Is Deep Blue designed to assist or replace human lifeguards?
Luke Cunningham: “Deep Blue is very clearly designed to assist, not replace, human lifeguards. The system acts as an extra set of eyes that never gets tired or distracted, continuously analysing video feeds and flagging early signs of distress or unusual behaviour in the water. “The intervention itself is always human. Regulators in most markets still require certified lifeguards on duty and ethically we fully support that. The role of AI here is to extend their awareness and buy them precious seconds, not to remove them from the equation.” Deep Blue was launched just before you enrolled on the EMBA. How has the programme aided the startup’s growth and development? Karl Baz: “The timing turned out to be ideal. As Deep Blue was moving from prototype to early deployments, I was moving into modules on strategy, marketing, finance and governance. Every little bit helped. So, when Luke and I sit down to review direction, we’re not relying purely on gut instinct; we can also draw on structured frameworks I’ve put through the EMBA filter and adapt them to Deep Blue’s reality.” Luke Cunningham : “Because Karl is coming in with tools he has just applied in the programme, it raises the quality of the questions we ask ourselves, which is often more important than having a perfect answer.”
How has the programme aided your own development as an entrepreneur?
“An MBA or EMBA can offer you a set of tools, a network and a structured way to think about decisions you will have to make”
Karl Baz: “It has given me more language and structure for things I was already doing intuitively. While I still believe in learning by doing, I can now connect those experiments to clearer frameworks in strategy, finance, organisational design and so on. “It has also improved my ability to translate between different stakeholders. I can talk to engineers about models and data, before turning around and talking to investors or regulators about risk, governance and outcomes, using a vocabulary they are comfortable with.” How were the AI models used by Deep Blue trained? Luke Cunningham : “We use modern computer vision models that are trained primarily on video from real pools and aquatic facilities, as well as the thousands of simulated drownings used by my other enterprise, lifeguard training company BlueGuard. The core dataset combines three elements: everyday pool activity across ages, densities and layouts to give the system an understanding of what ‘normal’ looks like; simulated and drill scenarios where trained staff act out distress patterns to expose the model to the kinds of events we hope never
happen in reality; and augmented and synthetic variants of those clips to cover different lighting, weather, camera angles and levels of crowding.” What are some of the biggest challenges you have experienced around using AI? Luke Cunningham: “There are several challenges. From the data perspective, you have to design training and evaluation around proxies, drills, near misses and behaviour patterns, while being honest about what the model has and has not seen. There is also the question of false positives versus false
32 Ambition • ISSUE 6 • 2025
Made with FlippingBook - Share PDF online