The Horse That Could Count and the Face That Did Not Know

In 1904 a Berlin newspaper reported on a horse that appeared to do arithmetic. The animal, called Hans, belonged to a retired schoolteacher named Wilhelm von Osten, who had spent years teaching it. Asked to add three and four, Hans tapped his hoof seven times and stopped. He was reported to handle fractions, calendar dates and musical intervals, and to answer correctly when strangers posed the question instead of his owner. A commission of experts that examined the horse in 1904 is reported to have found no evidence of deliberate trickery. Von Osten took no money for the displays. The public was delighted, and a small scientific puzzle was born.

The puzzle was solved in 1907 by Oskar Pfungst, a psychologist who worked under Carl Stumpf. Pfungst ran a series of tests and noticed a pattern the earlier examiners had missed. Hans answered correctly only when the questioner knew the answer and when the horse could see the questioner. Place a screen between them, or ask something the questioner did not know, and the performance fell apart. Pfungst concluded that Hans was reading tiny, unintended movements. As the taps approached the right number, the questioner's posture and head shifted slightly, and as the count reached the answer, the tension released. Hans had learned that the release was the signal to stop.

It would be comforting to treat this as a tale about a gullible public. It is better read as a tale about the horse's expertise. Hans had learned to read human beings more precisely than humans read themselves, and the people who cued him were entirely unaware that they were doing it. The horse had not learned arithmetic, but he had learned something harder to teach, which was how to read a face that did not know it was speaking.

The horse had not learned arithmetic, but he had learned something harder to teach, which was how to read a face that did not know it was speaking.

The episode is now taught as the Clever Hans effect, and it shaped how experiments are built. Researchers use double blind procedures, in which neither the experimenter nor the participant knows which condition is under way, precisely because an expectation can leak through a gesture. Yet it would be a mistake to conclude that animals cannot think. Pfungst's finding showed a real skill in Hans, and later work on animal cognition has had to separate genuine ability from cue reading, a task that remains difficult.

A sceptic might say the lesson is trivial, since any careful laboratory already guards against it. But the leak runs in both directions and well beyond the laboratory. A 1968 study by Robert Rosenthal and Lenore Jacobson, whose methods drew criticism, reported that teachers who expected certain pupils to flourish may have treated them differently in ways that helped. Interviewers, managers and doctors send signals of the same kind. If a horse could find an answer in a face that did not know it was giving one, the unsettling question is how much of what any person calls their own judgment was first handed to them in a glance they never noticed.