Article 1 1943-1956

The Idea of a Thinking Machine

From the first mathematical neuron to the workshop that named artificial intelligence.

Choose how you'd like to follow along

Sometime in the late 1930s, a teenage boy was sleeping outdoors in Chicago. He had run away from home in Detroit a few years earlier, around the age of fifteen, and he had been more or less on his own ever since. He had no money. He had no high school diploma. What he had was a head full of logic.

His name was Walter Pitts. And here is the strange thing about Walter Pitts: he may have been one of the most original minds of the twentieth century, and almost nobody has ever heard of him.

The story people tell about his childhood is almost too good to be true. As a boy, hiding in a library to escape some bullies, he is said to have pulled down a copy of Principia Mathematica - the thousand-page monument in which Bertrand Russell and Alfred North Whitehead tried to rebuild all of mathematics out of pure logic - and read the whole thing. He thought he had found errors. So he wrote to Russell. And Russell, the most famous philosopher alive, wrote back.

Hold that picture for a moment. A runaway kid, no money, no shoes to speak of, correcting Bertrand Russell. Because that kid is where this story begins. Not with a computer. Not with a robot. With a question that a homeless teenager could not stop turning over in his mind: what if thinking is just logic? What if the thing happening inside your head, right now, is underneath it all a kind of calculation - and if it is a calculation, then why couldn’t something other than a brain do it too?

That is the question this whole story is about. Can a machine think? It plays out over thirteen years, from 1943 to 1956. There are almost no computers in it, no screens, nothing you would recognize as artificial intelligence. What there is instead is a small, eccentric cast of people who looked at the human mind and saw a machine - and then set out to build one.

Pitts eventually found his way to a man named Warren McCulloch. McCulloch was a neurophysiologist, a doctor of the brain, and by temperament the opposite of the shy young logician - the kind of man who stayed up late drinking whiskey and arguing about the deepest questions he could find. And he was chewing on a question that fit Pitts perfectly. How does the brain - three pounds of wet tissue - produce something as clean and exact as logic? McCulloch had the biology. Pitts had the mathematics. And in 1943, the older doctor and the young runaway published one of the strangest, most important papers you have never read.

Its title was forbidding: “A Logical Calculus of the Ideas Immanent in Nervous Activity.” But the idea inside it was simple enough to change everything. Treat a single neuron, they said, as a tiny switch - it either fires or it does not, on or off, yes or no. Now wire a bunch of these switches together. What they proved is that a network of these all-or-nothing switches can compute any logical statement you can write down. In other words: the brain could be described as a kind of logic machine.

Sit with how audacious that is. They had taken the most mysterious object in the universe - the human mind - and suggested it might be, at bottom, an engineering problem. That single move is the seed that everything else grows from. Every artificial neural network that exists today, including the ones writing essays and answering emails right now, traces back to that 1943 paper by a brain doctor and a kid with no degree.

And here is the thing about that idea: it was in the air. Because at almost the exact same moment, in the exact same years, several other people were circling the very same territory - and they had no idea they were about to start a revolution.

One of them was Norbert Wiener. Wiener was the reverse of Walter Pitts in every way - a certified prodigy who had finished a PhD at Harvard by eighteen. And Wiener had noticed something. Take a torpedo that corrects its own course toward a moving ship. Take an animal reaching for food. Take your own hand picking up a glass of water. All of these look purposeful, even alive. But Wiener argued you could explain that purpose mechanically, with a simple loop: act, measure the error, correct, repeat. Purpose, he suggested, was not a mystery of the soul. It was feedback. In 1948 he gave this whole way of thinking a name - cybernetics - and a generation borrowed his vocabulary of control and feedback and information to start imagining machines that behaved as if they wanted things. People even built them: a contraption that rewired itself to stay balanced after you disturbed it; little robot tortoises that crept toward light and trundled back to their chargers when they ran low on power. Crude, twitching, weirdly alive.

Then there was the most charming genius of the bunch, a man named Claude Shannon. Shannon was famous at Bell Labs for riding a unicycle down the hallways, sometimes while juggling. He also, almost as a side project, invented the modern digital world. In 1948 he worked out the mathematics of information itself - he gave us the “bit,” the idea that any message at all can be reduced to a string of yes-or-no choices. But Shannon couldn’t leave it as theory. He built a mechanical mouse named Theseus that could learn its way through a maze and remember the path - one of the first machines anyone had built that visibly learned from experience. And he wrote the very first serious plan for how a computer might play chess. Games, it turned out, would become the proving ground for machine intelligence for the next seventy years, and Shannon drew the first map.

Underneath all of it, the machines themselves were finally arriving. A mathematician named John von Neumann had laid out the blueprint for the modern computer - one memory holding both the data and the instructions that work on it - and by 1948 the first true computers were humming to life. For the first time, the dreamers had somewhere to put their dreams.

And then, in 1950, the question got its sharpest voice. A British mathematician named Alan Turing wrote a paper that opened with four blunt words: “Can machines think?” And then he did something clever - he refused to answer. Those words, he said, are too slippery to argue about. So replace the question with a game. Put a person at a keyboard. On the other side are two hidden partners, one human, one machine, and the judge has to figure out which is which just by chatting with them. If the machine can fool the judge - if you genuinely cannot tell - then, Turing said, stop arguing about whether it “really” thinks, and just admit it is doing something we have no better word for. We call it the Turing test now. Seventy-five years later, we are still arguing about who has passed it.

