A recent report suggested that visitor numbers at the Giant’s Causeway had increased significantly. The explanation wasn’t a sudden surge in tourism, but something more technical. An AI system used to count visitors had been misinterpreting parts of the landscape, including rocks, as people.

On the surface, it’s an easy story to laugh off. A slightly clumsy piece of technology getting things wrong in a way that feels obvious once you see it. But it also highlights something more important about how AI is being used in practice.

These systems are often introduced to make things more efficient, whether that’s counting footfall, tracking usage, or helping organisations understand what’s happening in real time. The assumption is that once they’re in place, the numbers can be trusted. What this example shows is that the output can look entirely reasonable, even when the underlying interpretation is flawed. If no one questions it, the data becomes accepted as fact.

And that’s where things start to matter. These numbers don’t just sit in reports, they influence decisions around funding, staffing, planning, and resource allocation. A small error at the point of collection can quietly shape much bigger outcomes further down the line.

It’s also a reminder that AI doesn’t understand context in the way people do. A human would instantly recognise the difference between a person and a rock in that setting, but the system is working from patterns and probabilities rather than meaning. When something falls outside of what it expects, the mistake isn’t always obvious.

The challenge isn’t that AI makes errors, that’s expected. It’s that those errors can be difficult to spot, especially when the results look plausible. Which raises a slightly uncomfortable question: if we’re increasingly relying on AI to interpret the world for us, how often are we checking whether it’s actually getting it right?