CapuchinAI 1.0: Development of a machine learning-based touchscreen paradigm to test cognition in wild capuchins

Area of Science:

• Primate Cognition
• Artificial Intelligence in Ecology
• Field Methodology

Background:

• Studying wild primate cognition requires balancing ecological validity with experimental control.
• Traditional methods often lack the precision and scalability needed for in-depth field research.

Purpose of the Study:

• To introduce CapuchinAI, a novel field-deployable system for automated cognitive testing in wild primates.
• To demonstrate the feasibility of integrating AI, facial recognition, and touchscreen technology for ecological cognition studies.

Main Methods:

• Developed a system combining a YOLOv7 facial recognition model with a Raspberry Pi-driven touchscreen and reward apparatus.
• Deployed the system in the Taboga Forest Reserve for two weeks with habituated white-faced capuchins (Cebus imitator).
• Enabled autonomous task administration, data collection, and reward delivery contingent on performance.

Main Results:

• 16 wild capuchins interacted with the apparatus, with 10 learning to trigger rewards.
• 8 individuals formed and retained robust screen-reward associations, indicating rapid learning.
• The system demonstrated high facial recognition accuracy (>97%) and feasibility for autonomous operation.

Conclusions:

• CapuchinAI successfully bridges the gap between laboratory experimental paradigms and wild primate cognition research.
• The methodology allows for autonomous, standardized, and scalable cognitive testing in natural habitats.
• This approach facilitates the study of individual cognitive variation in wild populations, paving the way for comparative research.