AI learns to read horses’ facial expressions – new study explores emotion recognition in equines.
An open-access study published in PLOS ONE has shown that artificial intelligence can begin to distinguish between different emotional states in horses simply by analysing their facial expressions.
Using video footage collected by Ricci-Bonot and Professor Daniel Mills during controlled feeding scenarios, researchers tested whether AI could recognise four states: baseline, positive anticipation, frustration and disappointment. The team compared two approaches: a deep-learning model that analysed short video clips and a machine-learning model based on EquiFACS, the recognised system for coding equine facial movements.
The video-based AI performed best, correctly identifying emotional states with 76% accuracy, while the EquiFACS-based model achieved 69%. Both models found it most challenging to distinguish between anticipation and frustration, reflecting how subtle these states can appear even to trained human observers.
While not a ready-made tool for everyday use, this research marks an important step forward in automated emotion recognition, supporting future developments in welfare monitoring and the study of positive and negative equine emotional states.
Read the full open-access paper here: https://doi.org/10.1371/journal.pone.0302893
(Correction published February 2025: https://doi.org/10.1371/journal.pone.0319501)
