It is a truth universally acknowledged that lockdown was a boom time for dating apps. And now that the world is finally opening up again, single people are stampeding towards them in even greater numbers – Hinge has reported a 63 per cent spike in downloads since 2019 and a tripling of revenue in 2020, while alone saw more than 6.5 million people downloading Tinder.
But while this level of interest might be new, actually being on a dating app seems, anecdotally, to be the same old story: a largely fruitless cycle of swiping, matching, initial interest and near-inevitable disappointment. Nobody who’s spent any amount of time on them would be surprised to hear that Tinder and Grindr rank in the top 10 of apps most likely to leave users feeling sad (meanwhile, not a single dating or hook-up app made a parallel list of the 15 apps most conducive to happiness).
The big dating apps have proprietary matching algorithms that they’re famously cagey about, but most rely on a combination of stated preferences – what you tell the app you want in a partner, either explicitly in your settings or implicitly through engagement – and something called collaborative filtering. This means an app looks for patterns in who users have said yes or no to, then tries to work out how they resemble other users to make predictions about who’s ultimately going to like who. It’s similar to how TikTok selects videos for you to view and Amazon nudges you towards purchases.
5 per cent for women and just 0.6 per cent for men. Why, when recommendation systems do such a good job of streamlining every other area of our lives, is it so hard to build a dating app that reliably gives people butterflies?
It’s a question that’s been in the back of my mind for the ten or so years I’ve been on and off them. In fact, I ended up writing a book about it – in my novel No Such Thing As Perfect, Laura, the unlucky-in-love protagonist, is persuaded to sign up for Cupid, a new service that claims to use records of everything people have done online to accurately profile them and find their ideal partner. (Spoiler: things don’t go to plan.)
But when it comes to dating, these algorithms are doing a terrible job
“Machines can only work with what you give them,” says Samantha Joel of Western University in Canada, whose research focuses on how people make decisions about their romantic relationships. “Some things are highly predictable, some things are not – and we just haven’t found the right input to predict attraction.” A large part of the reason for this is that the complex weighing up of preferences that happens while we’re working out if we fancy someone isn’t a transparent process: people might be embarrassed by something they’re drawn to, or even entirely unaware that the preference is there, even as it shapes their behaviour.
A 2016 study of match chatspin rates across all sexual orientations on Tinder run by Queen Mary University of London revealed jaw-droppingly low rates of reciprocal interest: the match rate was 10
Someone joining an app, Joel explains, might list qualities like height and education level as non-negotiables. “But when they meet someone in real life, those aren’t actually the criteria they use to assess if there’s a spark there”patibility and desirability can be poles apart.
The original digital matchmaking programme came out of Harvard in 1965; a maths student called Jeff Tarr hired an IBM processor that weighed the same as a small elephant to process the answers to a ‘dating quiz’ he’d distributed among his classmates. Nearly six decades on, things look vastly more sophisticated. After presenting you with hundreds of questions OkCupid gives you a compatibility percentage with your matches, while eHarmony’s psychometric testing claims to pair people on the basis of “32 dimensions of compatibility”, including extroversion, altruism and adaptability. Hinge, meanwhile, although it’s a simpler ‘swiping’ app, takes things a step further and asks you for post-date feedback that it aims to incorporate into your future matches. But for Joel, all of these jazzy features are mostly window dressing. “I don’t think there’s been an improvement at all,” she says.