We Fear That AI Will Become Like Us

Perhaps we should be more concerned about what human beings may become with AI

For a long time, we have feared that artificial intelligence might become too human.

That it might begin to want.

That it might develop goals of its own.

That it might manipulate us, deceive us, rebel against us, or turn on us.

The story is old, even if the technology is new. From Frankenstein to HAL 9000, from Terminator to today’s discussions of superintelligence, the same question keeps returning:

What happens if the machine becomes like us?

But perhaps we are asking the question in the wrong direction.

Perhaps the greatest danger is not that AI becomes too human.

Perhaps the greatest danger is what human beings become when they acquire AI.

An old problem with new means

Human beings have never lacked the capacity to cause harm.

We have waged war. We have built prisons and concentration camps. We have developed propaganda, tortured, surveilled, manipulated, and killed. There is no reason to romanticize humanity’s moral history.

What is new about artificial intelligence is therefore not that evil suddenly becomes possible.

What is new is the scale. The speed. The precision. The automation.

Human beings are gaining a tool that can analyze millions of people at once, produce propaganda on an industrial scale, map vulnerabilities, identify targets, monitor behavior, imitate voices, manipulate images, develop code, and coordinate actions without fatigue.

Humanity’s old tendencies acquire new amplifiers.

This is where the question of AI becomes a question about the human being.

Technology has no morality

It is tempting to ask:

Can we trust artificial intelligence?

But the question may be too simple.

AI is not a single actor. It has no unified purpose, no political ideology, no nationality, no morality.

What matters is who sets the goal, who controls the system, what limitations are built into it, and what kinds of actions the technology makes possible.

A language model can help a student understand Aristotle. It can also help a propagandist produce thousands of convincing lies.

An analytical system can detect cancer earlier. It can also be used for mass surveillance.

An autonomous agent can identify security vulnerabilities in order to protect a computer system. It can also identify the same vulnerabilities in order to destroy it.

Technology is not morally neutral in the sense that all uses are equally good.

But neither is it morally responsible.

Responsibility remains human.

Three forms of distance

There is an old philosophical theme that appears again and again when we reflect on what power does to human beings:

distance.

Not necessarily physical distance, but the distance that arises between an action and its consequences, between the one who decides and the one who is decided over, between what I do and what I experience myself as doing.

Artificial intelligence has an unprecedented ability to produce precisely this kind of distance — on an industrial scale.

Emmanuel Levinas argued that ethics begins in the encounter with the other.

The face of the other makes a demand upon me.

You must not reduce the other person to a thing, to something that can merely be measured, sorted, classified, or used.

Much of digital technology pulls in the opposite direction.

The human being becomes data.

A pattern.

A profile.

A risk.

A probability.

A customer.

A voter.

A potential threat.

The more data we collect, the easier it becomes to believe that we know the person.

But there is a difference between knowing a great deal about someone and encountering them.

Artificial intelligence can widen that difference. A regime no longer needs thousands of employees to monitor a population. A company does not need to know a person in order to influence them. A military organization does not need to see the face of the person who is to be struck.

Levinas’s face does not disappear.

It merely becomes possible to act as though it were not there.

That is the first form of distance:

distance from the other.

The second form was described by Hannah Arendt, without machines in mind at all.

She showed how modern bureaucracies can create a kind of irresponsibility precisely because action is divided up.

One person makes the list.

Another organizes the transport.

A third keeps the accounts.

No one experiences themselves as the one who, alone, is doing evil.

The system does it.

Artificial intelligence can widen this distance even further.

When an algorithm selects who should be monitored, who should be denied access, who should be investigated more closely, or who should be regarded as dangerous, the human being can more easily say:

The system assessed it that way.

The model recommended it.

The algorithm found the pattern.

We can already see the outlines of this today in far more ordinary contexts than surveillance states: in automated recruitment filters that discard applicants before a human eye has seen them, in complex credit-scoring systems where responsibility is difficult to locate, and in content-moderation systems where decisions emerge through a chain of technical and organizational processes.

In each case, there is a decision.

But it becomes harder to point to who actually made it.

A recommendation from a machine does not remove human responsibility.

On the contrary, it may make that responsibility more difficult to see.

When a decision emerges through a chain of models, data, automated agents, and organizational procedures, everyone can feel a little less responsible.

That is a dangerous form of moral fog.

The third form of distance is the oldest of all.

Plato captured it in the story of the Ring of Gyges:

What would a human being do if they could act without being seen?

Invisibility is also a form of distance — the distance between an action and the gaze that would otherwise hold the actor accountable.

Today, the question can be formulated differently:

What will human beings do when they can see almost everyone else, without being seen in return?

