Can Artificial Intelligence Predict the Right Partner for You?

Finding love is big business (a multi-billion–dollar industry), and artificial intelligence is transforming that business. Dating apps are increasingly outsourcing user matching to AI. Recent examples range from Overtone, an AI-powered matchmaking service started by the former chief executive officer of Hinge, to SciMatch, a dating app that aims to use AI to detect compatibility from a single selfie. But can AI actually help people find love?

The promise of AI is to go deeper and end the days of endless swiping and unreturned messages. Now people can talk to AI about their hopes, dreams, and ideal partner preferences and AI can scan the pool of dating app users to find candidates who fit those criteria exactly. AI can also track behavior in the app to learn what people really want (something behavioral scientists call revealed preferences).

As a relationship scientist, I see both peril and promise in this new generation of dating apps. Despite long-heralded promises by dating app companies that their apps are based on relationship science, the truth is that there is no strong foundation of science upon which to build these matching algorithms—relationship scientists do not know what makes two particular people compatible with each other. Not yet, anyway. Sure, we know some of the factors that generally predict liking—physical attractiveness, positive personality traits, similarity, familiarity, reciprocity—but we have yet to predict with the kind of certainty dating apps might like to claim whether any two people will hit it off before they even meet.

So if a robust science of compatibility doesn’t exist yet, what is the AI matching based on? One of the first AI dating apps, Fate, uses “similarity and reciprocity of personality,” according to an interview with the founder. That sounds reasonable, and research from Amanda Montoya and her colleagues looking across 313 different studies of similarity and attraction does suggest that similarity may predict attraction—before people meet. Once people experience even a brief interaction, similarity has much less, if anything, to do with how much two people like and are attracted to each other, as seen in an analysis of actual and perceived similarity of personality traits in potential partners, by Sarah Humberg and her colleagues. A systematic inquiry from my research lab, led by Annika From, also yielded little evidence that romantic couples are happier if partners are more similar to each other.

What about “reciprocity of personality”? If this phrase refers to the idea that two people may have complementary personalities that meld well together, this theory of compatibility, unfortunately, has even less empirical support.

Perhaps it is not too concerning that we lack a strong, agreed-upon scientific basis for predicting compatibility. After all, one of the great promises of AI is that it will learn as it goes, based not just on what a user says they want, but how they actually behave (who they swipe right on, whom they actually chat with after matching). Emerging evidence from different fields suggests that AI may be able to detect complex patterns that humans have been unable to discern. In the medical field, AI has been used to create more precise and accurate predictions that are aiding diagnosis and treatment. From weather prediction to fraud detection, the use of AI has enhanced forecasting. So maybe it can do the same for our love lives?

If the problem has simply been that we have not been able to identify the complex patterns that make up human compatibility, then AI may, in fact, be the answer. If, however, the problem is that humans themselves are too complex, then AI may not get us any closer to being able to forecast love.

What do I mean by this? Well, researchers have found that people will list their preferred partner traits before taking part in a speed dating event, only to have all those preferences go out the window when they actually interact with people. In my own work, I have discovered that people will dismiss another user’s dating app profile for reasons like “same name as my brother” and “bad haircut.” Yes, they will rule someone out as the possible love of their life because of something as irrelevant as their name (give them a nickname!) or hairstyle (arguably our most changeable physical characteristic).

Why do I bring up these examples? They raise the question of whether AI even has the right information to rely on when gathering data based on people’s stated preferences or swiping behavior. So what data might AI need? In my lab, we hope to find some answers to this question with a series of studies that aim to combine dating-app data with real-life interactions. This project will allow us to see what chemistry looks like online and offline, uncover who has online chemistry that translates into real-world sparks, and identify the factors that make chemistry more likely between two people (if there are any meaningful predictors).

The truth is, researchers at the forefront of this work do not always agree on whether forecasting love is even possible. While many relationship scientists have devoted their careers to testing the complex ways in which personalities and preferences interact to predict compatibility, they have arrived at different conclusions. Some researchers believe it will be possible to build an algorithm to predict compatibility; these researchers would likely see the promise in AI. On the other hand, there are those who have concluded that compatibility is a dynamic process that evolves as two people interact with each other and cannot be predicted ahead of time. For these researchers, finding love is less about the perfect match and more about being in the right place at the right time and then building love together through shared moments, from special nicknames and in-jokes to regular rituals just the two of you share.

With the research projects in my lab, we hope to test these theories by having people meet different numbers of times. We’ve seen from our early work that pairing two random people together does not typically produce high levels of chemistry. That instant attraction seems to be relatively rare; but what if they met repeatedly? Could chemistry develop? Does it need some sort of initial spark to set it off, and can we create that spark with the right equipment, such as a well-placed question or activity that brings two people together? In other words, is chemistry possible with anyone if the right conditions were met, or is there some, as yet undiscovered, set of requirements for two people to be able to fall in love?

I don’t know the answer to these questions yet, nor does the field of relationship science, although many have theories. I see AI as both an exciting tool that might help us answer this question as well as a technology that people looking for love should employ with caution. When used as a way to explore and test models of compatibility, AI may yield invaluable insights. But we must always remember that AI is trained with specific beliefs about what makes two people compatible. If researchers do not yet know what makes two people compatible, then letting AI do all the work to vet potential partners means you are only going to see the people that AI thinks are right for you, while likely remaining in the dark about how AI decided they’re a good match. This gap in knowledge may keep you from meeting someone unexpectedly compatible with you.

An alternative approach to dating is to think about it as a time for exploration. Instead of only connecting with people who have similar facial structures or reciprocal personalities, try dating people with all kinds of different experiences, personalities, beliefs, and desires. If you go this route, Paul Eastwick offers a valuable piece of advice in his recent book Bonded by Evolution: for those who have serious dealbreakers (e.g., you know you want children, have strong religious views, or do not eat meat), go ahead and surround yourself with people who share those views. This will help you avoid facing potentially difficult lifestyle tradeoffs in the future if you do find yourself falling in love. But don’t eliminate someone just because of a bad haircut or a shared name. Instead, be open to exploring and then see for yourself who you connect with and who you do not. You might end up being surprised.