Can AI predict if your relationship will last?

Dating is a gamble even under the best circumstances. Most of the relationships people enter into don’t last, and it’s impossible to know up front if a particular relationship is worth the investment of time and emotions it will require. At least, it’s impossible for humans to know, but a recent study is using artificial intelligence to improve our ability to predict the survival of a relationship over time.

Bruno Arpino from the Department of Statistics from the University of Florence, and colleagues, developed a statistical model to evaluate the longevity of a relationship between two individuals and predict whether or not it will last. Their findings were published in the journal Demography.

The team used a machine learning technique known as Random Survival Forests, which takes in information to make predictions about how long a set of variables will remain in a particular state. In this case, it was looking at the probability of two individuals remaining in a relationship, based on a number of factors related to both individuals.

“The Random Survival Forests approach is able to account for many potential predictors of union survival at the same time and to account also for potentially complex ways in which these features might combine in influencing union dissolution or survival,” Arpino told SYFY WIRE.

Random Survival Forests is pretty good at making predictions, but its predictive power becomes challenged in social settings which have complex variables, like the relationships of two people. This process proved to be more successful at accurately predicting the fate of a relationship than previous methods like regression models, but the success rate was still low. That’s probably due to the number of variables which factor into relationship survival which weren’t captured by the model.

More than 2,000 couples were observed over a 12-year period, on average, during which time roughly 45% split up. They found that the most important factors related to relationship survival were the overall life satisfaction of both partners and percentage of housework carried out by individuals who identify as women. Because of the way factors interact with one another over time, the accuracy of the model’s predictions was higher over the short term and decreased with time.

“In principle, the method can be applied to couples that just formed or even that will form in the future. In our analyses it was easier to predict union dissolution that happened after one or a few years after union formation than dissolutions that happened 10 or more years later,” Arpino said.

That might be because factors which aren’t important at the beginning of a relationship become important down the line. For example, disagreements about how to raise children can be overlooked when you’ve just started dating or even while children are young, but they might become deal breakers when it comes time to implement one strategy over another.

The study found that many different factors can interact with one another over time, resulting in outcomes which might not be predicted by other methods. Even this method was limited in its predictive power by the number of factors it was able to consider.

In theory, this sort of information could be used at the beginning of a relationship, or even before individuals embark on a relationship, to determine how compatible two or more people might be over the short or long term, but Arpino cautions against that.

“Our goal was purely scientific. I know that apps that use similar approaches exist. However, I would not recommend using these kinds of apps other than for pure curiosity,” Arpino said.

We’re not sure we would want to trust our individual relationship outcomes to a computer algorithm, even if it were incredibly accurate. The mystery is part of the fun, and maybe interpersonal relationships shouldn’t come down to cold mathematics.