Before Machines: Calculation, Automata, and the Dream of Mechanical Reason

Part of The Long Human Road to AI series.

Imagine you are a navigator in the 1700s. Your ship’s position depends on logarithms, lunar distances, and hours of careful arithmetic. One digit copied wrong, one outdated table, and the vessel misses the harbor. Today we ask a phone. Then, people asked people—and those people asked beads, rods, tables, and gears.

The long road to artificial intelligence did not begin with transistors. It began with a much older problem: too many numbers to hold in memory, too much repetition to do reliably alone, and too much consequence to leave unchecked. Before anyone built a machine that could compute, people had to learn how to compute, how to organize that labor, and how to make thinking visible enough to check.

Point C1 Computation was performed by people before it became associated with electronic machines.

The work before the machine

Before electronic machines, a “computer” was often a person. Museums and labor histories record human computers performing calculation as organized work—supporting navigation, astronomy, ballistics, engineering, and later the machines themselves. The work could be skilled or repetitive; it was almost always institutional. By the late nineteenth century, much of it increasingly fell to women, a pattern the Smithsonian’s Human Computer Project documents across observatories, government offices, and research labs.

The WPA Mathematical Tables Project, run during the 1930s and 1940s in the United States, employed human computers to produce tables of mathematical functions. These tables were not neutral outputs. They were the product of planning, checking, division of labor, and error control. A table was only as trustworthy as the people who made it and the procedures that caught mistakes.

That social texture matters. When we call a modern laptop a “computer,” we inherit a word that once described clerical workers in rooms. The machine did not appear in a vacuum. It arrived after centuries of figuring out what calculation work looked like, who did it, and how to keep it honest.

Marks, counters, and visible memory

One early response to the memory problem was to give numbers a body. Counting boards and abaci let a person move counters along rods, wires, or lines and see the state of a calculation at a glance. The Computer History Museum and the Smithsonian both treat these devices as durable examples of physical aids for arithmetic, with forms that varied across place and time.

Point C2 Counting boards and abaci moved arithmetic into visible physical state that could be manipulated and checked.

Think of an abacus as a kind of visible working memory. Beads stand for values; position stands for place. A learner can watch a calculation happen, catch an error, and repeat the motion. But the abacus does not calculate by itself. It gives calculation a body that a person can see, change, and check. The human operator still supplies the rule, the goal, and the interpretation.

Tables, rods, and reusable steps

Once arithmetic became visible, the next trick was to avoid repeating it. Mathematical tables, logarithmic methods, and devices such as Napier’s bones broke complex calculations into smaller parts or reused work that had already been done.

John Napier’s calculating tools, described by the Whipple Museum, helped users decompose multiplication into additions and lookups. The rods are a teaching device as much as a labor-saving device: they make a procedure tangible. The Smithsonian’s collection notes that Napier also discovered logarithms, later embodied in the slide rule. Meanwhile, the NIST history of the Math Tables Project shows how far the table-making tradition extended into the twentieth century.

Point C3 Rods, tables, and logarithmic methods reduced complex calculations by decomposing or reusing prior work.

You can think of a mathematical table as cached work. It stores answers so that future workers do not have to derive them again. But a table is static and narrow. It does not decide which entry to use, and it cannot warn you when the world has changed. The cache metaphor is useful only as long as you remember its limits.

When gears carried the operation

The seventeenth century brought a new idea: arithmetic could be embodied in mechanism. Blaise Pascal’s 1642 calculating machine, designed for addition and subtraction, is preserved in replica form by the Science Museum Group. Gottfried Wilhelm Leibniz’s Step Reckoner, described by the Computer History Museum, extended the idea with a stepped-drum mechanism that influenced later four-function calculators.

Point C4 Seventeenth-century mechanical calculators embodied arithmetic operations in physical mechanisms.

These machines were remarkable, but they were not general-purpose brains. A Pascaline added and subtracted. A Leibniz machine could multiply through repeated addition. They automated specific operations, not judgment. They mark a shift from aided calculation to automatic operation, but the boundary between operation and intelligence is important. Fixed-purpose calculators are not artificial intelligence.

Leibniz himself went further in imagination. The Stanford Encyclopedia of Philosophy records his dream of a symbolic method in which reasoning errors could be exposed through calculation-like manipulation. That dream belongs more to the history of logic than to the history of gears; it foreshadows the next article more than it explains this one. Still, it shows that the idea of mechanical reason was alive before anyone could build it.

