Scholars: Scientific Outcomes from AI Tools and Models

Columbia University, New York NY

Investigators

Abstract

The research in this project investigates how scientists’ decisions – ranging from the type of AI tools and models to use, to how to train models and label behaviors – can shape what neuroscientists come to know about animal minds. Drawing on philosophy of science, neuroscience, and science and technology studies, the project analyzes how decisions about data, software design, and interpretation affect scientific outcomes. The study’s overarching goal is to improve how behavioral tracking tools are used, thereby making research more thoughtful and effective. The findings of this project will be of interest to scientists, educators, designers and users of AI. This project conducts a comparative assessment of several widely used AI-based behavioral tracking tools to investigate how different algorithmic approaches shape the study of behavior. The tools are built on distinct algorithmic foundations, ranging from supervised learning techniques that track visible features, such as body pose, to unsupervised models that infer behavioral patterns over time. Because many of these tools rely on machine learning, the project contributes to a better understanding of how artificial intelligence is now shaping scientific inquiry. The research examines how differences in these technical methods influence how behavior is organized, categorized, and interpreted. It addresses how behavioral knowledge is developed and defined, in turn offering pathways for developing more transparent, flexible, and representative research across species and settings. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.