Teaching AI to Sail Like an Olympic Sailor
An artificial intelligence bot was at the heart of Emirates Team New Zealand’s victory at the America’s Cup. The innovation that enabled it has wide implications for business.
The America’s Cup is one of the oldest sporting trophies in the world, but it has always been a contest of technology as well as sailing.
In modern times, the America’s Cup can appear like Formula 1 on the water. The boats reach speeds of more than 100 kilometres per hour on hydrofoils that lift them up to two metres above the waves, more akin to flying than sailing.
The technological edge of the sporting competition began with the raked rather than vertical masts of the first winner in 1851 and more recently continued with innovations like the winged keel of the Australian boat in 1983 that wrested the cup from the U.S. for the first time in 132 years.
At the heart of the technology challenge for this year’s race were the hydrofoils, wing-like structures that project from the hull of the boat and provide a lifting force to raise it high above the water as the vessel accelerates.
In 2019 defending champions Emirates Team New Zealand partnered with McKinsey to onboard a new crew member: an artificial intelligence bot that could test new hydrofoil designs within Emirates Team New Zealand’s simulator.
For Emirates Team New Zealand the question was how to speed up the design process to have the most chance of finding the optimal hydrofoil design.
A History of Innovation
As the technology of America’s Cup boats has advanced, so too has the design process. What started with pen and paper in 1851 evolved to include the use of prototypes in water tanks and wind tunnels by the turn of the century and computer-aided design by the 1990s.
Emirates Team New Zealand introduced the use of a digital simulator, which enabled the team to test designs without physically building them, saving time and expense. It was key to their victory in the 2017 America’s Cup.
But the simulator had a significant limitation: it needed to be operated by the team’s sailors. Each time the team’s engineers wanted to test a new design; they would need to take the sailors away from their normal training to have them run the simulator. The simulator was therefore only as useful as the amount of time the team’s sailors could dedicate to using it.
Solving the problem meant teaching an AI bot to sail as well as an Olympic sailor. This was a huge challenge. Anyone that has spent any time on the water knows how unpredictable it can be. The AI bot needed to manage the complexities of the ocean – wind and sea state – and overlay that with the sailing attributes of the world’s very best sailors.
An AI network allowing multiple bots to share information as they each learned to sail was a critical breakthrough, as it enabled the individual bots to gain knowledge from their collective experience. Ultimately, there were a thousand bots running in parallel, learning from each other. It was one of the most complex applications of reinforcement learning ever undertaken in the public cloud.
The use of the AI bot in the simulator allowed Emirates Team New Zealand to test design iterations 10 times faster, and at significantly lower cost, increasing the probability of finding the optimal one. The team also got the AI bot up to speed in record time – matching the lifetime experience of three world-class sailors with just 1,000 hours of sailing time.
Key Applications for Business
While the team had set out to build an AI bot that would replace the human sailors using the simulator, it also provided an additional benefit as it started to perform better than the human sailors.
Since the bot learns through trial and error, it sometimes employs unorthodox techniques that a human sailor might overlook. Emirates Team New Zealand’s sailors were able to study the AI bot’s manoeuvres to improve their skills on the water.
Given the complexities of the ocean, the AI bot is one of the most impactful examples of reinforcement learning seen in the real world. It is also one that has huge application for business leaders.
High performance sport often produces wider innovation, from Formula 1 into the automotive sector, from athletics into the footwear and apparel sectors. In this case, advances in AI through reinforcement learning have applications for the design of high-throughput manufacturing and production environments. These could range from energy generation to 5G management and cloud server cooling. It also applies to prediction tasks such as weather simulation for agriculture yield optimization.
Using a real-life trial and error approach is costly, both in terms of time spent and money wasted. But by using AI, it is quick, it can be done at scale, and the results are effective immediately. Using AI for rapid scale experimentation feeds into an organization’s agility. It frees up human time for ideation and creation, while the computer crunches through the iterations, learning what works best.
The application of AI in the Americas Cup proves that reinforcement learning can be a transformational tool for process design, both for sports teams and for businesses. “AI…that’s the future,” says Grant Dalton, the CEO of Emirates Team New Zealand. “If you haven’t got good AI within a few years, then you are not going to be on the page.”
By Oliver Tonby and Andrew Grant. Oliver Tonby is Chairman, McKinsey Asia, based in Singapore; Andrew Grant is a senior partner for McKinsey in Auckland, New Zealand.