So look at what is happening here. Learning - which had always seemed like the most human thing in the world - was starting to look like something you could build. In 1949 a psychologist named Donald Hebb gave it a rule: when one neuron keeps helping another one fire, the connection between them grows stronger. Neurons that fire together, wire together. Two years later a young graduate student named Marvin Minsky took that idea and built it out of vacuum tubes and surplus parts - a machine of about forty artificial neurons that learned to run a maze. The brain, rendered in wires.

By the mid-1950s, all these threads - the logical neuron, the feedback machines, information, learning, the first computers - were lying around, and nobody had tied them together or even given them a name. In the summer of 1956, a small group fixed that. The year before, four researchers had written a proposal for a two-month workshop at Dartmouth College, and to describe what they wanted to study, one of them, John McCarthy, needed a phrase. He coined one: artificial intelligence. The proposal made a bet so bold it still hangs over the field today - that every feature of intelligence could, in principle, be described so precisely that a machine could be built to imitate it. That same summer, two researchers named Allen Newell and Herbert Simon unveiled a program called the Logic Theorist that proved mathematical theorems on its own - and for at least one of them, found a proof more elegant than the one in the textbook. The first artificial intelligence program. A machine that did not just calculate. It reasoned.

So that is the founding. In thirteen years, a question a homeless teenager couldn’t shake had turned into a science with a name, a method, and its first working programs. The people who did it were almost giddy with confidence. And mostly, they had earned it.

But they had also set a trap for themselves. Because in 1958, a psychologist named Frank Rosenblatt would unveil a machine called the perceptron - a network that learned to recognize patterns straight from examples - and the newspapers would announce that machines were about to walk, talk, and reproduce themselves. That promise, and the long, cold winter that came when it didn’t arrive, is where the next chapter begins.

As for Walter Pitts - the runaway who started it all - the story does not end happily. He grew disillusioned, withdrew, and at one point burned a great deal of his unpublished work. He died in 1969, largely forgotten, his name missing from the textbooks of the field he helped invent. But the idea he and McCulloch lit in 1943 never went out. It is still burning. You are living inside it.

Listen to the article

The full article, read aloud. Runtime 10:21.

The video episode is coming soon

We're producing narrated video editions on YouTube - the same story, walking the timeline with visuals. In the meantime, read the article or listen to it.

Sources and show notes

Every claim in this article is drawn from the AI Library, where each entry links to its own primary source - the original paper, the official record, the actual document.

A Logical Calculus of the Ideas Immanent in Nervous Activity McCulloch and Pitts, 1943 - the paper that turned the neuron into logic. The first mathematical neuron model The first mathematical model of a neuron. Warren McCulloch Neurophysiologist who modeled the neuron as a logical switch. Walter Pitts Self-taught logician, co-author of the 1943 paper. Behavior, Purpose and Teleology Rosenblueth, Wiener, and Bigelow on feedback and purpose. Wiener publishes Cybernetics Wiener's 1948 book that named the science of control and communication. Norbert Wiener Founder of cybernetics. Cybernetics The study of control and feedback in machines and living things. Ashby's Homeostat, a machine that adapts to stay stable Ashby's machine that rewired itself to stay stable. Grey Walter builds the first electronic autonomous robots Grey Walter's light-seeking robot tortoises. Shannon's A Mathematical Theory of Communication Shannon's information theory and the definition of the bit. Turing's Intelligent Machinery report imagines trainable networks Turing's report imagining trainable unorganised machines. Neurons that fire together, wire together Hebb's learning rule: neurons that fire together wire together. Donald Hebb Psychologist who gave neural networks a theory of learning. Can machines think? Turing reframes the question Turing's imitation game - now the Turing test. Computing Machinery and Intelligence Turing's 1950 paper, Computing Machinery and Intelligence. The Turing Test The test that replaced 'can machines think?' with a behavioral test. Alan Turing Mathematician who reframed the question of machine thought. Shannon's Theseus, a maze-solving mechanical mouse Shannon's maze-learning mechanical mouse, Theseus. Shannon's blueprint for computer chess Shannon's blueprint for computer chess: evaluation and minimax. Claude Shannon Founder of information theory and computer-chess theory. Minsky and Edmonds build the SNARC, an early neural-net machine Minsky and Edmonds build the SNARC, an early neural-net machine. Marvin Minsky Built the SNARC; co-organizer of the Dartmouth workshop. The Georgetown-IBM machine translation demonstration The IBM 701 translates Russian to English in public. John von Neumann Architect of the stored-program computer. The workshop that named artificial intelligence The 1955 proposal that coined 'artificial intelligence.' John McCarthy Coined the term artificial intelligence. Logic Theorist, the first artificial intelligence program Newell, Simon, and Shaw's Logic Theorist - the first AI program. Allen Newell Co-builder of the Logic Theorist. Herbert Simon Co-builder of the Logic Theorist; later a Nobel laureate. The MIT symposium often called the birthday of cognitive science The MIT symposium often called the birth of cognitive science. Symbolic AI (GOFAI) Intelligence as the manipulation of symbols - the founding paradigm. The perceptron: a machine that learns from examples Rosenblatt's perceptron - the cliffhanger into Episode 2. Frank Rosenblatt Built the perceptron, the first network that learned from examples. Perceptron The first trainable artificial neuron.

Made with AI, sourced like a library

We are open about this: this article is produced with AI. The research, the writing, the narration - AI does the heavy lifting, with human review before anything is published. That is not a footnote; it is the point. CenterConsulting builds software with AI every day, and this series is the same method applied to storytelling.

What keeps it honest is the foundation: every article is built on the AI Library, our knowledge base where every fact links back to a primary source. The show notes above list this article's entries, so you can check everything yourself.