Mishari.AI: The story of artificial intelligence
In 1950 we asked if machines could think. We’re about to find out.
The first technology that teaches itself.
Old computers only did what we told them. AI learns from millions of examples, the way you learn by practicing.
The question.
Can machines think? Alan Turing asked in 1950, with a simple test: if you can’t tell a machine from a person by chatting, does it matter?
A name is born.
Artificial intelligence. A small group of scientists met at Dartmouth for the summer, gave the field its name, and hoped to make machines that learn and think.
One artificial neuron.
The Perceptron is built like a single brain cell and learns by adjusting itself when it’s wrong.
Newspapers said it would one day walk, talk and know that it exists.
The AI winters.
Big promises, tiny computers: twice the excitement ran ahead of the machines, and the money froze.
Now it’s back, with $582 billion invested in 2025 alone: a new spring, or a bubble?
AI winters, then AI money
Checkmate.
Deep Blue beats Kasparov. IBM’s computer outplays the world chess champion by checking 200 million positions a second: raw speed, not understanding.
Machines learn to see.
Deep learning arrives. A neural network called AlexNet recognizes images far better than anything before it, powered by huge collections of photos and video‑game chips.
Move 37.
AlphaGo beats Lee Sedol. In Go, a game with more positions than atoms in the universe, it plays a move no human would have chosen, and experts call it beautiful.
Pay attention.
Every word, all at once. A new kind of AI, the Transformer, reads a whole sentence in one go and links words that belong together, like “it” and “cat”.
Everyone meets AI.
ChatGPT arrives. About 100 million people try it within two months, faster than any app before it.
Time to reach 100 million users
The engine.
The secret is scale. Since 2010, the computing power used to train the biggest AI has grown four to five times every year.
| System | Year | Operations used for training |
|---|---|---|
| Theseus, a maze-solving robot mouse | 1950 | about 40 |
| Perceptron Mark I | 1957 | about 700 thousand |
| NetTalk | 1987 | about 28 billion |
| AlexNet | 2012 | about 4.7 × 10 to the power of 17 |
| AlphaGo | 2016 | about 1.9 × 10 to the power of 21 |
| GPT-3 | 2020 | about 3.1 × 10 to the power of 23 |
| GPT-4 | 2023 | about 2.1 × 10 to the power of 25 |
| GPT-6 Astra | 2026 | about 10 to the power of 27 |
Where we stand.
From answers to actions. AI now writes code, sees, speaks and reasons, and more and more it finishes whole jobs on its own.
Longest job AI can do on its own
It succeeds about half the time.
The hard questions
If a machine can do your job, what is your job for?
Who decides what an AI believes is right?
Can something understand without feeling anything?
What happens when it’s smarter than all of us, combined?
Will it make us more human, or less?
The future isn’t predicted.
It’s engineered.
Every chapter so far was written by people. The next one, we write together with the minds we’ve made.
Every message is read personally.