Week 4 · Lesson 3 · 50-60 minutes
Build and Question Recommendations
See how recommendation systems use similarity and your feedback to suggest what's next — and question how the same system can trap you in a filter bubble.
- Materials
- This lesson in a web browser · Paper and pencil, or a notes app
- This lesson
- In progress
Auto-saves on this device
What you'll be able to do
- Explain how a recommendation system uses similarity to suggest items.
- Describe how your feedback (likes, skips, watch time) trains recommendations.
- Explain what a filter bubble is and how recommendations can create one.
- Audit a recommendation feed and suggest ways to see a wider range.
Think about it
You watch one video about skateboarding tricks. Soon your whole feed is skateboarding, and videos about anything else disappear. How did the app decide to show you only this — and what might you be missing?
You'll learn both how this is built and why it's worth questioning.
Make a prediction
Predict: if a music app only ever played you songs similar to your three favorites, what kinds of music would you probably never discover?
Vocabulary
See the full course vocabulary- Recommendation
- A suggestion of something you might like next — a video, song, product, or post — chosen by an AI system.
- Similarity
- How alike two items or two people are. Recommenders suggest items similar to ones you liked, or that similar people liked.
- Feedback
- The signals you give a system — likes, skips, clicks, watch time — that it uses to learn your tastes.
- Filter bubble
- When a system keeps showing you similar things, so you mostly see one narrow slice of ideas or content and miss the rest.
Recommendations run on similarity
A recommendation system suggests items by measuring similarity. If you liked a video, it looks for other videos that are similar — same topic, style, or creator — or for videos that people similar to you also liked. Then it puts those at the top of your feed.
This is genuinely useful: it helps you find things you'd enjoy without searching. The system is answering 'what is most similar to what this person already liked?'
- Liked a cooking video → recommends more cooking videos
- 'People who bought this also bought…' on a shopping site
- A song station built from one artist you like
Your feedback trains the system
The system doesn't know your taste in advance — it learns from your feedback. Every like, skip, click, replay, and even how long you watch is a signal. Watch to the end, and it counts as a strong 'more like this'. Skip after two seconds, and it counts as 'less like this'.
So you are constantly training your own feed, often without meaning to. Pausing on a video long enough can nudge the system to show you more of that kind, whether or not you actually wanted more.
- Finishing a video → more of that topic
- Skipping quickly → fewer of that kind
- Liking a post → more from that creator and similar creators
Filter bubbles: when similarity narrows your world
Here's the honest trade-off. Because the system keeps showing you what's most similar to what you already liked, your feed can get narrower and narrower. This is called a filter bubble: you end up seeing one slice of content and miss other topics, viewpoints, and creators entirely.
Filter bubbles matter beyond entertainment. If a news or opinion feed only shows you views you already agree with, you might think everyone agrees, and rarely meet good ideas that challenge you. Filter bubbles aren't caused by 'bad' AI — they're a side effect of a system doing exactly what it was built to do: maximize similarity to your past behavior. Knowing this lets you push back — by searching for new topics, following different creators, or using controls that add variety.
- A feed that becomes only one hobby after a single video
- A news feed that shows only one side of an issue
- A shopping site that never shows you a cheaper or different brand
Worked example
How one click can shrink a feed
- You open a video app; the feed is mixed: sports, science, music, comedy, cooking.
- You watch one science video all the way through. That's strong positive feedback.
- The system finds videos most similar to it and moves them up, because similarity says 'show more like the thing they finished'.
- You watch a couple more science videos; each finish is more feedback pointing the same way.
- Now the feed is mostly science. The music and comedy videos, which you might also love, rarely appear — you've slipped into a filter bubble built from your own feedback.
- To widen it, you deliberately search a new topic, skip a few science videos, or use a 'not interested' control to send different feedback.
Takeaway: Similarity plus your feedback makes recommendations helpful, but the same loop can quietly narrow your feed into a filter bubble unless you act to widen it.
