Most of their "AI" is just applied statistics. Train a statistical classifier to classify "events a user wants to see" then run it in prod and elevate the re

Most of their "AI" is just applied statistics. Train a statistical classifier to classify "events a user wants to see" then run it in prod and elevate the results to notifications. Run an A/B trial (quite possibly p-hacked) to determine if people actually like it. Ignore the false positive rate (e.g. the rate at which you're annoying people) if you think it's arbitrarily small enough. When people ask how you determine which events to turn into notifications tell them The Algorithm decides it and let their imaginations, seeded by pop culture science fiction, run wild about what sort of superhuman synthetic intelligence you have pent up in your computers when really all you've got is a glorified bayesian spam filter.