Date: November 7, 2025

Speaker: Kristin Branson, Senior Group Leader at HHMI's Janelia Research Campus

Title: How can generative AI help us understand animal behavior

 A color photo of a woman smiling for a photo.

Abstract: The potential of big data and machine learning has generated tremendous excitement for data-driven biology research. Supervised machine learning has delivered critical progress in processing raw experimental data, for example video-based tracking of behaving animals. These advances have created a new bottleneck: the resulting datasets remain too large and complex for biologists to effectively analyze and interpret. Current approaches rely on hand-crafted features and dimensionality reduction for data visualization, which inevitably discard structure. In this talk, I will discuss an ambitious approach buoyed by the success of large language models in learning about the world through next-token prediction. With the goal of learning about animals' behavioral policies, we train large models to predict an animal's next movement. We show that, like LLMs, these models can autoregressively generate behaviors over longer periods, producing movements resembling those of real animals. However, unlike LLMs, which are directly useful as tools, a black-box model of animal behavior offers little value on its own. The critical second step is applying mechanistic interpretability and explainable AI techniques to understand animals' behavioral policies: What do animals attend to when? How is the external world represented internally? And what decisions and behaviors do these lead to?

Bio: Kristin Branson is a Senior Group Leader at HHMI's Janelia Research Campus, where she develops machine vision and learning paradigms to solve problems in practice, and help us gain scientific understanding of natural phenomena. She helped pioneer the use of machine vision and learning in animal behavior analysis. She is originally from San Diego, California, and has a Bachelors in Computer Science from Harvard (2000), a Ph.D. in Computer Science from UC San Diego (2007), performed her postdoctoral work at Caltech (2007-2010), and has been a Group Leader at Janelia since 2010.