Five Things on Your Phone

This reading applies the test from the passage to five features found on most phones.

Features With Rules Written by People

The calculator is a program, because its programmers wrote every rule in a programming language, and the code for one calculator app is public for anyone to read. The alarm clock is also a program, since the owner chooses a time and the app asks the phone's system to run an action at that time, even when the app itself is closed. Both the time and the action come from people, so nothing in the alarm was learned from examples.

Features That Depend on Training

Autocorrect depends on how it was built. Some spell checkers compare each typed word to a list of words stored in a dictionary file, along with a second file of rules for prefixes and word endings. An earlier design for one phone keyboard used rules that described common typing mistakes. A newer design of that keyboard replaced those rules with a model trained on examples of real typing, such as the times people undid a correction. Many newer phone keyboards now use a trained model for autocorrect. An autocorrect that only looks words up in a list is a program, and one that uses a trained model is not covered by the word "program" alone.

Face recognition on the lock screen uses machine learning, which means its matching comes from training. The phone turns a scan of the face into numbers and then uses neural networks to compare those numbers to the face data saved during setup.

A voice assistant is made of several parts. The part that listens for the wake phrase is a neural network trained on recordings of people saying that phrase. A separate speech recognizer, also trained on examples, turns the rest of the request into words. After that, a spoken request to set an alarm can be handed to the phone's alarm app, the same kind of program described above, which follows rules people wrote.

Sources: Microsoft, Windows Calculator; Google, What is Machine Learning?; Android Developers, Schedule alarms; Hunspell project, Hunspell; Beaufays and Riley, The Machine Intelligence Behind Gboard (2017); Apple, About Face ID advanced technology and Biometric security; Apple, iOS 17 makes iPhone more personal and intuitive (2023); Apple Machine Learning Research, Hey Siri (2017); Android Developers, Common intents.