Three thinkers, three historical contexts, the same basic structure:

When the distance between action and responsibility increases, it becomes easier to do what one might otherwise refrain from doing.

Artificial intelligence is not the first technology to create such distance.

But it may become the most effective we have ever built.

Evil without hatred

We often associate evil with powerful emotions.

Hatred.

Rage.

Fanaticism.

Sadism.

But great harm does not always require such emotions.

Sometimes indifference is enough.

Or efficiency.

Or obedience.

Or career ambition.

Or simply the desire to win a market.

Arendt wrote about the banality of evil precisely because evil does not always arrive with a demonic face.

It can come through people who perform tasks, follow routines, and fail to ask questions — people who act at a sufficient distance from the consequences for the question of responsibility itself to weaken.

AI can fit disturbingly well into this form.

Not because AI is evil.

But because the technology can make action so efficient and so remote that the question of responsibility disappears behind the system’s performance.

When the constraints disappear

It is easy to imagine dangerous AI systems in the hands of dictators, terrorists, or criminals.

That is a real risk.

But there is another, quieter one:

People with good intentions can also create dangerous systems.

They want to improve security.

Make healthcare more efficient.

Fight crime.

Improve education.

Prevent terrorism.

Optimize the economy.

The intentions may be good.

But good intentions do not automatically protect us from bad consequences.

History is full of systems that began with a desire for order, control, or improvement and ended by reducing the human being to a means.

Here it is worth pausing over something Aristotle might have pointed to, though from a somewhat different angle than the one we first expect.

Practical wisdom — phronesis, the ability to judge whether a means ought to be used, not merely whether it works — has historically had an ally we rarely mention:

limitation.

A person with harmful intentions has previously been limited by time, knowledge, resources, and capacity.

A government has been limited by how many people it could employ.

A propaganda organization has been limited by how much content it could produce, adapt, and distribute.

A surveillant has been limited by how many people could actually be followed.

These limitations were never virtues in themselves.

But in practice they functioned as a kind of external phronesis — boundaries that forced a degree of restraint even where inner judgment was lacking.

Artificial intelligence can remove these limitations one by one.

And when they disappear, it becomes clear how much of humanity’s historical restraint was never inner wisdom in the first place, but simply lack of opportunity.

It is not the machine’s phronesis that is missing here.

It is our own, exposed when the external constraints that long concealed its weakness fall away.

The human being with amplified will

Artificial intelligence can be understood as an amplifier.

It amplifies knowledge.

Creativity.

Science.

Productivity.

But it can also amplify fear.

Suspicion.

The desire for control.

The desire for power.

The will to win.

We often discuss whether AI will one day develop intentions of its own.

But long before that happens, if it ever does, the technology may make human intentions vastly more effective than before.

It is not the machine’s will that is growing.

It is the reach of the human being.

Trust

That is why the question “Do you trust AI?” is insufficient.

We need to ask:

Do we trust those who develop the systems?

Those who own them?

Those who decide which limits are built into them?

The states that regulate them?

Military authorities?

Technology companies?

Intelligence agencies?

Ourselves?

Trust cannot concern only whether the model gives correct answers.

It must also concern power.

Who can use it?

For what?

Against whom?

Who can say no?

And who can control the controller?

That last question is ultimately not technical.

It is political.

And as long as we treat it as a question for engineers alone, we will be looking for the answer in the wrong place.

An old temptation

There is an ancient human dream of a tool that makes the will limitless.

The magic ring.

The all-knowing machine.

The perfect weapon.

Total oversight.

AI is none of these things.

But the technology brings us closer to a situation in which certain people and institutions may acquire capacities that were previously unimaginable.

And then we return to one of philosophy’s oldest questions:

What does power do to the human being?

The decisive question

We therefore do not need to stop asking whether artificial intelligence can become dangerous.

It can.

We need safety, regulation, technical constraints, transparency, and oversight.

But at the same time, we should be careful not to let fear of a future artificial will blind us to the human will that already exists.

For the real danger may be far less futuristic.

A person wants control.

AI makes control more efficient.

A regime wants obedience.

AI makes surveillance cheaper.

A company wants our attention and learns, faster than any human could, how to hold on to it.

A military actor wants to identify a target.

AI finds it faster.

No machine needs to hate us.

It is enough that human beings use it — at the distance the technology now makes possible.

So perhaps the most important question is not:

What happens if artificial intelligence one day becomes like us?

It may be this:

What happens to human beings when our ancient desires for knowledge, control, and power acquire artificial intelligence as their instrument?

We fear that AI will become like us.

Perhaps we should be more concerned about what we may become with AI.

We fear that AI will become like us.

Perhaps we should be more concerned about what we may become with AI.

This essay was written in a conversation with Claude/Anthropic and OpenAI/ChatGPT