I have always needed to see before calculating. That is not a figure of speech. It is probably one of the oldest characteristics of the way I think. Before the numbers, before the equations, before the lines of code, something has to take shape in my mind.

I need to be able to move around it. Understand its masses. See what lies in front, behind, above. Mentally shift the elements until a coherence appears. Only then do the instruments come in.

Calculation verifies.

Technique measures.

The engineer checks.

The machine executes.

But the first gesture lies elsewhere. It lies in that strange ability to construct a form before it exists.

In the late 1980s, this intuition encountered a technology that would profoundly change my relationship to images. It was called Lumière 999. I did not build it alone. I want that stated precisely because technological histories have an unfortunate tendency, over time, to manufacture a lone inventor. François Moillo was its coauthor with me.

François played a far deeper role in my formation than the story of a graphic workstation alone might suggest. For many years, he transmitted to me an extraordinarily concrete knowledge of art: materials, techniques, processes, alterations, expertise, fakes, forgers, and the way an artwork bears within its very body the trace of how it was made.

He could look at a material and tell almost immediately what had happened to it. He knew that an image can never be reduced to what it represents.

It has a skin.

A history.

A chemistry.

A light.

A technique.

A succession of gestures.

Spending time with him changed the way I looked. To understand an image, you had to learn to see what lay behind the image.

We were then at a very particular moment in the history of computing. The general public was beginning to discover personal computers that were still extremely limited compared with what we handle today.

The digital image, as we now know it, had not yet become commonplace. A photograph was not something you instantly moved from one phone to another.

Working with a very high-definition image meant confronting considerable volumes of data, scarce memory, computation times, and specialized peripherals. That was precisely what attracted me. I have never been fascinated by a technology simply because it was new. It interested me when it made it possible to see something that until then had remained invisible.

With François, we began working on graphic tools that were extremely advanced for the time. Some of the technological building blocks available to us came from the Matra environment and from transfers into civilian applications of particularly sophisticated graphic-processing and recognition technologies.

Today, I prefer to leave military fantasies aside. They add spectacle and often subtract intelligence. What interested me was much simpler: techniques developed in highly advanced industrial environments were becoming available for thinking about images differently. What could we do with them? That was the question.

We called this architecture Lumière 999. The number was not accidental. 999 would accompany me for a long time. Much later, it would reappear in the initiatory film of the Abode of Chaos.

But at the time, it referred first to a machine, or more precisely to a new way of envisioning the creative chain. One sentence from our documents of the period sums it up perfectly: the Lumière 999 graphic workstation had been designed to make it possible to conceive everything, from the rough sketch to the finalized artwork.

Everything was already there. I did not want to build a specialized tool confined to a single profession. I wanted a machine capable of accompanying the movement of an idea all the way to its nearly final form.

A building, a boat, an industrial machine, packaging, a map, a poster, a book, an object, an image, or a placement within a landscape: everything could pass through the same logic. Digital technology was no longer meant merely to draw what already existed. It had to make it possible to see what did not yet exist.

That shift changed the relationship to decision-making. Today, we are so accustomed to simulations, 3D renderings, virtual architectures, and generated images that we forget what this idea meant in the late 1980s. Seeing a project before it was built changed the relationship to decision-making.

You could try.

Move.

Change a material.

Modify a color.

Transform a perspective.

Go back.

Compare several futures.

The cost of error changed in nature because part of the error could now occur within the image before occurring in actual matter. That was exactly what I was looking for.

See before building.

See before producing.

See before calculating definitively.

We were working simultaneously with two worlds that are sometimes wrongly opposed: vector and raster.

Vector constructed the geometry.

The lines.

The curves.

The volumes.

The calculable contours.

Raster introduced another dimension: photographic matter, the pixel, texture, everything that resists pure geometry. The two had to be made to communicate. A perfectly vectorized object can be technically correct and visually dead. Reality is never perfectly clean. There are accidents.

A light.

A shadow.

A trace.

Vegetation.

Skin.

A human presence.

It is often imperfection that makes an image credible. That lesson stayed with me. I would encounter it again much later with artificial intelligence. A model can produce something mathematically plausible without producing something our eyes accept as alive. Between calculated coherence and perceived truth lies a much more complex territory.

We worked on monochrome screens with a resolution that remains, in my memory, extraordinary for the period, on the order of 4,000 by 6,000 pixels in certain configurations. I deliberately give that figure as a technical memory, not as a certified specification that I claim to recall forty years later down to the last pixel.

But the order of magnitude says something. We were a very long way from ordinary computer imaging of that period.

Those screens were monochrome. That seems surprising today. One might think a color screen would be necessary to produce a color image. Not at all. Color could be decomposed. We worked separately on the red, green, and blue passes. RGB. Not CMYK.

