many people with fewer or different problems than ours are in fact perfectly lovely human beings who we would get on fine with, but we can't tell that. So we select for people who are like us because we can predict how they will react. This is homophily, the tendency to hang out with people like us.

In general, the question of who to be friends with is a classic example of the sort of thing I talked about in there's no single error rate, where the only error rate that is visible to you is the false positive rate. As I argued in you should try bad things this will tend to cause you to narrow your selection over time: If you only ever update to remove false positives, you will tend to increase the degree of homophily you exhibit.

This isn't necessarily a bad thing at the individual level, but bear in mind the way that bad rejections (false negatives) can contain systemic biases. My suspicion is that a lot of communities will tend to become more homogenous over time purely as a result of this effect: People who don't match the community expectations will tend to get filtered out, not necessarily because they're a bad fit, but because they don't act like the kind of people the community understands how to tell if they're a fit or not.