The interwoven topics of data mining, machine learning, artificial intelligence, and big data are common topics of discussion amidst some technical forums I frequent. For the handful of regular readers here, you know that most of those topics fall under the “#DoNotWant” category. It’s this sort of thing that causes things like the purchase of wild bird seed or the time of day you pay your bill to affect your credit rating. Data mining is how Target found out a teenage girl was pregnant before her parents. Self-driving cars are great until that realization hits that you’ll never own the car1 no matter how much you pay for it.
But despite my major aversion to all of these things, I still think there’s at least some good that’s possible. Watson is helping to diagnose strange diseases at rates and accuracies that rival the most skills doctors. A more crude form of AI has helped over 200,000 people successfully fight parking tickets. And I wonder if online dating could be one such area.
If you haven’t had the privilege of trying an online dating site, consider yourself fortunate. Nobody really likes it; it’s really the cesspool of humanity. Guys commonly open with crude remarks and make girls feel uncomfortable, girls commonly fill their picture slots with heavily-filtered duckface selfies at the expense of actually writing a profile, and the fact that ghosting is an actual thing done by both sides is almost as sad a reflection on our society as the fact that most of those conversations started with a “swipe right” in the first place. It’s really a pretty sad state of affairs when 59% of US adults consider this a “good way” to meet people to date.
So, in what sort of ways could AI help with this task? Let’s start by coming up with metadata with which to help influence matching. Are you a night person that can’t stand morning people who expect coherent thoughts (or worse – excitement) by 6AM? Are you a morning person with a group of friends who keep inviting you to things well past your bedtime? Logon times and frequency could help match this. What about profile length? Does a given user prefer lengthy profiles to succinct ones? Does the quantity of pictures of a potential match have an impact on their swipe direction? What about the use of a mobile app or website vs. a desktop, or the type of device? Does the user log in multiple times daily, or are they a once-a-month type? Is a given user more likely to send the first message? What about sending the last message (or failing to do so)? How long does an individual communicate through the service before providing e-mail or phone number? All of these are ways that the usage of a dating site itself could help narrow down the possibilities.
Then, there’s the more invasive sort of thing. Does a user prefer particular colors in photos? Are there common words in the profiles of their likes and dislikes? On the flipside, if a user is consistently declined, do they have anything in common with others who have been declined by the same set of users (especially ones who have been liked by that user) that can help optimize their profile? Could this information even help provide an “estimated interest level” and prioritize high-interest matches? Could users be penalized for ghosting or crude remarks, or conversely given preferred status for positive discussions? Let’s go for broke and incorporate an analysis of e-mail accounts and social media profiles, or even Amazon purchase histories, Netflix watching lists, and browser history to get the most complete possible profile of a person. Comparing that level of data between two people may make it entirely possible to get it right on the first swipe.
Yente is a well-known character in Fiddler on the Roof for being a matchmaker, and she was good at it in part because she knew everything about everyone, for good or for ill. The art of matchmaking requires more than simply a few photos and a laundry list of favorite foods and movies (or, heaven forbid, a collection of emojis) to be effective. I think that such a system could be effective, more so than the current situation as it presently stands. The question is whether letting a computer program mine vast amounts of data to become a 21st century Yente by determining the true nature of a person is the sort of tradeoff that makes the data collection worth it.
For some, it would be. And I submit that there is a fortune to be made as a result.
