The History of AI, Part 1 of 24. A long, story-driven walk through every major era of artificial intelligence, from ancient myths to modern agents. One new chapter every few days. Follow the series so you don't miss the next one.

Every few months someone announces that AI has "suddenly" arrived. A chatbot passes an exam, a model paints a picture, a program beats a champion. It feels like a bolt from the blue.

It isn't. The machines are new, but the dream behind them is thousands of years old: the dream that thinking can be captured, written down, and run by something that is not a person. If you want to understand why today's AI looks the way it does, why it keeps getting hyped, why it keeps disappointing us, and why it keeps coming back, you have to begin long before electricity.

This is the first of 24 articles that will take us from bronze giants and clockwork ducks all the way to modern AI agents. Each one stands on its own, but together they tell a single story. I'll keep the math light, the code short, and the people front and center, because the history of AI is first of all a history of strange, brilliant, stubborn humans.

Grab a coffee. This first chapter is about the world before computers.

What do we even mean by "AI"?

Before we dig in, let's pin down a slippery word. "Artificial intelligence" has no single agreed definition, and the definition has shifted every decade. Historians of the field often point to four different ambitions that get lumped together:

  1. Systems that think like humans. Model how minds actually work.
  2. Systems that act like humans. Be indistinguishable from a person in conversation or behavior.
  3. Systems that think rationally. Follow the laws of logic and reasoning.
  4. Systems that act rationally. Do the right thing to achieve a goal, whatever is happening inside.

These four goals pull in different directions, and the whole history of AI can be read as researchers swinging between them. Early thinkers were obsessed with number 3: if we could just write down the rules of correct reasoning, a machine could follow them. Later, number 4 took over, and today we mostly build systems that are judged by whether they work, not by whether they resemble a human mind.

Keep these four in your head. They will keep returning.

Part 1: The ancient dream of artificial beings

Bronze giants and golden servants

The oldest AI stories aren't about computers. They're about craftsmen who make things that move and think.

In Homer's Iliad, composed roughly 2,700 years ago, the god Hephaestus, the divine smith, is described walking with the help of golden maidens he forged himself. They are described as having intelligence in their hearts, voice, strength, and skills learned from the gods. Think about that: an eighth-century BCE poet imagining artificial servants with minds, built by a master engineer.

Then there is Talos, the bronze giant said to guard the island of Crete by hurling boulders at approaching ships and circling the island three times a day. In the version told in the Argonautica (a Greek epic from the third century BCE), Talos has a single vein of ichor, the blood of the gods, sealed at the ankle by a nail. The sorceress Medea defeats him by tricking him into removing it. If you squint, it's a story about a machine with a single critical failure point.

The Greeks also gave us Pygmalion, the sculptor who carved a statue so perfect that he fell in love with it, after which the goddess Aphrodite brought it to life. This is a different kind of AI myth: not the tireless worker but the created companion. It would echo for centuries in stories, plays, and eventually films.

Golems, clay, and the dangers of creation

Other traditions told similar stories with sharper warnings.

In Jewish folklore, the golem is a being shaped from clay and animated by sacred words, often a name of God inscribed on its forehead or placed in its mouth. The most famous tale, attached to sixteenth-century Prague and the rabbi Judah Loew, tells of a golem built to protect the Jewish community. It grows too powerful and has to be deactivated by erasing a letter from the word on its forehead, changing "truth" (emet) into "death" (met).

Even this old story contains something that sounds uncannily modern: a servant that follows instructions literally, lacks judgment, and becomes dangerous precisely because it does what it was told. AI safety researchers today would recognize the plot.

In the Chinese text attributed to the philosopher Liezi, there is a tale of the artisan Yan Shi, who presents King Mu of Zhou with a mechanical man that can sing, dance, and even flirt with the court ladies. The king is so annoyed by the figure's wandering eye that he threatens to execute the craftsman, and Yan Shi saves himself by taking the figure apart to prove it is made of leather, wood, glue, and lacquer. It's a gorgeous little parable about the unease people feel when a made thing seems too alive.

