Can AI judge dressage? The technology exists – but is the sport ready for it?
As the sport wrestles with questions of fairness, welfare and public trust, Oscar Williams explores whether AI could help judges and whether it may force dressage to confront deeper questions in this exclusive article for H&H subscribers
Dressage judging is a hell of a job. Judges are asked to do the near-impossible: absorb dozens of movements in real time, weigh technical correctness against harmony and expression as well as several other criteria, and distil it all into marks awarded in seconds before moving on to the next movement.
To add to the pressure, those performances are often dissected online with the benefit of slow-motion replays and freeze-frame screenshots – and even then, the comment sections rarely agree on what the score should have been.
Now, as the sport wrestles with questions of welfare, credibility and public trust, artificial intelligence (AI) judging is increasingly being pushed forward as a possible solution.
Supporters argue it could ease the cognitive load on judges and bring greater consistency and transparency to scoring, while identifying conflict behaviours or technical faults. Sceptics worry that a sport built on feel, nuance and interpretation risks being flattened into data.
AI could track every stride a horse takes. But could a machine really judge dressage?
The science of judging bias
Dr Inga Wolframm, professor of sustainable equestrianism at Van Hall Larenstein University of Applied Sciences in the Netherlands, has spent years studying the cognitive demands placed on judges. She is also working with the German AI technology company EQAD to develop tools designed to support more objective decision-making in horse sport.
Under the FEI judging handbook, a single movement can theoretically require judges to consider multiple pages of descriptors before awarding a score – something Inga argues is simply unrealistic within the time available.
In research analysing more than 500 scores from seven CDI5* competitions, she found officials were more likely to award higher marks not only to riders from their own country, but also to those from the countries of their fellow judges – patterns she describes as a “bias cascade”, where small advantages accumulate over time to influence rankings, starting order and ultimately future scores. Rather than a single source of bias, she argues, it operates as a wider ecosystem of interacting effects.
But she stresses this isn’t a failing of individual judges; rather, it’s a natural consequence of the brain searching for shortcuts when confronted with more information than it can realistically process.
“At the end of the day, however hard you try, human beings cannot avoid bias because our brains simply cannot process all the information,” she says. “It feels as though we can see everything – you look at a horse and think: I can see the horse, what’s the problem? But in reality, you only see the areas you happen to focus on.”
Time pressure only intensifies that problem.
“You’re thinking: is this a 6.5 or a 7, a 7.5 or an 8? And it’s a rider you saw two weeks ago who won a competition – what are you going to do? You only have seconds to decide because the next movement is just around the corner. So, you’re going to give an 8,” Inga says.
“If it happens once, it’s not a problem. But with 33 movements – and 11 of them counting double – those small decisions quickly add up to bias.”
Her research has tried to understand how judges cope with this information overload using eye-tracking technology. Inga pulls up a presentation summarising some of her results: using Tobii eye-tracking systems, it maps the gaze patterns of 20 judges of varying experience as they watch grand prix tests on video.
Some focus on the hindquarters, others on the rider or the front of the horse. But they’re all looking at different things – each taking in only part of the picture rather than the whole.
“There’s so much to see, and so little time, that judges have to choose,” Inga explains. “What they focus on may come from experience or habit. But everyone looks at something slightly different – not because they’re incompetent at all, but because they’re human.”
Under the FEI system, judges aim to score movements objectively against directives such as rhythm, elasticity and balance, converting that judgement into marks on a 0-10 scale. But applying those principles to a moving horse in real time inevitably involves interpretation – subjectivity is built in.
“On the one hand, we say we accept that judges see different things,” Inga says. “But in reality, we don’t accept it at all. Social media explodes the moment judges give different scores.
“To me, there are only two choices. Either we accept that judges all see different things and celebrate the expertise they bring – or we decide we want an element of complete objectivity, and then we need technology to help us.”
What AI can already measure
Any mention of AI interventions in sport can understandably get people’s backs up. But most proposals do not involve replacing judges; rather, measuring what happens in the arena in far greater detail than the human eye can manage in real time.
Dr David Stickland, a research physicist and co-founder of Global Equestrian Technology (GET), says the first step is simply tracking how the horse moves. Using modern video analysis, their system can already track more than 50 skeletal points on the horse and rider, capturing their position many times per second.
“That enables you to analyse motion: the rhythm, the time between different feet hitting the ground, behind-the-vertical angles, how engaged the hind end is, whether the horse is uphill or downhill,” he says.
It doesn’t judge in the human sense. It measures – where limbs are positioned, how quickly they move, the angles of joints and body positions. From those measurements, the software can begin to recognise patterns that correspond to the movements judges are already trying to evaluate.
“One big criticism of AI is that the computer says something or produces a score and we don’t know why,” says Eddy Schuurmans, a dressage rider and breeder himself, as well as a theoretical-experimental psychologist and co-founder of GET.
But any AI system has to be taught first – and that process carries its own risk. AI models learn by analysing large datasets and identifying patterns. In dressage, that would likely mean training software on thousands of historical performances and the scores awarded by judges.
The system could simply learn to replicate the same biases researchers are already trying to address – reproducing past judging patterns rather than questioning them.
“That criticism can be reduced by identifying the observables first, and only then translating those observables into points.”
That’s why the earliest applications are likely to focus on areas where the answer is unambiguous – where there is no historical bias baked in.
“The initial goal wouldn’t be for the computer to say ‘this is a 7.5 or a 7.3’, but to give the judge more information while leaving the interpretation to the human,” David says. “It’s easy to miss a glitch in the changes. We can nail those things today — we can say there were exactly 15 changes, that one was a double, or the hindlegs were stuck behind on two of them.”
