The Dream

Transcript

Dartmouth, summer 1956

In the summer of 1956, a handful of men gathered on the top floor of the mathematics building at Dartmouth College, in the small town of Hanover, New Hampshire. They had asked for money to fund ten researchers for two months. They believed that might be enough to make real progress on a rather large problem: how to make a machine think.

It was not enough. It still isn’t. But that summer gave the problem a name, and the name stuck. Artificial intelligence.

This is The History of AI. A short series about a very old idea, told one chapter at a time. It is written and produced by Walma, and narrated by a synthetic voice. Which, as you will hear, is rather the point. Episode one: The Dream.

Bronze giants and paper wheels

The dream is much older than the computer.

The ancient Greeks told stories of Talos, a giant made of bronze, built by the god Hephaestus to guard the island of Crete. Three times a day he walked its shores, hurling boulders at ships that came too close. Talos was not born. He was made. And for the Greeks, that was the unsettling part.

For centuries the dream stayed in stories. Then, slowly, it moved into mathematics.

In the thirteenth century, on the island of Majorca, a philosopher named Ramon Llull built a device from paper discs. Each disc carried a set of concepts. Turn the discs, and the concepts combined into new statements. Llull believed that reasoning itself could be done mechanically, by combining symbols according to rules.

Almost four hundred years later, the German philosopher Gottfried Wilhelm Leibniz read Llull, and took the idea further. Leibniz imagined a universal language of thought, so precise that arguments could be settled the way sums are settled. When two philosophers disagreed, he wrote, they would simply pick up their pens, sit down, and say to each other: let us calculate.

Leibniz also built a machine that could multiply and divide. It jammed often. But the idea was now out in the world: that thought might be a kind of calculation. And if it was, then a machine could do it.

Ada Lovelace draws a line

In 1843, in London, a twenty-seven-year-old mathematician named Ada Lovelace published a set of notes on a machine that did not exist.

The machine was the Analytical Engine, designed by her friend Charles Babbage. It was to be built of brass gears and driven by steam, and programmed with punched cards borrowed from the weaving looms of the day. Babbage never finished it. But on paper, it was a general-purpose computer, a full century early.

Lovelace saw something in it that Babbage himself had not dwelt on. The engine, she wrote, did not have to work only with numbers. Numbers could stand for other things. Musical notes, for instance. Such a machine, she suggested, might one day compose elaborate pieces of music.

And then, in the same notes, she drew a line. The Analytical Engine, she wrote, has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform.

A machine can follow. It cannot create. That single objection would hang over the field for the next hundred and eighty years. It hangs there still.

Alan Turing and the imitation game

The person who answered Lovelace most directly was a young English mathematician named Alan Turing.

In 1936, at the age of twenty-three, Turing described an imaginary device: a machine that read and wrote symbols on an endless strip of tape, one square at a time, following a simple table of rules. Then he proved something remarkable. One such machine, given the right instructions, could imitate any other. A single universal machine could carry out any computation at all.

That is the idea behind every computer built since.

During the Second World War, Turing worked at Bletchley Park, helping to break German ciphers. After it, he returned to the question that interested him most. In 1950, he published a paper that opens with one of the most famous sentences in science. I propose to consider the question: can machines think?

Turing thought the question, put that way, was too vague to answer. So he replaced it with a game. A judge exchanges written messages with two hidden players: one human, one machine. If the judge cannot reliably tell which is which, Turing argued, we have no good reason to deny that the machine is thinking.

He called it the imitation game. We call it the Turing test.

He took up Lovelace’s objection by name, and offered a dry reply: machines take me by surprise with great frequency. And he made a prediction. By the end of the century, he wrote, one would be able to speak of machines thinking without expecting to be contradicted.

Turing did not live to see it tested. He died in 1954, at the age of forty-one.

Neurons as logic

While Turing was thinking about machines, others were thinking about brains.

In 1943, in Chicago, two men published a strange and difficult paper. Warren McCulloch was a neurophysiologist in his forties. Walter Pitts was a self-taught logician, barely twenty, who had run away from home as a teenager. Together they proposed that a nerve cell could be treated as a simple logical switch. It either fires, or it doesn’t. Connect enough of these switches together, they argued, and the network can compute anything that logic can express.

It was a crude picture of the brain. But it was the first mathematical model of a neural network. Nearly every system that we call AI today descends, by a long and winding road, from that paper.

The Dartmouth workshop

And so we come back to that summer in New Hampshire.

By the mid 1950s, the first electronic computers existed. They filled rooms, and they were used mainly for arithmetic. But a few young researchers suspected they could do far more. One of them was John McCarthy, a mathematics professor at Dartmouth, twenty-eight years old.

McCarthy wanted to bring these people together. He needed funding, and to ask for funding, he needed a name for the subject. In a proposal dated August 1955, written with Marvin Minsky, Nathaniel Rochester and Claude Shannon, he chose one: artificial intelligence.

The proposal rested on a single bold conjecture. That every aspect of learning, or any other feature of intelligence, can in principle be so precisely described that a machine can be made to simulate it.

Leibniz would have recognised it at once.

The workshop itself was a looser affair than planned. People came and went. The funding was about half of what they had asked for. There was no great breakthrough, and McCarthy later admitted that the summer fell short of his hopes.

But two of the visitors, Allen Newell and Herbert Simon, arrived with something to show. They had written a program called the Logic Theorist. It could prove theorems in mathematical logic, and in one case it found a proof more elegant than the one published by Russell and Whitehead. Simon had already told his students, a few months earlier, that over Christmas he and Newell had invented a thinking machine.

Lovelace had said a machine could never originate anything. Here, perhaps, was the first small piece of evidence against her.

What was promised, and what arrived

The men who left Dartmouth that summer were confident. Within a generation, some of them would say, machines would match the human mind. Herbert Simon predicted that within ten years, a computer would be the world’s chess champion.

It took forty.

That gap, between what was promised and what arrived, is the real story of artificial intelligence.

Next time: the golden years, the first machines that seemed to talk, and the long winter that followed.

The History of AI is produced by Walma. We help companies in Europe put AI to work, safely. You will find us at walma.ai. Thank you for listening.

Narrated by a synthetic voice. Voice and music generated with ElevenLabs.