So in 2019, the team partnered with McKinsey and its artificial intelligence arm QuantumBlack to develop an AI bot which would be able to “sail” thousands of potential boat designs with potential new hydrofoil designs. A key element of this project was the need to train the boat to sail like an Olympic-level athlete and ensure that any performance gains from hydrofoil options were attributable to design changes, not sailor-based variation.
McKinsey and QuantumBlack began by migrating the existing simulator into a cloud-based virtual environment where tests could be run at scale. Deep reinforcement learning was then applied to refine the skills of the AI bot, ensuring it could respond to conditions such as wind speed and direction, as well as trading-off immediate and long-term goals. Multiple bots were created that could run in parallel and apply machine-to-machine learning.
After 1,000 hours of virtual sailing, the bot began to beat human sailors which allowed the team to retire the simulator. Testing iterations increased tenfold across 2019 and 2020 and a new bonus was recognised – the bot made manoeuvres that improved on human performance in the simulator, creating a competitors they could repeatedly race against. Constraints were also put in place to ensure any design changes would be capable to being deployed and used in real-world racing.
Emirates Team New Zealand won the America’s Cup in 2021 for the fourth time in a row. The AI bot means boat designs can now be iterated and tested in hours, not days, while the near-perfect competitor means human sailors can constantly refine their skills in the simulator.
The DataIQ Grand Prix is given to an entry that has already won an award and is the highest-scoring of all candidates. “Making boats fly with AI” achieved a near perfect score from the judging panel and was the unqualified stand-out.
