The Channel That Took Over

For almost all of recorded history, people met their partners through a small number of channels: family, neighbourhood, church, work, and friends. These were different in detail and identical in structure. In every case the intermediary was somebody embedded in both people's lives.

Survey research on how American couples met shows that changing. Across the twentieth century, meeting through friends rose to become the dominant channel; meeting through family and church declined. Then meeting online rose steeply, and around 2013 it passed meeting through friends among heterosexual couples, for the first time in the data.

That is a particular kind of event. New channels usually add themselves to the list. This one displaced the incumbent, and the authors of the research describe it as the internet taking over roles that family and friends used to play.

Which raises a question that is more interesting than the usual complaints about apps: who is the intermediary now, and what do they want?

A friend making an introduction wants it to work. Their incentive is aligned almost perfectly with yours: they will see both of you again, they will be told if it went badly, and their reputation as a judge of people is at stake. The service is free and the quality control is social.

A platform's revenue depends on continued use. A user who meets somebody in week one and leaves is a short relationship with the company. A user who searches for two years, subscribes, buys boosts and keeps coming back is a good customer.

This is not a conspiracy and it does not require anybody at these companies to behave badly. It is just two parties optimising different objectives, and the divergence is structural.

Then there is the matching itself, which is the industry's central marketing claim.

A substantial review of the field concluded that there is no compelling evidence that any online dating matching algorithm predicts whether two people will do well together. The reviewers went further and argued that these algorithms are unlikely to work even in principle.

The reasoning is the important part, because it is not an accusation of dishonest data. Algorithms match on what can be collected from two profiles before anybody meets: similarity of personality, similarity of attitudes, stated preferences. Relationship research finds that these are weak predictors of long-term outcomes compared with how a specific pair actually interacts, how they handle conflict, how they respond to each other's bids for attention, what happens between them in a room.

You cannot get that from two profiles. It does not exist yet when the matching is done.

And the previous lesson makes it worse. Even the preference data the platforms do collect is self-report about the causes of attraction, which people cannot reliably produce. So the input is both incomplete and partly fictional, and no amount of model sophistication fixes an input that is not there.

Which leaves the platforms doing something real and much less impressive than the pitch: they produce a very large pool of people who are available and looking, with basic filters, at low cost. That is a genuine service and it is the opposite of matchmaking. It is a bigger room.