The Thinking Machine

Cold Open 0:00

Vienna, 1770. In a candlelit room at Schönbrunn Palace, the imperial court gathers around a cabinet.

Behind it sits a life-sized figure in Ottoman robes.

Its inventor opens the doors. Gears. Cylinders. Clockwork. He closes them.

The figure turns its head, lifts its arm... and moves a chess piece.

Two hundred and forty-six years later, in Seoul, another machine faces one of the greatest Go players alive: Lee Sedol.

On move thirty-seven, the machine plays a stone so strange one commentator thinks it's a mistake.

Lee Sedol takes nearly fifteen minutes to answer.

Two rooms. Two boards.

One question: can a machine think?

To get from that room to this one, we have to go back further, to a man certain it could not.

I. Is Thinking Just Counting? 1:26

In 1637, René Descartes imagined a machine built to look and move like us.

It might speak words, he wrote, even cry out if you touched it.

But it could never "arrange them variously so as appositely to reply to what is said in its presence".

Speech was the proof of a mind. Remember that test.

Not everyone agreed.

In Paris, an English exile, Thomas Hobbes, attacked Descartes's ideas in print, and Descartes answered him curtly.

Hobbes would later write: "Reason, in this sense, is nothing but Reckoning".

If thinking is arithmetic, a machine could do it.

In September 1647, Descartes called on a frail young man in Paris: Blaise Pascal, who at eighteen had begun building a calculating machine.

They argued for two days, about whether empty space could exist.

Descartes later wrote that Pascal had too much vacuum in his head.

Pascal's own verdict: "The arithmetical machine produces effects which approach nearer to thought than all the actions of animals. But it does nothing which would enable us to attribute will to it".

No will.

Then Gottfried Leibniz built a machine that could multiply,

and dreamed of settling every argument by saying: "Let us calculate."

He even wrote numbers with just two digits: one and zero.

No one could settle it. So instead... they built illusions.

II. Can a Machine Fool Us? 3:38

In Paris, crowds flocked to Vaucanson's mechanical duck.

It flapped, it ate, and then, to everyone's delight, it seemed to digest.

A century later, by a magician's account, the secret was pre-loaded breadcrumbs, dyed green.

In Switzerland, the Jaquet-Droz family built a boy who writes whatever message you program into him.

He still writes.

And the Turk? It toured, on and off, for more than fifty years, and reportedly beat Napoleon.

In 1836, Edgar Allan Poe watched it play.

A true machine, he argued, like Mr. Babbage's, follows fixed steps. Chess does not.

His verdict: "It is quite certain that the operations of the Automaton are regulated by mind, and by nothing else."

He was right, and he wasn't the first to suspect it.

Inside, by candlelight, sat a human chess master. The thinking machine was a man in a box.

In 1854, the Turk burned in a Philadelphia museum.

One witness claimed that through the flames he heard it cry out: check... check.

The Turk had been a lie.

But the machine Poe mentioned was real, and its inventor dreamed far bigger.

III. Can a Machine Create? 5:35

As a boy, Charles Babbage was enchanted by a silver dancing automaton.

"Her eyes were full of imagination, and irresistible."

In Lyon, a loom ran on punched cards: change the cards, change the pattern.

If cards could weave flowers, why not numbers?

Babbage designed an engine you could program. No one would pay to build it.

But one person saw what it could become: Ada Lovelace, the daughter Lord Byron left behind, raised by her mother on mathematics.

She wrote that "the Analytical Engine weaves algebraical patterns just as the Jacquard-loom weaves flowers and leaves".

And that it "might compose elaborate and scientific pieces of music".

Her notes held what many call the first computer program.

Then she drew a line herself: "The Analytical Engine has no pretensions whatever to originate anything."

In 1816, the summer Ada's father fled England, he rented a villa on Lake Geneva.

There, Mary Shelley began a story about a man who builds a creature he cannot control.

Ada Lovelace died in 1852, at thirty-six, the same age as her father.

Babbage died nineteen years later, his engine still on paper.

Yet the dream outlived them. A shoemaker's son, George Boole, had already written logic as algebra.

