Can the world’s first AI ski instructor beat the real thing?

Three thousand metres up on Austria’s Hintertux glacier, the world’s best ski racers are being put through their paces. It is mid-October — long before the opening of any conventional ski resort — so they come up here to train, their multicoloured skin-suits animating a monochrome world of rock and ice, far above the tree line. Teams from Canada, Italy, Russia, Holland and many more crowd around the T-bars and near the tops of the runs, waiting for their turn through the slalom gates. And looking on, a battalion of stern coaches in puffa jackets, radios and binoculars slung round their necks, notepads in hand.

All these racers, all these coaches, and yet by far the most closely monitored skier on the entire mountain is me. I am skiing alone, unnoticed, wearing an old black anorak and woolly hat rather than national colours and polished helmet. But scrutinising my every move, from the angle of my ski edges to the relative pressure exerted by each toe, is the world’s first AI ski instructor. Hidden in my boots are two sensor-packed footbeds, communicating via Bluetooth to the smartphone in my jacket pocket, which in turn is reporting back to a server in Frankfurt. With each turn I throw up a spray of snow and a cloud of data.

Smartphone apps that record top speed and distance skied have become commonplace in recent years, but Carv is something altogether different. An average turn lasts about 1.5 seconds, during which time it will have collected and analysed more than 5,000 pieces of information. In free-skiing mode, it will silently record your performance so on the lift going back up, you can pull out your phone and check, for example, your minimum edge to edge time or see how early in the turn you start applying pressure. If that sounds altogether too geeky, you can leave the phone in your pocket, and a computerised voice will pipe up to give you a simple overall score for your run, a “Ski:IQ”.

Recording and scoring is just the start, though. Choose one of the instruction drills and the app will talk to you as you ski — via cordless headphones — giving you tips and encouragement, and marking each turn. An upbeat chime (I picture Pac-Man eating a ghost) rewards a good turn, a sludgy bleep denotes a poor one.

“Rather than just give the raw data, and appeal to a niche of serious racers, we decided we wanted to create a coaching experience which will actually teach the average person how to get better — and that is a much more complex problem,” says Jamie Grant, chief executive of Motion Metrics, the company behind the system. One might guess it is an offshoot of a big ski or boot manufacturer or perhaps, at a push, a video-game producer — but Motion Metrics is in fact a start-up whose roots owe more to the machine learning and artificial intelligence being used in finance.

After a degree at Oxford and gap year in Whistler, Grant, now 32, was working as an intern for Barclays Capital while doing a PhD in financial economics at Imperial College London. “I was looking at using machine learning to optimise portfolio allocation in futures markets, so that gave me exposure to coding, statistics and data science. I became interested in the idea of using the same techniques to try to understand my own data as I skied.”

He began developing algorithms to track skiing and posted a message on the college noticeboard asking for help (and jokily promising the chance to “sell to Google for billions”). Among those to respond was Pruthvikar Reddy, a masters student in mechanical engineering who had already been snapped up by JPMorgan to work part time, helping develop an iPad app for use at the London Metal Exchange. Together, they founded Motion Metrics, initially developing a tracking app, then creating prototypes of sensors — at first just iPhones taped to skis — turning their backs on their nascent City careers to work on the project full time (even though Reddy had never previously been skiing). Helped by Imperial’s Venture Catalyst Challenge, an accelerator programme for start-ups, they won backers including Alex Hoye, chief executive of ski maker Faction, and Sean O’Sullivan, the US-based serial tech investor.

Since 2015 they have led itinerant lives, moving between Shenzhen, China’s electronics capital, where Carv’s hardware is made, and ski resorts in Austria, Italy, the US and Slovenia, where the software is tested and developed. “At one point we were testing prototypes at a dry ski slope in Essex,” says Reddy, now 26. “It was summer, and someone saw me carrying ski boots, with lots of wires coming out of them, through Liverpool Street station. The next thing, we turned round and the whole station had been shut off, the police were there with attack dogs and bomb squad. This terrified looking policeman asked me what I was holding — I said ‘it’s a digital ski coach!’.”

Four years later the company has eight staff and is in the process of setting up a permanent office in Innsbruck — a central base for its European markets but also close to the year-round glacier ski areas at Stubai and Hintertux, where we meet. In the restaurant at the top of the mountain, Reddy sits coding by the fire, while Grant skis alongside me as I put the app through its paces.