Real machines in the ancient world

Myths are one thing. Real engineering is another, and the ancients did a surprising amount of it.

Hero (Heron) of Alexandria, working in the first century CE, wrote about steam-powered devices, coin-operated dispensers, and automatic theater scenes that played out using ropes, pulleys, and falling weights, essentially programs encoded in physical string. Greek and Hellenistic engineers built water clocks and moving figures.

Perhaps the most astonishing is the Antikythera mechanism, a hand-cranked bronze device recovered from a shipwreck, dated to roughly the second or first century BCE. It used a train of gears to model the motions of the sun, moon, and calendars, and to predict eclipses. It is a mechanical computer in the loosest, truest sense: a physical mechanism whose internal state represents something about the world and can be used to calculate.

Fragments of the Antikythera mechanism, a Greek geared device from about 150–100 BC (National Archaeological Museum, Athens)
Fragments of the Antikythera mechanism, a Greek geared device from about 150–100 BC (National Archaeological Museum, Athens). Image: Zde, CC BY-SA 4.0, via Wikimedia Commons.

In the medieval Islamic world, the engineer Ismail al-Jazari wrote The Book of Knowledge of Ingenious Mechanical Devices in 1206. It described mechanical clocks, water-raising machines, and, most famously, a programmable automaton band: a boat carrying mechanical musicians whose performance could be altered by moving pegs on a rotating drum. Changing the pegs changed the music. That is a program, stored as physical pegs, controlling a machine's behavior. It is also the same basic idea we'll meet again with the Jacquard loom and, eventually, punch cards.

None of these devices was "intelligent." But they established something crucial: behavior can be stored in a physical arrangement of parts and replayed.

Part 2: The dream of mechanical reasoning

Moving statues are one half of the AI story. The other half is about thought itself, and whether it could be formalized.

Aristotle and the first rules of reasoning

Around 350 BCE, Aristotle wrote down the first systematic theory of logical reasoning. His syllogisms are patterns of argument where the conclusion follows necessarily from the premises:

  • All humans are mortal.
  • Socrates is a human.
  • Therefore, Socrates is mortal.

The insight isn't the particular claim about Socrates. It's that the argument is valid because of its form, no matter what the words mean. Replace "humans" with "robots" and "mortal" with "replaceable" and the logic still holds. If validity depends only on form, then in principle a mechanism that manipulates forms could check or even produce valid arguments, without understanding any of them.

That is the seed of symbolic AI, the approach that dominated the first decades of the field, and we'll spend several articles on it. Aristotle didn't plant a tree, but he planted the seed.

The Stoic philosophers later developed a logic of whole propositions ("if it is day, then it is light; it is day; therefore it is light"), which is closer to what modern computing uses.

Ramon Llull's rotating wheels

Now we jump to the thirteenth century, to one of the strangest characters in this story: Ramon Llull, a Majorcan philosopher and mystic who believed he could prove Christian truths to anyone using pure logic.

His tool was the Ars Magna ("Great Art"), a system built from concentric paper wheels. Each wheel carried letters standing for fundamental concepts such as goodness, greatness, eternity, and power. By rotating the wheels you generated every possible combination of concepts, and each combination represented a candidate statement. In effect, it was a mechanical combinatorial generator of ideas.

Was it useful? Mostly not. Was it a profound idea? Absolutely. Llull proposed that new knowledge can be generated by systematically combining symbols according to rules, without any inspiration. A tiny version of this idea is what happens when a language model strings together tokens.

Jonathan Swift would later mock this very idea in Gulliver's Travels (1726), describing a professor on the island of Laputa who built a huge frame of wooden cubes covered in words that students turned with cranks to generate books and sciences "without the least assistance from genius or study." Swift meant it as satire. It reads today like an uncomfortably accurate preview of automatic text generation.

Hobbes: "Reasoning is nothing but reckoning"

By the seventeenth century, philosophers had started to treat thought as a kind of calculation. In Leviathan (1651), Thomas Hobbes wrote that reasoning is "nothing but reckoning," meaning adding and subtracting. Whether or not he intended it literally, the sentence became a rallying cry for everyone who thinks that thinking is computation.

