I've recently started taking dating more seriously. This has forced me to be significantly more explicit and considered about what it is exactly that I look for in people. While initially motivated by romantic contexts, this clearly also both applies to and benefits from considering non romantic contexts. While I previously thought I had good models for what it is that I looked for in my personal relationships, dating online quickly made me realise my models of what made me compatible with people were pretty incomplete. This post introduces an entire category of errors in my models.
When I think about what I look for in people, a natural starting point is to study the people I associate with in my day to day life already. These people have been sampled from an extremely narrow distribution. This distribution is likely non representative in many ways. Some of these ways might be explicitly known to me and already informing my actions. For example I might know that I get along better with people who like cats than dogs, and so select more cat loving friends than dog loving friends.
Other ways might be implicit and unknown to me. Here is an incomplete list of ways such implicit properties might come about:
- Suppose you have some strong hidden preference $x$ that you lack introspective access to, but which divides the population in two. You might end up having a stronger draw to people who have $x$ or a push away from people who don't have $x$, but be unable to explain why in each individual case. This might lead you to your social circles over representing $x$, with you still blind to what $x$ actually is. By definition, $x$ is not represented in my models of what I look for in people. I have little reason to think $x$ might not be important. This one is least egregious, as this is most plainly an introspection skill issue.
- Suppose traits $x$ and $y$ are correlated in the population, and you filter for $x$ consciously (meaning you know what $x$ is). You have now implicitly also filtered on $y$, but need not know what $y$ is. $y$ might be important.
- Suppose you only in your day to day life encounter a subdistribution of the population and that subdistribution has trait $x$, you've now implicitly filtered for trait $x$ but again don't know what $x$ is. $x$ might be important.
- Suppose there is some trait that has low variance across the distribution of people you spend time with. In extreme cases such a trait might not even seem like a property that varies amongst people at all - you might not realise the trait is a trait. But even if you do track it, you might under-represent how important that trait is. If everyone around you has $x$, you might think $x$ is not that important - it may take seeing not $x$ to realise $x$ really is quite important.
Where I am going with this is there is a Chesterton's fence like situation occurring here -- there are properties of my social circle that are important for getting along well with me that I myself am either not aware of, or aware of but unaware of the importance of. If I stray too far from the distribution I am used to, these can break hard -- and worse than that, I might not even notice immediately when they do break.
Online dating can take you quite far out of distribution and cause this problem to manifest. The particular preferences I have in mind while writing this eventually graduated from implicit to explicit, but have left behind a stronger belief in this Chesterton's fence like preference problem in their wake.
I don't think I've seen this consideration on online dating discussed before (though also haven't looked hard). This is an error of a similar kind to the one Cate Hall describes here. I've made two general updates from this realization:
- I now hold much more general uncertainty over my models of what I look for in people. I think there likely exist a bunch more properties that I still remain blind to (the importance of). I think it's probably possible to discover more of these over time, but hard to fully enumerate the list.
- I am less excited about dating online and a bit more excited about meeting people in real life, where fewer of these unknown unknowns might bite. People I meet in real life are sampled from a distribution much closer to the narrow distribution of my friends, often because I meet people through my friends, and my friends' distributions are close to mine. This is only a modest update: there are other reasons to be excited about dating online; e.g. high information value.
Thanks to Elinor Oren for feedback on this post that massively improved its clarity.