SALT LAKE CITY — Despite the claims of some dating websites, researchers find tests of algorithms conceived by cutting-edge computers still fail to predict attraction.
While such algorithms have been able to accurately predict traffic and the weather, University of Utah researchers found poor results when it came to predicting desire and attraction.
“We found we cannot anticipate how much individuals will uniquely desire each other in a speed-dating context with any meaningful level of accuracy,” says study lead author Samantha Joel in a press release. “I thought that out of more than 100 predictors, we would be able to predict at least some portion of the variance. I didn’t expect we would find zero.”
A University of Utah psychology professor, Joel says they had some limited success with predicting someone’s overall desirability. But when it came to individual interactions, their ability to predict desire was less than one percent accurate.
“We tried to do it and we couldn’t do it,” Joel says. “Dating can be hard and anxiety provoking and there’s a market there for a short cut. What if you didn’t have to kiss all the frogs? What if you could skip to the part where you click with someone? But our data suggests that, at least with the tools we currently have available, there isn’t an easy fix for finding love.”
Despite the research team’s findings, large matchmaking companies like eHarmony have and continue to use machine learning as part of their matchmaking process. With larger data sets available to them, it’s conceivable that they could have greater success than the researchers had with their limited speed-dating study.
According to a Fortune magazine article on eHarmony’s use of machine learning, the site collects demographic data, psychographic data, and behavioral data. This information, combined with data from research on couples who met through the site, is then fed into machine learning algorithms by in-house data scientists and psychologists.
It’s hard to imagine there isn’t some difference in the effectiveness of the algorithms on these larger scales and the demonstrated ineffectiveness in the University of Utah study. Exactly how big the gap may be is hard to say.
While suggesting it may be possible to make headway in the future, Joel’s fellow researchers expressed skepticism about the current ability of any algorithms to help with matchmaking.
“It may be that we never figure it out, that it is a property we can never get at because it is simply not predictable,” says co-author Paul Eastwick of the University of California. “Romantic desire may well be more like an earthquake, involving a dynamic and chaos-like process, than a chemical reaction involving the right combination of traits and preferences.”
The study was published this week in the journal Psychological Science.