Emirates Team NZ taps McKinsey for America's Cup edge

In a sport where the smallest of design features could prove the deciding factor, Emirates Team New Zealand turned to McKinsey to gain a competitive edge in its bid to retain the America’s Cup.

When the first America’s Cup was awarded in 1851, little could the yachtsmen have imagined that come the event’s 36th edition its competing crews would be supported by an army of intelligent robots. But when Emirates Team New Zealand sets out to defend the Cup in the waters off Auckland this week its meticulous preparations will have been underpinned by an advanced AI bot developed in collaboration with the world’s leading consultancy McKinsey & Company.

Without in any way detracting from the exceptional skills of highly talented and dedicated sailors, it’s fair to say that the contemporary America’s Cup competition is fundamentally one of high-tech design innovation. McKinsey senior partner Brian Fox, himself a competitive sailor, puts it succinctly; “Every boat in the America’s Cup is designed with a computer simulator. Whichever team has the best simulator, and uses it most effectively, gains the advantage.”

In the long build up to the event, Team New Zealand and Cup challenger Luna Rossa Prada Pirelli of Italy convened to establish the boat design parameters and stock components for the 36th America’s Cup, ensuring some level of sporting competitiveness between the two 75-foot vessels while leaving key areas in their designs open to tweaking. The issuing of the ‘Class Rule’, now close to three years ago, could effectively be considered the first shot of the starter’s gun.

Not just another crew member

Shortly after, Team New Zealand drafted in McKinsey to help in the hunt for its newest ultra-human crew-member – one “that could sail thousands of boats at a time”. With often millions of dollars spent on shaving milliseconds off performance times – these are the fastest monohulls on earth after all – prototyping and testing potentially revolutionary design features doesn’t come cheap, and nor is the process particularly speedy when capped by budgets and human limitations.

One area in which the 2021 Class Rule allowed for tinkering was in relation to the boats’ hydrofoils, the mechanism which lifts the entire craft out of the water and allows for speeds of up to 100 kilometres per hour. In the words of McKinsey, such an opportunity for modification, if brilliantly seized, can translate into a huge game-day advantage. It would be the consulting firm’s role to maximise Team New Zealand’s hydrofoil testing capacity, thus allowing for greater innovation.

“The simulator had been key to the team’s victory in 2017; the sailors had used it to test new boat designs without having to actually build them,” the firm notes. “But that simulator required multiple team members using it simultaneously for it to work properly. This was a logistical challenge, given the sailors’ scheduled practices, travel, and competitions.” McKinsey’s solution was to build an AI bot which could independently run design models through the simulator.

The first step was to bring in the team from McKinsey’s analytics arm QuantumBlack, with its data analysis and machine-learning experts from Australia and the UK building the initial infrastructure to migrate and run Team New Zealand’s simulator in the cloud. From here, things start getting especially technical, with the firm then adopting a newly innovative approach known as “deep reinforcement learning”, to essentially teach the AI bot how to become a professional sailor.

“The technique allowed the bot to learn dynamically and gain greater accuracy through continuous feedback,” explains Nic Hohn, QuantumBlack’s Chief Data Scientist for Australia and project co-lead. “When you start, the AI agent knows nothing and learns by trial and error using countless variables – and is refined again and again. Since the bot keeps experimenting, if you coach it to learn in the right way, it compresses into hours what would take a human years to understand.”

While one bot may have replaced the intensive labour of a massive number of humans, that bot would ultimately in turn benefit from the support of its own team, with McKinsey establishing a parallel network of thousands of bots to share information among one another as they each learned the ropes of high-octane sailing. This collective, huge-scale level of learning was a critical breakthrough, the firm says, dramatically reducing the time and cost of the project.

Deep reinforcement learning

The end result? Within two weeks the bot had mastered the basics of sailing in a straight line, before then graduating to more complex manoeuvres. A month or so later it was prevailing against human sailors, and could then start testing out different hydrofoil variations on its own – on rapid repeat for 24 hours a day – without bothering their schedules. Soon, the crew would be learning sailing techniques from the bot, and now, the latest Team New Zealand entry has been described as “scary”.

Ultimately, Team New Zealand’s design process was hastened ten-fold through its collaboration with McKinsey. Furthermore, the successful application of deep reinforcement learning has the consultants excited about its vast cross-industry potential. To highlight the significance, QuantumBlack co-founder and Chief Scientist Jacomo Corbo raises the concept of game-tree complexity and the board-game Go, famously conquered by Google’s DeepMind technology.

“This is one of the most complex deployments of deep reinforcement learning in the public cloud,” said Corbo. “One way to think about the difficulty of the problem: game-tree complexity roughly tallies the size of the space one has to navigate while accounting for the set of possible game paths, or the sequence of decisions, that need to be taken. Go, an extremely complex board game, has a game-tree complexity of 170 – our sailing problem has a game-tree complexity of nearly 2900.”