AIThe Story of AI
Chapter Four · 2012–2017

The Deep Learning Revolution

One image-recognition contest in 2012 set off the biggest AI explosion in history.

Imagine teaching a computer to recognize a cat. For decades, scientists had to write thousands of rules: "cats have whiskers, pointy ears, fur..." It barely worked! Then in 2012, three scientists named Geoffrey Hinton, Yann LeCun, and Yoshua Bengio (later called the "Godfathers of Deep Learning") helped prove a better way: instead of giving computers rules, you show them millions of examples and let them figure out the patterns themselves. This is called a neural network, and when you stack lots of layers together, it's called deep learning.

The huge breakthrough happened at a contest called ImageNet, where computers competed to identify objects in photos. A program called AlexNet, built by Hinton's student Alex Krizhevsky, crushed the competition, making about half as many mistakes as anyone else! Their secret? They used GPUs, the same graphics chips that make video games look amazing. GPUs could do millions of math problems at once, which is exactly what neural networks need.

After AlexNet, everything exploded. Suddenly computers could understand pictures, voices, and even play games better than humans. Big tech companies raced to hire AI scientists, and AI started showing up everywhere in daily life.

What Is a Neural Network?

Think of your brain. It has billions of tiny cells called neurons that pass messages to each other. When you see a dog, certain neurons light up and shout "DOG!" to your brain.

A neural network is like a math version of that. Imagine a giant assembly line with many layers of workers. The first workers look at tiny pieces (like edges and colors). They pass their notes to the next layer, who spot bigger things (like ears and tails). The final layer puts it all together and says, "That's a dog!"

It's not really a brain — it's just a LOT of math — but it learns the same way you do: by seeing tons of examples and slowly getting better.

§

Five Years That Changed Everything

2011
Apple Launches Siri
The first famous voice assistant on iPhones.
2012
AlexNet Crushes ImageNet
Shocks the world and starts the deep learning revolution.
2014
Google Buys DeepMind
Google buys the London AI lab for about $500 million.
2014
Alexa & GANs
Amazon releases Alexa with the Echo speaker; Ian Goodfellow invents GANs — AI that creates brand-new images.
2015
Better Than Human
Computers beat humans at recognizing images for the first time (Microsoft's ResNet).
2016
AlphaGo vs Lee Sedol
DeepMind's AlphaGo beats the world champion at the ancient game of Go.
2016
Google Assistant
Google launches its own voice assistant.
2016–2017
Self-Driving Cars
Tesla rolls out Autopilot and Google's self-driving project becomes Waymo.
§

Three Stories From the Boom

01

AlphaGo's Move 37

During game 2 against Lee Sedol, AlphaGo made a move so weird that human experts thought it was a mistake! Commentators were shocked. But it turned out to be brilliant — the kind of move no human had ever played in 2,500 years of Go history. Lee Sedol had to leave the room for 15 minutes to recover. AlphaGo won the match 4-1.

02

Made For Video Games

The chips that train powerful AI today were originally invented to make Mario, Halo, and Minecraft look cool. AI scientists basically "borrowed" gaming technology and changed the world with it!

03

AI Dreams Up Faces

Thanks to Ian Goodfellow's GANs (Generative Adversarial Networks), computers can now invent photos of people who don't actually exist. Goodfellow supposedly came up with the idea while arguing with friends at a pub in Montreal!

PIXELS EDGES SHAPES PARTS "DOG!"
"Deep" learning means many layers stacked together. Each layer learns something a little more complex, until the network can recognize a whole object.

The ImageNet Drop

Error rate on the famous image-recognition contest, 2010–2017.

30%20%10%0% OLD METHODS AlexNet human level (~5%) 2010201120122013201420152017
In one year, AlexNet cut errors nearly in half. By 2015, AI was beating humans at recognizing images.
It's not really a brain — it's just a LOT of math — but it learns the same way you do: by seeing tons of examples and slowly getting better.
— What a neural network really is