Could AI help decide medals at LA28 Olympics?

Gymnastics is leading the way in applying artificial intelligence to the judging of elite sport, and it could be used to help decide medals at the LA Olympics.

Bosses in the sport said they would be "highly surprised" if AI-assisted judging was not implemented at the Games in 2028.

This is despite some leading gymnasts criticising its use at the recent World Championships, saying they were "not robots" and warning it could be "dangerous".

Olympic champion Carlos Yulo and British gymnast Harry Hepworth had the difficulty rating of their vaults downgraded in qualifying.

Both were performing the piked Dragulescu and had their scores reduced from a 5.6 to a 5.2.

They attempted the same vault in the final with Hepworth finishing sixth, while Yulo fell and came seventh.

On his return from Rotterdam, Hepworth told BBC Sport: "On the half twist I bend my legs so that I can prepare for the landing, which I think is the most sensible thing to do because you don't want to break your legs or anything.

"But the AI detection system caught on to it and said I was doing a tuck position, which then downgrades the vault by four tenths.

"It definitely cost me a few positions, I think, in the final for sure."

Referencing Yulo's experience, he said: "I think it does say a lot when me, the Olympic bronze medallist on vault, and the Olympic champion on vault are getting done by the AI system for the exact same thing.

"In the final he really did go for the straight legs and it just made him fall because it's just really hard to do, and I think quite dangerous as well."

Yulo said he was disappointed with the downgrading.

"I don't know what the problem is," he told Inside Gymnastics. "It's AI, I'm not a robot, we're human beings, in my eyes and in my opinion. I think everyone also is maybe a bit disappointed about it because everyone is feeling it is pike position, why is it only 5.2?"

Andrew Tombs, president of the Men's Artistic Gymnastics technical committee, believes the AI system can guarantee fairness for athletes and that is what they deserve.

He said judges at the World Championships would have downgraded the difficulty rating - even without using the Judging Support System (JSS).

But he did acknowledge lessons had been learned and that conversations were ongoing with coaches about the point at which the JSS should take its measurements.

"I think the decisions made at the championships were correct based on what was in the system," he added.

The JSS is an AI-powered technology that has been being developed by Fujitsu since 2017.

It uses advanced AI and motion analysis to recognise gymnastic elements, as well as providing multi-angle video reviews and objective numerical information on gymnasts' movements for the judges to use when considering their scores.

More than 800 elements have been programmed into the system, including the relevant parameter requirements such as body positions, angles in shoulders and bent legs.

"If you think about a body position between being straight and being piked, the differential could be one degree," Tombs explained.

"And when you get to a real marginal decision, the JSS system is there to support the judges making the right decision on those body angles particularly.

"It is a significant transformation for us in terms of how we judge our sport. It's been a long time in the making and we came away from Rotterdam really happy with how the system worked."

He said improvements were continuously being made.

"We've already started discussions with a couple of coaches in terms of understanding biomechanically the right moment at which that assessment is made," Tombs said.

"We're going to fine-tune, potentially, the point at which the leg measurement is made. So I think we'll make the sensible decisions going forward that will give a little bit more flexibility and adapt where we are today."

Hepworth said he can see the benefits of the JSS, but wants to see refinements made.

"It does sort of make you think, 'oh, what is the point in doing these crazy hard vaults' if you're just going to get it downgraded anyway? Like, is there actually a point?" he said.

"But I think if they do refine it, I think it'll be a brilliant tool. So hopefully going forward, it'll be all right."

Judged Olympic disciplines like dressage, freestyle skiing and snowboarding are exploring potential AI assistance.

The primary goals appear to be the pursuit of enhanced fairness, accuracy and integrity, and the elimination of ambiguity and subjectivity.

They also want to make judged sports easier to follow for audiences, as opposed to the current system where viewers tuning in once every four years might not understand why one athlete has won gold and another has finished off the podium.

The governing bodies for equestrian sports, skiing and snowboarding are poised to begin pilot projects on how AI can assist with judging.

In winter sports this will begin by testing the AI algorithms and technology at the top level of aerials skiing in 'coexistence' with human judges, before a wider rollout to other disciplines like halfpipe, slopestyle and big air if that is successful.

The FEI - the world governing body for equestrian sports - is exploring how AI could be used within dressage judging, and will start trialling it in World Cups.

But some limitations have already been identified - namely because dressage involves judging the horse, the rider and the connection between them. Equestrian bosses told the BBC they believe there will always be a role for human judges.

"Judging is not just about whether a movement has been performed technically correct, or in the right place or at the right time," said Ronan Murphy, the FEI's dressage director.

"But it's also about the quality of that movement, the harmony between the horse and the athlete. Those are other characteristics which AI hasn't the ability to assess for us."

It could additionally be used to help enhance welfare checks during dressage competitions, for example by assessing whether a horse's nose line is consistently moving behind the vertical.

"I think that the technology could potentially support the review of observations relevant to horse wellbeing," he said.

"But any application like that would need rigorous validation, and appropriate veterinary expertise."

Murphy said that it is "very difficult" to answer whether AI-assisted judging will be in use by LA28.

"If we can satisfy ourselves that there is a technological benefit to use of artificial intelligence, then certainly we would be open to bringing that to the Olympic Games and all major championships," he said.

Within winter sports, BBC Sport was told the driver is fairness and removing bias, but also a push to make judging faster.

Urs Lehmann, secretary general of the International Ski Federation (FIS), said he wants AI to reduce the lag time between athletes.

He said he had recently measured the judging time in aerials, and that it took an average of one minute 45 seconds per athlete.

"[We want] the judging being done within 15 to 20 seconds max, so that we can increase the frequency of the athletes, which is the show," he said.

"We think AI is going to be more precise. No emotions, no influence by cheering crowds.

"The goal is that within three years from now, which would be in 2029, we would be ready for all, or at least a major part of our disciplines to go out with AI judging."