An 18th-century machine could play chess on its own, beating Napoleon and Benjamin Franklin

A machine that seemed to think

In 1770, at the court of Empress Maria Theresa in Vienna, the official and inventor Wolfgang von Kempelen presented a device unlike anything his audience had seen. A life-sized figure carved and dressed as an Ottoman Turk sat behind a wooden cabinet, one hand resting near a chessboard set into its top. Kempelen opened the cabinet doors to show the interior packed with gears and machinery, closed them again, wound the mechanism with a key, and invited a member of the court to play. The figure moved its own arm, lifted and placed the pieces, nodded when it put a king in check, and, more often than not, won.

The effect was overwhelming. Here, apparently, was a machine that could not merely imitate a lifelike motion, as the celebrated mechanical ducks and musicians of the age did, but could reason its way through the most cerebral of games. For the next eighty-four years the Automaton Chess Player, soon known simply as the Turk, would travel the courts and theaters of Europe and America, drawing crowds who left convinced they had watched a machine think.

They had not. Inside the cabinet, in a space the opened doors were carefully arranged never to reveal, sat a living human being: a skilled chess player who followed every move and directed the figure's hand. The Turk was one of the most successful and long-running illusions in history, and, in a way its builder could not have foreseen, one of the most instructive.

How the illusion worked

The genius of the Turk lay less in its arm than in its choreography of concealment. The cabinet was divided into compartments, and the operator sat on a sliding seat that let him move his body from one section to another as the doors were opened in sequence. Whichever compartment a spectator was shown stood empty of anything but machinery; the human occupant had already slid into a closed part of the box. When Kempelen or, later, Maelzel held a candle inside to prove the space was hollow, the light fell exactly where no one was hiding.

Following the game from inside a sealed cabinet was solved with equal ingenuity. Each chess piece was fitted with a small magnet, and beneath the board a matching set of magnets was suspended so that, on a small internal board, the hidden player could see the exact position of every piece above him without any line of sight. He worked out his move, then operated a pantograph, a linkage of connected levers, that translated the motion of his own hand into the movements of the Turk's arm on the table above. A candle inside supplied light, and the design allowed for ventilation and for masking its smoke.

The reveal of the machinery was not a refutation of the trick. It was the trick.

None of this required futuristic engineering, only careful cabinetmaking and a strong chess player willing to fold into a box for an hour. That, ultimately, is the point. The Turk contained no mechanism that could choose a move, because no such mechanism existed anywhere in the 18th century. What it contained was a person, and the elaborate apparatus existed to hide that person and relay their decisions to the figure's hand.

Fire, and the confession that followed

Maelzel died at sea in 1838, and the Turk changed hands, eventually coming to rest in Philadelphia in the collection of the physician John Kearsley Mitchell, who restored it and exhibited it privately. Its touring days were over. On 5 July 1854, a fire that spread from a neighboring building reached the National Theater, formerly the Chinese Museum, where the automaton was stored, and the Turk was destroyed. Reports have it that its last owner heard the figure's “last words” consumed in the flames, a fittingly theatrical end for a machine that had always been a piece of theater.

The fire did not bury the secret; it freed it. With the object gone and nothing left to sell, those who knew how the Turk worked began to describe it plainly. Mitchell and, in 1857, his son, the physician and author Silas Weir Mitchell, published detailed accounts in the chess press of the sliding seat, the magnetic indicator board, the pantograph arm, and the roster of hidden operators who had given the Turk its play. Their descriptions matched what Racknitz had deduced in 1789 and what Poe had argued in 1836. There was no residual mystery left to solve.

The operators themselves became, in retrospect, the real story. Strong players such as Johann Allgaier, who is credited with the game against Napoleon, and William Schlumberger, Maelzel's operator during the American years, had been the Turk's intelligence all along. The machine had never selected a move in its life. Human beings had, from a cramped and candlelit box, for the better part of a century.

Where the evidence lands

On the central claim, that the Mechanical Turk was a self-operating machine capable of playing chess on its own, the verdict is debunked, and it has been settled for well over a century and a half. The Turk was a cabinet illusion operated by a concealed human chess master, a fact deduced in print as early as 1789, argued by Poe in 1836, glimpsed by eyewitnesses, and finally confirmed in full by the people who had run it after the machine burned in 1854. No credible authority regards it as a genuine automaton, and the technology to build a real chess-playing device did not exist until the 20th century.

What endures is the metaphor. The Turk has become shorthand for a particular kind of deception: a smooth automated surface concealing human labor underneath. That is why Amazon reached back to it in 2006 to name a crowdwork platform, describing the arrangement as “artificial artificial intelligence,” and why critics of the modern AI industry keep invoking the Turk when they point to the annotators and moderators whose hidden work props up systems sold as autonomous. The eighteenth-century original was a frank fraud and today's human-in-the-loop labor usually is not, so the comparison is a caution rather than an accusation. But the caution is a real one, and it is old: when a machine seems to be thinking, it is worth asking who, exactly, is inside the box.