Neuroscience & Mind
The Clever Hans Problem and Research on Animal Cognition
As Denyse O’Leary will likely discuss in her new series “Animal Minds,” a number of research studies claim to show that under some circumstances animals are capable of abstract thought. I believe that those conclusions are wrong, and that animals are not capable of abstraction.
Animals think particularly, not abstractly. That is, animals think about particulars (individual things) and about perceptions connected to the particulars, but they are not capable of abstracting universal concepts from particulars and contemplating the universal concepts in isolation from individual things that evoke them. For example, my dog certainly thinks of my kindness and responds by wagging her tail, but there is no evidence she is capable of contemplating kindness as a concept, abstracted from my particular behavior.
That is not to say that animals can’t be clever. Animals can be very clever — sometimes more clever than humans. Try to build a spider web yourself. But animal cleverness is always tied to particular things and situations and instincts. A spider builds a very effective web to catch his meals. There is no evidence the spider contemplates the beauty of spider webs or that it understands “web” as a metaphor for a complex set of associations.
In animal cognition research, it is of the utmost importance to distinguish clever perceptive thinking from abstract thinking. Animals can be trained to do very clever perceptive things, and some of these clever things seem like abstract thinking. A parrot could be trained to recite Godel’s Theorems, but the parrot knows nothing of mathematical logic. To be trained in perceptual cleverness is not to think abstractly.
The pitfall of mistaking training in perceptual thought for abstract thought is clearly illustrated by a famous example in animal cognitive research — the case of Clever Hans.
Wilhelm Von Osten was a German mathematics teacher and phrenologist at the turn of the 20th century who was convinced that animals were capable of abstract thought. He tutored several animals in the hope of demonstrating the truth of animal reason. The cat and the bear didn’t turn out well (he actually tried), but he claimed to have success with Hans, an Arab stallion. Osten drew mathematical problems on a blackboard, and, after a while, the horse seemed to understand some of what was written, and could even answer mathematical questions (by counting with his hoof). Hans responded to a “6” on the blackboard with six taps of his hoof. He responded to “What is the square root of nine” with three taps. He was able to answer correctly consistently, over a broad range of problems. He was even able to calculate dates — if asked “if Tuesday is the 4th of the month, what date will the following Friday be?” he would tap his hoof seven times. Osten announced that Hans’ mathematical skills were those of the average 14-year-old child.
Osten exhibited Hans across Germany to large crowds (the horse was even featured in the New York Times). The horse’s accuracy was about 90 percent, and the exhibitions attracted much attention from the scientific community.
In Germany, the educational establishment formed a Hans Commission, staffed by scientists who studied Hans’s abilities. They observed the exhibitions carefully and studied Hans extensively. The Hans Commission concluded that Hans’ abilities were genuine, and there was no evidence for deception. The horse was capable of abstract mathematical reasoning.
However, a leading psychologist — Oskar Pfungst — remained skeptical, and he was invited by the Commission and by Hans’ owner to examine Hans in even more detail. This is what Pfungst found:
When his owner asked Hans a question, he did very well.
When someone other than his owner asked Hans a question, he also did very well.
However, when someone who did not know the answer to the question asked Hans a question, Hans almost never got the answer right.
When Hans was asked a question in such a way that he could not see the person asking the question, Hans almost never got the answer right.
The precondition for Hans’ correct answer was that the questioner had to know the answer and Hans had to be able to see him.
Pfungst carefully studied the questioners. He noted that the questioner always gave subtle verbal or physical clues — a tone of voice at the right moment, a grimace, a tensing of muscles — and Hans’ would tap his hoof until he saw the clue to stop. All of this was quite unconscious on the part of the questioners — volunteers did it, as well as Hans’ owner. But Hans was not capable of abstract mathematical reasoning. He was capable of perceiving remarkably subtle particular clues. He was trained to respond to particular clues. This perhaps should come as no surprise in a horse — an animal that was selected, bred and trained to respond to the subtle cues of a human rider.
Hans was exceptionally skilled at perceiving cues from humans. He had no abstract mathematical ability at all.
This sublime perception of unconscious cues is familiar to mentalists who can do amazing feats of perception on audience members who are unaware that they are subtly signaling their thoughts.
The Clever Hans problem is well known to researchers in animal cognition, but it haunts the field. The difficulty is this: animals are capable of prodigious feats of perception — think bloodhounds who follow vanishingly subtle scents for miles — and animals are capable of high levels of training. Subtle perception reinforced with meticulous training can give the appearance of abstract thinking when in fact none is present.
An example of behavior that would suggest abstract thought would be Hans’ apparent ability to calculate dates using the concept of a calendar. If an animal were to respond correctly, given a date and day of the week, to a question about the date of an ensuing day of the week, the inference that the animal was able to abstract the concept of “calendar” from the present and calculate according to a calendar would be reasonable. But in the absence of perceptual clues no animal has ever done it, or has ever done anything like it.
The Clever Hans effect is not merely an effect caused by the ability of the animal to see the trainer — it is an effect that can taint computer and flash-card training, even when the trainer is not visible to the animal. Just as a parrot can be trained to recite the words of a complex abstract theory, an animal can be trained to group objects according to type, color, shape, use, etc. Yet there is no reason to infer that the animal understands the concepts of the categories–the animal is merely responding to perceptual clues. Color, size and shape are cues just like grimaces.
Abstract thinking is not mere memory of similarities between particulars reinforced by training. Worms can be trained to avoid or tolerate encumbrance of a particular kind — that does not mean that the concept “encumbrance” means anything to a worm.
Abstract thinking is the ability to think conceptually without particulars. There is no convincing evidence for abstract thinking in animals. Research that is claimed to show animal abstract reasoning invariably shows sublimely trained perception, not genuine abstract thought.
Animals rely on particulars to think, and their perceptual skills can be prodigious. They do not think abstractly.
Image: Clever Hans, by Karl Krall (Karl Krall, Denkende Tiere, Leipzig 1912, Tafel 2) [Public domain], via Wikimedia Commons.