History Technology

How Did Machine Learning Begin: The Real History From 1943 To ChatGPT

September 22, 2026

Introduction: How Did Machine Learning Begin

How did machine learning begin.

I remember the first time I asked this question.

It was 2016. I was in my 2nd year of CSE. My professor was teaching us neural networks.
I raised my hand and said “Sir, who invented this? When did it start?”

He smiled and said “2012. With AlexNet and deep learning.”

I nodded. But something felt wrong.

How can something as big as ChatGPT start in 2012?

That night I went home and started digging. Old papers. Old books. Old interviews.

What I found shocked me.

Machine learning didn’t start in 2012. It didn’t start in 2000.
It started in 1943. During World War 2. On a piece of paper.

This article is the story of how did machine learning begin.
Not the textbook version. The real version. With mistakes, failures, and comebacks.

Let’s go back 80 years.

Chapter 1: How Did Machine Learning Begin Number 1 The Crazy Idea In 1943

Two Guys, One Paper, And A Dream About Thinking Machines

Year: 1943. The world is at war.

In Chicago, there is a neuroscientist named Warren McCulloch.
And there is a 19 year old kid named Walter Pitts. He was so smart he dropped out of high school.

They met and asked a weird question: “What if we could build a brain?”

Not a real brain. An artificial one.

In 1943 they wrote a 5 page paper. Most people ignored it.

But inside that paper was the first idea of an artificial neuron.

Their idea was simple.
A neuron gets inputs. If the inputs are strong enough, it “fires” and gives an output.
1 or 0. On or off.

They drew it on paper. They even proved it could do logic like AND, OR, NOT.

This was the very first step.

So when people ask how did machine learning begin, the answer starts here.
With two guys and a pencil. No computers. No data. Just an idea.

The problem? There were no computers to run it.

Chapter 2: How Did Machine Learning Begin Number 2 The Perceptron That Could See 1957

Frank Rosenblatt And The Machine That Learned By Itself

Jump to 1957. Cold War. America vs Russia.

At Cornell University, a psychologist named Frank Rosenblatt was frustrated.
He said “We have the idea from 1943. But it can’t learn.”

So he built something.

He called it “The Perceptron”.

It was not code. It was a real machine. Huge.
It had 400 photocells as “eyes”. Wires everywhere. It looked like a sci-fi robot.

And this machine could learn.

Here is how:
You show it a picture of a triangle. It guesses.
If wrong, it changes its internal numbers called “weights”.
You show it again. It does better.

After 100 pictures, it could tell triangle from square.

The New York Times went crazy. Headline: “Navy Reveals a Machine That Learns”

Rosenblatt did interviews. He said “In 10 years, perceptrons will walk, talk, and be aware of themselves.”

Everyone believed him. Government gave millions of dollars.

For a moment, it looked like how did machine learning begin and end in 1957.

But then came the punch.

Chapter 3: How Did Machine Learning Begin Number 3 The Winter That Killed Everything 1969

How One Book Stopped AI For 10 Years

1969. MIT.

Two very smart guys: Marvin Minsky and Seymour Papert.
They wrote a book called “Perceptrons”.

Inside the book they proved something brutal.
“A single layer perceptron cannot solve simple problems like XOR.”

XOR is: 1+1=0. 0+1=1. Sounds dumb. But the perceptron couldn’t do it.

What did that mean? It meant the perceptron was too weak.

The government read the book. And stopped funding.
Companies stopped hiring. Students changed majors.

This was the “First AI Winter”. From 1969 to 1980.

For 11 years, almost nobody worked on neural networks.

I talked to a professor who lived through it. He said “We had to lie on our resumes.
If we wrote ‘neural networks’, no one would hire us. We wrote ‘statistics’ instead.”

So how did machine learning begin and then almost die?
Because people promised the moon, and delivered a flashlight.

Chapter 4: How Did Machine Learning Begin Number 4 The Algorithm That Saved It 1986

Backpropagation: The Math That Made Learning Real

1986. Things were dead for 17 years.

Then 3 researchers did something.
David Rumelhart, Geoffrey Hinton, and Ronald Williams.

They published a paper. Title: “Learning Representations by Back-Propagating Errors”

It was 15 pages. And it changed everything.

The idea is simple, but genius.

Step 1: Let the network guess.
Step 2: Check how wrong it was.
Step 3: Send that error backwards.
Step 4: Tweak every weight a little bit so next time it’s less wrong.

This is called “Backpropagation”.

Now a network with 3, 4, 5 layers could learn. Not just 1 layer.

Hinton later said “We knew this was big. But nobody cared for 10 years.”

Why? Because computers in 1986 were slower than your phone calculator.
Training took 1 month.

But this 1986 paper is the real answer to how did machine learning begin.
Because without this, there is no ChatGPT, no Tesla, no Google Photos.

Geoffrey Hinton is now called “The Godfather of Deep Learning”. And he started here.

Chapter 5: How Did Machine Learning Begin Number 5 The 90s: When ML Got Boring And Useful

Decision Trees, SVM, and Banks

1990s. The internet was born. Suddenly we had data. Lots of it.

