![]()
Chapter 1
The AI Marriage
Whom should you marry?
This may be the most consequential decision of a personâs life. The billionaire investor Warren Buffett certainly thinks so. He calls whom you marry âthe most important decision that you make.â
And yet people have rarely turned to science for help with this all-important decision. Truth be told, science has had little help to offer.
Scholars of relationship science have been trying to find answers. But it has proven difficult and expensive to recruit large samples of couples. The studies in this field tended to rely on tiny samples, and different studies often showed conflicting results. In 2007, the distinguished scholar Harry Reis of the University of Rochester compared the field of relationship science to an adolescent: âsprawling, at times unruly, and perhaps more mysterious than we might wish.â
But a few years ago, a young, energetic, uber-curious, and brilliant Canadian scientist, Samantha Joel, aimed to change that. Joel, like so many in her field, was interested in what predicts successful relationships. But she had a noticeably different approach from others. Joel did not merely recruit a new, tiny sample of couples. Instead, she joined together data from other, already-existing studies. Joel reasoned that, if she could merge data from the existing small studies, she could have a large datasetâand have enough data to reliably find what predicts relationship success and what does not.
Joelâs plan worked. She recruited every professor she could find who had collected data on relationshipsâher team ended up including eighty-five other scientistsâand was able to build a dataset of 11,196 couples.*
The size of the dataset was impressive. So was the information contained in it.
For each couple, Joel and her team of researchers had measures of how happy each partner reported being in their relationship. And they had data on just about anything you could think to measure about the two people in that relationship.
The researchers had data on:
- demographics (e.g., age, education, income, and race)
- physical appearance (e.g., How attractive did other people rate each partner?)
- sexual tastes (e.g., How frequently did each partner want sex? How freaky did they want that sex to be?)
- interests and hobbies
- mental and physical health
- values (e.g., their views on politics, relationships, and child-rearing)
- and much, much more.
Further, Joel and her team didnât just have more data than others in the field. They had better statistical methods. Joel and some of the other researchers had mastered machine learning, a subset of artificial intelligence that allows contemporary scholars to detect subtle patterns in large mounds of data. One might call Joelâs project the AI Marriage, as it was among the first studies to utilize these advanced techniques to try to predict relationship happiness.
If you like guessing games, you can try to predict the results. What do you think are the biggest predictors of relationship success? Are common interests more important than common values? How important is sexual compatibility in the long term? Does coming from a similar background as a mate make you happier?
After building her team and collecting and analyzing the data, Joel was ready to present the resultsâresults of likely the most exciting project in the history of relationship science.
Joel scheduled a talk in October 2019 at the University of Waterloo in Canada with the straightforward title: âCan we help people pick better romantic partners?â
So, can Samantha Joelâteaming up with eighty-five of the worldâs most renowned scientists, combining data from forty-three studies, mining hundreds of variables collected from more than ten thousand couples, and utilizing state-of-the-art machine learning modelsâhelp people pick better romantic partners?
No.
The number oneâand most surprisingâlesson in the data, Samantha Joel told me in a Zoom interview, is âhow unpredictable relationships seem to be.â Joel and her coauthors found that the demographics, preferences, and values of two people had surprisingly little power in predicting whether those two people were happy in a romantic relationship.
And there you have it, folks. Artificial intelligence can now:
- defeat the worldâs most talented humans at chess and Go;
- reliably predict social unrest five days before it happens merely based on chatter on the internet; and
- inform people of an emerging health issue, such as Parkinsonâs disease, based on the odors they emit.
But ask AI to figure out whether a set of two human beings can build a happy life together. And it is just as clueless as the rest of us.
WELL . . . THAT SURE SEEMS LIKE A LETDOWNâAS WELL AS A truly horrific start to a chapter in my book with the bold thesis that data science can revolutionize how we make life decisions. Does data science really have nothing to offer us in picking a romantic partner, perhaps the most important decision that we will face in life?
Not quite. In truth, there are important lessons in Joel and her coauthorsâ machine learning project, even if computersâ ability to predict romantic success is worse than many of us might have guessed.
For one, while Joel and her team found that the power of all the variables that they had collected to predict a coupleâs happiness was surprisingly small, they did find a few variables in a mate that at least slightly increase the odds you will be happy with them. More important, the surprising difficulty in predicting romantic success has counterintuitive implications for how we should pick romantic partners.
Think about it. Many people certainly believe that many of the variables that Joel and her team studied are important in picking a romantic partner. They compete ferociously for partners with certain traits, assuming that these traits will make them happy. If, on average, as Joel and her coauthors found, many of the traits that are most competed for in the dating market do not correlate with romantic happiness, this suggests that many people are dating wrong.
This brings us to another age-old question that has also recently been attacked with revolutionary new data: how do people pick a romantic partner?
In the past few years, other teams of researchers have mined online dating sites, combing through large, new datasets on the traits and swipes of tens of thousands of single people to determine what predicts romantic desirability. The findings from the research on romantic desirability, unlike the research on romantic happiness, has been definitive. While data scientists have found that it is surprisingly difficult to detect the qualities in romantic partners that lead to happiness, data scientists have found it strikingly easy to detect the qualities that are catnip in the dating scene.
A recent study, in fact, found that not only is it possible to predict with great accuracy whether someone will swipe left or right on a particular person on an online dating site. It is even possible to predict, with remarkable accuracy, the time it will take for someone to swipe. (People tend to take longer to swipe for someone close to their threshold of dating acceptability.)
Another way to say all this: Good romantic partners are difficult to predict with data. Desired romantic partners are easy to predict with data. And that suggests that many of us are dating all wrong.*
What People Look for in a Partner
The major development in the search for romance in the early part of the twenty-first century has been the rise of online dating. In 1990, there were six predominant ways that people met their spouses. The most frequent way was through friends, followed by: as coworkers, in bars, through family, in school, as neighbors, and in church.
In 1994, kiss.com was founded as the first modern online dating site. One year later, Match.com was started. And, in 2000, I excitedly set up my profile on JDate, an online Jewish dating site, confident that I had discovered the cool new thing . . . only to quickly real...