Descartes draws a line

René Descartes made things more complicated. In the Discourse on Method (1637), he argued that the body, including the bodies of animals, is a machine, a kind of intricate clockwork. But he insisted that the human mind was different.

Intriguingly, Descartes also proposed what amounts to an early thought experiment about distinguishing machines from people. He argued that even if a machine looked like a human and could imitate our actions, we could tell the difference using two tests. First, a machine could never use words or signs flexibly enough to respond appropriately to anything said in its presence; at best it could produce canned responses. Second, a machine could never act from general understanding the way a person does, only from the particular arrangement of its parts.

That is, more or less, a 1637 version of "does it generalize?" and "can it hold a real conversation?" Three centuries later, Alan Turing would turn the first test into a formal game. Four centuries later, we're arguing about it again with chatbots.

Pascal, Leibniz, and the first calculators

While philosophers argued, engineers started actually building calculating machines.

In 1642, the nineteen-year-old Blaise Pascal built the Pascaline, a gear-driven device that could add and subtract. He made it to help his father, a tax official, who spent his days adding up long columns of figures.

Then came Gottfried Wilhelm Leibniz, perhaps the most important figure in this chapter. Around 1673 he built the Stepped Reckoner, a machine that could add, subtract, multiply, and divide. It was temperamental, but it showed that mechanical arithmetic wasn't limited to addition.

But Leibniz's deeper contribution was conceptual. He dreamed of two things:

  • A characteristica universalis: a universal formal language in which all human knowledge could be expressed precisely.
  • A calculus ratiocinator: a logical calculus, a way of computing with those symbols, so that arguments could be settled by calculation.

He imagined that when two philosophers disagreed, they would no longer argue. They would sit down and say, "Let us calculate."

Leibniz also wrote up the binary number system in 1703, noticing that everything could be expressed with only 0s and 1s. He even saw a mystical beauty in it. He had no idea that two and a half centuries later, binary would become the language of every digital machine on Earth.

Leibniz never built his reasoning machine, but he defined the program. Much of the history of AI is people trying, with varying success, to build what Leibniz imagined.

Part 3: Clever frauds and honest machines

The digesting duck and the chess-playing Turk

Joseph Racknitz's cutaway drawing of the Mechanical Turk, showing how a hidden operator might have sat inside
Joseph Racknitz's cutaway drawing of the Mechanical Turk, showing how a hidden operator might have sat inside. Image: Nhthanhtruc103, CC0, via Wikimedia Commons.

The eighteenth century was the golden age of the automaton: intricate clockwork figures designed to amaze. In 1739, the French inventor Jacques de Vaucanson unveiled a mechanical duck that could flap its wings, eat grain, and, to audiences' delight, appear to digest it and excrete waste. (The excretion was actually pre-loaded, so the digestion was fake.) He also built a flute-playing figure that really did produce music by blowing air through its mouth.

Then there is the most famous AI hoax in history. In 1770, Wolfgang von Kempelen presented to the Habsburg court a chess-playing machine: a wooden cabinet with a turbaned mannequin sitting behind a chessboard. This was the Mechanical Turk. It beat many opponents, reportedly including Napoleon Bonaparte and Benjamin Franklin, and it toured Europe and America for decades.

It was a fake. Hidden inside the cabinet was a skilled human chess player, concealed behind a clever arrangement of mirrors and mechanisms. The truth wasn't confirmed until the 1800s, and the original was destroyed in a fire in Philadelphia in 1854.

Why does this matter? Because it established a pattern that repeats throughout AI history: a dazzling demonstration, an eager public, and a gap between what is claimed and what is actually happening inside. (Amazon later named its crowdsourcing marketplace "Mechanical Turk" in a knowing wink: it is literally humans hidden inside the machine.) Whenever you read a breathless AI headline, it's worth asking who or what is behind the curtain.

