← Paper page A Sample Paper on Learning Systems and the People Who Use Them PDF ↗

A Sample Paper on Learning Systems
and the People Who Use Them

When a system learns from what people choose, what happens to the people?

Jordan Lee, Sam Rivera, Morgan Ellis · EXCONF 2026 · a fictional sample

Press → or click to step through

Three players around one recommender

creators people recommender ranks what to show choices flow back as training data

The system learns from the very choices it shapes.

Illustration. Colours keep one meaning in every deck: people blue, creators orange, the system purple, good outcomes green.

The worry: the loop narrows what people see

Round 1 topic Atopic Atopic Btopic C a varied menu ✓
Round 10 topic Atopic Atopic Atopic A one topic crowds out the rest

How much does this narrowing cost the people the system serves?

Schematic, not data.

Result: a little exploration keeps most of the value

share of the best possible welfare that the loop keeps best possible no exploration 55% a little 89% a lot 75% illustrative numbers for this sample deck

Without exploration, the loop locks in early guesses.

Too much exploration wastes attention; a little goes a long way.

Try it: how fast does the menu narrow?

topics still shown after each round rounds →
exploration

Illustration with made-up curves: replace this slide with an interaction built on your paper's numbers.

What this means

  1. 1Feedback loops can narrow choicesA system that learns from what it shows can lock in early guesses.
  2. 2A little exploration helpsKeeping a few varied items in view preserves most of the value.
  3. 3Build your own deckCopy this file, keep the engine in static/explainers/_deck/, and replace the content.

Paper page · PDF