Researchers realized: “Neural nets are cool but too slow.”

So they invented other ways to learn from data.

1. Decision Trees: If income > 50000 then approve loan. Simple. Banks loved it.
2. Support Vector Machines: SVM. Great for faces. Used in early digital cameras.
3. Random Forest: Take 100 decision trees and vote. Still used by Amazon today.

This was “Classical Machine Learning”.

No brains. No neurons. Just math and statistics.

Google used SVM to rank websites.
Credit card companies used decision trees to catch fraud.

Machine learning left the university lab and went to Wall Street.

So if you ask how did machine learning begin, the 90s answer is: “In banks.”

Chapter 6: How Did Machine Learning Begin Number 6 The Day Everything Changed 2012

AlexNet, GPUs, and The Deep Learning Revolution

2012. ImageNet Competition.

The task: Look at 1.2 million images and classify 1000 objects. Cat, dog, car.

For 5 years, the best score was around 25% error.

Then a team from University of Toronto entered.
Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton.

They built “AlexNet”. 8 layers deep. Trained on 2 gaming GPUs for 1 week.

Their error rate: 15%.
Second place: 26%.

The room went silent. Then everyone started clapping.

Why did it work in 2012 and not 1986? 3 things came together:

1. Backpropagation: From 1986
2. GPUs: Nvidia made chips for gamers. AI people stole them.
3. Data: The internet gave us millions of labeled images.

After that day, Google hired the whole team. Facebook hired Yann LeCun. Microsoft hired everyone.

This is when most people think how did machine learning begin.
Because this is when it started working.

Chapter 7: How Did Machine Learning Begin Number 7 The Transformer That Created ChatGPT 2017

“Attention Is All You Need”

2017. Google Brain. 8 researchers in a room.

They were tired of old neural networks. They were too slow.

So they wrote a paper. 8 pages. Title: “Attention Is All You Need”

They invented something called “Transformer”.

Old models read like humans. Word by word.
Transformers read the whole sentence at once and decide “what words to pay attention to”.

Example: “The animal didn’t cross the street because it was too tired”
The model pays “attention” to “animal” when reading “it”.

1 year later, OpenAI used this to build GPT-1.

2019: GPT-2.
2020: GPT-3.
2022: ChatGPT.

And suddenly your mom could use machine learning.

So how did machine learning begin?
It began in 1943. But it went viral in 2022.

Chapter 8: The People Behind How Did Machine Learning Begin

You can’t tell this story without names. Here are the 7 people:

1. Warren McCulloch + Walter Pitts 1943: Drew the first artificial neuron
2. Frank Rosenblatt 1957: Built the first machine that learned
3. Geoffrey Hinton 1986: Invented backpropagation. Without him, no AI
4. Yann LeCun 1998: Invented CNNs. Used in every phone to read your handwriting
5. Andrew Ng 2008: Put ML courses on Coursera. Taught 2 million people
6. Alex Krizhevsky 2012: AlexNet. Started the deep learning boom
7. Ilya Sutskever 2017: Co-founder of OpenAI. Pushed transformers

These 7 people are why you are talking to me right now.

Chapter 9: Why Did It Take 80 Years

If the idea was in 1943, why did ChatGPT come in 2022?

Because we were missing 3 things:

1. Data

In 1950 there was no data. In 2020 we have all of Wikipedia, books, and Reddit.

2. Compute

In 1950 a computer cost $2 million.
In 2020 a $1500 GPU is 1 million times faster.

3. Algorithms

We needed 40 years to invent backpropagation and transformers.

So how did machine learning begin? With an idea.
And it took the world 80 years to build the tools to run it.

Chapter 10: Full Timeline of How Did Machine Learning Begin

Chapter 11: FAQ About How Did Machine Learning Begin

1. How did machine learning begin and who invented it

It began in 1943 with McCulloch and Pitts. But modern ML began in 1986 with Geoffrey Hinton.

2. What was the first machine learning algorithm

The Perceptron in 1957 by Frank Rosenblatt.

3. Why was there an AI winter

Computers were too slow and researchers promised too much in the 1960s.

4. What caused the 2012 deep learning boom

GPUs + Big Data + Backpropagation. AlexNet proved it.

5. Is ChatGPT the end of machine learning history

No. We are just in chapter 7. AGI is probably chapter 8.

Final Thoughts: How Did Machine Learning Begin And What’s Next

I started this to answer one question for my professor.

How did machine learning begin.

The answer is not 2012. Not even 1990.

It began in 1943, during a world war, with two guys who just wanted to understand how we think.

It failed. It came back. It failed again. Then GPUs saved it.

Now in 2026, it is in your phone, your car, your doctor’s office.

The people in 1943 dreamed of a thinking machine.
We are 80 years in. And we are close.

Maybe in 10 years your kids will ask “How did AGI begin?”
And the answer will start with “ChatGPT, 2022.”

What part of this history surprised you most?
Comment below. I read every comment.

And if you found this useful, share it with someone who thinks AI started with Chat