Programs you can hold: the Jacquard loom

In 1804, Joseph Marie Jacquard introduced a loom controlled by chains of punched cards. Holes in the cards determined which threads were lifted, so by swapping cards you could weave different patterns using the same machine. The pattern was stored, not in the loom's structure, but in swappable data.

This was a huge deal. It separated the machine from the instructions, the hardware from the software. Al-Jazari's pegged drum had hinted at it; Jacquard made it practical and industrial. It also caused riots, as weavers feared for their jobs. Sound familiar?

Part 4: Babbage, Lovelace, and the machine that could have been

The Difference Engine and the Analytical Engine

The great leap came from an irascible English mathematician named Charles Babbage. In the early 1820s, frustrated by errors in printed mathematical tables, he began designing the Difference Engine, a mechanical calculator for producing tables of polynomial values accurately. Government funding ran out and the full machine was never completed in his lifetime (a working version was finally built from his plans at the Science Museum in London in 1991 and it worked perfectly).

But Babbage's next idea was far bigger. From the 1830s, he designed the Analytical Engine, and it reads like a blueprint for a modern computer:

  • a "store" (memory) to hold numbers,
  • a "mill" (a processor) to perform operations,
  • conditional branching and loops,
  • and instructions supplied on punched cards, borrowing from Jacquard.

It was designed to be general-purpose: not a machine for one calculation but for any calculation you could express as a sequence of steps. It was never built. The engineering was too demanding for the metalwork of the time, and Babbage kept redesigning it. But on paper, it was the first general-purpose computer.

Ada Lovelace's remarkable footnotes

In 1842, the Italian engineer Luigi Menabrea published a description of the Analytical Engine in French. Ada Lovelace, the mathematically gifted daughter of the poet Lord Byron, translated it into English in 1843 and added her own notes. Her notes were three times longer than the original article.

Ada Lovelace, portrait by Margaret Sarah Carpenter (1836)
Ada Lovelace, portrait by Margaret Sarah Carpenter (1836). Public domain, via Wikimedia Commons.

In Note G, she laid out a detailed step-by-step procedure for the engine to calculate a sequence of Bernoulli numbers, widely regarded as the first published computer program. But her bigger contribution was a vision. She realized that the engine's symbols didn't have to represent only numbers. She wrote that the engine might act upon other things besides number, for example, if the fundamental relations of pitched sounds in the science of harmony were susceptible of expression, the engine "might compose elaborate and scientific pieces of music of any degree of complexity or extent."

That is a startlingly early glimpse of general-purpose symbolic machines, machines that manipulate any information, not just arithmetic.

But Lovelace also included a famous caution, which is now called Lovelace's Objection:

The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform.

This is the great skeptical argument about machine intelligence: a computer only does what it is told; it cannot be creative or original. Alan Turing would address this objection head-on in 1950. And you can feel its echoes in every modern debate about whether an AI model is "really creative" or only recombining what it saw in training.

Interestingly, whether Lovelace was right is still contested. Modern machine-learning systems are programmed with how to learn, not with the specific answers; they sometimes surprise their creators. Both sides of the argument can quote her.

Part 5: Logic becomes algebra

While Babbage and Lovelace worked on hardware, another thread was quietly being woven on the software side: the idea that logic itself could be turned into mathematics.

George Boole

George Boole, a self-taught English mathematician who became a professor in Ireland, published The Mathematical Analysis of Logic in 1847 and then his masterwork, An Investigation of the Laws of Thought, in 1854. His central insight was audacious: logical reasoning can be expressed as algebra.

He used the symbols 1 and 0 for "true" and "false" (or "everything" and "nothing"), and defined operations that behave like AND, OR, and NOT. Statements about the world could be manipulated by symbol-pushing rules, the same way you manipulate equations.

This is Boolean algebra, and it is not an exaggeration to say that every digital circuit and every line of conditional code you've ever written traces back to it. Whenever you write if user.is_admin and not user.is_banned:, you are using Boole's algebra.

