AI Is a Mirror, Not a Mind

A recent Communications of the ACM article asks why Nietzsche might matter in the age of artificial intelligence.

The connection initially sounds like the kind of thing that happens when we run out of contemporary metaphors and start recruiting dead philosophers. Nietzsche didn't know anything about language models, and I doubt much is gained by pretending he somehow anticipated them.

There is, however, a question he took seriously that feels newly useful.

What happens when something that once helped organize meaning stops carrying the weight we placed on it?

Nietzsche's declaration that "God is dead" wasn't simply an announcement that some people had stopped believing in God. He was concerned with what followed from the weakening of a framework through which European culture had organized morality and meaning. Removing an answer doesn't remove the questions the answer had been carrying.

AI obviously isn't the death of God.

It is putting pressure on some much smaller arrangements that we've nevertheless allowed to carry a surprising amount of identity.

For a long time, asking someone what they did for a living told us quite a bit about how to begin understanding them.

She's a journalist.

He's a programmer.

She's an illustrator.

He's a teacher.

Those descriptions never contained the whole person, but they compressed a great deal. They suggested training, competence, daily activity, social role and sometimes status. People spent years acquiring the abilities that made those descriptions credible.

Then a machine produces competent prose, writes functioning code, generates an illustration or explains calculus.

That doesn't make the machine a journalist, programmer, illustrator or teacher. Neither does it make the human role meaningless.

It does make the relationship between the role and the capability less obvious.

If producing code was part of the evidence that someone was a programmer, what happens when millions of people can produce useful code with assistance from a machine?

Perhaps nothing important.

Calculators didn't eliminate mathematicians. Cameras didn't eliminate painters. Search engines didn't eliminate people who know things. Technologies routinely change which parts of a skill are scarce without eliminating the larger practices surrounding them.

AI may do the same.

Still, I think it's worth noticing the pressure because many of us have allowed competence to answer questions it was never quite capable of answering.

I can write this, therefore I'm a writer.

I can build this, therefore I'm a programmer.

I know more about this subject than most people around me, therefore this knowledge tells me something important about who I am.

Those aren't absurd conclusions. What we repeatedly do becomes part of our lives, and other people understandably come to recognize us through it.

The trouble appears when the capability and the identity become indistinguishable.

AI separates them experimentally.

A person who has never programmed can now produce some working software. Someone who doesn't draw can generate an image. Someone who hasn't spent years learning prose can produce an acceptable business letter in seconds.

The artifact no longer tells us as much about the person who produced it.

That creates a practical problem in education, employment and authorship, but it creates a more interesting personal question too.

If the artifact becomes weaker evidence of identity, what evidence remains?

My first instinct is to answer with persistence.

What do you return to?

What do you continue doing when the novelty disappears, when nobody is impressed, when the work becomes tedious or when the first version fails?

There is something useful in that question, although I wouldn't turn it into another definition of identity.

People return to harmful things too. Habits persist. Fear persists. Obligations persist. Someone can spend thirty years doing work they dislike because it feeds their family. Another person can abandon a lifelong practice and discover that leaving it was one of the more important changes they ever made.

Persistence is evidence.

It isn't essence.

Maybe that's the more interesting consequence of having familiar labels weakened. We don't necessarily discover the "real" person underneath them. We get less certainty about how much the labels were explaining in the first place.

A journalist is still a journalist, but perhaps journalism was never identical to producing sentences.

A teacher was never merely the person capable of supplying the correct answer.

A programmer wasn't reducible to remembering syntax.

An illustrator wasn't simply a mechanism for causing pixels to appear in the correct arrangement.

Once a machine can perform portions of those activities, we're forced to make finer distinctions about what the larger practice contains.

Some of those distinctions may flatter us.

Others won't.

There may be work we regarded as uniquely expressive that turns out to be fairly reproducible. There may also be apparently ordinary parts of a role whose importance becomes visible only after we automate something around them.

A local reporter knows that a town board approved a project two years ago and remembers the argument that preceded it. A teacher notices that a student who normally talks has become unusually quiet. A programmer recognizes that the requested feature solves the wrong problem.

AI may assist with any of those things.

The point isn't to draw a permanent boundary and declare everything on one side human.

The boundary will move.

The useful question is what becomes visible as it moves.

This is where Nietzsche becomes interesting to me without needing him to have predicted any of this.

When an inherited explanation loses some of its authority, we have choices.

We can defend the old category more aggressively.

We can declare the category meaningless.

Or we can examine what the category had actually been carrying.

That applies to moral categories too, although I would be careful here. The fact that technology changes doesn't mean good and evil have become obsolete or that morality must be replaced by some new principle of adaptability. Some moral commitments may be valuable precisely because they resist changes in circumstance.

At the same time, applying a principle to a new situation can reveal that we didn't understand its boundary as well as we thought.

AI is producing plenty of those situations.

What counts as authorship when a machine contributed half the sentences?

What does plagiarism mean when generated language isn't copied from a single identifiable source?

What responsibilities belong to someone who deploys a system whose individual outputs they didn't choose?

When should a machine be permitted to make a decision rather than merely advise a person making it?

Those questions don't prove that inherited morality has collapsed.

They give inherited principles new cases to encounter.

Sometimes the principle will survive almost unchanged.

Sometimes we'll discover that we had confused the principle with the particular form through which we were accustomed to seeing it.

That is enough uncertainty for me. I don't need to conclude that AI is leading us toward greater authenticity.

It might.

It might also produce enormous quantities of performance.

People can use AI to make themselves appear more knowledgeable, more productive, more articulate or more interesting than they otherwise would. Institutions can use it to generate the appearance of human attention without supplying much human attention at all.

The same technology that weakens one performance of identity can make another easier.

So I don't think AI is a mirror either.

A mirror suggests passive reflection. These systems transform what we give them according to learned patterns, instructions, interfaces and the circumstances in which we use them. They can expose assumptions in our institutions, but they can also introduce behavior that wasn't present in any simple sense before.

That makes the encounter more interesting than reflection.

We built systems capable of performing activities that we had attached to human roles, and now those systems are changing the environment in which the roles themselves develop.

The result will probably be uneven.

Some professions will change dramatically. Others will absorb AI as another tool. Some people will find that work they regarded as central to their identity matters less to them once a machine can perform parts of it. Others may discover the opposite: removing mechanical parts of the work leaves them with a clearer view of why they cared about the practice in the first place.

We won't know which interpretation applies merely by watching what AI can do.

We'll have to watch what people do next.

That's where I still find something Nietzschean in the moment.

Not because God died twice.

Not because machines have completed some historical movement toward self-authored meaning.

Something simpler has happened.

One of the things we used to point toward and say, "That tells me who I am," has become less reliable as an explanation.

We can replace it with another answer immediately.

Or we can leave the question open long enough to see what else we were carrying.