Can a computer learn?
A one-hour, hands-on crash course in AI & machine learning
π Dr. Rachid Zeghlache
Assistant Professor of Artificial Intelligence
American University in Dubai
Hands up π
Who used AI today?
πUnlocked a phone with your face
π΅Got a song recommended
πΊοΈChecked traffic on a map
π§Didn't see any spam
You probably used AI a dozen times before breakfast β without noticing.
π€ AI or Not?
Drag each card. Shout your answer first!
β open
The big idea
Rules vs. learning
βοΈ Normal program
π Rulesπ¦ Data
β¬
Computer
β¬
β
Answers
A person writes every rule.
"IF temperature < 20 THEN heat on"
π§ Machine learning
π¦ Dataβ
Answers
β¬
Computer
β¬
π Rules
You show it examples.
It works out the rules itself.
Words you'll hear
π€ Artificial Intelligence
π Machine Learning
π§ Deep
Learning
AI β any computer doing something that looks smart. Even a chess program with hand-written rules.
Machine learning β AI that learns from examples instead of following written rules.
Deep learning β machine learning with huge neural networks. Behind face unlock, ChatGPT and self-driving cars.
π― Teach the Computer
Cats, dogs or hamsters? Tap to add examples.
β open
Plot twist
Where do the examples come from?
From people. Millions of photos, drawings and sentences β each one labelled by a human.
For the next few minutes, you are the data labeller. βοΈ
βοΈ Doodle Trainer
Draw β save with a label β draw a new one β does it guess?
β open
π² Box Labeller
π bonus
Box every object and label it β like the people who train self-driving cars.
β open
ποΈ How a Computer Sees
A picture is just a grid of numbers.
β open
Deep learning, in one picture
Layers on layers of filters
π· pixelsγ°οΈ edgesπ· shapesπ± "cat!"
π§ Build a Brain
Design the layers. Train on 10,000 digits. Draw one.
β open
πΈ Train a Camera AI
Hold a button to record examples. Then test it.
β open
Why did it get confused?
An AI only knows its data
34.7%
error rate for darker-skinned women in commercial face-analysis softwareβ¦
0.8%
β¦versus lighter-skinned men.
Gender Shades study, Buolamwini & Gebru, MIT Media Lab, 2018
π Same problem as our demo
The training photos were mostly of one kind of face. The AI wasn't "mean" β it just never learned the others.
Fix: more varied data, and people who check.
That's why AI needs people from every background building it β including you.
π Object Detector
π bonus
A real AI trained on 200 000+ hand-labelled photos. What can it spot?
β open
From pixels to words
Can a computer read?
To a computer, a sentence is just numbers too β words become counts, and counts become patterns.
π Mood Reader
Type anything. Is it happy or sad? Watch every word vote.
β open
π¬ Mini ChatGPT
It read Alice in Wonderland. Now it predicts the next word β again and again.
β open
π©Ί AI Doctor's Assistant
569 real biopsies. Which mistake is worse?
β open
Beyond the classroom
What can AI do?
π©ΊMedicine
Spots signs of disease in X-rays and eye scans, helping doctors catch problems early.
π§¬Science
AlphaFold predicted the 3D shape of ~200 million proteins β a 2024 Nobel Prize in Chemistry.
π¦οΈWeather & climate
AI weather models make a 10-day forecast in under a minute β a job that takes supercomputers hours.
π¦»Accessibility
Live captions for deaf users, and apps that describe the world out loud to blind users.
π£οΈLanguage
Translate between 100+ languages, chat, summarise, help write code.
π¨Art & music
Generate images, songs and video from a sentence.
πNature
Identifies animals in millions of wildlife-camera photos and whale songs in ocean recordings.
βοΈGames
AlphaGo beat the world Go champion in 2016 with moves no human had played.
β¦and what to watch out for
AI can be wrong β confidently
π€₯Hallucinations
Chatbots can invent facts, quotes and sources that sound 100% real. Always double-check.
πDeepfakes
Fake videos and voices of real people. If it's shocking, check where it came from.
βοΈBias
Unfair data β unfair AI. You saw it with the camera.
πPrivacy
AI is trained on data β sometimes yours. Think before you share.
AI is a tool. It's only as good as its data β and the people using it.
Remember three things
What you learned today
1
AI learns from examples
Not from rules someone wrote. You trained seven yourselves.
2
Data is everything
Humans choose and label it. Bad or narrow data β bad or unfair AI.
3
Stay curious & critical
AI is powerful and can be wrong. Check it β and build it better.
Your future
Jobs in AI β and AI in every job
π οΈMachine Learning Engineer
Builds AI models and puts them inside real apps used by millions.
πData Scientist
Turns messy data into answers and predictions β sport, business, health.
π¬AI Researcher
Invents the next generation of AI. Usually via a master's and a PhD.
ποΈComputer Vision Engineer
Teaches cameras to see: self-driving cars, drones, factory robots.
π¬Language-AI Engineer
Chatbots, translation, voice assistants β the Mini ChatGPT, for real.
π©ΊAI in Healthcare
Reads medical scans, speeds up drug discovery β like the AI doctor's assistant.
π¦ΎRobotics Engineer
Robots that learn to walk, grasp and help β at home, in hospitals, in space.
βοΈAI Ethics & Safety
Makes AI fair, safe and legal β the bias problem you saw today, as a job.
π AI + anything: medicine, law, finance, climate, art, sport, games β every field now needs people who understand AI.
Start this weekend β for free
The tools to master, step by step
- 1
π Maths you already do
Statistics, probability, algebra β the ideas behind every model.
at school Β· Khan Academy
- 2
π Python
The language of AI. Clear, free, used everywhere.
Python Β· Google Colab (nothing to install)
- 3
π Data tools
Load, clean and chart data.
NumPy Β· pandas Β· Matplotlib
- 4
π€ Machine learning
Train your first models in a few lines.
scikit-learn
- 5
π§ Deep learning
Neural networks for images, sound and text.
PyTorch Β· TensorFlow / Keras
- 6
π€ Ready-made AI
Use and fine-tune thousands of models.
Hugging Face
- 7
ποΈ Share your work
Every project you post is part of your portfolio.
Git Β· GitHub
- 8
π Practise on real problems
Real datasets and friendly competitions.
Kaggle
π The secret: build one small project a month β a game bot, a mood detector, a sports predictor. Projects teach more than any course.
Want to build them for real?
Every AI today, in a few lines of Python
Tap any card to see the code Β· all of them run free in Google Colab π
Keep playing at home
π
Thank you!
Questions? Ideas? What would you teach an AI?