Skip to content ↓

Monkey see, one-two-three: innate number sense discovered

The chart behind Professor of Neuroscience Earl Miller in this photo-composite shows examples of the stimulus displays used to train monkeys to judge whether successive displays contained the same number of items.
The chart behind Professor of Neuroscience Earl Miller in this photo-composite shows examples of the stimulus displays used to train monkeys to judge whether successive displays contained the same number of items.

MIT scientists who taught monkeys the numbers one to five believe the effect this had on the monkeys' brain cells may shed light on how people process number concepts.

The Picower Center for Learning and Memory researchers - Professor of Neuroscience Earl K. Miller, postdoctoral fellow Andreas Nieder and postdoctoral associate David J. Freedman - published their findings in the Sept. 6 issue of Science.

"Several studies have suggested that nonverbal human infants and monkeys have basic numerical abilities that provide a foundation on which the higher numerical aptitude of adult humans is built," said Miller, associate director of the Picower Center.

"This study shows how and where these abilities are encoded in the monkey brain and, by extension, the human brain," he said. "The neural properties we find in the monkey brain may be very similar or even identical to those that provide the foundation of numerical abilities in humans."

The researchers hope that by identifying the neural networks in the brain that underlie fundamental numerical competence, they can gain insight into what makes intelligent behavior. This in turn may lead to therapies that alleviate cognitive deficits in humans, or changes in the way mathematics is taught to children.


Over seven months, Miller, Nieder and Freedman painstakingly taught a four-year-old and a five-year-old rhesus monkey the numbers one through five. Then they taught the monkeys, both males, the concept of sameness; finally, they trained them to release a lever if two visual images looked identical.

The images were gradually changed, but the monkeys were rewarded if they responded when the images had the same number of items (dots, triangles, squares, etc.).

After the monkeys learned to perform judgment tasks using one through five objects, the researchers identified the individual neurons that reacted to the different numbers of items. They found that a neuron that reacted strongly to a given number of objects also reacted to other numbers of objects, but progressively less so.

An individual neuron showed peak activity to a given number - for example, the number three. It would respond less actively to the numbers two and one. "This means that neurons showed tuning curves' whose peaks were centered on a given number. This progressive change in activity - the farther away from the neuron's preferred number, the less active the neuron - indicated that the neurons were preserving the ordinal relationship among numbers, not treating them as independent categories," Miller said.

Nieder added, "This was critical - we know that three is larger than two and less than four, and it appears that the neurons also 'knew' this."

"It was surprising to see the close correspondence between neural activity and the monkeys' behavior," said Miller. "It was exciting to see how the properties of neurons can so readily explain behavioral observations of something as high-level and cognitive as numerical ability."


Why look at how the brain processes numerical concepts? "The ability to abstract notions of quantity and number is fundamental to intelligent behavior," Miller said. "They are not only important for mathematics, but for survival in general. Social animals such as primates can make decisions like fight or flee by judging the relative number of friends vs. foes. In foraging, choosing a larger alternative can contribute to survival."

Studies of humans with brain damage and functional imaging studies of human brains have suggested that the prefrontal cortex might play a central role in numerical judgments.

Previous results from the Miller laboratory show that neurons in the prefrontal cortex represent abstract perceptual categories, and number may be regarded as an abstract category of any collection of objects. But the exact neural mechanisms underlying numerical abilities have been a mystery.

"The properties of the neurons can explain fundamental characteristics of numerical competence in both humans and monkeys," Nieder said. "Because the neurons showed similar activity to adjacent numbers (such as two and three) and progressively dissimilar activity as numbers are progressively farther apart (such as two and five), this may be how we preserve the ordinal relationship between the numbers in our own brains."

This property also may help explain other characteristics of numerical judgments, such as the numerical magnitude effect, in which people find it harder to discriminate between small differences in numbers as the numbers get larger.


A biological representation of numbers in our brain could impact how mathematics is taught in schools.

"Knowing more about how the brain processes numerical abilities and what the differences are between brains that are good at numbers and those that are not may lead to better strategies for teaching," Miller said.

There also are brain disorders such as acalculia, in which people have impaired mathematical abilities. "Knowing how the brain normally processes numbers may lead to therapies that can help alleviate such disorders," Nieder said.

In addition, this study confirms once again that many animals have rudimentary numerical competence. Even though these representations are only approximate and become increasingly imprecise for larger numbers, they may be biological precursors of humans' highly developed mathematical abilities.

This work is funded by the National Institute of Mental Health, the Human Frontiers Science Program, and the RIKEN-MIT Neuroscience Research Center.

A version of this article appeared in MIT Tech Talk on September 11, 2002.

Related Topics

More MIT News