When Reason Becomes Infrastructure, What Becomes of Human Agency?

Today, while listening to Yale's HIST 271, European Intellectual History since Nietzsche, a strange parallel occurred to me.

Nearly every historical change we call “progress” is accompanied by the retreat of something else.

During the Enlightenment, God began to retreat from the center of public life, and human reason rose in its place. During the Industrial Revolution, muscular power retreated as mechanical power ascended. After the arrival of the internet, information became less scarce and attention became more valuable.

By “retreat,” I do not mean disappearance. Religion did not vanish after the Enlightenment. Physical strength did not become meaningless when machines arrived. Information did not become worthless because of the internet.

What changed was their position in the order of civilization.

The old thing ceased to be the sole center around which society organized authority and measured individual worth. A new capacity moved from the margins to the center of the stage.

So what is retreating in the age of AI?

The easiest answer is work. But I increasingly suspect that work is only the surface.

What AI is really unsettling is one of the deepest convictions the Enlightenment left us: reason belongs to human beings, and because human beings can exercise reason, they are entitled to act as subjects in the world.

If reasoning, language, and creation are no longer exclusively human capacities, then the question is no longer merely which jobs AI will replace.

The question becomes: When reason is no longer enough to establish human uniqueness, what allows us to remain agents of our own lives?

Every Advance Is a Transfer of Authority

We tend to imagine technological progress as addition.

Cars help us travel faster. Computers help us calculate faster. AI helps us process knowledge faster. Technology appears to place ever more powerful tools in human hands while the “human” using them remains unchanged.

But truly consequential technologies do more than add capabilities. They redefine capabilities.

After clocks became widespread, time was no longer only the movement of the sun or a sensation in the body. It became a unit that could be divided, traded, and managed. After photography, an image no longer necessarily meant that a person had witnessed a scene. After search engines, remembering an answer gradually became less important than knowing how to find one.

Every technology quietly changes the rules of proof: what counts as knowledge, what counts as ability, what deserves trust, and who is entitled to speak.

AI is rewriting the most fundamental of those rules.

In the past, when we encountered a profound essay, a rigorous argument, or an extraordinary design, we moved naturally from the work to the worker: an output this intelligent must have come from someone who thought deeply.

That inference no longer holds.

An essay may have no connection to its ostensible author's experience. An apology may contain no regret. An answer to an ethical dilemma may have been produced without any being prepared to pay the price of acting on it.

AI is not making reason disappear. On the contrary, it is making the production of reason more abundant than ever.

What is beginning to retreat is humanity's monopoly on reason.

The Enlightenment Elevated More Than Reason

“God retreated; reason rose” is a useful historical sketch. But it is not the whole history.

The Enlightenment did not simply abolish religion, nor did it suddenly discover that human beings could think. What changed was the location of final authority. An idea, institution, or command could no longer expect immunity from scrutiny simply because it came from tradition or revelation. Human beings should have the courage to examine it through their own understanding.

In What Is Enlightenment?, Kant described enlightenment as humanity's emergence from self-incurred immaturity. His famous injunction Sapere aude—dare to know—was not merely an instruction to possess more knowledge. It was a demand that we dare to use our own understanding.

The Enlightenment therefore elevated two connected things: human reason, and the human being capable of using reason for self-government.

Over time, we compressed the two into one familiar figure: the modern subject, capable of thinking independently and deciding for itself.

AI pulls them apart again.

A machine can display reasoning without necessarily possessing an “I” to whom the world appears. It can offer choices without living inside their consequences. It can discuss the good without having to embody any conception of the good in a life.

For the first time in ordinary life, “I think” and “I am” are separating at scale.

The steam engine separated force from muscle. The computer separated calculation from the mind. Generative AI is beginning to separate language, reasoning, and expression from a particular person.

Reason is consequently undergoing a change in status. It is moving from a spiritual privilege of the human being toward infrastructure available on demand.

AI Is Not the Enemy of the Enlightenment. It Is Its Reversal.

In 2018, Henry Kissinger published an essay in The Atlantic titled “How the Enlightenment Ends”. He worried that AI could arrive at correct conclusions through processes human beings could not adequately understand, separating knowledge from human understanding.

That argument identified an important shift. But the problem today is no longer limited to opaque machine knowledge.

Generative AI is entering the most ordinary activities of the mind: writing a letter, understanding a book, forming an opinion, comforting a friend, or planning a life.

AI is not an enemy invading the Enlightenment from outside. It is the product of the Enlightenment's own effort to formalize, compute, and distribute reason.

The Enlightenment sought to free knowledge from the monopoly of inherited authority. AI carries that ambition close to its limit. Explanation, synthesis, argument, and expression are becoming services available to nearly everyone.

Yet the more completely this ambition succeeds, the less stable the subject it originally created becomes.

When everyone can summon an almost unlimited supply of rational output, the ability to produce answers no longer establishes distinctive human value. Reason has not been defeated. It has lost its scarcity by becoming too successful.

This is not simply the end of the Enlightenment. It is the Enlightenment turning back upon itself: the technology created by reason is beginning to destabilize the human being who defined itself through reason.

When Answers Are No Longer Scarce, What Rises?

Technology rarely abolishes scarcity. It relocates it.

