Why you can't stop scrolling: recommendation AI
Big idea: recommendation AI chooses what you see next, and it is usually trying to keep you watching.
The story in short
Happy watched one cricket catch on Mummy’s phone, and three hours later he had watched 200 videos, ending with people dropping watermelons from rooftops. Mitthu watched 41 parrot videos. The children tried an experiment: three tablets, three completely different worlds (Bunny’s space videos, Riya’s cooking, Arjun’s exam tips). At dinner, Kabir chachu explained that the recommender’s main number is “how long will this keep you watching?” In the society, Gowda thatha only ever saw filter-coffee videos and Jyoti aunty only instant coffee, until they swapped phones.
The idea
- Recommendation system: AI that chooses what to show you next.
- Signals: clicks, watch time, skips and rewatches: clues it learns from.
- Collaborative filtering: “people who watched this also watched…”
- Objective: the number the system is designed to push up, often watch time.
- Rabbit hole: small steps towards more and more extreme content.
- Filter bubble: seeing mostly what matches what you already like or believe.
- Autoplay: a design choice that makes it harder to stop.
Family rules: who’s choosing?
- Turn autoplay off.
- Decide first what you will watch, and set a timer.
- Teach your feed: use “not interested”, and search for new topics on purpose.
- Check your feelings: if you feel worse, stop.
Remember: choose what you watch, before it chooses you. Your watch history is a diary you didn’t write.
Check yourself
You watched one cooking video and now your whole feed is cooking. What is the feed doing?
Which family rule helps most with "just one more video"?
Gowda thatha and Jyoti aunty each only saw their own kind of coffee videos. What is this called?
More for you: 🎒 Students · 🧑🏫 Parents and teachers