What was Vaucanson's Digesting Duck, and did it really digest?

In 1739, crowds in France gathered to watch a copper duck eat. The bird stretched its neck toward a visitor's hand, gulped down grain, flapped its wings and, a little later, produced something that looked very much like what a real duck leaves behind. Its builder, Jacques de Vaucanson, said the inside held a small chemical laboratory that broke the food down.

The duck was one of the most famous machines of its century. It was also, in one important respect, a trick. Both facts make it worth knowing about in 2019, when we are again being shown machines that seem to do something lifelike.

Who built it, and when

Jacques Vaucanson was born in Grenoble, France, in 1709, the son of a glove-maker. (The "de" in his name was added later, by the French Academy of Sciences.) As a young man he studied anatomy with a surgeon named Le Cat, and he became fascinated by the idea of building machines that imitated the body's functions, such as breathing, blood flow and digestion.

His first big success was a life-size figure that played the flute. He presented it to the Royal Academy of Sciences in Paris in early 1738, along with a written explanation of how it worked; a royal censor approved that text for printing on June 12, 1738. He then built two more automata: a shepherd who played a pipe and drum, and the duck. According to the most common account, the duck was unveiled on May 30, 1739, in France.

In 1742 the three machines went on show in London, in a room at the opera house in the Haymarket, with showings at set hours each afternoon. That same year an English translation of Vaucanson's explanation was printed, together with a letter in which he describes the duck in his own words.

How the duck worked, in Vaucanson's words

Built to the scale of a real duck, the machine was cased in gilded copper. In his letter, Vaucanson lists what it could do:

• stretch out its neck to take grain from your hand, then swallow it with quick, greedy movements of the throat;
• drink, splash in water with its bill and make a gurgling noise;
• rise on its legs and turn its neck left and right;
• move wings in which, he says, every bone and joint of a real wing was copied.

Once wound up, he adds, it ran through all these actions without anyone touching it again. He deliberately left the mechanism uncovered, so that viewers could see the moving parts and would not suspect any hidden trick.

On digestion he was careful with his words. He did not claim, he wrote, a perfect digestion that turned food into blood and nourishment. He claimed only to imitate three steps: swallowing the grain, dissolving it, and making it come out visibly changed.

The trick inside

Even that modest claim was not true. In 1844 the French stage magician and automaton maker Jean-Eugène Robert-Houdin examined the duck. He found that the grain went into one container and stayed there, while the "droppings" were pellets of breadcrumbs dyed green, loaded in advance into a second compartment. Nothing was dissolved at all. Robert-Houdin admired it as a piece of artifice he would have been glad to use in his own magic act.

Here is the gap between what was shown and what happened, side by side:

Most of the duck was genuine engineering of a high order. The single feature that made headlines was staged.

Why it mattered

The duck made Vaucanson famous. The philosopher Voltaire wrote in 1741 that without Le Maure's singing voice and Vaucanson's duck, France would have little left to remind it of its glory. Vaucanson himself grew tired of his automata and sold them in 1743; all of the originals are now lost, and the duck is thought to have burned in a museum fire in 1879.

His more lasting contribution came next. In 1741 he was put in charge of inspecting France's silk industry, and in 1745 he designed a loom that could be controlled automatically using punched cards. Weavers resisted it, but decades later Joseph-Marie Jacquard refined the idea into the famous Jacquard loom. In the 20th century, punched cards became a standard way to feed data and instructions into computers. The road from Vaucanson's workshop to modern machines runs through that loom more than through the duck.

The duck also left a question about demonstrations. Builders of lifelike machines have often needed to impress patrons and the public, and a staged detail can win attention that honest engineering does not. In 2019, with robots and AI systems shown in polished videos, that question is as live as ever.

How can you tell a real AI breakthrough from a clever demo?

Start by asking what exactly is being shown, and what is only being said. Vaucanson's duck really swallowed grain; the "laboratory" existed only in his description. With an AI system, separate what you see it do on screen from claims about how general, reliable or independent it is.

Next, look for independent checks. Robert-Houdin's inspection settled the duck's secret. For AI, the equivalent is a published paper, open code, or results on a public test that other researchers can rerun. The AlphaGo matches against top Go professionals, played in public, are a good example of a claim that was tested in the open.

Finally, notice who is doing the selecting. A demo video shows the best attempts, not the average one. Ask how often the system succeeds, not only whether it can, and you will be much harder to fool.