Eight wild capuchins in Costa Rica have learned to tap an artificial intelligence-powered touchscreen to earn pieces of dried banana. Researchers set up a wooden box in the rainforest, fitted with a touchscreen and a camera-linked food dispenser, and watched to see how the monkeys would respond.
The device, called CapuchinAI, was built by teams from Emory University and the Georgia Institute of Technology. They wanted to find out whether wild primates would learn to use a machine on their own, and whether those interactions could generate useful data without taking the animals out of their environment. After two weeks in the Taboga Forest Reserve, at least some of the monkeys had shown they were eager to try.
Sixteen capuchins approached the box, ten triggered a food reward, and eight figured out that touching the blue-lit screen meant a banana treat. The system’s AI, based on the YOLOv7 model and called MultipleCapuchins, could identify individual monkeys with over 97% accuracy using facial features. The device does not yet run fully personalized tasks, but it can now automate cognitive testing for wild animals in the field.
Unlike traditional animal cognition studies, which often require capturing animals and moving them to labs, CapuchinAI lets the monkeys remain in their natural setting. They choose whether to participate, and when or how often to return. That preserves their social dynamics and daily routines, which are hard to replicate in captivity.
The first experiment was simple: when the camera detected a capuchin, the screen displayed a blue shape. A single touch released a slice of dried forest banana. For two days, the box sat untouched. Then a male named Trompudo broke the silence, hitting the box before eventually touching the screen. The food reward drew in other monkeys, and soon several were experimenting with the device.
The monkeys reacted in different ways. Some quickly understood cause and effect, while others watched from a distance before trying it themselves. A few even used their mouths on the touchscreen, showing their flexibility and the unpredictable ways animals learn. Automated field tests can reveal what a species can do, and how individuals differ in curiosity and problem-solving.
The study’s authors point out that success with the touchscreen does not mean a monkey is more intelligent overall. A monkey’s social rank and hunger can affect participation; its past experience also matters, as does its willingness to approach new objects. The technology is a tool for collecting data, not a definitive measure of intelligence.
CapuchinAI’s facial recognition already works well, but in this pilot it mainly confirmed that a capuchin, rather than another animal, was present. The next step is to automate personalized tasks, giving each monkey challenges suited to its history and skill level. Future tests will examine memory and impulse control, while also testing cognitive flexibility beyond the simple touch-for-food setup.
This approach is part of a wider trend in wildlife research: using technology to reduce human interference and improve data quality. A similar idea is behind the return of large mammals to Chernobyl’s exclusion zone, where the absence of people, rather than radiation, has allowed wildlife to thrive.
For now, CapuchinAI’s biggest achievement is practical: it shows that wild primates can take part in controlled experiments without leaving the forest. The device’s software can also limit rewards per animal, preventing dominant monkeys from taking over. Still, the researchers say no machine can replace insights gained from long-term fieldwork and close observation of social behavior.
Scaling up means involving more animals and running longer studies, with tasks designed to test primate cognition. For now, a handful of capuchins and a wooden box suggest that animal cognition research can take place in the forest as well as the lab.
