Coding and AI classes in Lytham St Annes

First the learner draws the town. One OpenStreetMap query returns 1,299 roads, from the A584 down to residential streets, 208.6 km of them, inside a rectangle drawn around the town. Each road is painted onto a grid of 10 metre pixels, 500 rows by 770 columns. Buildings came back too, but only 2,621 of them are mapped, far fewer than the town has, so they are left out rather than pretending the gaps are empty land. The result is a picture in which 20,441 pixels, 5.31% of the grid, are road, joined into 4 connected pieces.

Now the map has to become 30% narrower. Deleting every third or so column is the obvious method. A smarter one deletes whichever straight columns hold the least road. Seam carving, published by Shai Avidan and Ariel Shamir in 2007, does something cleverer: it removes a wiggly path of pixels, one per row, where each step may move one pixel left, right or straight down. To find the cheapest such path it gives every pixel a cost, here the amount of road in the 3 by 3 square around it, then works down the rows keeping, for every pixel, the cheapest way to reach it from the top. That running total is dynamic programming. Tracing back from the cheapest pixel on the bottom row gives the seam; remove it, recompute, repeat.

At 30% narrower, evenly spaced deletion throws away almost a third of the road and shatters the network from 4 pieces into 882. The emptiest straight columns do far better, because some columns of the rectangle hold little or no road, but a straight column still has to cross every road that runs east to west. Seams bend round roads, so they keep 96.4% of the road pixels and leave 36 pieces. None of the three is free: every method loses something, and the count of pieces tells the learner where.

Being able to answer that question with code is what keeps young people in Lytham St Annes in charge of AI tools, and it starts with learning to program. The longer argument is in what learning to code still gives a teenager in 2026.