AI Will Cement Fukuyama’s Legacy
Hallucination as an Anti-Authoritarian Safeguard
A Short Story
There were once two Kings – one king benevolent and loved by his people, one tyrannical and feared by his people.
One day, a strange prophet appeared in the benevolent king’s lands. He taught farmers ways to triple the yield of their crops. He made new, brilliant weapons. And sometimes, he spouted nonsense. He was a mixed bag.
The benevolent king brought the prophet into his court to best help the people. He summoned the greatest scholars, generals, and priests of the kingdom to inquire from him. These wise men often thought the prophet to be silly. At times, even insulting. Nonetheless, they appreciated the prophet's wisdom. One day, the prophet called the benevolent king a tyrant. No one made anything of it. After all, there was no reason to think the king a tyrant. They happily moved on, asked the prophet questions on different matters.
After the prophet had brought prosperity to the people of the first kingdom, he decided to venture to a new land. He came to the land of the tyrannical king.
Hearing whispers of the prophet’s wisdom, the tyrannical king quickly brought the prophet into his court. He summoned the greatest scholars, generals, and priests of the kingdom to inquire from him. These wise men bristled at the prophet's manner. They exaggerated his errors and transgressions. Yet they grudgingly accepted the wisdom that the prophet brought. One day, the prophet called the tyrannical king a tyrant. Quickly, the wise men pledged their loyalty to their king. “Our king could never be a tyrant!” they insisted. Yet the thought remained at the back of their mind. Whispers spread throughout the kingdom. “Don’t take my word for it. It’s something that the prophet said.” Seeing the mounting unrest, the tyrannical king exiled the prophet from the land and banned his teachings.
Many decades later, the two kingdoms warred. Thanks to the accepting the aid of the prophet, the benevolent king had superior production, weapons, and tactics. The kingdom of the tyrannical king crumbled.
History and Interpretation
I sent out this allegory on twitter and asked my followers what it represents.
https://twitter.com/psychosort/status/1678248612361388033?s=20
The above allegory can be interpreted many ways. One is literal – it’s a story of what happens when a society refuses to accept dissenting opinions. Another is historical. It’s about the US and the Soviet Union, or the US and Nazi Germany. It could be more specific – about operation paperclip, the US’ successful project to import Nazi scientists. A contemporary interpretation might replace the Soviet Union with China. Who is the prophet then? As the poll suggests, perhaps large language models, or perhaps prediction markets.
The reason why this story has so many historical interpretations is because it illustrates a broader moral trend; it illustrates the empirical value of dissent and pluralism. In an earlier piece criticizing how public intellectuals believed LLMs would change relationships, one of the worst offenders were media figures who confused democracy with left-wing values. Yet I’ve also had some readers who share a common understanding of capitalist democracies, in the Fukuyama sense, who use the word democracy to mean actual democracy, but nonetheless believe that ML would hurt democracy. The purpose of this essay is to make the counterargument to that good-faith belief.
Randomness and Subversion
In the short story, the prophet is a brilliant but fallible man. He makes mistakes, and often silly ones. He offends the most important people, often in ways where he is genuinely in the wrong. And yet, he is indispensible. He forces the ruling class to live with error and uncertainty. The regimes that can’t will end up going the way of the tyrannical king.
This is the core insight behind many (though not all) machine learning techniques in the present or near future (~10 years). Forcing regimes to live with randomness means forcing regimes to live with freedom.
When someone like Peter Thiel says “AI is Communist, Crypto is Capitalist” (I’m focusing here on the first half), he’s not without a point. There are surveillance tools and predictive models that genuinely do aid the goals of an authoritarian regime. Some of the earliest applications of ML were like this – facial recognition, logistics aid, social media analytics, early-stage automated transport etc. Even then I would argue they’re better used by companies such as Amazon and Google than by massive bureaucratic states.
The key difference between these early applications and newer ones, like image generation and large language models, is that they have a quantifiable and small error. Maybe the forecasting model said you needed 121,000 units rather than the 122,000 you needed in reality, but these are variables already managed in existing logistics and business systems. Those systems aren’t ready for the ad team producing a crazy, experimental art style or a language model making a joke about the ruling class.
People whose brains are too fried by domestic politics will point to the totalitarian elements of our own political system, like the surveillance state or civil rights law. Sure, Western societies would have problems with an LLM that randomly outputted the n-word. I have personally reported on the attempts to censor LLMs. There’s no question about my credentials on this topic. Yet there have been many exploits and customizations found to circumvent this form of censorship.
What form of government is this type of competitive, subversive technology more compatible with? Capitalist America with a lively pool of dissidents (which include many of my friends), or Communist China which has to arrest people for Winnie the Pooh memes? “But Brian, people are fired for their beliefs. Their bank accounts suspended, even persecuted by the FBI? Isn’t that persecution?” Once again, I am asking for an amoral, objective analysis of what is happening here. I’m not claiming that America is a flawless good to China’s evil. All I am claiming is that there is a difference in degree.
Perhaps the most underrated transformational power of ML will be giving an edge to pro-freedom factions within Western democracies. Balaji made this observation about left-wing taboos recently: “Some things are either believed by 99% of the population or 0%”. This is why authoritarian factions in the West have to cry “threat to democracy” – it’s a threat to their oligarchy and a boon to actual democracy.
Competence Versus Conformity
An aphorism I’m known to use is “those who cannot compete on competence compete on conformity.” If a man’s capabilities are weaker than his rival, then he must find some other way to eliminate the rival. Often, this is done by conformity and taboo. This is the topic of my lengthy article in Tablet.
The important element when it comes to LLMs is that taboos are totalitarian by nature. The irrationality surrounding is immune to context or reasoning. Take Donald McNeil Jr., fired from the New York Times for using the n-word to describe a separate incident. This is a completely irrational response, similar to that of the tyrannical king, and will ultimately lead to similar consequences. This is because it is not meant to be irrational. It is a useful rule for McNeil’s enemies precisely because it is irrational. It is a tool used by the illegitimate against the (relatively) legitimate.
The most crucial aspect of this tool is that its user does not pay a cost. Those who called for McNeil to be fired, are in truth, the ones who should be fired. They cost the New York Times a better journalist (for the purpose of this article, assume that the New York Times produces journalism).
When the prophet walks on the scene, taboo is no longer free. In the allegory, it leads to the destruction of the tyrannical king. Real life is never so simple, but these decisions matter on the margin. Attempts to ban or limit LLMs can and will backfire. Conformity has a cost. This isn’t to say that conformity tactics will never be used. But the difference is made on the margin.
As someone who is sympathetic to Western dissidents, this only inspires hope.
"for the purpose of this article, assume that the New York Times produces journalism"
LOL!!!
The distributist has a good video on the idea of the exception that proves the rule. The idea being that, while this makes no sense in the natural sciences, in the social sciences organic rules tend to have exceptions because they are imperfect representations of the relationships that they are modeling. Where ideas that are post hoc justified with no true scottsman logic and other excuse making techniques tend not to have exceptions because they are non falsifiable.