The Tool That Replaces Judgment

Débora Kaz ·

“I cannot stand seeing artificial intelligence images anymore.”

I read this sentence often.

Sometimes I agree.

I am tired too of the repetition of characters with the same expression, perfectly smooth skin, dramatic lighting, grandiose architecture, and a fog that seems to have been permanently added to the staff of visual creation.

There is a saturated aesthetic. Everything looks cinematic, loaded with meaning, and ready to receive a grave soundtrack. After a few seconds, almost nothing remains.

But the expression “AI image” has begun to contain very different situations.

First, there is a problem of identification.

A person recognizes certain visual clichés and concludes that artificial intelligence was involved. They may be right or wrong.

Smooth skin, diffuse light, symmetrical compositions, excessive detail, and certain deformations have begun to function as clues. In some cases, they are. In others, they belong to digital practices that existed before generative systems.

Works produced without AI are accused of using it because they have a certain finish, texture, or visual organization.

There are artists who worked that way before the technology existed. Now they have to demonstrate procedural innocence, as though every highly polished surface had acquired a criminal record.

There are also uses of AI that do not display the appearance normally associated with it.

The system may participate in only one stage. It may be mixed with photography, video, modeling, drawing, archive material, and editing. It may generate an element that is later deformed, erased, reassembled, or almost completely hidden.

In that case, someone who says they are tired of “seeing AI” may not realize they are looking at it.

That does not make the work better. The invisibility of the process is not an artistic quality.

It only shows that visual perception does not provide the technical certainty some people believe they possess.

Then there is a second problem, more important to me.

Even when someone correctly identifies the use of AI, that information begins to function as the end of the reading.

The image used artificial intelligence, therefore it must be empty, without concept, without direction, without authorship. Provenance becomes a sentence.

At this point, we are no longer discussing whether someone correctly recognized the process.

We are discussing what they do with that information after recognizing it.

A person may know that generation was involved and still fail to perceive how that material participates in a video, an installation, a performance, or a broader research process.

They may identify the medium precisely and still misread the work.

The opposite happens too. Someone may know nothing about the technique used and perceive a tension that the specialist overlooked.

Knowing how an image was produced helps. It does not exhaust what the image produces when it is seen.

Remaining in front of it without reducing everything to the first available category already requires some willingness. And images rarely arrive alone. They come accompanied by captions, disputes, accusations, and declarations of purity. Before looking, we often already know which side we are supposed to be on.

“It looks like AI” is not always about AI

When someone says that an image “looks like AI,” they may be talking about something else.

It looks generic.

It looks too smooth.

It looks made only to impress.

It looks as though it does not know why it exists.

Or it simply looks bad.

Artificial intelligence has become the available name for a set of discomforts that existed before it.

Digital art produced polished, empty images long before generative models.

Advertising did too.

Stock-image libraries built an empire on people smiling in front of computers without anyone really believing in that happiness.

AI accelerated banality and made it more spectacular. Now it can have wings, monumental ruins, glowing eyes, and an enormous moon behind its head and still remain banal.

The public, however, did not invent this rejection alone.

Saturation is real.

Thousands of images are published without selection, editing, or context. The result appears, provokes a small burst of enthusiasm, and is immediately thrown into the world.

People using AI also helped build the fatigue.

It is not possible to flood networks with repetitive images and then treat every irritation as a critical failure on the part of the viewer.

Generation became easy. Publishing did too. Between one and the other, there should be some criterion, although enthusiasm has rarely been known for its editorial skills.

Before, technical difficulty already performed part of the selection. An image cost time. A video required resources. A variation could take hours.

Now the selection needs to happen afterward, and perhaps we have not yet developed much discipline for it.

Abundance produces fascination. Every result seems to deserve existence simply because it was possible to obtain it.

I have been carried away by this myself.

One image opens onto another. One solution suggests a path. Suddenly there is an entire series, and only later do I realize that quantity was hiding the weakness of the initial question.

It is easy to accuse beginners of lacking direction.

Harder is recognizing that experienced artists can also be seduced by technology. Repertoire does not protect us completely from enthusiasm. Sometimes it merely helps us construct a better justification afterward.

What we call force

Saying that an image has concept, direction, or force does not solve everything either.

These words can become another label.

Who decides that a work has force? The artist? The critic? The public? The curatorial text?

An image does not become profound merely because it is accompanied by a paragraph about memory, identity, and the collapse of subjectivity.

Some works receive a concept afterward, as though someone were placing an intellectual frame around them to prevent the image from noticing its own loneliness.

I distrust that.

I also distrust the opposite demand, that every idea must be immediately visible.

Some works depend on context. Others reveal themselves slowly. There are works that seem fragile in isolation and gain meaning when seen as part of a set.

The point is not to replace the automatic rejection of AI with the automatic acceptance of any image accompanied by a conceptual justification.

It is to prevent technical information from ending the reading.

An image made with AI can be delicate, superficial, violent, decorative, pretentious, powerful, or forgettable.

The same variety that has always existed in other media.

Even so, the debate often behaves as though every language had its own moral personality.

Painting would be human. Photography, documentary. Digital, artificial.

AI, fraudulent.

These classifications begin to fail as soon as actual works appear.

Evaluating an image also does not mean discovering an objective essence.

There is no universal detector of artistic relevance.

Knowledgeable people disagree. Artists fail to recognize the value of works that later become important. Critics build careers defending works that time receives with almost cruel indifference.

Judgment fails, including my own.

I can reject an image because I am tired of the aesthetic it uses and miss something interesting. I can value a work because I recognize its references and confuse familiarity with quality. I can attribute coherence to a work simply because I know the artist's previous research.

There is no completely safe position.

But there is a difference between being wrong after looking and refusing to look because the technical sheet has already supplied the answer.

When the presence of AI is enough to disqualify the work, we are no longer facing a reading. We are applying a classification.

It is faster and also more comfortable.

Some critics accuse artificial intelligence of standardizing images and then standardize their perception of all of them.

The machine repeats aesthetics. The observer repeats judgments. In the end, no one needs to confront difference.

Perhaps the greatest challenge is not learning how to identify synthetic images.

It is recovering the capacity to sustain attention in an environment that rewards immediate responses.

To look at composition, context, relationships, continuities, and failures. To ask what the presence of that technology changes in the experience, instead of merely confirming that it was present.

This does not guarantee that we will reach a better conclusion.

Perhaps it will only make the conclusion more difficult, and that difficulty is also part of reading.

Not every use of AI deserves defense. Not every refusal is conservative. Not every powerful image immediately reveals how it was constructed, and whoever believes they recognize force also carries preferences, repertoires, and fatigue.

The discussion becomes less orderly this way.

It is probably a sign that we are getting closer to the problem.