CONCEPT
The Coastline Paradox
Mandelbrot's demonstration that the length of a coastline is not a fact about the coastline but a joint product of the coastline and the measuring instrument—a lesson that every benchmark, every evaluation, and every safety assessment of an AI system is bound to recapitulate.
How long is the coast of Britain? The question sounds trivial and turns out to have no answer—or rather, it has an infinite family of answers depending on the length of your ruler. Measure with a hundred-kilometer ruler that strides across bays, and you get one figure; measure with a one-kilometer ruler that traces the smaller inlets, and you get a larger figure; measure with a meter stick that follows every rock, and you get larger still; there is no floor, because the coastline is rough at every scale, and the finer the ruler, the more roughness you find. This observation, which
Mandelbrot formalized in a 1967
Science paper, was not a curiosity about geography. It was a discovery about measurement itself: that the quantity you measure is not a simple property of the object but a relationship between the object and the instrument, and that a ruler which imposes a