Horse & Hound just dropped a piece asking whether AI could judge dressage. The comments section is exactly what you’d expect - half the community excited about objectivity, the other half horrified at the thought of machines scoring art.
But here’s what I think both sides are missing: the technology isn’t the hard part. The hard part is something dressage has avoided for decades.
The Bias Problem Is Real
Professor Inga Wolframm’s research has made this uncomfortably clear. After analyzing over 500 scores from seven CDI5* competitions, she found judges are predisposed to give higher marks to riders who share their nationality and to combinations who’ve previously performed well.
This isn’t judges being corrupt. It’s judges being human.
“The reason these biases exist is because our brains cannot compute all the information they are being confronted with,” Wolframm explained. Judges are theoretically required to consider up to seven pages of descriptors per movement - and they have seconds to do it, with no pause button.
So the brain takes shortcuts. It defaults to patterns. And those patterns favor the familiar.
What AI Could Actually See
Here’s where it gets interesting. AI systems can already:
- Track 50+ skeletal points on horse and rider in real time
- Count every flying change with precision
- Measure exactly how long a horse spends behind the vertical
- Detect asymmetries in movement invisible to the human eye
But the most compelling application isn’t scoring movements. It’s reading pain.
Research on the Equine Grimace Scale has shown that horses display consistent facial indicators when experiencing discomfort - tension around the eyes, changes in ear position, nostril shape, muzzle tension. These signals are subtle and easy to miss when you’re watching a horse blast down centerline in an extended trot.
An AI system doesn’t get distracted by flash. It doesn’t confuse tension with expression. It just measures.
The Flamboyance Problem
This is the part that keeps me up at night.
As humans, we’re drawn to spectacle. The huge movers, the dramatic extensions, the horses that make your jaw drop. And those horses often score well - even when the work isn’t technically correct.
Meanwhile, the quieter horse moving in genuine balance, with a soft topline and no flash, gets overlooked. Not because judges don’t know better, but because our brains are wired to notice what’s exciting.
AI doesn’t have that wiring. It could measure actual outline correctness - whether the horse is truly through, whether the poll is the highest point, whether the hind legs are carrying rather than pushing. None of that requires subjective interpretation. It’s geometry.
But Here’s What AI Can’t Do
Feel.
The best judges aren’t just measuring angles. They’re reading the relationship between horse and rider. They’re sensing whether the horse is offering movement freely or being manufactured into a frame. They’re weighing harmony against technical execution in ways that defy quantification.
That’s not going away. That shouldn’t go away.
What I hope for isn’t AI replacing judges. It’s AI giving judges better information - objective data they can use alongside their trained eye and feel.
Imagine a world where judges get a real-time overlay showing:
- Time spent behind the vertical
- Pain indicator flags
- Stride regularity metrics
- Straightness measurements
Not as the score - as context. Data to check against what their eye is telling them.
The Real Question
Wolframm put it perfectly: “To skip where we decide on what we find important and jump straight into AI would be premature.”
AI is only as good as the information you put in. And right now, dressage hasn’t agreed on what it’s actually trying to measure.
What does “optimally extended” really mean? Is a 7.5 versus an 8 about mechanics or impression? When we say we want “expression,” are we describing genuine freedom or trained spectacle?
The technology exists. The question is whether the sport is ready to have the harder conversation about what it actually values.
Because whatever we teach AI to measure, that’s what will get rewarded. And what gets rewarded shapes training. And training shapes the horses we breed.
The computers aren’t the scary part. The mirror they’ll hold up to our sport is.
For more on how I think about training that prioritizes correctness over flash, check out my course From Stiff to Supple in 28 Days.
