How does a phone know a dog?
You take a photo of your dog. Your phone says "Dog" and puts it in an album.
Nobody wrote a rule for every dog in the world. Big dogs, small dogs, black dogs, wet dogs.
The phone learned from millions of photos. This is a big change in how computers work.
Key words
| Word | What it means | Example |
|---|---|---|
| Artificial intelligence | Computer programs that do tasks that usually need human thinking. | A chatbot |
| Machine learning | A kind of AI that learns patterns from examples, not fixed rules. | Photo labels |
| Training data | The examples a program learns from. | Millions of labelled photos |
| Model | What a program has learned from its training data. It is used to make a guess. | A dog photo model |
| Generative AI | AI that makes new text, pictures, video or sound. | A tool that writes an email |
| Augmented reality (AR) | Pictures added on top of the real world on a screen. | A game that shows an animal in your street |
| Virtual reality (VR) | A fully computer-made world that you see in a headset. | A tool to practise flying a plane |
| Extended reality (XR) | The name for AR and VR together. | Both kinds of headset |
The cells students fill in are left blank in this preview.
Quick check: rules or learning?
Have a go
Quick check: rules or learning?
3 questionsYour answers are marked here and are not kept. A college account keeps every student’s marks.
Machine learning
Artificial intelligence (AI) is a big idea. Machine learning is one kind of AI.
In normal programs, a person writes the rules. In machine learning, the program finds its own rules from training data.
It is used for spam filters, shop suggestions, map traffic guesses and face unlock.
Try it: a rule written by a person
message = "Win a free prize now"
spam_words = ["free", "prize"]
count = 0
for word in message.lower().split():
if word in spam_words:
count = count + 1
print("Spam words:", count)
# Change the message and run it again.
A program learns what it is shown
data-callouts.warning.label
If the training data is poor or unfair, the program will be poor or unfair too. People must test it and check the results.
Generative AI
Generative AI is trained on huge amounts of text, pictures or sound. It then makes new content when you ask.
It can write a plan, draw a picture or write code. It saves time. But it can be wrong while sounding sure. It can also make a fake picture or voice that looks real.