How apps use AI

What's going on inside an app that uses AI? Let's open one up. An app that uses A I is mostly a normal app, with buttons, screens and data, just like the apps you learned about in the tech course. Inside, there's an A I model that does one special job, like recognising a picture, predicting something, or writing text.

Inside an AI app

Here's how it works. First, the input. For example, a farmer takes a photo of a cassava leaf. The photo goes to the model, which was trained on thousands of labelled leaf photos. The model gives an output. Mosaic disease, eighty seven percent sure. Then the app takes an action. It shows advice on what to do next.

The same pattern everywhere

You'll find this pattern everywhere. A spam filter takes an email in, gives a spam score out, and moves spam to a folder. Maps take traffic data in, and predict how long your journey will take. Voice typing takes your speech in, and gives text out. And your keyboard takes your words in, and suggests the next word.

New word: Confidence score

Notice the eighty seven percent. That's called a confidence score. It shows how sure the model is about its answer. Good apps use this score to decide what to do next. If the confidence score is low, say fifty five percent, a good app doesn't pretend to be sure. It says something like, I'm not sure, please show this leaf to an expert. Keeping a person involved for hard cases is called having a human in the loop.

Quick check

Quick check. A plant app is only fifty one percent sure about a disease. What should a good app do? Say it's unsure, and suggest asking an expert. Honest apps are safer apps. Options: Say it is unsure, and suggest asking an expert; Pretend to be completely sure; Delete the photo; Show a random answer.

Recap

A I apps are normal apps, with a model inside. They follow a pattern. Input, model, output, action. Confidence scores show how sure the model is. And when it's unsure, a good app asks a person.

I've read this lesson