In 1891 a Berlin schoolteacher named Wilhelm von Osten started teaching his horse arithmetic. By 1904 Hans could add, subtract, give the day of the week for any date, and answer in front of strangers by tapping a hoof, and Von Osten never charged a ticket for it. That year The New York Times covered him as a serious story, not a novelty item, because the ordinary explanations were running out.

Germany's board of education appointed a commission to settle it: thirteen people, including a cavalry officer, a circus manager, and the director of the Berlin zoo. On 12 September 1904 they reported that Hans was not being cued by any of the tricks a trainer usually relies on. Newspapers read this as the commission declaring that a horse could think, which is not what a "no known trick found" verdict says, and the philosopher who chaired it, Carl Stumpf, spent real effort afterward explaining the difference to a reporter who had already printed it the other way.

Stumpf handed the actual investigation to his assistant, Oskar Pfungst, who did something the commission had not: he tried to make Hans wrong on purpose.

The variable nobody had isolated

Pfungst's first useful move was to ask questions that nobody in the room, including Pfungst, knew the answer to. Under those conditions, the horse that could supposedly do arithmetic could not reliably do arithmetic. His second move was to control what Hans could see. The early attempts were clumsy — Von Osten held a slate up to block his own face, and it worked; a second person holding that same slate in the same place did not — until proper blinders, fifteen centimetres square, gave a result clean enough to publish. Across 102 trials, when Hans could see the person asking the question, he answered correctly 89 percent of the time — fifty of fifty-six. When he could not see them, that fell to six percent: two right answers out of thirty-five.

The questioner did not have to be Von Osten. Any stranger got the same 89 percent, on one condition — the stranger had to already know the answer. Hans was not reading his owner's private signals. He was reading the small change that happens in a body watching for a number it already expects, released the instant the correct tap arrives. Von Osten had spent thirteen years teaching Hans to count. What Hans had actually learned was to read a tell his teacher did not know he was making.

The part that outgrows the debunking

Pfungst worked the mechanism out, described it in enough detail to make the whole trick predictable, and then turned the method on himself. In his own account — a chapter he titled "The Author's Introspections" — he wrote that even after he had learned the signal well enough to call forth any answer he wanted from Hans at will, he went on giving the older, unwitting version of that same signal, unprompted, without meaning to. Knowing exactly what his body was doing did not retire the habit. Switching it off, he wrote, was not something knowledge did in a moment; unlearning it took the same kind of sustained practice that learning it in the first place would have.

That is a stronger result than proving the horse was not thinking. A trick a person can retire the moment they see through it is a habit. A signal that keeps running after the person producing it has fully diagrammed it — one that takes deliberate practice to unlearn, not just the fact of knowing — is closer to a reflex. Von Osten spent thirteen years teaching a horse to count. It took one autumn of investigation to discover that the horse had been teaching a stranger something about his own body that understanding it did not immediately switch off.