The problem AI technology can’t solve alone
For more in-depth use, even the most sophisticated technology runs into a deeper problem. AI can measure movement with remarkable precision – but before it can do that meaningfully, the sport must decide exactly what it wants to measure.
“AI is only as good as the stuff that you put in,” Inga says. “If we don’t have a clear understanding of what it is that we’re looking at, AI can’t solve it for us.”
In other words, technology cannot compensate for vague definitions in the rulebook.
“There’s structure to the directives we have, but we need to understand: how do we test for this? What does ‘optimally extended’ actually mean? And what’s the difference between that and ‘extends very well’?
“You need to define those things – if we can’t separate what it is we want to assess, how can we possibly be accurate and objective?”
That is why, for Inga, a code of points matters so much.
What is a code of points?
A code of points is a scoring framework used in sports such as gymnastics, whereby a performance is broken down into clearly defined technical elements – each awarded a specific value – rather than being assessed as a whole.
In artistic gymnastics, the World Gymnastics has introduced an AI-assisted judging support system using multiple cameras to build a 3D model of an athlete’s routine, measuring angles and identifying elements as a judging aid rather than replacing human officials.
Gymnastics codes of points even include diagrams illustrating exactly what a correct element looks like – giving AI systems something concrete and unambiguous to measure against.
“Before you can apply AI, you need to determine what it is you actually want to measure,” she says. “Once you know what you’re analysing, then you can decide which elements technology can measure and which elements should remain with the human judge.”
One attempt to create that kind of framework already exists in dressage. In 2018, the FEI Dressage Judges Working Group developed a proposed code of points, designed to break movements down into clearer technical elements – but it has never been put to use. An FEI spokesperson confirmed it remains under consideration, but with no implementation timeline in place.
Correctness or spectacle – what does dressage actually want to reward?
Judging has always balanced two elements: technical correctness and aesthetic impression. The directives in the FEI rulebook emphasise rhythm, suppleness, contact and impulsion – principles rooted in the training scale. Yet the horses that capture attention at the highest levels of dressage are often those with the most extravagant natural movement. Inga believes more objective measurement could help rebalance that equation.
“If I don’t have a grand prix horse and I can only spend ten thousand on a horse, my horse might not move as spectacularly,” she says. “But if I can demonstrate that what I’m doing with that horse is correct – that the rhythm is there, the contact is there – then surely that should be rewarded.”
In theory, that could level part of the playing field. But it raises a harder question: is objective correctness actually what the sport wants to reward? Dressage has always been both sport and spectacle. The extravagant movement and charisma of horses like Totilas are a large part of what has drawn audiences to the arena.
A judging system that measured correctness more precisely might reinforce the principles of the training scale – but it could also reshape the type of horse and performance that succeeds. Because what judges reward does not only determine placings. It shapes how riders train their horses – and, ultimately, the horses that are bred for the sport.
This is where the debate around AI intersects with the sport’s social licence to operate. Supporters of technological judging argue that objective data could help enforce clearer welfare standards – setting defined thresholds for things like prolonged conflict behaviours or horses being behind the vertical.
But the familiar question resurfaces: what counts as unacceptable? If technology can calculate exactly how long a horse spends behind the vertical, someone still has to decide where the line is drawn.
How soon could AI dressage judging happen?
So, how close is all of this, really? Pose-detection systems capable of tracking skeletal points and extracting detailed movement data already exist, and are widely used in sports analysis and biomechanics – from analysing running technique in athletics to tracking joint movement in physiotherapy.
David and Eddy are confident the technology can be adapted to the complexity of a dressage arena, though challenges remain, including testing it in real competition conditions and the cost of implementation.
“We’ve demonstrated proofs of principle,” David says. “What is missing is an organisation that decides: let’s make this happen.”
Eddy believes that when those obstacles are overcome, the technology will arrive gradually rather than all at once. “If you look at self-driving cars, everybody thought we would have them quickly. Instead, we got assistance systems first. You can see judging in the same way.”
He pauses. “But for me, it’s not a matter of whether it will come or not. The question is when – and by whom it will be controlled.”
That question of control points to something deeper than technology. The debate about AI in dressage judging is not, in the end, really about AI at all. Systems capable of tracking movement, measuring angles and analysing biomechanics already exist. What remains unresolved is what the sport actually wants those measurements to represent.
“I don’t want to replace judges with AI,” Inga says. “I see it as a tool that helps formalise decisions – an objective baseline while judges still assess harmony, expression and the overall impression.”
It’s a reasonable vision.
Before the sport decides whether a machine can judge the horse, it may first have to decide what judging is meant to measure in the first place.
- To stay up to date with all the breaking equestrian news throughout the year, subscribe to the Horse & Hound website
You may also be interested in:
How AI is changing the game in what we know about our horses
Anna Ross: ‘Everyone’s watching now, and willingness to question ourselves has to be part of the deal’
Richard Davison: ‘Dressage needs to implement the code of points’
‘Judges are being asked to do the impossible’: research into bias in dressage
Is dressage judging getting tougher? The data behind the debate
Subscribe to Horse & Hound magazine today – and enjoy unlimited website access all year round
Oscar joined Horse & Hound in October 2023 and is the magazine’s dressage editor and sports manager, overseeing coverage of equestrian sport. After studying equine science at Myerscough College, he spent four years working for leading dressage rider Emile Faurie, competing at the 2015 National Dressage Championships and travelling with the yard to CDIs including Aachen and Saumur. He holds a master’s degree in Literature from York St John University (2021).