Researchers at KAIST have created an artificial intelligence system that learns and analyzes animal behavior much like it processes human language. The model was able to independently detect social behavior problems in a mouse model of autism, according to the research team.

Key takeaways

  • An AI model from KAIST processes animal movement sequences as if they were language tokens.
  • The system identified social behavior deficits in an autism mouse model without being specifically trained to look for them.
  • This approach may allow researchers to study brain disorders in a more objective and interpretable way.

Understanding animal behavior as language

Traditionally, studying animal behavior in a lab involves researchers watching videos and manually coding actions. This process is slow, subjective, and can miss subtle patterns. The KAIST team took a different approach. They treated sequences of mouse movements, like sniffing, grooming, or approaching another mouse, as if they were words in a sentence.

The AI model learned the structure of these behavioral “sentences” by identifying which actions tend to follow others. This is similar to how large language models learn grammar and meaning from text. Once trained on typical mouse behavior, the model could spot unusual patterns.

Spotting autism related social deficits

In tests, the AI was shown videos of mice that had genetic changes linked to autism. The model flagged specific social behaviors in these mice as different from the typical patterns it had learned. For example, it detected that the autism model mice spent less time in certain social interactions and had unusual sequences of movements during encounters with other mice.

Importantly, the system was not told what autism related behaviors look like. It learned what “normal” behavior was from the data and then flagged deviations. The researchers report that this approach could provide a more objective and quantifiable way to measure behavioral symptoms in animal models of neurodevelopmental disorders.

Implications for neuroscience research

The ability to translate complex behavior into readable patterns could change how scientists study conditions like autism, depression, or schizophrenia in animal models. Currently, behavioral tests can be labor intensive and may not capture the full range of symptoms.

By using an AI that understands the “grammar” of behavior, researchers might detect subtle changes that human observers could miss. The team at KAIST suggests this method could be extended to other animal species and used to screen potential treatments for brain disorders.

Frequently Asked Questions

How does the AI model “read” mouse behavior?

The model breaks down continuous video footage of mouse movements into discrete actions, such as sniffing, approaching, or grooming. It treats these actions as tokens, similar to words in a sentence. By analyzing how these tokens are sequenced, the AI learns behavioral patterns and can identify when a sequence is unusual.

Can this AI be used for human behavior analysis?

The study focused on mouse behavior, but the underlying method might be adaptable. The researchers note that the principles of analyzing movement sequences as language could apply to human behavior in controlled settings. However, human behavior is far more complex, and more research is needed before such tools could be used in clinical assessments.

What makes this different from other AI behavior analysis tools?

Other AI tools may classify specific actions but do not learn the overall structure of behavior. The KAIST model is unique because it learns the relationships between actions, treating behavior as a language with rules and context. This allows it to detect subtle deviations without needing a predefined list of what to look for.

This is an original report by Vital Signs Today, informed by reporting from Medical Xpress. Read the original source.

This article is for information only and is not medical advice. See our Medical Disclaimer.