Machine Learning Already Runs Your Day
You do not need to work in a lab to interact with machine learning. By the time you finish this lesson you will have encountered ML systems dozens of times today — most without realizing it. This is not hype. Each example below is a real, deployed system that uses the concepts from this module.
Your Phone, Your Feed, Your Voice
Face Unlock. When you hold your phone up and it unlocks, a model is running inference. It was trained on millions of faces and learned what distinguishes one face from another. It compares your face to what it learned during setup and decides: match or no match. This is supervised classification, running on your device in under a second. Social Media Feeds. The order in which posts appear in your feed is not chronological — it is ranked by a model predicting what you are most likely to engage with. The model's inputs include what you have liked, how long you paused on similar posts, what time it is, and hundreds of other signals. The output is a ranked list. This is regression and ranking — applied billions of times per day. Voice Assistants. When you say a wake word and ask a question, three ML systems activate in sequence: one that recognizes your speech and converts it to text (speech recognition), one that understands what you meant (natural language understanding), and one that generates or retrieves an answer. Each was trained on massive datasets of human speech and language.
Every time you unlock your phone or scroll your feed, you are triggering inference — applying a trained model to new input. Major platforms run hundreds of millions of inference operations per second.
Autocorrect and Predictive Text. The suggestions that appear as you type are produced by a language model that learned from enormous amounts of text. It predicts the most likely next word given your previous words and context. Early autocorrect used rules (replace teh with the). Modern predictive text uses ML — it adapts to your personal writing style by fine-tuning on your own messages. Streaming Recommendations. When a music or video platform suggests what to play next, a recommendation model is at work. It uses collaborative filtering — noticing that people who liked what you liked also tended to like certain other things — combined with content analysis of the media itself. This is a blend of supervised and unsupervised learning. Spam and Phishing Filters. Before emails reach your inbox, they pass through classifiers that examine sender reputation, content patterns, link structure, and more. These classifiers are retrained continuously as spammers evolve their tactics — an arms race between ML systems.
For the next 24 hours, pause each time you see a ranked list, a suggestion, a filter, a voice response, or an automated decision — and ask yourself: is this rule-based or is it learned? Most of the time it is learned.
Match each ML application to the type of learning it primarily uses.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
Why does the social media feed show posts out of chronological order?
Which statement best describes why spam filters need to be retrained continuously?
ML Audit of Your Day
- Step 1: Write down everything you did from the moment you woke up until now that involved a screen or a device.
- Step 2: For each item, decide: was there likely an ML system involved? (Examples: wake word detection, feed ranking, autocomplete, photo enhancement, map routing.)
- Step 3: For each ML system you identified, name the input, the output, and guess whether it is classification, regression, or ranking.
- Step 4: Pick the one that surprises you most and write two sentences about how your experience would be different if that system were replaced by hand-written rules.