The Trinary Dream Endures

This is a post from Robin Sloan’s lab blog & notebook. You can visit the blog’s homepage, or learn more about me.

Since the beginning, there has been an alter­na­tive vision for computing, not binary but trinary, also called ternary. (“Tri­nary” sounds so much better to me.)

Tri­nary didn’t make any headway in the 20th century; binary’s direct map­ping to the “on”/”off” states of elec­tric cur­rent was just too effective, or seductive; but remember that elec­tric cur­rent isn’t actually “on” or “off”. It has taken a ton of engi­neering to “simulate” those abstract states in real, phys­ical circuits, espe­cially as they have gotten smaller and smaller.

Tri­nary is philo­soph­i­cally appealing because its ground-floor vocab­u­lary isn’t “yes” and “no”, but rather: “yes”, “no”, and “maybe”. It’s prob­ably a bit much to imagine that this archi­tec­tural dif­fer­ence could cas­cade up through the layers of abstrac­tion and tend to pro­duce soft­ware with subtler, richer values … yet I do imagine it.

Tri­nary might still have its day. You can train a capable and super-efficient lan­guage model using weights of only -1, 0, and 1, and I believe many models in the future will use this architecture.

Viva la “maybe”!