Whitney Wolfe Herd has a vision for modern romance. More than a decade after founding Bumble, in 2014, she’s back at the dating-app company—and this time, she wants to get things right. For too long, she argues, people have been swiping in the dark: evaluating other multifaceted beings on the basis of a few pictures and superficial bits of description, being evaluated in turn, feeling judged and empty. Now, she says, she’s seeking a new way to inject some warmth and humanity into the process—using, as she recently told The Wall Street Journal, “the world’s smartest and most emotionally intelligent matchmaker.” She’s talking about AI.
The titans of online dating have heard the message loud and clear: Their customers are burnt out and dissatisfied, like department-store patrons who’ve been on their feet all day with nothing to show for it. So a growing number of apps are aiming to offer something akin to a personal shopper: They’re incorporating AI not only as a tool for choosing photos and writing bios or messages, but as a Machine-Learning Cupid. Wolfe Herd’s new app, she says, will ask people about themselves and then use a large language model to present them with matches—based not on quippy one-liners or height preferences, she told the Boston radio station WBUR, but on “the things that matter most: shared values, shared goals, shared life beliefs.” (According to the Journal, she’s working with psychologists and relationship counselors to train her matching system accordingly.) A new app called Sitch, meanwhile, asks users questions and then gets AI to serve them bespoke suitor options. Another, Amata, has people chat with a bot that then describes them briefly to other singles, essentially taking them out to market. On Monday, Meta announced that Facebook Dating is launching an “AI assistant” that can help singles find people who match their criteria—and a feature called “Meet Cute” that presents people with a weekly “surprise match” to help them “avoid swipe fatigue.” At The Atlantic Festival last week, Spencer Rascoff—the CEO of Match Group, which owns major dating apps including Hinge and Tinder—told my colleague Annie Lowrey that Tinder is experimenting with surveying users and, based on their responses, presenting one custom prospect at a time. “Like a traditional matchmaker,” he said, this method is “more thoughtful.”
That certainly sounds nice. But is the idea truly groundbreaking? Maybe not. Several of the oldest online-dating sites have long asked patrons to fill out questionnaires, which Wolfe Herd herself told WBUR can be as laborious a process as “filling out doctor-office reports.” And more information hasn’t always meant deeper or more successful matchmaking. In 2013, OkCupid—which still has users answer questions and gives prospects a compatibility score—ran a series of experiments, and found that it mattered less whether the site deemed a duo compatible and more whether it told them they were compatible; when OkCupid informed pairs with a low “compatibility score” that they had a high one, they were more likely to keep chatting than couples who’d had a high score and were told they had a low score. And the writing on profiles seemed to matter little: When people rated profiles that didn’t show any text, the evaluations were roughly the same as when the text was there. When the company took pictures off, site activity tanked. “OkCupid doesn’t really know what it’s doing,” Christian Rudder, one of the site’s co-founders, concluded in a blog post about the findings. “Neither does any other website.”
Of course, the dating-app questionnaires of today aren’t the same ones people were completing in 2013. And although major apps already use machine learning to note users’ preferences and to suggest prospects, it’s possible that as AI improves and as dating sites collect more personal information from users, the result could eventually be more fine-tuned matches. But exactly how these algorithms are meant to anticipate human chemistry remains unclear. Unless dating companies have access to some new and groundbreaking information, one big problem remains: Romantic compatibility is largely still a mystery. People tend to couple with those who are demographically similar to them, yet when it comes to people’s personalities, tendencies, and “values”—that vague but relentlessly used term—decades of research have revealed no simple rule for what makes people click. As Eli Finkel, a Northwestern University psychology professor, once told me, a real-life spark is unpredictable partly because it depends somewhat on chance: What one person just happens to say might resonate with the other one, or lead to a topic that proves conversationally fruitful—or not. At the moment, only one true test of chemistry exists: Two brave souls have to meet and see what happens.
Psychologists will continue learning about human thought and behavior. But their findings don’t always translate to clear matchmaking takeaways. Take attachment theory, which Bumble’s new AI will supposedly incorporate. Research does back up the idea that people vary in their tendencies toward “secure attachment” (an ability to trust in other people’s love and goodwill) and insecure attachment, whether of the “anxious” variety (clingy, reassurance-seeking) or the “avoidant” one (remote, self-protective). Amir Levine, a Columbia University psychiatry professor and a co-author of Attached: The New Science of Adult Attachment and How It Can Help You Find—And Keep—Love, told me that the broad strokes way this might apply to pairing people up: Secure attachment is like type O blood; it works well for everyone. (Must be nice.) But not enough securely attached people exist to go around—especially, he said, because they often get “snatched up” early. So what about everyone else? Anxious and avoidant types can set each other off; anxious-anxious pairs can get “dysregulated,” as Levine put it, “like two cats in a tree—and they’re both hissing at each other, and there’s no one to help them come down.” Avoidant-avoidant duos, with all their sturdy walls up, might never form much of a bond at all.
The point isn’t that single people should flee from any whiff of insecure attachment. It’s that romance doesn’t really work this way: We don’t all exist in perfect attachment buckets, or in any kind of buckets at all. And even if we did, they wouldn’t reduce love to a calculable equation. When Levine co-wrote Attached, he wasn’t presenting a basis for choosing partners. He was arguing that we should be aware of our tendencies, and of the fact that not everyone moves through the world in the same way—and that understanding other people’s particular needs could make it easier to meet those needs and to express your own.
All of that takes work—the kind of work that AI dating promises, implicitly or explicitly, to render unnecessary. Sometimes those promises seem plainly dystopian. Wolfe Herd, in a Bloomberg Live interview last year, predicted that someday soon people would rely on their AI “dating concierge” to do courtship for them—that it would not only identify people to meet but would take it from there, replacing all the embarrassment and exhilaration of human flirtation with the come-ons of a machine that feels and risks nothing. Yet even for people who wouldn’t want tech companies reaching tendrils so far into their intimate life, matchmaking-AI ventures might dangle a subtly alluring idea: that a more perfect algorithm would lead to a more perfect partner, a more perfect union; that it can release you, like a trap door, from romantic fatigue.
But the success of a relationship doesn’t only hinge on whom you find; it depends also on you. You are the one who can use principles such as attachment theory—for self-reflection. You have far more control over your own behavior, after all, than you ever will over a romantic prospect’s. And besides: Would you really want human connection to be so straightforward that a machine could crack it, just like that? For now, love evades understanding—which means that finding someone will remain, much of the time, a pain in the ass. It also means that when a connection is made, it will be so distinctive that no one ever could have predicted it.
When you buy a book using a link on this page, we receive a commission. Thank you for supporting The Atlantic.
