80 Years of AI

Quick story. AI didn't just show up last year. It's been brewing for eighty years. We'll fly through it from World War Two right up to the AI on your phone today. Buckle in.

Nineteen forty-three. Two scientists, McCulloch and Pitts, write down a math equation for a single brain cell. Just one neuron, on paper. Sounds tiny, but that little equation is the seed of every AI today.

Seven years later, Alan Turing — the guy who helped crack Nazi codes in the war — asks one bold question. Can machines think? He doesn't try to define thinking. He just makes a test. If a computer can fool you in a chat, it's thinking-enough. That's the Turing Test.

Nineteen fifty-six. A bunch of scientists meet at Dartmouth College for a summer workshop. They give the field its name — Artificial Intelligence — and confidently predict it'll be solved in a few years. It took seven decades. But the name stuck.

Fast forward to the seventies. Meet Shakey. First-ever robot that could see with a camera, plan a path, and actually move stuff around. Slow as a snail, sure. But Shakey is the great-grandparent of every robot vacuum, every self-driving car. It started here.

But the brag — quote, AI in a few summers — well, it didn't pan out. Computers were too slow. Neural networks couldn't do enough. Money got pulled. Researchers left. They called it the AI Winter. AI nearly died right here.

Nineteen ninety-seven. IBM builds a supercomputer named Deep Blue. It plays the world chess champion — and wins. Garry Kasparov, the best human alive at chess, gets beaten by a box. Nobody calls AI dead anymore.

Two thousands. Internet goes wild. People post billions of photos, songs, posts — all that data is suddenly perfect food for AI to learn from. Plus graphics cards from video games turn out to crunch neural network math at lightning speed. Two ingredients lock into place.

Two thousand twelve. A team enters an image-recognition contest with a neural network called AlexNet — and crushes it. Twice as accurate as anything else. People who'd written off neural nets twenty years earlier suddenly snap to attention. Deep learning is real.

Then in twenty sixteen — Google's AlphaGo plays the world champion at Go. Go is a board game with more possible moves than atoms in the universe. No regular program could ever crack it. AlphaGo learned by playing itself, millions of times. And it won.

Twenty twenties. AI shifts gears. It stops just recognizing things and starts making them. Type a sentence — get a picture. Hum a tune — get a full song. Describe an app — get the code. We call this generative AI.

November twenty-twenty-two. A company called OpenAI puts ChatGPT online. Anyone could chat with it. In two months — one hundred million people had tried it. Fastest growing app in history. Your grandparents heard about AI on the news. That's when AI showed up at the dinner table.

Okay your turn. I'll show you a famous AI moment. You pick the decade it happened in. Six rounds. Don't sweat the wrong ones — the dates are tricky.

And here we are. Today. AI is in your phone, your games, your school, your medicine. Trillions of dollars are being spent. New stuff drops every month. [pause:400] And — kind of wild — you are growing up right in the middle of this story. The next chapters? Your generation is the one writing them.

Cool. So now you know how AI got here. Next question — how does it actually learn? Turns out there are three different styles. Let's go.