NOW
The machines have started to write themselves.
This is where the story has arrived. To understand it, you have to go backward — past every answer, down to the first idea.
scroll down to rewind ↓
Consequence
Everyone could suddenly talk to it.
A chat box put the whole field in a billion hands overnight. The conversation felt new — but it was only the visible tip of a much older effort.
Because, before that —
Six words rearranged the field.
“Attention is all you need.” The Transformer let a model weigh every word against every other at once. Ashes of old architectures reassembled into something that scaled.
Because, before that —
A network learned to see.
On a pile of labelled photos, a deep net crushed the vision benchmark and ended the argument. Two graphics cards did what rooms of code could not.
Because, before that —
The human hand began to withdraw.
Rules written by people gave way to patterns found by machines. Slowly, the hand that reached in to correct every step un-reached — and let the data pour back.
Because, before that —
A machine out-played the champion.
Deep Blue beat the reigning grandmaster at chess. Brute search proved a machine could win a game we thought needed a soul.
Because, before that —
The network learned how to correct itself.
Backpropagation spread the blame for every mistake backward through the layers. Learning stopped being magic and became arithmetic.
Because, before that —
A book nearly ended it.
A famous proof showed a single perceptron could not solve simple problems. Funding froze. The first long silence began.
Because, before that —
A summer workshop gave it a name.
At Dartmouth, a handful of researchers coined the phrase “artificial intelligence.” The word arrived before the thing did.
Because, before that —
One man asked, “can machines think?”
Turing replaced the unanswerable question with a game of imitation. Every model since has been, in some way, an attempt to pass it.
The beginning