3 questions: Using AI to help Olympic skaters get a chance | MIT News

Olympic figure skating seems simple. Athletes cut across the ice, then fly through the air, circling overhead, before landing on a single 4-5mm wide spot. To help figure skaters achieve quadruple axels, Salchows, Lutzes, and the elusive quintuple without looking at the slightest stress, Jerry Lu MFin ’24 developed a tracking system called. OOFSkate that uses artificial intelligence to analyze video of a figure skater’s jumps and make recommendations on how to improve. Lu, a former researcher at MIT Sports Labhas been assisting Team USA’s top broadcasters with their technical operations and will work with NBC Sports during the 2026 Winter Olympics to help analysts and TV viewers better understand the complex scoring system in figure skating, figure skating and figure skating. He will be using AI technology to explain friendly refereeing decisions and show how challenging these games can be.

Meanwhile, Professor Anette “Peko” Hosoi, founder and faculty director of the MIT Sports Lab, is starting a new study aimed at understanding how AI programs assess the performance of beauty in figure skating. Hosoi and Lu just talked to them MIT News about using AI in sports, whether AI systems can be used to judge Olympic figure skating, and when we might see a figure skater get a chance to compete.

Question: Why should you use AI for skating?

Lu: Skaters can always push, high, fast, and hard. OOFSkate is all about helping skaters find a way to rotate a little faster in their jumps or jump a little higher. The system helps skaters catch things that might pass the eye test, but that may allow them to pinpoint specific areas of high probability. The artistic side of skating is more difficult to assess than the technical aspects because it is subjective.

To use the mobile training app, you just need to take a video of an athlete jumping, and it will spit out physical metrics that dictate how much rotation you can do. It tracks those metrics and builds on all other elite and former athletes. You can see your data and see, “This is how the Olympic champion did this, maybe I should try that.” You get a comparison and an automatic classifier, which shows you that if you did this trick at the World Championships and it was judged by an international panel, this is probably the level of result they would give you.

Hosoi: There are a lot of AI tools coming online, especially things like pose estimators, where you can estimate bone poses from video. The challenge with these position meters is that if you have a single camera angle, they do very well in the camera plane, but they do very poorly in depth. For example, if you try to criticize someone’s form in fencing, and they walk towards the camera, you get very bad data. But with figure skating, Jerry has found one of the few places where depth challenges don’t really matter. In figure skating, you need to understand: How high did this person jump, how many times did they spin, and how well did they land? None of them rely on depth. You’ve got an app that shows raters are doing really well, and that doesn’t charge a penalty for things they do wrong.

Question: Can you envision a world where AI is used to explore the artistic side of figure skating?

Hosoi: When it comes to AI and aesthetic testing, we have new work underway thanks to a grant from the MIT Human Insight Collaborative (MITHIC). This work is in collaboration with Professor Arthur Bahr and IDSS graduate student Eric Liu. If you ask an AI platform for aesthetic evaluations like “What do you think of this painting?” will respond with something that sounds like it’s coming from a person. What we want to understand is, in order to reach that assessment, do AIs go through the same kind of reasoning processes or do they use the same intuitive concepts that humans go through to reach, “I like that painting,” or “I don’t like that painting”? Or are they just couples? Are they imitating what they heard someone say? Or is there a conceptual map of aesthetic appeal? Figure skating is a good place to look for this map because skating is judged on beauty. And there are numbers. You can’t walk around a museum and score, “This painting is the 35th.” But in skating, you have data.

That brings up another even more interesting question, which is the difference between beginners and experts. It is known that experts and novices will react differently when they see the same thing. A professional judge may have a different view of skating than a member of the general public. We try to understand the difference between the reactions of experts, beginners, and AI. Does this reaction have a common place of origin, or does AI come from a different place than both the expert and the student?

Lu: Figure skating is interesting because everyone working in the field of AI is trying to find AGI or general artificial intelligence and trying to build this extremely noisy AI that replicates humans. Working on the use of AI in sports such as water skiing helps us understand how people think and deal with judgment. This has downstream implications for AI research and companies building AI models. By getting a deeper understanding of how current AI models work with these games, and how you need to train and fine-tune these models to make them work for certain games, it helps you understand that AI needs to move forward.

Question: What will you be watching at the Olympic figure skating competitions in Milan Cortina, since you have been studying and working in this area? Do you think someone will get you a quint?

Lu: For the winter sports, I work with NBC on the cross-country skiing and snowboarding championships to help them tell a data-driven story about the American people. The goal is to make these games relevant. Skating looks slow on TV, but it’s not. Everything should seem easy. If it seems difficult, you will probably be punished. Skaters need to learn to skate very fast, jump very high, float in the air, and land well on one foot. The data we collect can help show how tough skiing is, even though it should look easy.

I’m glad we work in an Olympic sports environment because the world watches it once every four years, and traditionally it’s a sport that’s focused on training and driven by talent, unlike a sport like baseball, where if you don’t have a good tracking system you don’t maximize the value you have right now. I’m glad we’re starting to work with these Olympic games and athletes and make an impact here.

Hosoi: I have been watching the Olympic figure skating competitions since I could turn on the TV. They are always unbelievable. One of the things I will practice is to identify the omission, which is very difficult to do if you are an uneducated “judge”.

I also did some back-of-the-envelope calculations to see if a quint was possible. Now I am absolutely sure that it is possible. We will see one in our lifetime, if not soon. Not in this Olympics, but soon. When I saw that we were very close to quint, I thought, what about six? Can we do six rounds? Probably not. This is where we begin to come up against the limits of human power. But five, I think, are available.