WHEN AI PRESSES the touch-sensitive computer screen, the Arabic numerals 1, 3, 5, 7 and 9 pop up in random order. She correctly taps the numbers in ascending order, earning a handful of raisins as a reward. Ai, a chimpanzee, also reveals a lightning-fast short-term memory for numbers. When white boxes mask the numerals right after each is displayed, she still strikes the boxes accurately—unlike some of the human volunteers who take the test.
Ai's videotaped demonstration, at a recent conference, is the culmination of two decades of work by Japanese researcher Tetsuro Matsuzawa of Kyoto University. Matsuzawa has been training and observing Ai since 1978. Ai proved to be a star pupil, and she and Matsuzawa get together almost daily for a mixture of play and research. The 30-year-old chimp participates enthusiastically—as does her six-year-old son, Ayumu.
Humans can easily memorize small chunks of information, such as strings of numbers, a phenomenon that psychologist George A. Miller wrote about in his 1956 landmark paper, “The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information.” How might our closest primate relatives develop an understanding of numbers—and what might that process tell us about our own abilities to grasp such abstract concepts? Matsuzawa has sought answers in his work with Ai.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Scientists have long known that in the wild, primates can count to at least three: a single chimp or a pair will always flee if they encounter an unfamiliar chimp while hunting. Only a chimp in a group of at least three members is willing to pick a fight. But coping with larger numbers, represented with abstract numerals, is a far more challenging cognitive task—one that, until now, has been considered the exclusive purview of people.
At the conference, Ai's videotaped efforts on number-order tests seems unbelievable to the delighted human audience. In the masking trials, she excels: the scientists watching at the conference are barely able to note more than one or two of the numbers before all of them are covered up. In one video, a human tries his hand at the test with the white squares and scores only a few hits. But the ape, working “blindly,” almost always types a sequence of four correctly and even handles five numbers with some accuracy. Another video shows Ai, just after a test begins, becoming distracted by the noise of other chimps fighting. After pausing for 20 seconds to listen, she returns to the screen and then calmly notes and types in all the masked numbers in the correct order.
Math Classes
Behind this steady performance are many years of patient training. First, Matsuzawa taught Ai the meaning of the cardinal number—the correspondence between each number and a quantity of objects. Ai learned to associate a quantity of one to nine objects with the corresponding numeral appearing on the screen. Next, Ai achieved a certain understanding of numerical order, gradually learning that the numbers rise from 0 through 9. Once Ai had grasped that 0 comes before 1 and not the other way around, she was presented with a 2—and so on, one by one, up to 9.
Ai thus can count, but only in a limited sense. She has never appeared to develop the abstract sense of generalization that human children, even at the tender age of three, are beginning to grasp—that for each number that exists there is another, larger one: one plus one is two, two plus one is three, and so on. Ai has only learned to associate each Arabic numeral with a corresponding quantity of objects and to put these numbers in order.
How can Ai, who needed many years of careful teaching to learn the numbers, be so much faster than humans at sorting when the digits appear in random order on the screen? Typing them in, in the correct order, after they flash on the screen one after another for a fraction of a second each, seems to be child's play to her—although for the audience at the conference, it proved challenging to do.
Matsuzawa believes that numerical abilities in humans came at the expense of other abilities in a kind of evolutionary trade-off—one between having the skill to quickly note things, on the one hand, and developing more complex cognitive abilities, on the other. For apes in the wild, it is a matter of life and death to take in a number of stimuli simultaneously and arrange them in the right order: over there lurks a leopard, above another chimp swings through the trees, and beneath hangs a piece of ripe fruit. We end up “paying” for our human abilities to grasp abstract quantitative concepts—which are far beyond those of apes—with greatly reduced perceptual abilities.
Human brains are better than those of other primates in artificial environments that are rich in social signals. For people, the focus is on correctly interpreting social signals, avoiding misunderstandings in speech, and making careful calculations in response to any given situation, rather than merely reacting immediately to all the objects we see.
