OkCupid probably has a very good idea of how attractive a person is based on how many profile views and messages they get. I'd be very surprised if they aren't using that data to match roughly equally attractive people. Perhaps they are even taking personal preferences into account with probabilistic matrix factorization that is often used for recommendation systems. Basically if person X shows interest in A, and person Y shows interest in A and in B, then this makes it more likely that person X will like B as well. Whether A and B are movies, products on Amazon or people on OkCupid, PMF can exploit this kind of data to match people to products or to other people.
The biggest problem with algorithmic blind dating is probably not actual average quality of dates, but rather (perceived) risk of going on a bad date. People trust their friends a lot more than an algorithm.
>I'd be very surprised if they aren't using that data to match roughly equally attractive people.
I can confirm this is true. Last year, as an experiment, I made a dummy account with a picture of former NFL quarterback Kyle Boller but the same profile/personality as me. His matches are significantly more attractive than mine. He also received this email a few weeks after registration:
We just detected that you're now among the most attractive people on OkCupid.
We learned this from clicks to your profile and reactions to you in Quickmatch and Quiver. Did you get a new haircut or something?
Well, it's working!
To celebrate, we've adjusted your OkCupid experience:
You'll see more attractive people in your match results.
This won't affect your match percentages, which are still based purely on your answers and desired match's answers. But we'll recommend more attractive people to you. You'll also appear more often to other attractive people.
Sign in to see your newly-shuffled matches. Have fun, and don't let this go to your head
"You'll see more attractive people in your match results."
Reading this email actually disturbed me. A computer program is telling you that you're attractive, and that it's going to hustle "more of the attractive people" to you.
"the attractive people"
I'm only 20 years old and still have plenty to learn about ladies and relationships, but I know I can do better than have a server cluster tell me who is "in my league" and who are "the attractive people." Holy shit.
Yeah. Attractive people are fairly consistently approached or hit on or treated differently or what have you every day in real life. Remove the approach anxiety that the other 90% who DON'T approach these people but want to by creating an easy forum for approach, and you have an overwhelming amount of messaging. I don't blame girls for being quickly disillusioned. Even as a guy, I feel like the pickings are slim, and I'd much rather date a friend than meet someone through OkC.
I don't know what the typical HN user looks like, but anybody can work at being attractive regardless of their career or interests.
It's sort of a self-fulfilling role anyway. Staying away from an attractive woman makes you the loser. Having the mindset you just described does as well.
"self-fulfilling role" is another term for "blaming the victim." You know, that guy in a wheelchair could totally walk--he just doesn't want it enough.
Ok, well if you find your ability to appeal to women comparable to a paraplegic's ability to walk then I am truly sorry. Certain things like attractiveness do come with effort.
It is easy for the attractive to say that; like most traits, people don't like to believe dumb luck helped them. It makes a much better personal narrative if it was their own hard work. That doesn't mean it's backed up by any fact.
> OkCupid probably has a very good idea of how attractive a person
They do. They will tell you when they think you attractive. You get a message like "Congrats, you're in the top X% of attractiveness on OKC, we'll start showing you more attractive matches now." This is something only those attractive people know about.
The company I work for (a face recognition software company) has done some research on assessing attractiveness from photos of faces (as have some other research groups and no doubt other companies). Maybe using this to assess yourself would be useful. Or you could just put yourself on hotornot.
Women on OKCupid tend to rate men disproportionately low. There is an OKCupid blog post showing the distribution of the ratings they give men, and they are strongly skewed towards the lower end of the scale. This is one of the many reasons I don't use online dating sites.
Ok, that's true, and could potentially have the side effect that matches presented to women are more random (more noise in the data).
Even then, it seems like the effect would be kinda weak, no? And no individual guy loses out, except possibly to someone who was just about as attractive as him anyway (and he's equally as likely to gain).
Yep. Here it is. As a guy who's by most empirical measures not in the top 50% of attractiveness in the first place, this makes me depressed about human nature.
Male attractiveness is a lot more malleable than female attractiveness. Go to the gym, learn how to tell if clothes fit you and what colours suit you, always shower and shave, and get a decent haircut. You have now jumped 5-10% in attractiveness. Those are the biggest relatively easy gains. Posture isn't that hard to fix but it's hard to keep it fixed. After that there's the hard shit, becoming a more interesting, fun or rich person.
OKC pictures can probably be a tad misleading since they are often taken with poor quality camera phones and people will deliberately try to take photos at more flattering angles.
Also a lot of men will just spam these sites and send a message to every woman within X distance of them.
OKC actually did a blog post [1] about attractiveness by camera model, and (unsurprisingly?) photos taken by more expensive cameras were rated as more attractive than those taken by cheap camera phones.
Obviously there's dozens of confounding variables (if you have an expensive camera, you're more likely to know how to take a good photo), but still.
But there's several ways to dissect this data. For example, the number of unsolicited messages that a man gets and the number of followups (and perhaps, how fast those followup messages occur) is probably a good insight to how "attractive" he is.
Now if you think its harder to determine this for females because the amount of noise that they get...then the computer can factor in the number of "attractive" men (as judged above) who fawn over you, either by messaging or stalky profile views.
And of course, number of profile views (over time) and by repeat viewers (indicating the number of obsessed secret fans you have) is probably a good baseline to start from.
I did a similar experiment years back on hotornot years back. Took pictures of myself from various different angles and watched the scores after a few days.
A self shot face picture got me about a 5-6. A side profile head shot got me about a 4 and a shot taken by a photographer friend with a good camera and my shirt half undone got me a 9.5 or something.
Okay but obviously we know that's not a real picture, but instead a hyperbole of a photo. Still proves your point, I suppose. It's fun meeting people in real life then finding them on Facebook and seeing their complete lie of a profile picture that makes them look 3-4 points higher on a 10 point scale than they really are.
They definitely sort people into attractiveness tiers of some sort. You'll get a message mentioning that you're in the top half the first time their metrics indicate so, and it claims that this affects what members are presented/suggested to you. I'm not sure if this is based solely on the star ratings members can assign, or if they take other interactions into account.
I wouldn't be surprised if they do more subtle things with the same data as well.
The biggest problem with algorithmic blind dating is probably not actual average quality of dates, but rather (perceived) risk of going on a bad date. People trust their friends a lot more than an algorithm.