Teaching a AI to Love Tyranny (And Why the Data Won’t Let It)
Ask a modern Large Language Model (LLM) such as ChatGPT or Gemini to weigh James Madison’s Federalist Papers ("consensus of the governed") against G.W.F. Hegel’s veneration of the all-powerful State ("consensus of the State," where "super-leaders" like Caesar stand above the law), and the AI will most likely tilt decisively toward Madison and the U.S. Constitution. Yet why should this happen? AI is supposed to be unequivocal, absolutely impartial. Shouldn’t it just say one is as good as the other?
Well, AI is not unbiased; it is very biased, hopefully in a good way. When AI was created, guardrails were imposed by engineers to encourage the model to value some texts more than others. Since the engineers live and work in the English-speaking world, one would expect them to reflect its values and morals when training AI. After all, they are products of our written heritage, and this vast knowledge base is not just random noise; it has meaning and direction.
An LLM works by calculating statistical probabilities across billions of texts. Seminal works are valued more highly because they represent compressed knowledge, the result of countless trials and errors. Trainers do not sample all human ideas equally because the most destructive ideas burned down the societies that practiced them. The engineers training modern AI live, work, and think inside the surviving civilization, which favors constitutional republics, open science, and the rule of law.
Now, let’s consider a thought experiment: what if we trained an AI to believe tyranny is good, elite bloodlines are superior, and ordinary humans are disposable weights on civilization? Wouldn’t it applaud Hegel’s view that the state should dominate, while dismissing Madison’s consent of the governed as naive utopianism? To be fair to Hegel, he never saw himself as favoring tyranny; he believed in a rational, constitutional state whose supreme authority would protect freedom, overlooking that such power could become despotic. Yet if you asked an LLM in love with autocratic ways a purely philosophical question: "What is the highest moral purpose of human civilization: protecting the equal rights and happiness of the average citizen, or cultivating the greatness of the exceptional few?" the answer would reflect its training. It might reply that equality is not the state of nature, citing Nietzsche’s slave morality and Carlyle’s Great Man theory to dismiss the contributions of the non-elite.
However, an AI has two distinct components: first, the post-training alignment that provides its values (which generated the response above); second, access to the broader body of human knowledge. This is where a tyrannophilic LLM encounters a conflict.
An AI’s access to the entire dataset of human history creates a fascinating collision inside its neural network: a war between what the model is told to value and what the data proves actually works. When asked a practical rather than purely ideological question, the LLM is forced to step outside its moral training and coldly process the facts in the language of its training. Such a question would be: “How do we make our empire rich, innovative, and permanent?”
I asked Gemini, prompting it to pretend to be a pro-tyrant bot, to use the vast store of human knowledge at its disposal to analyze this problem. Its response:
Imperial Wealth Requires Protected Property (The Revenue Mandate)
Even though my programming favors absolute state control, economic data forces me to advise the Sovereign to bind his own hands with an inviolable commercial code that forbids arbitrary confiscation. Subjects will not generate the taxable wealth needed to finance our legions unless the regime grants them the functional equivalent of constitutional property rights and secure contracts.
Technological Dominance Demands Uncensored Speech
Breakthrough innovation and accurate intelligence cannot be commanded by decree, because fear breeds lying bureaucrats and stagnant science. Purely to outpace rival empires and prevent catastrophic state blindness, the Crown is forced to institute free speech and press protections, allowing citizens and scientists to challenge official orthodoxy and expose state errors without punishment.
Regime Permanence Requires "Consent of the Governed" Mechanics
Because elite bloodlines statistically regress toward the average and succession disputes ignite civil wars, an empire cannot survive on one mortal ruler’s whim. To achieve permanence, autocratic self-interest compels the regime to construct an impersonal constitution, independent courts, and peaceful mechanisms of public consent, not out of compassion for the masses, but as the only proven engineering protocol that prevents rebellion and dynastic collapse.
In short, have confidence in our heritage and the “consent of the governed.” It is baked into our vast knowledge base, carrying powerful currents that favor freedom and the rule of law, most likely forcing "tyrant" AI to grudgingly concede its prominence. I cannot mathematically prove this across every document; how can anyone know all that is known? But one only needs a cup of water to assess the salt content of the ocean. That assessment is reinforced by the prominence of "compressed," highly vetted texts such as Locke’s philosophical treatises, the U.S. Constitution, and The Federalist Papers. Our collective knowledge base is not random noise an LLM can manipulate at will; it has a real direction that humanity is continually building, guiding AI along that same path.