A circular flow with four stages. Stage 1: you watch, like, or skip something (feedback). Stage 2: the system updates its guess of your taste. Stage 3: it finds items most similar to what you liked. Stage 4: it shows you those items, which shapes what you watch next — and the arrow loops back to Stage 1. The caption notes that the loop is helpful but tends to tighten around a narrow set of items over time, forming a filter bubble.
The loop learns fast, but each turn can pull your feed toward a narrower set of similar items.
Before: a balanced feed with five topics in roughly equal shares — sports, science, music, comedy, and cooking. After: following one finished science video and a few more, the same feed is now mostly science with only tiny slivers of the others. The shift shows how similarity and feedback can turn a varied feed into a narrow one, and points to actions (search new topics, skip, mark 'not interested') that restore variety.
Activity
Recommendation-System Builder
Build a content-based recommender: rate items, choose which features matter, read recommendations that each explain themselves, and run a filter-bubble experiment.
- Rate a few activities you like or don't. The recommender builds a profile from your ratings.
- Adjust which features matter, then read the recommendations — each one explains why it appeared.
- Run the filter-bubble experiment: rate one topic only, then add a different topic and compare.
This catalog is made up. Don't enter personal information — just rate the built-in items.
Low data: Rate at least two items so the recommender has enough to learn from.
3 · Recommendations
Design a Simple Robot
score 0.0
robots · activity · beginner · hands-on
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
Build a Cardboard Rover
score 0.0
space · activity · beginner · hands-on
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
Animal Habitats Book
score 0.0
animals · book · beginner · reading
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
Make a Cloud in a Jar
score 0.0
weather · activity · intermediate · hands-on
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
Space Station Exhibit
score 0.0
space · exhibit · intermediate · video
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
Planets Picture Book
score 0.0
space · book · beginner · reading
Why you're seeing this
- This appeared because there isn't enough rating information yet to prefer anything specific.
4 · Filter-bubble experiment
- Like a few items from just one topic and look at the feed's topics below.
- Snapshot it, then like an item from a different topic and compare.
- Turn on Explore mode to see how variety changes the feed.
Topics in your current feed (4 of 6 topics)
Current feed topics: space 3, robots 1, animals 1, weather 1.
Check your understanding
Answer these to check that you can explain and question recommendations.
Question 1. What does a recommendation system mainly use to decide what to suggest next?
Question 2. Which of these are ways to widen a feed and escape a filter bubble? (Choose all that apply.)
Select all that apply.
Question 3. Which choice best explains what is happening in this scenario?
After Sam finishes several videos arguing one side of a school debate, his feed stops showing the other side entirely, and he starts to feel like everyone agrees with him.
0 of 3 answered
Try it yourself
Design a bubble-buster feature
Invent one feature for a video or music app that helps people escape filter bubbles while keeping recommendations useful.
- Describe the feature and what button or control the user would see.
- Explain how it changes the feedback or similarity the system uses.
- Name one downside — for example, some users may not want more variety — and how you'd handle it.
Success looks like
- A clear feature with a control the user can use.
- An explanation of how it widens the feed.
- One honest downside and a response to it.
Reflection
Saved on this device as you type
Please don't include private information. Your answers are saved only in this browser and are never sent anywhere.
Recap
Recommendation systems use similarity and your feedback to suggest what's next, which is helpful but can narrow your feed into a filter bubble unless you act to widen it.
- Recommenders suggest items similar to what you (or people like you) liked.
- Your likes, skips, and watch time are feedback that trains your feed.
- The same loop can create filter bubbles; searching and varied feedback widen the feed.
Grades 7–8 extension
Who benefits from your feedback?
Recommendation systems are often designed to keep you watching, because more watch time can mean more profit for the company. That goal doesn't always match your goal of learning widely or spending your time well.
Explain how a company's goal (more watch time) and a user's goal (variety, well-spent time) can pull in different directions, and suggest one honest design change that would serve users better even if it lowered watch time.
Finish this lesson
- Attempt the knowledge check — not done
- Save a reflection (recommended) — not done
Attempt the knowledge check to finish this lesson.