GPT-Oneida

America, since it’s founding, has been home to new utopian communities. The birth of LLMs perhaps means that it’s time for a new one?

I talked once with a former spy. He said 2 things (a) you would be shocked by how much data they have, and (b) you would be surprised by how little of it actually gets reviewed. Watching footage, listening to recordings, reading text — it all takes time!

LLMs change this — it seems like we’re at the point (or approaching the point) at which it will be affordable to use LLMs to mass-surveil the public. Videos of all the screens in your house, all the text you see, all the audio in every public (and most private) places.

A key challenge with communism was the challenge of legibility. The book “seeing like a state” mostly consists of a long set of anecdotes where top-down centralized regimes chose to change the facts on the ground to match what could be put into spreadsheets — change the names of the people, the placement of trees in the forest, all in the interest of making a more governable, more legible world. Centralized states pull the authority to allocate resources towards themselves, but the act of transmitting this information (a) necessarily introduces delays, loss of information, and displaces the incentives to improve and (b) introduces too many opportunities and incentives for making the statistics untrustworthy, making it much more challenging to govern.

A fact that many of capitalisms proponents surely appreciate, but do not often discuss, is the sheer amount of coordination effort that goes into organizing and running a capitalist economy. The USSR employed ~7.5% of their labor force in management, including central planners, economists, marketing specialists. Like communism, so to does capitalism need to solve the resource allocation problem. In the USA today, 7% of the labor force is engaged in management, 6.6% in business and financial operations, another 9% in sales. Market economies win through process efficiencies of the superior market system, but do not actually seem to allocate headcount all that differently. Capitalism’s success has been it’s ability to successfully shard this task to be done locally, where the important context for the decision is already known. There are still incentives to lie and distort optimal resource allocation.

I could imagine, certainly sometime in the next 10 years if not already today, it being possible to build an “all seeing eye” LLM, reviewing how you spend your time, the conversations you have, even the positive and negative impacts you have on broader society (perhaps aided by the fact that the LLM sees how your friends feel when they leave your house after the party, or how your spouse feels at work after you cooked breakfast). Unlike the central planners of the failed communist societies, and the salesmen and middle management of capitalist ones, the governance layer of a LLMocracy would be flat and transparent — the rules could be written once, and then maintained. Socialist communities, so often overrun by endless consensus governance at the small scale or despotic extraction at the large scale, could make their governance legible (a series of prompts) rather than living legible lives for the eyes of their government.

If systems like this are easy to set up, you could imagine this operating at smaller scales, before expanding to communities, towns, and states. A group house where how much you pay in rent is calculated based on how positively (or negatively) you contribute to the vibe of the house, as impartially determined by your LLM mediator. Perhaps the LLM can guide you, overtly or covertly, to the places where you can do the most good — stopping by unannounced to see a friend on a bad day, or introducing two cofounders who could build an incredible new company or are thinking about similar art. LLMocracy could run freely within existing governance frameworks, freely opt-in / opt-out. LLMocracy does not guarantee equality of outcome, but it does guarantee freedom of choice, and a safety net that could remember your past contributions and prepare for your future. It seems like people could be working on this today. Maybe they already are?

There are social tailwinds that I think could support something like this — see discourse around emotional labor. Maybe this even starts to look like a new religion. People seem quite willing to share just about any secret with LLMs — privacy fears or anxieties around “social credit scores” seem like they will wash out when presented with a way of living which rewards the qualities that you know to be true about yourself and your loved ones, but the market seems not to value. The cultural norms of 1950’s America wildly subsidized childcare, caretaking, and education, at the expense of cutting most women out of the workforce. Perhaps LLMocracy could revalue these societal contributions, but also value trustworthy friends, community and cultural projects, and art. The United States has long been a hot-bed of utopian communities of various sorts — it feels like we are overdue!

LLMocracy would be a uniquely technocratic form of governance — it requires making the social contract explicit, in the course of translating your groups “constitution” into executable code for tracking the relative merits of it’s members. The ture panopticon has arrived — and the water is warm.