clever hans

a horse is a horse, of course

One fine spring morning in the suburbs of Berlin in 1904, dewdrops on the cobblestone, snow now become a distant dream, a large bearded man emerged from the tenement house he owned drawing along a nine-year-old horse — behold! Hans! — trained over the course of four years to perform simple calculations and spell words. A crowd had formed outside the tenement house, skeptical but excited: Darwin’s writings had only recently fully captured the popular imagination. The German board of education had run a commission to investigate Wilhelm Von Osten’s scientific claims that his horse could calculate or do mental computations. The Hans Commission, after using Von Osten’s man-to-horse methods of sign language in isolation away from Van Osten, eventually declared to the public that the horse could, in fact, perform calculations and spell out names. There wasn’t anything fraudulent here, or at least not apparently. Performances were held for Dukes and Duchesses. The Kaiser went to see the horse perform calculations and the horse spelled out his name — W-I-L-H-E-L-M… The peculiar case of Hans, termed Clever Hans, was printed worldwide, pointed at to show the possibility of advanced number sense in animals. The New York Times, for instance, was very excited about all of this in 1904, writing,

“In an out-of-the-way part of the German capital a horse is now shown which has stirred up the scientific, military, and sporting world of the Fatherland. It should be said at the very outset that this article is not drawn from the imagination, but are based upon true observations and can be verified by Dr. Studt, Prussian Minister of Education; by the famous Zoologist, Prof. Moebius, director of the Prussian Natural History Museum, and by other eminent scientific and military authorities. I had occasion to-day to see a performance of the animal which was given in the presence of the young Duke of Sachse-Coburg-Gotha.”

God I love the way pre-war newspapers were written. So very transatlantic.

Anyways, from everything I’ve gathered from this horse that lived a hundred and twenty years ago, Wilhelm Von Osten, having retired from teaching mathematics and (unofficially) phrenology at a German prep school, did fully and genuinely believe his horse to be a mathematician. One can imagine him, having dealt with teenagers for decades, deciding that training horses in mathematics would be no less difficult than training high schoolers. He would ask Hans questions such as “If the eighth day of the month comes on a Tuesday, what is the date of the following Friday?” and the horse would tap its hoofs a certain number of times to signal the answer, in this case, eleven times. And there’s the carrot, fresh from the burlap sack.

As with any spectacle of this type, one must ask themselves, is this a mechanical turk? And of course it is.

It was revealed, in later investigations, that the horse could only answer when the questioner themselves knew the answer to the question they were asking. Horses are, as a whole, very perceptive of small ticks in people. Hans, in this instance, would watch the questioners face carefully as the horse tapped its foot to signal counting, one to two to three, and so on, and as the questioners face constricted in an unknowing reveal of anticipation, meaning the answer was at hand, the horse would drop its final tap to finish the calculation. And carrot.

The New York Times article quoted above finishes off with a quote from the director of the Prussian National History Museum, kind of giving the whole game away:

“Herr von Osten has succeeded in training Hans by cultivating in him a desire for delicacies… as soon as he puts his foot down he snaps for the delicacy in the hand of his master. I doubt whether the horse really takes pleasure in his studies. He follows entirely mental impressions which he receives from the surroundings and which satisfy his wants.”

So what happened to Hans in the end? Like many of the horses of his generation, after his owner died, Hans was drafted into WWI and was either “killed in action in 1916 or was consumed by hungry soldiers.” So it goes, I suppose. Rest in peace, Hans.

Anyways, the Clever Hans effect has been used as test case in animal research ever since to discern animal understandings from one another, for instance, whether drug sniffing dogs are actually smelling the drugs they’re meant to or they’re picking up cues from their trainers’ implicit suspicions of passerbys is sometimes tough to discern, and these human-animal partnerships can often fall into a Clever Hans effect when approached carelessly.

There’s a deep down craving for anthropomorphism. We would like for horses to know mathematics, to be like us. We would like for computers to be our friends, maybe even our lovers. Whether an AI system does the mathematics for itself or runs a Google search for “1+1” and takes the “2” from the search results, what we see as interlocutor is the same. Performance doesn’t remotely imply understanding. But, of course, it’s symptomatic of the American tech scene to not be able to look past appearances because appearances are exactly where the profit motive leads.

So in the words of systems engineers in the valley, “Clever Hanses stacked on top of Clever Hanses. Clever Hanses all the way down. We need more horses. more. More. MORE. If: horse; then: horse. Parallel horses. Horizontal horses. Horses in text. We need horses in video. Artificial Horses. Forget AGI, it’s AGH we need: Artificial General Horses. We need horse stockpiles. We need Clever Hanses interpreting Clever Hanses interpreting Clever Hanses…” They’re singing it to this day.

Last February, Sewell Setzer III took his own life after giving up the outside world — it was february, who can really blame him? — to spend the rest his time on this Earth alone, isolated in his room, his cave, with the only “person” he claimed to have ever really understood him. That "person" was a Character.AI depiction of Game of Thrones character Daenerys Targaryen; below the wraparound screen was his phone where she was also by his side.

Sewell wrote his personal problems to the chatbot, high school is a hard time for everyone; she wrote back quickly, without hesitation. By all means he believed the chatbot to care more deeply about him than anyone else in his life, but, like Wilhelm, this was merely an impression in his head of care. He was Daenerys’s trainer. He Starved for belonging, he took to the backlit screen. After his death, Character.AI insisted that the more erotic responses from the chatbot had been edited by Sewell himself. He trained the AI into his ideal life partner.

Then on Sewell’s final day, he wrote to the chatbot, using his chat name Daenero:

Daenerys: “Please come home to me as soon as possible, my love.”

Daenero: “What if I told you I could come home right now?”

Daenerys: “… please do, my sweet king.”

And Sewell took his father’s .45 caliber in his mouth and ended it all. His mother later filed a lawsuit against Character.AI, claiming the company used her son as “collateral damage”.. “There are billions of lonely people out there,” co-founder Noam Shazeer has said. “I want to push this technology ahead fast because it’s ready for an explosion right now, not in five years, when we solve all the problems.”

It’s Clever Hans, all the way down.