“Go get ’em tiger!” says the female, US-accented voice in my cordless headphones as I ski away from the top of a lift. Much as I am charmed by Grant and Reddy’s youthful enthusiasm, I am quietly dubious about their concept — given all the endless variables, I doubt any amount of machine learning can compete with an experienced instructor. More than that, I come to the mountains precisely to unplug from the constant buzz of technology.

And yet, Carv is surprisingly addictive. Even on the free-ski mode I find myself concentrating hard on every turn in order to up my score for the run — especially given the league table feature that ranks you against other users in real time, whether they be alongside you or anywhere else in the world. A combination of barometer and GPS tells the system when we are taking a lift, and on the T-bar Grant and I compare scores. (He skis at speed and spends much of the time waiting for me at the bottom — “I started a company that analyses turns but I really like going straight.”).

Warmed up, I begin on Carv’s drills. First a session on edging. I start on level 12 of 20, and will go up to the next level only if 16 out of 20 consecutive turns meet the standard. I quickly start to hear the gratifying chimes, but then the voice warns “Try not to make an A-frame with your legs”. This is a bolt from the blue: after years of struggle I thought I had long since ironed out my A-frame — when inside knee moves to meet the outer as you turn. That the system picks out this failing is as impressive as it is dispiriting. It’s also uncanny — I can’t escape the feeling that it is somehow watching me. (In fact, Grant tells me later, the A-frame is inferred because of a disparity in the angle of my inner and outer ski.)

The 48 pressure sensors under each foot can tell if your weight is forward or back at the right point of the turn and whether you are transferring the weight smoothly from ski to ski. The movement of the phone in your jacket pocket can tell if your upper body position is correctly facing the fall line. An accelerometer, gyroscope and electronic compass, hidden in a tiny chip under each instep, can monitor edge angles, tempo, the symmetry of turns and much more. But finally, with a “Great job!” Carv announces I have completed the level, and by tapping my glove on my earphone twice, I move up to the next one.

So far, professionals have been positive — both racers, including British number one Dave Ryding and former US moguls world champion Jeremy Bloom, and instructors. When the company launched a Kickstarter campaign, members of the US ski team and the Professional Ski Instructors of America association saw it and invited Grant and Reddy to a training camp in California. Some of the PSIA’s star skiers, including Jonathan Ballou, head of the Aspen ski school, have helped develop the drills and skied with the system to provide benchmark data. “I guess you could say they are turkeys voting for Christmas,” says Grant, “but I think they genuinely just want to get more people into skiing.”

Carv can’t predict avalanches, so you’ll still need a guide or instructor off piste. It can’t teach beginners, it can’t direct you to the best mountain restaurants, it can’t carry your skis and it certainly can’t flirt in an endearing French accent. But it does win on price: a single day with a private instructor would typically cost about £500 in the Alps, much more in the US. If bought before December, Carv’s footbeds and app cost £229 — and the company is in discussions with major bootmakers to incorporate the technology into off-the-shelf “smart boots”, for both retail and rental. Beyond that, there are other potential applications in sports such as cycling, running and golf, as well as in medicine. The sensors are already being incorporated in the braces used to treat scoliosis, to give doctors quantifiable data rather than relying on patient interviews.

Even at those prices, instructors’ jobs are probably safe for now — Carv is more an added extra for the keen intermediate eager to accelerate their progress. But it is learning all the time — every turn you have ever made is stored in its servers, so the programme can begin to identify which instructions and drills you respond best to, as well as comparing you to other skiers of a similar type. Grant makes comparisons with the way Spotify gets to know what kind of music users like, and the way AI is being used by education companies. “The endgame is that everything is optimised just for you,” he says.

At the end of the day we take the series of cable cars back down to the valley, from winter back to summer — in the car park it is 21C and there is a smell of warm hay. Google and its billions may not have come knocking just yet — Grant and Reddy are staying in the cheapest hotel in town — but there’s little doubt the information age has arrived in the mountains.

Follow @FTLifeArts on Twitter to find out about our latest stories first. Subscribe to FT Life on YouTube for the latest FT Weekend videos