Here's a taste of how simple the idea is. This little Python program checks whether a logical statement is a tautology, meaning it's true no matter what the inputs are, simply by trying every combination, which is a mechanical process Leibniz would have loved:

from itertools import product def is_tautology(expr, variables): """Check an expression for every combination of True/False inputs.""" for values in product([False, True], repeat=len(variables)): env = dict(zip(variables, values)) if not eval(expr, {}, env): return False, env # found a counterexample  return True, None # "If P implies Q, and P is true, then Q is true" (modus ponens) modus_ponens = "(not ((not p) or q) or not p) or q" print(is_tautology(modus_ponens, ["p", "q"])) # (True, None)  # A fallacy: "If P implies Q, and Q is true, then P is true" affirming_consequent = "(not (((not p) or q) and q)) or p" print(is_tautology(affirming_consequent, ["p", "q"])) # (False, {'p': False, 'q': True}) <- counterexample found

There's no understanding inside that program, only a loop over truth values, and yet it correctly separates a valid argument from a classic fallacy. That is the entire promise of formal logic in a nutshell: valid reasoning without comprehension.

Logic machines

If logic is algebra, can a machine do it? The economist and logician William Stanley Jevons thought so. In 1869 he built a "logic piano", a keyboard-operated wooden machine that could work through logical problems with up to four variables, and presented it to the Royal Society in 1870. It's regarded as one of the first machines that solved a problem in formal logic faster than a person could. It was a curiosity at the time, but it was a real mechanical reasoner.

Frege, Russell, and the quest for foundations

In 1879, Gottlob Frege published his Begriffsschrift ("Concept Script"), a far more powerful logical notation that introduced quantifiers like "for all" and "there exists." That's the foundation of predicate logic, still the backbone of knowledge representation systems we'll meet again in the days of expert systems and the Semantic Web.

Then Bertrand Russell and Alfred North Whitehead produced the monumental Principia Mathematica (1910-1913), an attempt to derive all of mathematics from logical axioms. It famously takes hundreds of pages to prove that 1 + 1 = 2.

Around the same time, the mathematician David Hilbert set out a grand program: to show that mathematics could be placed on a complete and consistent foundation, and that there should be a mechanical procedure, an algorithm, to decide whether any given mathematical statement was provable. That last demand is called the Entscheidungsproblem ("decision problem"). It's a Leibnizian dream with a precise modern form.

And here is where the story turns. In 1931, Kurt Gödel proved that any consistent formal system rich enough to express arithmetic must contain true statements it cannot prove. In 1936, Alan Turing and Alonzo Church each independently showed that Hilbert's decision problem has no general solution. To prove it, Turing invented an imaginary machine, the Turing machine, which became the mathematical definition of "computation" itself.

That is the story for Part 2 of this series. For now, notice how far the thread has run: from Aristotle's syllogisms to Leibniz's "Let us calculate," to Boole's algebra, to Hilbert's dream and Gödel's blow. All of it was a conversation about what can be done by following rules.

Part 6: Machines enter fiction and fear

At the same time that mathematicians were laying the groundwork, writers were exploring what it might mean to succeed.

In 1818, Mary Shelley published Frankenstein, whose scientist builds a living being and then recoils from it. It's the modern template for the "creation turns on its creator" story. It was written in an age of galvanism and experimentation with electricity, and it asked a question that hasn't gone away: what do creators owe their creations, and what do they owe the world if the creation goes wrong?

In 1863, Samuel Butler wrote an essay called "Darwin among the Machines," arguing that machines were evolving much faster than biological life, and might someday supplant humans. He was reacting to the Industrial Revolution and Darwin's On the Origin of Species, and he sounds eerily contemporary.

Then in 1920, the Czech writer Karel Čapek wrote the play R.U.R. (Rossum's Universal Robots), which premiered in 1921. It introduced the word "robot" to the world, from the Czech robota, meaning forced labor, a word suggested to him by his brother Josef. In the play, mass-produced artificial workers eventually revolt and wipe out humanity. The word, and the fear, spread worldwide within a few years.

