Building a light chip is like a game where you have to guide a beam through a maze of tiny mirrors and beam splitters. Each element acts like a railroad switch: it either lets light pass straight through or directs it to an adjacent track. The goal is to produce a specific light pattern at the output while using as few 'switches' as possible.
Usually, engineers spend years finding the shortest route for each task. But here, the program was taught to do it on its own. It was given only the rules of the game, how light passes through a single device, and asked to come up with a general recipe for any number of beams. The program sifted through hundreds of thousands of options and found surprisingly simple moves. For example, for matrices that flip the light pattern like a mirror reflection, it managed with just 2N−3 elements instead of the usual N(N−1)/2. For large circuits, this yields nearly a twofold saving.
But the key is that the program learned to see the structure of the problem and adapt its solution. If the matrix is 'holey' (with many zeros), it automatically discards unnecessary beam splitters. With 95% emptiness, the savings reached 38%—and this directly means fewer expensive components in a real chip. This approach turns optical circuit assembly from a manual art into an automated pipeline where the computer itself finds the best path for light.
The simplest beam splitter is a half-silvered mirror like those used in shop windows. But in chips, they are etched onto glass as thin as a hair, and they can split a beam with precision to a fraction of a percent.