The dream of self-moving mechanism

Not all mechanisms calculated. Some simply appeared to act. Automata—clockwork figures, musical machines, mechanical animals—have long invited audiences to project life or agency onto fixed motion. The Science Museum’s stories of early robots and Britannica’s survey of automata describe mechanical objects that are “relatively self-operating after being set in motion.”

Point C6 Automata made mechanism appear self-directed, inviting audiences to project life or agency onto fixed motion.

This projection is a recurring theme in the history of artificial intelligence. When a machine moves on its own, we are tempted to see intention. The movement is real; the intention is ours. Fixed motion is not sensing, learning, or understanding. Automata are theater, not cognition.

The Mechanical Turk, the famous chess-playing “machine” of the eighteenth and nineteenth centuries, is a cautionary example. As the Computer History Museum notes, it appeared to play chess but concealed a skilled human player. It is useful as a reminder that apparent machine intelligence can hide human labor—not as a milestone in autonomous reasoning.

Against this backdrop, the Antikythera mechanism looks different. Where automata performed, the Antikythera mechanism calculated and displayed celestial cycles. Research published in Nature and summarized by NYU’s Institute for the Study of the Ancient World describes it as a geared astronomical calculator or display mechanism for lunar phases, calendar cycles, and heavenly motions. Modern imaging projects, such as the one at the Max Planck Institute, continue to reveal inscriptions and reconstruct its functions—but reconstruction is not complete certainty.

Point C5 The Antikythera mechanism was a sophisticated geared astronomical calculator or display mechanism.

It is tempting to call the Antikythera mechanism an “ancient computer.” That label only works if we are careful about what we mean by “computer”: a device that calculates and displays astronomical cycles, not a general-purpose machine or an artificial mind. The mechanism is impressive precisely because it is a narrow, geared answer to a specific problem, not because it predicts like modern modeling.

Patterns, cards, and controlled sequence

By the early nineteenth century, the idea of controlling a machine from outside was taking a new form. Joseph-Marie Jacquard’s punched cards stored weaving patterns; changing the cards changed the pattern without rebuilding the loom. The Smithsonian’s punch-card spotlight and the Science and Industry Museum’s Jacquard loom story both note that Charles Babbage admired this approach and suggested punched cards for his own computing designs.

Point C7 Jacquard punched cards controlled textile patterns and influenced later ideas about machine input and control.

Think of Jacquard cards as stored pattern instructions. They show how a machine can be controlled by an external encoded sequence. But the sequence controlled weaving, not general symbolic computation. The cards were not software in the modern sense, and the loom did not reason.

Babbage’s Difference Engine and Analytical Engine sit at the threshold. The Science Museum and the Computer History Museum both emphasize that Babbage designed automatic computing engines but failed to build complete versions in his lifetime. Later reconstructions, such as Difference Engine No. 2, were built from his original drawings. The designed-versus-built distinction matters: Babbage’s engines were visions as much as physical machines.

Point C8 Babbage’s engines mark a threshold between mechanical arithmetic and designs for automatic computing machinery.

The Stanford Encyclopedia of Philosophy’s “The Modern History of Computing” treats Babbage as part of a longer chronology of computing machines. In his designs, one can already see the pieces that would matter later: instructions, memory-like stores, and controlled sequences of operations. The pieces were not yet a computer in the modern sense, but they were no longer just a faster abacus.

The threshold

By the middle of the nineteenth century, the long prehistory of computing had separated out several ideas that had once been tangled together: the symbol, the rule, the operator, and the mechanism. People had learned to externalize arithmetic into visible state, to decompose complex work into reusable steps, to embody operations in gears, and to control machines with encoded instructions.

None of this made artificial intelligence inevitable. Each step solved a narrow problem: navigation, astronomy, textile patterns, reliable tables. The thread running through them is not progress toward a predestined machine. It is the gradual discovery that parts of thinking can be represented, checked, delegated, and mechanized—while other parts remain stubbornly human.

The next article in this series asks what happens when rules and symbols become formal enough for a machine to execute in principle. That is the story of logic, computation, and the mathematical road to AI. Before we get there, it is worth remembering that every early computer—every abacus, table, calculator, automaton, and punched card—was also a record of human need, human labor, and human ingenuity. The machines came later.