Its true and false would become Leibniz's one and zero.

Then, in 1921, a Czech play gave the world the word robot, from the Czech for forced labor.

And the robots rise up.

IV. Can Machines Think? 7:55

The world had a word for its nightmare, but no theory.

In 1936, Alan Turing imagined a machine that could imitate any other.

After the war, Turing began a paper: "I propose to consider the question, 'Can machines think?'"

His answer was Descartes's test, turned into a game: if you can't tell a machine from a person in conversation, why say it can't think?

He even answered Ada Lovelace, by name, under the heading "Lady Lovelace's Objection".

His reply: "Machines take me by surprise with great frequency."

In 1952, he ran a chess program by hand, playing the machine himself.

The Turk had hidden a man in a machine. Turing became the machine.

That same year, the country he had helped save prosecuted him for being gay, and made him undergo hormone treatment.

In 1954, he was found dead. He was forty-one.

Two years after his death, the field got its name at Dartmouth: artificial intelligence.

Its founders hoped for real progress in one summer.

Then a young psychologist, Frank Rosenblatt, unveiled a learning machine modeled on the brain.

The New York Times said it was expected to "walk, talk, see, write, reproduce itself and be conscious of its existence".

His most famous critic, Marvin Minsky, had gone to the same Bronx high school, and thought it a dead end.

At MIT, Joseph Weizenbaum's ELIZA played therapist.

His secretary, he wrote, asked him to leave the room so she could talk to it.

It shocked him.

Then Minsky and Papert published a book proving what simple perceptrons could never learn. The money moved on.

Two years later, Rosenblatt died in a boating accident. Winter had come.

V. What If It Learns? 10:50

And the winter deepened.

A British government report found AI had not produced "the major impact that was then promised".

Funding froze on both sides of the Atlantic.

Through the cold, a few kept Rosenblatt's idea alive, and showed how artificial neurons could learn from their mistakes.

But the money chased expert systems, Leibniz's dream as a product, until the market for their special computers collapsed in a single year.

A second winter fell. For years, many researchers avoided calling their work AI.

Still, rules could play chess. Garry Kasparov beat IBM's Deep Blue.

In the rematch, the machine played a quiet move that looked too human.

Kasparov's first instinct was Europe's in 1770: someone must be inside.

He accused IBM of cheating and lost the match. No one was inside.

Deep Blue didn't learn; it searched, by rules people gave it.

Learning needed what Rosenblatt never had. The internet brought data.

Video games brought power: graphics chips doing many sums at once.

On two such chips, in a bedroom, two of Geoffrey Hinton's students trained a network that crushed the field on Fei-Fei Li's ImageNet.

Hobbes had said reason is nothing but reckoning. No one wrote this machine's rules. It learned them.

And so, back to Seoul. Game two. Move thirty-seven. This time, we see what they saw.

European champion Fan Hui: "It's not a human move. I've never seen a human play this move. So beautiful."

Lee Sedol won one game, with a move commentators called divine.

For once, a human had surprised the machine.

Three years on, he retired: "Even if I become the number one, there is an entity that cannot be defeated."

One frontier remained: Descartes's.

A year after Seoul, a Google paper, "Attention Is All You Need", introduced the transformer, a network that learned which words matter to which.

Five years later, ChatGPT took the world by storm: a reported million users in five days, an estimated hundred million in two months.

A machine could finally arrange words to reply to whatever was said to it.

Three hundred and eighty-five years after Descartes said it never could.

Epilogue: So, Can a Machine Think? 14:31

Today, AI no longer just generates words and images. It acts.

Agents plan, browse, write and run code, and carry out some tasks on their own.

Two centuries after Lake Geneva, Mary Shelley's fear has a new shape: machines that act.

Some of AI's own builders have warned of the risks.

Now AI is stepping off the screen. Descartes imagined a machine built to look and move like us.

Robots shaped like us are learning to move.

They need what every child learns: how the world works.

Enter world models: AI that learns to predict what happens next.

The machines count, converse, surprise, and begin to move.

But open the cabinet today, and the oldest question is still inside. Whether anyone is in there.

Whether, in Pascal's word, it has a will.