Embedded Files
AI Gone Wild: The Present and Future of Machine Learning Methods to Study Primate Cognition and Behavior in the Field
AI Gone Wild: The Present and Future of Machine Learning Methods to Study Primate Cognition and Behavior in the Field
Federico Sánchez Vargas1, Sai Rakshith Potluri2, Dora Biro3, and Marcela Benítez1
1 Department of Anthropology, Emory University, 2 Department of Computer Science, Georgia Institute of Technology, 3 Department of Biology, University of Rochester
Lab experimentation has rendered great insights into primates’ cognitive abilities, yet falls short of demonstrating how these highly intelligent animals perceive, understand, and choose to act in the environments they evolved in. The key challenge thus far is that while behavior is directly observable, the cognitive abilities and operations behind it are not. Up until now, technical limitations have prevented us from directly testing for cognitive abilities in the wild — but emerging machine vision (AI) models now promise to allow us to investigate, for the very first time, how primates’ minds work in their natural settings. How might such technology be leveraged to advance the study of cognition in the wild? Here, I will present on various ways in which machine learning can be and already is being used to automate or facilitate processes such as behavior coding, body pose estimation, and image classification, among others. I will specifically focus on two ongoing projects: 1) the use of DeepLabCut, a user-friendly and rapidly trainable model for marker-less pose estimation, and Anipose to track the gaze of multiple common marmoset monkeys (Callithrix jacchus) in a 3-dimensional space participating in unconstrained, naturalistic foraging; and 2) the development of a novel facial recognition model to be implemented at cognitive touchscreen testing platforms in a wild population of wild-face capuchins (Cebus imitator) that will allow for the collection of multi-timepoint data on cognitive performance across various individuals from multiple groups. Crucially, this data can be collected even in the absence of a researcher, as the model will automate monkey ID, individualized testing administration, and reward dispensation. Finally, I will discuss crucial future directions for machine learning research into wild primate cognition, highlighting its promise for more effective and standardized protocols for interspecific comparative work.
Lab experimentation has rendered great insights into primates’ cognitive abilities, yet falls short of demonstrating how these highly intelligent animals perceive, understand, and choose to act in the environments they evolved in. The key challenge thus far is that while behavior is directly observable, the cognitive abilities and operations behind it are not. Up until now, technical limitations have prevented us from directly testing for cognitive abilities in the wild — but emerging machine vision (AI) models now promise to allow us to investigate, for the very first time, how primates’ minds work in their natural settings. How might such technology be leveraged to advance the study of cognition in the wild? Here, I will present on various ways in which machine learning can be and already is being used to automate or facilitate processes such as behavior coding, body pose estimation, and image classification, among others. I will specifically focus on two ongoing projects: 1) the use of DeepLabCut, a user-friendly and rapidly trainable model for marker-less pose estimation, and Anipose to track the gaze of multiple common marmoset monkeys (Callithrix jacchus) in a 3-dimensional space participating in unconstrained, naturalistic foraging; and 2) the development of a novel facial recognition model to be implemented at cognitive touchscreen testing platforms in a wild population of wild-face capuchins (Cebus imitator) that will allow for the collection of multi-timepoint data on cognitive performance across various individuals from multiple groups. Crucially, this data can be collected even in the absence of a researcher, as the model will automate monkey ID, individualized testing administration, and reward dispensation. Finally, I will discuss crucial future directions for machine learning research into wild primate cognition, highlighting its promise for more effective and standardized protocols for interspecific comparative work.
Page updated
Google Sites
Report abuse