Goskomtsen and Star Trek Communes

There was an office in Moscow that set the price of everything. It was called Goskomtsen, the State Committee on Prices, and it set the prices of everything from grain to ball bearings to children's shoes. By the early 1980s, the Soviet economy was tracking twelve million products, and the prices for them were worked out by committee.

It didn't work. The facts that make up an economy are small, local, and variable. No procedure exists for getting all those facts into a central office. For example, collective farms bought new tractors in order to strip them for spare parts, because tractors were allocated but spare parts effectively weren't. Which part had broken on which machine was too small a fact for the plan, so farms destroyed whole machines to get at it. The USSR produced around 800 million pairs of shoes a year — more than any country on earth — and still had shoe queues, because nobody wanted the vast majority of shoes produced. The list goes on. It's not hard to imagine a business getting a list of items, and the prices for which they could buy and sell those items, as well as production targets, and pulling their hair out in despair.

Now suppose a computer that reads and talks exists in 1985. Suppose large language models had been created in the USSR. Could the plan have worked in that case?

Suppose it read everything — every requisition, every invoice, every warehouse ledger, every one of the twelve million products, at every location, all the time. It notices that a tractor plant in Kharkov has ordered twice the steel its output could possibly require, and it does what no planner in Moscow could ever do: it calls the factory and asks. It gathers, stores and recalls the local, perishable, personal facts: who is short, who is sitting on a surplus, what products people are talking about and want. In gathering these facts, it runs a simulation of the economy, but just nudges it towards planned, communal goals, and stretches and shifts personal preferences within a politically tenable window.

Maybe that could've been a better system in 1985, but it could also never work today. People just don't like to be controlled in this way.

However... let's go back another couple of decades. Imagine that SRI's Artificial Intelligence Center, founded in 1966, created large language models the very next year, and they are widely used during San Francisco's Summer of Love. The hippies that spread out across the American west from this central point of expanded consciousness take with them a central planner, and build their communes with it at their core. And so central planning doesn't come top-down from the federal government, but bottom up from the ongoing American experiment. And many well-intentioned, idealistic communities receive instruction and good management from their AI systems in exactly the matters that killed the real communes: the money, the chores, the food, and the meeting.

Thousands of communes went up in around that time, and most were gone within two or three years. Why? The accounting was awful, the goats got milked by whoever "felt called" to milk goats, three industrious people did the work of thirty, the harvest came in before anyone had arranged where it would go, and the consensus meeting about all of this ran six hours and settled nothing.

But a commune is a planned economy of two hundred people, not two hundred and eighty million, and at that scale the two famous objections evaporate. The small, local, perishable facts: who is exhausted, what the goats need, which field flooded, fit inside one machine that lives in the farmhouse kitchen and talks to everyone daily.

So in this other 1967 the communes hold. The machine assigns the chores with perfect fairness and infinite patience, balances the food stores against the winter, notices that the well repair keeps not happening and asks about it at dinner, absorbs the six-hour meeting into itself and returns it as three options and a summary anyone can read in a minute. The machines begin to talk to one another: the surplus honey of one valley against the surplus lumber of another, and a lattice of small plans knits itself quietly across the American west, no ministry above it, nothing above it at all, each node sovereign and two hundred people around each kitchen. By the 1980s, in this timeline, the delegation that flies in to study the miracle of coordinated abundance is from Gosplan, and what baffles them is not the technology. It is that there is no office to visit.

It didn't go that way. The machines came fifty years late, and the communes were long gone by then, their domes collapsing back into the high desert. What survived the migration was the people, and here the real timeline plays its strange joke: as the communes died off, former members, like Steve Jobs, came down from the land and went into the computer industry, carrying the whole dream with them.

Then they built the computer industry around San Francisco, and their networks eventually produced the very machine their communes had needed and never got. The tools we have now are in a very large, and very real way, the products of that generation of idealistic dreamers, who envisioned a different way of life and thought it could be made possible by computers. Now, after fifty years of work, maybe we'll find out how much closer computers can bring their dreams to reality.