The machine could figure out who the most desirable people in the bunch were based on certain characteristics like physical attractiveness, Joel says. But when it came to predicting which people would be a good fit for each other, the machine failed spectacularly.
"It predicted 0 percent [of the matches.] Some of the models we ran got a negative percentage, which means you're better off just guessing," Joel says. "I was really surprised. I thought we would be able to predict at least some portion of the variance — like extroverts or liberals like each other."
The result is a little unnerving to scientists, too.
"They're saying [real attraction] is something over and beyond what we know about what makes someone attractive," says Robin Edelstein, a psychologist at the University of Michigan who studies relationships and was not involved in the work. If the results suggest that attributes psychologists would think attract certain people are effectively useless when it comes to making matches, then what is actually going on when two people are drawn to one another?
That question has left Joel and other psychologists scratching their heads. "It's a very elusive, mysterious thing. I don't think people even know themselves what it is about a specific person," Edelstein says. "I don't know if it's about specific questions or specific traits."
There are a few flaws in the study, though. "One concern is that they're testing in a relatively small undergraduate sample," Edelstein says. College students plucked from the same campus are probably more similar to one another than those out in the wider dating world, and there isn't much scientific evidence that similar people are more attracted to one another, Edelstein says. Without a bigger range of personalities, Joel's algorithm might not have come across that magic combination of traits and preferences that makes that special someone stand out to another person.
And 350 participants isn't a great study size, either, though that doesn't worry Chris Danforth, a computational social scientist at the University of Vermont who did not work on the study. If something isn't showing up in a small study population but did in a huge data set, it just might not be very important, he says. "Would there be predictive utility with a larger data set? I'm guessing yes, but only in the constrained sense the result might not be relevant," he says.
It's also possible that the researchers just didn't look at the right thing.
It's hard to say what, though. After including over a hundred traits guided by scientific literature in the study, Joel is left with only wild guesses. "Maybe there's something very idiosyncratic about the interaction that's more than the sum of its parts. Maybe it's based on things like how tired were you that day? Did they like the shirt you are wearing?"
She adds, "Maybe we could predict attraction if we really had all the variables and situation-specific variables."
When researchers begin using their imaginations, they rattle off an inexhaustible number of potential variables that might affect attraction. That would make predicting attraction much like predicting the weather; romance could be chaos. If that is true, it'll be a long time before algorithms can make accurate predictions, if they ever are up to the task, Danforth says. "This feels like the absolute edge in terms of difficulty."
That doesn't inspire much faith in the algorithms at dating website like eHarmony or OKCupid. "It's disappointing. There isn't that shortcut we want there to be," Joel says.
On the other hand, she says the study only looked at whether their participants had an initial attraction that would start a relationship, not long-term compatibility. Limiting the pool to people with similar views might help with that, like the way eHarmony does, even if it does nothing for attraction. Neither eHarmony nor OKCupid provided a comment for this story.
But in Western culture, at least, you still need someone you're initially attracted to in order to get to the long-term relationship, Joel says. After this study, she doesn't think using mathematics is the way to figure that out – at least not today. "I no longer have faith in matching algorithms," she says. To know if sparks are going to fly, Joel says, nothing is more telling than an old-fashioned face-to-face.
Angus Chen is a journalist based in New York City. He is on Twitter @angRChen.
Copyright 2017 NPR. To see more, visit http://www.npr.org/.