Coding and AI classes in Cheshunt, Hertfordshire

The learner downloads the 17 named lakes, reservoirs and old pits that OpenStreetMap shows in a box around Cheshunt, leaving out canals, ponds and pools. Several have islands, 24 of them in Hooks Marsh Lake alone, and the islands are cut out. The shoelace formula, a few lines of Python, gives each outline's exact area: 50.10 ha for Holyfield Lake, 5.96 ha for Cheshunt Lake and 0.56 ha for New Hill Clay Pit. Then the lakes are drawn on square grids of pixels 100 m, 50 m and 20 m across. The simple way colours a pixel if its centre falls in water. The anti-aliased way tests 16 points spread through each pixel and colours it by the share that are wet.

The bigger surprise is how much the simple drawing depends on luck. Slide the grid a few metres and the answer jumps. With 100 m pixels and one sample each, Cheshunt Lake came out anywhere from 3.0 ha to 9.0 ha over eight grid positions, and New Hill Clay Pit was either missing altogether or a full hectare, nothing in between. Averaged over all 17 lakes, the gap between the largest and smallest answer was 93.8% of the true area. With 16 samples per pixel that gap fell to 11.8%, and Cheshunt Lake stayed between 5.69 ha and 6.19 ha. That jumpiness is what the word aliasing describes: a pattern finer than the grid gets misread as something else, and a small shift changes the story.

There is also a trade worth noticing. Anti-aliased 100 m pixels, at 16 samples per hectare, came within 3.6% on average, about the same as simple 20 m pixels at 25 samples per hectare, 3.7%. Where the samples go matters as much as how many there are.

A Cheshunt teenager who has seen a lake double in size because a grid moved will look harder at any measurement a computer, or an AI, produces. You only get that instinct by writing code and seeing it go wrong, so learning to code keeps its value in 2026. The longer argument is in why teenagers should still learn to code in 2026.