Back to How Love Really Works.
Why Can't Machine Learning Predict Romantic Compatibility?
Paul Eastwick says studies with psychologist Samantha Joel tested whether compatibility at a first meeting can be predicted from what two people report about themselves. Matching on deal-breakers does not work well, and similarity is not a strong predictor of who will like whom.
Speed daters had reported their own traits, preferences, goals, and desires. Machine learning used one person's self-report and the other person's self-report to guess which pairs would be especially compatible when they met. For compatibility, the prediction was exactly 0 percent. Some people really liked each other and some really did not. Self-reported information still could not explain ahead of time why two people ended up compatible.