The Skeleton Puzzle
In a dusty museum room, a student stares at a table meant for a dinosaur skeleton. But most of the bones are gone. Only a few leg bones and vertebrae sit on the surface. It looks like a mistake, yet the head curator set it up this way to test the student.
Usually, apprentices learn on skeletons that are nearly complete. If a toe is missing, they just copy the shape of the one next to it. It is easy, but it makes them lazy. They stop thinking about the animal's biology and just match patterns locally.
The boss changed the rules. She hid most of the bones, leaving only a quarter on the table. She explains that without nearby clues, the student cannot just guess. To fill the massive gaps, they must understand the whole creature. A single hip bone has to tell them exactly how long the leg should be.
This changes the work. The student only has to clean and examine the few bones actually there. They ignore the empty space for now. Since they aren't wasting time on the missing parts yet, the first step happens three times faster.
Once the shards are clean, the real test begins. The student must sculpt the missing pieces from scratch using their knowledge. Because it is so hard, they have to learn the deep rules of the skeleton. It proves they know the anatomy far better than if they had just connected the dots.
It turns out computers learn the same way. When we hide most of an image, the machine has to imagine the whole picture from just a few hints. By removing the easy answers, we teach machines to understand the essence of a scene rather than just memorizing the surface.