Dating apps such as Hinge, Bumble, eHarmony, Match and OKCupid claim to use algorithms to match people based on compatibility. Are their methods based on scientific evidence and can dating apps even calculate how compatible two people can be?
A new dating app, Laguna, has just been launched, and is available on Apple Play. “Online dating has become a maze of endless swiping and superficial connections,” says Sonny, co-founder of Laguna with 16 years of machine learning experience. “Laguna cuts through the noise and helps people find genuine connections based on deep compatibility.” By answering just three confidential questions, users receive in-depth compatibility insights that delve into personality traits, values, and psychological wiring. It claims: “Our intelligent search doesn’t just match keywords; it understands concepts and traits, delivering results that truly align with your desires.”
I am very optimistic about the use of AI in dating (see my predictions here), and expect that as AI evolves it will be much better than the existing algorithms on dating apps such as Hinge, Bumble, Tinder and the like at finding us compatible people to date. However, I do have questions about the use of personality traits, values, and psychological wiring as the predictors of compatibility. Did this model use the latest research on compatibility?
What are the ingredients of good relationships?
One of the best papers in recent years comes from Dr Samantha Joel (2020) and 85 colleagues from around the world in the journal Proceedings of the National Academy of Sciences, which used machine learning to test a large number of individual and relationship level variables on relationship quality. 11,196 couples from 43 dyadic longitudinal datasets from 29 laboratories formed the data. The main findings were:
Four relationship-specific variables (people’s own judgments about the relationship) predicted 45% of their current satisfaction and 18% by the end of the study: (1) how satisfied and committed they perceived their partners to be; (2) how appreciative they felt toward their partners; (3) sexual satisfaction and (4) how they handled conflict. Individual variables and partner’s perceptions and personality traits did not improve the predictions. In other words, personality did not predict whether a couple would stay together. Values and “psychological wiring” were not any of the variables tested, presumably because they did not feature in many, if any, of the 29 datasets. (For anyone not familiar with the world of research in psychology, both the 45% and the 18% make important contributions to our understanding – there is always a lot of “noise” in the data from millions of variables that have a tiny effect that even machine learning is not yet able to identify or process, but that will be specific to each couple.)
This figure, reproduced with permission from the first author, shows the model used and the range of variables investigated:
Other research that I use in my coaching work to help couples comes from Drs John and Julie Gottman of the Gottman Institute. Their research is carried out with couples longitudinally, meaning over various time points to see how things develop and change as the relationship progresses, or whether there is any difference before and after their therapeutic interventions. The validity of their work with hundreds of thousands of couples is strengthened by the use of three types of data: various physiological measures, observations by researchers and the subjective perceptions and experiences of the couples. They are able to predict with high levels of accuracy (on average 91%) whether a couple will stay together (without intervention/support) after observing them in conflict, based on their conflict style.
Some couples fear that they are not compatible enough for their relationship to work. They may think that compatibility is about being similar or about often they argue. But the research shows that all couples argue and it’s about how you argue and whether you repair rifts after they’ve happened (as opposed to e.g., not talking to each other, bearing a grudge or pretending it never happened) that counts.
Four conflict styles predict relationship failure
The Gottmans call these the four horsemen: criticism, contempt, defensiveness and stonewalling. Contempt is the worst and predicts this with 94% accuracy after seven minutes of observation. The Gottman’s therapeutic interventions show that no couple can avoid some unsolvable conflicts. These often reflect differences in values (e.g., whether to send your children for religious instruction, or planning versus being spontaneous on holiday). What makes the difference in whether you can stay together or not is about how you deal with those unsolvable conflicts. This is not something that can be easily predicted by current algorithms on dating apps – although there certainly is potential for AI to measure and intervene on our conflict styles in real life! What this means for your dating choices, though, is that you do not need to avoid all differences in values.
Communicating about values and dreams is another ingredient of successful relationships
Other factors identified by the Gottmans include communicating about and validating each other’s values and developing shared dreams – this would mean that Laguna’s assumption that people need pre-existing shared values isn’t valid.
Emotional closeness and responsiveness are vital
In line with another evidence-based and highly successful approach, Dr Sue Johnson’s Emotion-Focused Therapy, there is also an emphasis in the Gottman therapeutic programme on emotional closeness and responsiveness, warmth and admiration.
Relationship skills are what count
So overall, research demonstrates that whether a relationship works or not is in fact based mostly on relationship skills rather than, e.g., similar or complementary traits. So predicting success based on these skills – rather than than on compatibility – may a better way of helping couples to make (and stick with) relationship choices. Comparing their constructs to those of the Laguna dating app, while values are part of the Gottman’s model, personality traits are not, and neither is their third dimension, “psychological wiring”.
What is “psychological wiring”? It is not, as far as I understand it, an operationalised variable that can be tested, nor do I know how it could be measured. It refers to the way our brains are shaped by our experiences – but we know that our wiring is plastic and constantly changing in response to new experiences. While trauma (and especially developmental and attachment trauma) may influence how we deal with e.g. conflict, challenges to our capacity, triggering words and actions, and our expectations about relationships, I don’t think AI is yet able to predict the unique interaction between two people and the millions of social and emotional experiences that have shaped their brains and bodies – to say nothing of how those experiences have interacted with each other.
Can an app predict how we will experience each other in real life? Polyvagal theory, pioneered by psychiatrist Prof Stephen Porges, provides a neurobiological foundation for our emotions, attachment, communication, and self-regulation, and emphasises the need for smiling and a gentle tone of voice in order to establish a safe connection, which are essential for healthy relationships, and especially for staying regulated during conflicts or challenging times. An AI-based app cannot yet predict whether two people are able to do either of these during a real-life social interaction.
Until then, I hope people dating will continue to use the resources of their own wisdom as well as their support groups and communities and qualified health professionals (and yes, AI therapists are useful) to make ongoing judgments about compatibility. Dating apps, when used in a healthy way, are a useful tool for meeting people, and it’s good to understand how they work so we can get the most out of them, but it’s not yet time to rely on them as matchmakers.
References
Gottman, J. M. (2008). Gottman Method Couple Therapy. Clinical handbook of couple therapy, 4(8), 138-164.
Joel, S. et al (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117 (32), 19061-19071
Johnson, S. (2011). Hold me tight: your guide to the most successful approach to building loving relationships. Hachette UK.
Porges, S. W. (2021). Polyvagal safety: Attachment, communication, self-regulation (IPNB). WW Norton & Company.
Related blog posts on algorithms used by dating apps:
You must be logged in to post a comment.