Back to Glossary

Machine Learning

Synonyms: ML, statistical learning, pattern recognition, trained models, “the algorithm”

Do not index

Definition

Machine learning (ML) is a branch of artificial intelligence where software learns rules from examples. Feed it 10,000 emails labeled “spam” or “safe,” and it works out its own sorting rules for email number 10,001.
The output is a model, a system that makes educated guesses. It’s right most of the time and confidently wrong the rest (it has famously mixed up chihuahuas and blueberry muffins).
 

Use cases

When designers ignore how ML fails, users either trust it blindly or quit it.
  • The cold start flop: A new user opens a music app and gets a stranger’s playlists. The model has no data on them yet, so the first session feels broken.
  • The confident wrong tag: A photo app labels your mom as your aunt, and there’s no fix button. Users stop trusting every tag after that.
  • The one-click spiral: One sourdough video turns someone’s feed into a bakery for a month. The model read a casual click as a strong preference.
 

How it's used in practice

  • Check if ML is even needed: If a simple rule works (sort by date, filter by price), skip the model.
  • Design for zero data: Ask 2 or 3 preference questions at signup, or show popular picks until the model learns.
  • Show confidence where it matters: Use soft labels such as “Suggested for you” or “Is this Sarah?” on low-confidence results.
  • Give users a correction path: Add “Not this person” and “Not interested” controls, then send those corrections to the data team.
  • Prototype with real outputs: Test 50 real model results with users, bad ones included. Static mockups always show the perfect answer.
🪄
Ask your data scientist who falls into the error rate. A 95% accurate model that fails mostly for one accent or skin tone is a fairness bug waiting for a headline. Design the fallback for that group first.
 

Challenges & limitations

  • Black box outputs: Many models can’t explain their choices, so honest “Why am I seeing this?” copy is hard to write.
  • Bad data, bad results: Thin or biased training data produces skewed outputs, and no amount of UI polish fixes that.
  • Models change over time: Retraining shifts behavior, so a flow that tested well in March can feel off by June.
 

Free resources

Share this post

Get free UX resources

Get portfolio templates, list of job boards, UX step-by-step guides, and more.

Download for FREE
 
 

The best email 📮 for growing 🌱 designers

Honest notes about the work behind the work. Read in 2 minutes, weekly. Free forever.
 
 
     
    notion image
    Join 13,045 designers and get tactics, hacks, and tips.