CapuchinAI tested with wild capuchins: the station records visits and dispenses banana when the screen is touched
CapuchinAI tested with wild capuchins: the station records visits and dispenses banana when the screen is touched
On July 28, a team from Emory and Georgia Tech published a paper on a portable station for white-faced capuchins at the Taboga reserve in Costa Rica. Over two weeks, 16 animals approached the station. Ten learned to touch the screen to receive a reward, and eight learned that touching it produced a piece of banana.
A field biologist may observe a single capuchin for years but cannot always present the same task again and compare the animal’s responses. In a laboratory, the researcher selects a participant, starts the task, and records the result. CapuchinAI brings this sequence of actions into the forest: the same task, a video of every visit, and the time of participation.
The wooden enclosure contains a camera, a 15-inch touchscreen, a Raspberry Pi, and a dispenser filled with pieces of dried banana. When the camera detects a capuchin’s face, it begins recording. The animal touches the blue screen, the relay starts a motor, and the motor turns the dispenser. Each encounter produces a linked record of the video, the touch, and the time, without a person stationed at the feeder.
The pilot study tested whether free-ranging animals would return to the screen. Five of the eight capuchins retained the association between touching the screen and receiving a reward after a one-week break. These repeat visits give the authors records of animals that learned the same action, rather than isolated observations.
In the CapuchinAI paper, the authors describe a separate model trained to recognize six males from 2 160 photographs. It correctly identified the individual in 97,5% of held-out images. The field device distinguishes capuchins from other animals and records every visitor. The authors are using videos of the sixteen participants to develop a model that will link each animal to its previous responses.
Portable screens have already been used with wild primates. In a 2023 experiment, eight groups of vervet monkeys completed an associative learning task. CapuchinAI adds a camera and records each visitor. The next version of the station will use these records to recognize a specific capuchin and present the next stage of its task.