Chapter 3 of 8
You’ve Been Using AI All Along
Long before anyone typed a question into ChatGPT, machines were filtering your email, guarding your credit card, and finishing your words.
Updated as AI changes. Last reviewed October 1, 2026.
Some of it is true machine learning. Some of it is just clever rules. Telling them apart is the whole skill, and it is the reason the words on these badges matter. Pick anything below and see what it was really doing, then see if you can tell the difference on your own.
“…I think for the first time, people are seeing the power of what’s been working behind the scenes for a long time.”
“All the data points that are just constantly being absorbed from our smart lights to our smart thermostats, to our smart doorbell, to our car’s geolocation, to our cell phone’s location, to our watch knowing if we’re standing up or sitting down, and our heart rates, and our wearable rings, and just all these things.”
Fifteen things you already used
- A person wrote the rules. It does not learn.
- Predicts from patterns in a lot of data, without learning as it goes.
- Finds a known shape, like a face, inside an image.
- Improves its own predictions from data, with nobody rewriting the rules.
-
Each ghost follows its own chase rule.
-
A neural network scores every swipe against your normal pattern.
Machine learning fico.com : FICO Falcon, 1992 (opens in a new tab) -
Swaps common typos for the right word from a list.
-
Calculates the shortest route across a road map.
-
Ranks pages by the links pointing to them.
-
Amazon recommends items often bought together.
-
Guesses your word from keypresses using a dictionary ranked by how common words are.
-
Predicts your rating from people with similar taste.
-
Bumps, turns, and spirals by fixed rules.
-
Learned what spam looks like from millions of labeled messages.
-
Finds faces in the frame and focuses on them.
Pattern recognition dpreview.com : Nikon face-priority AF, 2005 (DPReview) (opens in a new tab) -
Learned translation from already-translated text.
-
Pools phone speed data to predict traffic and arrival times.
Statistical, later ML blog.google : Google Maps traffic and ETAs (opens in a new tab) -
Predicts the rest of your search from what others search.
-
Face recognition suggests which friend is in the photo.
1980
Pac-Man ghosts
Is this AI?
Ten questions, no email, nothing saved. Harder than it looks. The reveals are where the real distinction lives, so read them even when you get it right.
-
MapQuest printing your route (1996)
Not AI
A routing algorithm: fixed math, no learning. But you still offloaded your sense of direction to it.
-
Your email spam folder (2003)
AI
Machine learning, trained on millions of messages people marked as spam.
-
T9 texting (1999)
Not AI
A ranked dictionary lookup. Clever, not learning.
-
Your bank texting "Did you make this purchase?" (1992)
AI
A neural network flagged the swipe as unusual for you.
-
Roomba (2002)
Not AI (at launch)
Fixed reactive rules. Later models added mapping.
-
Netflix "Because you watched…" (2000)
Borderline
Statistical prediction from other people’s ratings. Early recommendation engines sit right on the line.
-
Face ID (2017)
AI
A neural network matching your face.
-
AutoCorrect (1993)
Not AI
A list of common typos. Today’s phone keyboards do learn; the 1993 version did not.
-
Pac-Man ghosts (1980)
Game AI
Rules that feel like personality. That feeling is the point.
-
ChatGPT (2022)
AI
Generative AI built on a large language model. See Say It Right for what that means.
Not everything smart is AI, and not all AI is ChatGPT. That is why the words matter.
