The Love Algorithm. Practical and ethical questions on machine learning applied to online dating

Online dating used to be a place for those looking for transgression or for individuals whose quest for love wasn’t — and in some cases still isn’t — approved by society.

Things have radically changed, and looking at data from 2010, more heterosexual couples met online than in bars, church, and college. Only those that met through friends were more numerous, but it is likely that in 2018 online-made couples will surpass the “traditional” ones. It is a safe bet to predict that the gap is destined to widen (for same-sex couples, over 70% of them met online in 2010).

It is not difficult to see the reasons for this growth. Online dating allows every user to access an almost infinite pool of potential partners, far wider than any user’s social circles (family, friends, co-workers and other associations). It is also faster, cheaper and doesn’t come with the overwhelming feeling of rejection that the same amount of real-life interactions would entail. It also offers the theoretical possibility of a better quality match, in light of the sheer number of options and the ability of quickly filtering down to the most relevant ones.

Machine learning will be the determining factor for the speed and efficacy of this filtering process. It is not at all difficult to envisage a system that “learns” from the user’s previous decisions and then extends these decisions across the pool of potential partners, essentially re-sorting the position of each potential partner in the queue based on the user’s real preferences, placing the most relevant ones on top.

Similar analytical tools have already been successfully implemented in a variety of industries, e.g., retail (Amazon) and entertainment (Netflix), in addition to dedicated niche fields that deal primarily with data collection, indexing and retrieval solutions, like eDiscovery. However, what needs to be assessed is how this technology should be applied when it comes to online dating.

How markets shape algorithms

To start, we need to acknowledge the fundamental differences that exist between a commodity market (e.g., retail, entertainment) and a matching market (dating, recruitment).