Where did the idea of artificial intelligence come from?

People were imagining artificial minds thousands of years before anyone wired a circuit. Ancient stories gave us crafted servants and statues that came to life. Skilled builders made moving figures that seemed alive. Philosophers asked whether thinking itself might follow rules, like arithmetic. When the first computers appeared, they landed in a world that already had a long list of hopes and fears about machines that think.

Knowing that history is useful right now. Last month OpenAI showed GPT-2, a program that writes surprisingly fluent paragraphs, and the headlines quickly reached for very old images: the obedient servant, the creature that escapes its maker. Those images come from somewhere, and seeing where helps you judge today's news with a clearer head.

Myths and legends: minds made by hand

Greek mythology already had artificial helpers. Hephaestus, the smith of the gods, was said to have made golden servants, and the sculptor Pygmalion fell in love with a statue, Galatea, that came to life. Medieval stories added darker versions: alchemists who tried to grow a tiny artificial person (the homunculus attributed to Paracelsus) and the Golem, a figure brought to life by Rabbi Judah Loew.

Fiction kept the theme going. Mary Shelley's Frankenstein asks what a maker owes to the being he creates. Karel Čapek's play R.U.R. (Rossum's Universal Robots) put factory-made artificial workers on the stage.

Notice what these stories share: the artificial being is either a perfect servant or a threat. That pair of expectations still shapes how people react to every new AI result.

Automatons: machines that looked alive

Builders also made real devices. Records describe humanlike mechanical figures by Yan Shi in ancient China and by Hero of Alexandria in the Greek world. In 1206 the engineer al-Jazari described automatons whose behavior could be changed, an early hint of programming.

These machines did not think. They followed gears, water, weights and cams. Yet they taught an important lesson: something can look intelligent while running a fixed routine. Keep that in mind; it comes back in every chapter of AI history.

Philosophers: is reasoning a kind of arithmetic?

A different line of thought asked what thinking is made of. Ramon Llull (1232–1315), a philosopher from Majorca, designed logical machines meant to combine basic truths mechanically. Gottfried Leibniz, centuries later, dreamed of a universal language in which arguments could be settled by calculation. Thomas Hobbes put it bluntly in Leviathan (1651): reason is "nothing but reckoning."

Here is that idea in a small table, using a classic example:

If thinking could be broken into steps like these, a machine might perform them. That hope drove much of the AI research of the 20th century.

Lovelace and Turing: from dream to test

In 1843 Ada Lovelace translated a description, by the engineer Luigi Menabrea of Turin, of Charles Babbage's Analytical Engine, a programmable calculating machine that was designed but never built, and added long notes of her own. She saw that it worked like the Jacquard loom, which wove patterns from punched cards, and guessed that it could handle symbols other than numbers, even musical notes. She also drew a firm line: in her words, "The Analytical Engine has no pretensions whatever to originate anything." It could only do what people knew how to tell it.

In 1950 the British mathematician Alan Turing answered her directly. His paper "Computing Machinery and Intelligence" opens with the question "Can machines think?", swaps it for a practical game of typed questions (now called the Turing test), and lists objections, including one he named after Lady Lovelace. He also suggested building a "child machine" and teaching it, instead of programming a finished adult mind. In 1956, a summer workshop at Dartmouth College gave the field its name.

What do ancient myths and early machines teach us about AI today?

First, they show that expectations come before the technology. When a system like GPT-2 writes a convincing paragraph, people tend to sort it into the old boxes of servant or monster. A calmer question is more useful: what does this system actually do, and what does it fail at?

Second, the automatons warn us that appearance is not understanding. A clockwork figure can seem alive; a program can seem to understand. Researchers still debate how much of today's language software understands and how much is very good pattern-following, and the honest answer is that it depends on the task.

Third, Lovelace's objection is still worth asking. Modern systems learn patterns from large amounts of data instead of following hand-written steps, so they can produce things nobody explicitly told them to write. Whether that counts as originating anything is a fair question to keep open, and a good one to bring to every AI demo you see.