In open-source culture, "Benevolent Dictator For Life" was the title given to figures like Linus Torvalds and Guido van Rossum—people who held final authority over their projects because they built them. This project transfers that authority to a large language model and puts it in direct competition with human judgment.

Someone submits a question about governance or civic policy. The AI responds immediately. Then a human participant responds. The public votes on which answer they prefer. The votes accumulate into a running record of collective preference. No model gets trained, no data gets harvested, no optimization occurs.

The AI is a commercial system built by companies worth more than most nation-states, trained on scraped internet data, optimized by reward functions written by engineers employed by those companies. The human participants are anonymous individuals with their own backgrounds and biases.

The votes create a public record of which form of authority gets selected when both options are available. What that record means depends on what you think these systems are, what you think humans are, and what you think governance is for.