This matters for our story because culture shaped the field before the field existed. When researchers later began building real AI, they were already surrounded by expectations of thinking machines, and by fears of them. Those stories still color how we talk about AI today, for better and for worse.

What the pre-computer era got right and wrong

Let's step back and take stock.

What this era got right:

  • Behavior can be stored. From al-Jazari's pegs to Jacquard's cards to Babbage's punched instructions, people understood that you could separate what a machine does from how it is built.
  • Reasoning has structure. Aristotle, Boole, and Frege showed that parts of thought can be expressed in formal rules.
  • Reasoning might be calculation. Hobbes and Leibniz argued that this reasoning could be reduced to computation, and Leibniz said we should one day just compute disagreements away.
  • General-purpose machines are possible. Babbage and Lovelace described a machine whose power isn't tied to a single task.

What it got wrong, or at least overlooked:

  • Logic isn't everything. The real world is messy, ambiguous, and uncertain. Pure logic is brittle in that world. Much of the later history of AI is the story of discovering this the hard way.
  • Appearances deceive. The Turk reminds us to separate performance from reality.
  • Rules are not enough. Everyday intelligence, recognizing a face, understanding a joke, riding a bicycle, doesn't feel like following explicit rules at all. Eventually, AI would find that it had to learn rather than be told.

Those last points will drive the next 23 articles.

Quick glossary

A few words from this chapter that will keep coming back:

  • Automaton: a self-operating machine, from the Greek for "acting of its own will." Hero's theaters and al-Jazari's musicians were automata.
  • Syllogism: a three-part argument (two premises, one conclusion) whose validity depends only on its form. Aristotle's invention.
  • Algorithm: a precise, step-by-step procedure for solving a problem. The word comes from the name of the ninth-century Persian mathematician al-Khwarizmi.
  • Boolean algebra: the algebra of true and false (1 and 0) with AND, OR, and NOT. Boole's gift to every digital circuit.
  • Formal system: a set of symbols and rules for manipulating them, in which proofs can be checked mechanically.
  • Entscheidungsproblem: Hilbert's "decision problem": is there a mechanical procedure that can decide whether any mathematical statement is provable? Part 2 tells how it was answered.

Key takeaways

  • The dream of artificial minds is ancient: Greek myth, Jewish folklore, and Chinese legend all imagine made beings.
  • Real engineering followed: ancient gear computers, medieval programmable automata, and Renaissance calculators.
  • The core ideas of AI, that reasoning can be formalized and computed, were stated clearly by Aristotle, Hobbes, and Leibniz long before any computer existed.
  • Babbage designed a general-purpose computer in the 1830s, and Lovelace wrote what many call the first program and the first serious reflection on what machines can and cannot originate.
  • Boole turned logic into algebra, which would later become the language of digital circuits.
  • Fiction supplied the fear. By 1920, the word "robot" and the story of rebellion were already part of the culture.

Try it yourself

If you'd like a small exercise to carry into the next chapter, try extending the tautology checker above:

  1. Add support for implies(a, b) as a function so you don't have to write (not a) or b by hand.
  2. Test the classic syllogistic forms: Modus Ponens, Modus Tollens, Hypothetical Syllogism, and the fallacy of denying the antecedent.
  3. Ask yourself: is this program reasoning? Where, exactly, would you say understanding lives, if anywhere?

That last question is the one that has fueled seventy years of argument, and it's where the series goes next.

Next up

In Part 2, we meet the man who turned all of these ideas into a precise mathematical definition of computation, and then asked a question that still defines the field: "Can machines think?" We'll follow Alan Turing from the Turing machine, through wartime codebreaking at Bletchley Park, to the Imitation Game.

If you enjoyed this chapter, follow the series so you get each new part as soon as it's published. And if you have a favorite pre-history story I left out (the Antikythera mechanism? Greek fire? Hero's automata?), tell me in the comments.

See you in Part 2.

☕ If this series helped you, you can support my writing on Ko-fi.

Support me on Ko-fi