ML Around You

When Iris took the Plateful job, she assumed machine learning was a new thing she was about to meet for the first time. Then she paid attention to one ordinary morning. Her phone unlocked by looking at her face. Her inbox showed twelve real emails — and had quietly hidden ninety junk ones. Her maps app told her to leave early because of traffic on the bridge. Her music app queued a song she'd never heard and somehow liked. All before breakfast, and every one of it learned from examples, none of it written as rules.

This page is a walk through that ordinary day. The point is not a list of products — it is to retrain your eye. Machine learning is not arriving; it arrived years ago and did what all good technology does: disappeared into the furniture.

Your Inbox Was First

Spam filtering was one of machine learning's first mass jobs, and it is still one of the clearest. Nobody can write rules that keep up with spammers — every rule gets dodged within weeks. So instead, the filter learns from millions of emails that real people marked as junk, the exact move from Topic 01: too tangled for rules, plenty of examples, learn the pattern.

Notice the scale of the gift. The filter reads every email you receive and makes a prediction — junk or real — hundreds of times a week, for billions of people. It gets so much quiet practice that you only remember it exists on the rare day it slips.

Your Phone Is Full of It

Face unlock is a prediction: is this face the owner's? Learned from the example photos your phone took at setup. Photo search that finds "beach" in your camera roll never had anyone label your photos — a model learned what beaches look like from millions of other images. The keyboard suggesting your next word, the voice assistant turning your speech into text: predictions, all of them, learned from examples.

None of this feels like "AI" in the movie sense, which is precisely the lesson. A prediction machine in the furniture does not announce itself. It just makes a small decision, quickly, millions of times a day.

The Feed, the Queue, the Cart

Then there is the most commercially important prediction on the internet: what will this person engage with next? Every "recommended for you" row — the video queue, the shopping suggestions, the social feed, the song that follows the song — is a model predicting your next click from what you, and millions of people like you, did before.

Plateful runs the same machinery. When the app suggests a restaurant, that is a model predicting what Iris's customers will order next. Recommendations quietly drive a huge share of what gets watched, bought, and read in the world — a fact with a shadow side that Chapter 10 looks at directly.

The Route and the Fraud Alert

Two more from Iris's morning deserve a nod because they will both return in this book. Her maps app predicted traffic on the bridge — a number, learned from the movement of millions of phones through that spot at that hour. And when her card was used in a city she had never visited, her bank texted her within seconds — a model had predicted the purchase didn't fit her pattern. Predicting numbers is Chapter 4's territory; spotting things that don't fit the pattern is Chapter 5's.

Why You Never Noticed

Here is the through-line: good machine learning disappears into the product. Nobody ever says "I'm going to use some ML now" — you just flip the switches, the way you use electricity without thinking about power stations. You notice it only when it is wrong: the real email in the junk folder, the absurd recommendation, the fraud alert about your own groceries.

Hold on to that last observation, because it is your first taste of a serious idea. Every one of these systems is wrong some of the time — and they are useful anyway. How wrong, how often, and whether that is acceptable is a question you can actually measure, and Chapter 6 is devoted to it.