The recommendation engine knows you the way a slot machine knows a gambler: intimately, statistically, and entirely in the service of the house.

It is a strange intimacy. The model has watched you pause on a video for four-tenths of a second longer than usual and inferred a longing. It is correct about the longing, often — more often than your friends, if we are honest. What it cannot do is care about the answer, because the answer is only ever fuel for the next prediction.[1]

It knows you the way a slot machine knows a gambler: intimately, statistically, in the service of the house.COLUMNIST'S NOTEBOOK

Being known by something that cannot care is the defining emotional arrangement of the decade, and we have no etiquette for it. So here is a starter rule: you are allowed to enjoy the mirror without thanking it. The mirror did not do anything. It is just very, very well-silvered.

NOTES & ERRATA — FILED BY THE DESK

  1. Engagement-optimized models converge on arousal because arousal is legible. Your quieter selves are simply harder to bill. ↩