CONCEPT
The Method of Exhaustion
Archimedes's technique of trapping an unknown quantity between two converging sequences—the ancestor of the integral calculus and, in spirit and structure, of the machine learning training loop—which establishes truths about the unreachable by ruling out every alternative, and whose deepest lesson is that perfect approximation is fully compatible with categorical difference.
The method of exhaustion is how
Archimedes established exact truths about curved figures—circles, parabolas, spheres—using only finite polygons. He would inscribe a polygon inside the curve and circumscribe another outside it, obtaining a lower and an upper bound on the unknown area or volume. Then he would add more sides, tightening the bounds, squeezing the truth from above and below until no alternative value could survive the pressure. The curve could never be captured directly; it could only be approached, bounded, and inferred. Archimedes attained certainty about a limit he could not construct. It took nearly two millennia before Newton and Leibniz turned the method into the calculus. The training loop of a modern
neural network—gradient descent correcting itself example by example, crawling toward a minimum it approximates but never exactly reaches—is this same procedure wearing modern clothes. Each step is finite,