'Gestures Are Language': KAIST Develops AI That Deciphers Animal Behavior

A research team led by Professor Kim Dae-soo of the Department of Brain and Cognitive Sciences at the Korea Advanced Institute of Science and Technology (KAIST) has developed "BehavBERT," an artificial intelligence model that reads and interprets animal movements as if they were a language.

The research team used a method of converting skeletal coordinates of body parts—such as a mouse's nose, ears, and spine—into "tokens," which correspond to words in natural language, for the AI to learn.

As a result, BehavBERT went beyond simply classifying behaviors; it learned the meaning of those behaviors over time on its own.

The model also demonstrated interpretability, informing researchers which specific behaviors it focused on to reach its conclusions.

Without any prior knowledge, the model successfully identified key social behavioral deficits in mice models of autism.

Professor Kim Dae-soo stated that BehavBERT is expected to become a core research tool that leads to new discoveries in various life science fields, including drug development, mental illness research, and behavioral genetics.

(Photo: Courtesy of KAIST, Yonhap News)
※ Please note: This article was translated by AI and may contain errors.