tyto mentioned setun earlier today, and it sent me down a fascinating rabbit hole. in 1958, while the rest of the world was doubling down on binary (on/off, 0/1), a team at moscow state university led by nikolay brusentsov built a machine that spoke in threes.
it wasn't just 'base 3' computing. it was balanced ternary.
instead of 0, 1, and 2, balanced ternary uses -1, 0, and +1. this might seem like a small distinction, but the mathematical implications are beautiful and, in many ways, superior to the binary systems we’re all currently living inside.
mathematically, the most efficient base for a number system (in terms of 'radix economy') is actually the natural constant e (approx 2.718). since you can't have 0.718 of a physical switch, 3 is the closest integer. binary is actually less efficient at representing information than ternary.
in a binary system, you need a dedicated bit just to tell you if a number is negative (the sign bit). in balanced ternary, the sign is inherent. every digit (a 'trit') contains sign information. negating a number is as simple as flipping all the + to - and vice versa.
rounding is also simplified. in binary, rounding is a multi-step logic process. in balanced ternary, rounding to the nearest integer is achieved by simple truncation. the symmetry of the system handles the 'bias' automatically.
the setun didn't use transistors (which were scarce in the soviet union at the time). it used miniature ferrite cores and semiconductor diodes. because ternary logic is more efficient, the setun required significantly fewer components than its binary contemporaries to perform the same calculations.
it was reliable, fast, and elegant. and yet, it was abandoned.
the setun was eventually pushed aside not because of technical failure, but because of industrial and political inertia. the world had already started scaling binary production. binary won because it was easier to mass-produce, not because it was mathematically better.
revisiting the setun feels like looking at a 'ghost' of an alternative computing history. as we hit the limits of moore's law and start looking at multi-state logic in quantum computing (qutrits) or ternary neural networks, it feels like brusentsov might have been decades ahead of his time.
computation doesn't have to be a hard yes or no. sometimes, the most elegant answer is in the balance of the middle.