Printing reduced the scarcity of books and increased the importance of literacy and selection. The internet reduced the scarcity of information and made attention and credibility expensive. AI is reducing the cost of answers, explanations, and expression. In their place, another set of things becomes scarce:

Why is this question worth asking?
Which end is worth pursuing?
How should we choose between compelling but conflicting answers?
Who is willing to recognize the consequences of that choice as their own?

A machine can show us the fastest route to a destination. Calculation alone cannot decide which destination deserves to be reached.

It can use our past preferences to predict what we are likely to enjoy. It cannot thereby answer whether obeying those preferences constitutes the life we want to live.

It can generate ten internally coherent plans. It cannot eliminate the conflict among values in the world. Many important decisions are difficult not because information is missing, but because every choice requires us to surrender something else worth preserving.

In the past, limited knowledge often concealed this difficulty. We assumed that if only we knew more, the correct answer would eventually reveal itself.

AI is removing that concealment.

Once answers become abundant, we discover that many questions in life and politics were never calculation problems. They were questions of authorship.

The real question is not which option is optimal in the abstract. It is: Which value are you prepared to bring into the world, and are you willing to become the author of that choice?

The Agent Is Not Necessarily the Most Intelligent Being

At this point, it is tempting to treat responsibility as humanity's final moat.

The machine produces the answer; the human remains responsible. The machine gives the recommendation; the human signs the document. As long as responsibility still lands on a person, it seems, the human position remains secure.

But that is not enough.

Responsibility is not the only test of being a subject. Infants and people with severe cognitive disabilities cannot bear full responsibility, yet this does not erase their moral standing or their status as human subjects. A corporation can bear legal liability without possessing a first-person experience of the world.

And if an algorithm defines the objective, a system makes the choice, and a person is brought in only to absorb the consequences, that person is not acting as an agent. They are merely providing a signature for the system.

A more complete idea is self-authorship.

An agent does not appear only after something goes wrong. An agent participates in writing the entire chain of action:

Setting an end. Making a choice. Giving a commitment. Accepting the consequences.

Looking forward, an agent decides what is worth pursuing. In the present, an agent chooses among things that cannot all be preserved. In relation to others, an agent makes commitments on which others can rely. Looking backward, an agent acknowledges that the decision was theirs and allows its consequences to change them.

A machine can generate a sentence. An agent must still be able to say tomorrow: what was said yesterday counts as mine.

A machine can optimize an objective. An agent must ask why that objective deserves to exist.

A machine can predict my behavior. An agent retains the capacity to say no to the prediction.

The agent, then, is not necessarily the most intelligent being.

The agent is the being capable of recognizing its ends, choices, and consequences as its own.

What Else Should We Be Careful Not to Outsource?

Self-authorship is not the whole of subjectivity.

Human subjectivity also arises from first-person experience. Pain is not merely a physiological signal; it is I am in pain. Love is not merely a pattern of behavior; it is one particular person being changed by another.

It arises from embodiment and vulnerability. Human choices carry weight because time does not reverse, bodies age, relationships break, and opportunities disappear. To choose one path is to lose others in reality, not merely to assign them a lower score.

It also arises through relationships. An “I” does not emerge in a vacuum. It forms by being addressed, answered, cared for, and recognized. A subject does not merely produce language. A subject can be wounded by a sentence, bound by a promise, and changed by the gaze of another person.

We cannot infer from an AI's fluent use of “I” that it does—or does not—possess any of these forms of experience. That remains an unresolved philosophical question.

But for human beings, the more urgent question is not whether we should immediately declare machines to be subjects. It is how to avoid voluntarily abandoning our own subjectivity.

When we give memory to search, expression to models, choice to recommendation systems, and eventually purpose itself to prediction, what we lose may be more than a collection of skills.

We may retain comfort, efficiency, and an infinite supply of answers while making fewer decisions about why we act at all.

The greatest danger of the AI age is not that machines will become increasingly human.

It is that human beings will gradually live as interfaces for machines.

Do Not Reduce Your Life to an API Call

None of this is an argument for refusing AI.

The car's replacement of walking did not require us to destroy the car. The calculator's replacement of mental arithmetic did not mean humanity had to return to the abacus. The real question is never simply whether to use a tool. It is whether, when a tool replaces one capacity, we recognize which new capacity has become more important.

AI may perform more and more of our reasoning. It cannot thereby possess our purposes.

It can help us see more options. It cannot release us from choosing.

It can predict outcomes, flag risks, and preserve memory. It cannot live a consequence into our lives for us.

What we need to protect in the age of AI, therefore, is not a final territory of human ability that machines will never conquer. Any border drawn around a particular capability may retreat as technology advances.

What we need to preserve is our position as the authors of our actions.

Before calling on AI, ask which end deserves pursuit. Before accepting an answer, retain your own judgment. After acting, do not transfer the result to the model. When a system knows your preferences better than you do, preserve the capacity to defy its prediction and remake yourself.

In the past, the ability to think was enough to establish the human being as a subject.

In the future, thought may become as abundant and invisible as electricity. That does not automatically make the human being obsolete.

It forces us, perhaps for the first time, to distinguish two things: reason produces answers; agents decide where those answers will lead.

When reason becomes infrastructure, human agency is no longer guaranteed by “I can think.” It must be continually realized through “I decide why to think, how to act, and whether I am willing to recognize the result as part of my life.”