Equestrian art meets artificial intelligence: dressage judges are supported

Since Heike Kemmer no longer gallops into the dressage arena with her horse, but increasingly distributes grades herself behind the judges’ table, the 2008 team Olympic champion has not only changed her perspective externally. “As a dressage judge you have to pay attention to an incredible number of things,” she says. “One blink of an eye and you’ve missed something.” For example, if the horse only makes 13 instead of 15 flying canter changes. If you as a judge forget to count, at the same time you make sure that the horse doesn’t throw its rump back and forth, but stays upright, and that the rider sits elegantly and as motionlessly as possible in the saddle, so that it looks as if everything is happening on its own (which of course it is). doesn’t) – he can start to sweat.

Each judge in Grand Prix tests has to award around 40 individual marks in around six minutes – and sometimes 30 times in one day with only short breaks, depending on the size of the starting field. An English study claims to have found that people, even if they understand enough about dressage, are chronically overwhelmed because it is almost impossible to keep track of all the criteria used to evaluate the dance between rider and horse.

Because it should be a dance, lively, harmonious, detached from all earthly heaviness; Horsemanship, as it is often called. And awarding grades for art is of course a difficult challenge. Especially since not only the horses and the people in the saddle, but also the tastes and sensitivities in the five judges’ houses vary. One had too much coffee in the morning, the second drank too much beer at the equestrian ball, the third is longing for his cigarette. Not to mention human sympathies, respect for big names, animosities and occasionally solid business interests. Dressage sport is a million-dollar business, young four-legged talents are traded for seven-figure sums. High judge scores have a concrete impact on the horse owner’s account.

It seems like a good idea to use artificial intelligence (AI), which provides objective data, to relieve judges of some of their work. This has been in the works for several years. The Black Horse One company based in Vaterstetten, Bavaria, founded by computer scientist Daniel Göhlen, has developed a program that registers technical details and can evaluate the overall precision of a ride. For this purpose, several cameras are used at the same time, the results of which are combined.

Two systems have already been tried out at the CHIO in Aachen and the World Cup tournament in London

The AI ​​has already been tried out at the CHIO Aachen and last weekend, with another company’s program, at the World Cup tournament in London. The rider’s path is closely followed: Does he ride correctly from point to point, and doesn’t he cut corners? Does the horse really step on the spot during the piaffe or does it cheat its way forward a few steps? During the canter pirouette, does it turn “on the plate”, as it is said among riders, or in a generous circle, which is of course easier. The World Equestrian Federation (FEI) also promotes the idea of ​​adding an additional grade of “precision” to the overall assessment.

Those affected – riders, judges, trainers – are positive about the idea. “This is not an indictment for the judges, but modern technology can help to reduce complexity wherever grades have to be awarded in quick succession. The judges can then concentrate on what the AI ​​cannot depict,” says Klaus Röser, chairman of the dressage committee of the German Olympic Equestrian Committee (DOKR). Isabell Werth, eight-time dressuro Olympic champion, thinks: “We can objectify the discussion if the AI ​​shows technical inaccuracies. Then no one can turn a blind eye to it anymore. The AI ​​can contribute to the transparency of the grades, I think that’s a good thing.”

An AI could detect some irregularities, including constantly pricking spores

Dressage national trainer Monica Theodorescu can also imagine the use of AI in technical details, but “the question arises as to how the AI ​​data will subsequently be incorporated into the result”. The AI ​​could of course do much more than count kicks and check lines, at least if you “feed” it accordingly. For example, you could record how often the horse’s nose line falls behind the vertical line – contrary to the rules. This should be on, or better in front of, the vertical to enable the horse to move upwards. The AI ​​could also check how far the horse strides and how actively it uses its hind leg. It could show the rider’s seating errors: the raised heel, which seriously disrupts the overall picture, constantly pricking spurs, restless hands and legs, the throwing of the upper body during canter changes or the infamous “reclining seat”, in which the rider leans against the back of an imaginary chair seems to support. The judges can notice all of this, but it is often not reflected in the grades. But those affected shy away from this step.

In addition to purely technical data, the judge should have control over the overall assessment, according to the unanimous opinion. “You can’t put a template over rider and horse,” warns Monica Theodorescu. One horse is rather small at 1.60 meters high at the withers, the other is a giant at more than 1.90 meters, like Sheldon Cooper, who won the Louisdor Prize for young Grand Prix horses in the Frankfurt under Carina Harnisch at the weekend Festhalle won. A horse like this naturally needs more space, for example in the traversals, than its more delicate competitor.

The fear that people at the judge’s table could one day be replaced by artificial colleagues is not yet widespread, but many people believe that caution is warranted. “Of course we cannot replace the principles of our education and judge training,” says Monica Theodorescu. “No AI can judge the overall picture, the smoothness and harmony,” says Heike Kemmer, who is currently undergoing a series of judge training courses. Computer scientist Daniel Göhlen has developed an app in which anyone, whether a dressage expert, a walker or no rider at all, can take part. Surprisingly, the results are usually close to those of the extensively trained judges. Who would be surprised if AI could do this in five years?