Week 8: Recommendation Algorithms

A feed doesn't have to remove your choices to narrow your world — it only has to keep offering you more of whatever you already picked. This week is about what that quietly does to what you even get shown.

12 April 2027

Open the slides

This week’s question#

A recommendation feed never has to say no to you. It doesn’t need to block a genre or remove an option to shape what you watch — it only needs to notice what you already picked, and quietly offer you more of that next time. Do five small, individually reasonable picks in a row change what you’re even being shown by the sixth?

Feedback loops don’t need intent to narrow#

The mechanism here doesn’t require a feed to be trying to trap you. A simple rule — “show more of what this person picked before” — is enough on its own, because every pick becomes an input to the next round’s options, and the next round’s options shape what’s available to pick from. There’s no step where variety gets deliberately removed. There’s just a rule that responds to your last pick, applied repeatedly, and repetition is what does the narrowing.

This is worth separating clearly from taste. Liking comedies more than documentaries and ending up shown mostly comedies looks the same from the outside as a feed narrowing on its own momentum, independent of whether your underlying interest ever actually changed. The two get tangled together constantly — genuine preference feeding a system that then amplifies exactly that preference until it looks less like a preference and more like the whole menu.

A familiar scene#

You watch one workplace comedy on a whim, on a slow evening. Within a few sessions, your home screen is visibly heavier on comedies than it used to be — not because you told it your genre preferences, and not because you watched anything else in between. One pick, repeated back to you as more of the same, is often all it takes to notice the shape of the feed itself start to shift.

The interactive experiment: The Narrowing Feed#

Below is a small, five-round version of that loop. Each round shows four shows from four categories; pick one. The next round’s mix updates based only on what you just picked, following a fixed, simplified rule — not a real streaming algorithm, just enough of the same shape to make the mechanism visible.

What just happened?#

Every round’s tile mix was generated purely from your own pick history, fed through a fixed weighting rule set in advance — nothing here adapted to guess your “real” taste, and no live model or randomness was involved. If the categories on screen narrowed noticeably by round five, that’s the mechanism working exactly as designed: a handful of individually small picks compounding into a visibly smaller menu, with no single step where anyone decided to remove an option.

If you deliberately varied your picks across categories, you likely saw the mix stay closer to its starting diversity — worth noticing, since it shows the narrowing isn’t inevitable, only a predictable response to a pattern of picks, not to any one pick in isolation.

What this demonstration can’t tell you: a four-category, five-round toy catalogue with a fixed formula is a teaching simplification, not the algorithm any real streaming platform actually runs, and your reaction to it says nothing about your real-world viewing habits or how “in a bubble” you personally are. What it can show, reliably, is the shape of the mechanism — that this kind of feedback loop narrows a feed mechanically, given nothing more than a rule and repetition.

Reflection#

Think about your own most-used streaming or video app’s home screen right now. If you tried to remember what it looked like the very first time you used it, would you expect it to look meaningfully different from today’s — and if so, was that difference something you asked for, or something that accumulated from picks you made without thinking about where they’d lead?

Next episode#

This week was about a feed narrowing in private, based only on your own picks. It said nothing about what changes once a pick isn’t private at all — once you can see, before choosing, what other people already chose. That’s next episode’s question.