In The News: Using AI to Police Horse Welfare

In a 2024 study, researchers used machine learning models to predict emotional states in horses — an approach with potentially far-reaching consequences.

_{Image by Ruth Weitz from Pixabay}

Animal welfare is an increasingly hot topic, and equestrian sports are no exception. Numerous horse-riding practices are currently criticized for being detrimental to equine well-being. The policing of horse welfare at events is also under intense scrutiny, especially in light of the multitude of scandals at last year’s Olympic Games. But what can be done?

The team of Phelipon, Lansade, and Razzaq believes AI could be another valuable tool in the toolbox for event officials.

In their study, the researchers used three indicators of horse well-being:

• Tail swishing
• Head behind the vertical posture
• Certain facial expressions, such as the opening of the eyes or mouth, and the position of the ears

The dataset consisted of 1,036 images of horses, divided into two classes: comfortable (546) and uncomfortable (490). A doctoral student in ethology specializing in the identification of horse emotions, along with a technician in ethology, categorized the images using the three factors listed above to train the computer models.

Next was a substantial amount of computer science and mathematics that, to be completely honest, I don’t fully comprehend. However, the outcome is a set of images and datasets classifying each horse studied.

The study concludes with what the researchers identify as its limitations — most of which revolve around the lack of a more refined computer program and publicly available photos.

As a lifelong horse person, I have to say this study has more limitations than just the number of images fed into the model — and honestly, it could even be outright dangerous.

I’m all for improved animal welfare. I didn’t sugarcoat my opinion about Charlotte Dujardin.

But here’s the thing… I also coach a competitive youth robotics team. We compete in the FIRST Tech Challenge, and the team is very good. We placed second in our division at the World Championship in April. Our robot uses machine learning and AI vision systems to complete tasks during matches. And guess what? The models make a lot of mistakes. The computer messes up a lot.

This year’s game involved picking up red and blue rectangular blocks off the ground. At the championship, the overhead lighting made the blue blocks so dark — and a nearby bare aluminum structure so blindingly reflective — that our camera had to be recalibrated every single morning. Even then, it still made errors.

So now we’re supposed to believe these AI vision systems can be hand-tuned by a programmer at every horse event, every day, multiple times a day? Will the afternoon horses be judged more harshly simply because shadows fall across their eyes and ears at odd angles relative to the camera?

And that’s just the tech side. What about the fiery red chestnuts? The boss mares? My buckskin mare pins her ears almost the entire time I ride her — especially during polo. That’s just her GIT ‘ER DONE GAME FACE.

_{Image by tcdeveloper from Pixabay}

It’s also worth noting that emotions can be — and often are — fleeting. I get mad sometimes, especially at horse events. I’m hot. I’m sweaty. I’m tired. I’m hungry. I’m thirsty. I have to pee. Our horses experience all those same feelings — and probably more. So yes, they’re likely more tense. Yes, they get annoyed with us sometimes. And that probably shows in their expressions and body language.

That’s why, to me, the idea of being flagged by a computer for “abuse” based on a single photo — without any context — is horrifying.

What do you think, Horse Nation?

Go riding.

Amanda Uechi Ronan is an equestrian, author, and wannabe race car driver. Follow her on Instagram @au_ronan.