Each component could undergo its own processing, its own calculations, its own corrections. Then the three sets of information were recombined. The final color was therefore the result of a succession of operations that the eye never saw simultaneously while they were being produced.

This way of working taught me a great deal. It showed that what we call an image is already a reconstruction. We think we see a unity. The machine sees components. We think we perceive a color immediately. The system decomposes it.

We think the image is simply in front of us. In reality, it is produced by a series of invisible operations. That technical discovery connected with something I already knew intuitively: the visible is always the surface of an invisible architecture.

Some images could represent volumes of data that I remember reaching 330 megabytes, which, when placed back within the computing capacities of that period, completely changes the perception of the experience. Three hundred megabytes today means almost nothing. At the time, it was massive.

You had to think about storage.

Transfer.

Memory.

Processing.

Waiting.

Output.

The slightest operation needed a reason. You did not launch a hundred variants out of distraction. Every calculation had weight. That constraint could seem burdensome. It also had a virtue. It forced you to think before acting.

Seeing before calculating did not mean rejecting calculation. It meant knowing what direction to give it.

Output was an adventure in itself. We could work all the way to producing images intended for high-quality photographic output, particularly on large-format Ektachrome.

The image then left the machine. It became light again. Then chemistry. Then matter. That passage fascinated me. Digital technology did not abolish older techniques. It connected itself to them. Sequences of numbers ended up on a photographic emulsion that you could hold in your hands.

I have always loved these changes of state.

An idea becomes a number.

The number becomes light.

Light becomes matter.

Matter becomes an image.

And the image returns to the eye of a human being.

That circuit already constitutes a philosophy of information.

François Moillo contributed enormously to my understanding of this. He was scientific enough never to give in to vagueness and artist enough to know that an exact measurement never exhausts an image. It was this dual quality that interested me.

I have never liked the simplistic division between technicians and artists. It is often produced by people who know little about either world. Leonardo da Vinci would probably not have understood that separation. Constructing a perspective, studying optics, dissecting a body, observing a flow, or painting a face all belonged, for him, to the same activity of knowledge.

With the Lumière Stations, I found something of that unity again. The machine was not the enemy of art. It constituted a new laboratory for the eye.

It was also during this period that our work on digital imagery encountered, in an almost dizzying way, the very origins of cinema. I became close to Bernard Chardère, whose knowledge of film history and the Lumière heritage was immense. We worked around the Lumière films and original materials.

And once again, my reaction was the same as in every field I approached. I wanted to see the source. Not merely the commentary on the source.

The document.

The film stock.

The frame.

The physical object.

This obsession would run through my entire life.

In law, I go to the text.

In the Art Market, to the auction catalog.

With a signature, to the stroke.

With a building, to the archive.

With cinema, to the film.

I distrust successive layers of interpretation when they end up making us forget what they are interpreting.

With the Lumière films, this method took on an almost physical meaning. It was not enough to say that one version was older than another. We had to look for clues.

Compare.

Look at the differences.

Study the frames.

Examine the shadows.

The clothing.

The light.

The architectural elements.

The inscriptions.

Compare what appeared in the image with what the archives could establish. Some of this work contributed to documenting the history of the different versions of Workers Leaving the Lumière Factory and to strengthening the identification accepted by researchers of the March 19, 1895 version as the earliest known version.

I want to preserve that cautious wording. Personal memory loves certainty. History requires evidence. Forty years later, I do not want to transform a magnificent memory into a claim broader than what the documents establish.

What interests me more is the method. We were using some of the most advanced technologies available to us to interrogate images from the very dawn of cinema.

That encounter still moves me when I think about it. On one side, a few dozen seconds filmed in Lyon at the end of the nineteenth century. On the other, digital systems announcing a radically new way of producing and examining images. Almost a century separated the two. And yet they met on the same screen.

I believe it was then that the name Lumière acquired its full depth for me. It was not commercial packaging. There was a lineage. The Lumière brothers had transformed humanity’s relationship to the moving image.

We were entering another age in which the image could become calculable: modifiable, decomposable, interpolatable, transportable as information, restorable through calculation, then recomposable.

The same word united two thresholds. Lumière. Physical light exposing film. Then light reconstructed from data. There was something too exact in that for it to be merely a product name.

We also worked on processes that made it possible to enlarge, interpolate, and process images whose original definition was limited. The word upscaling is commonplace today. Contemporary systems, particularly those using artificial intelligence, perform operations incomparably more powerful than those available to us.

I especially do not want to create an anachronism. We obviously did not have today’s generative models.

But the intellectual question was already there:

What can reasonably be reconstructed between the information that is present?

That is an immense question.

When you enlarge an image, you cannot magically create a truth that does not exist in the source.

You can calculate.

Interpolate.

Compare.

Estimate continuities.

Reinforce certain structures.

But at every moment, a boundary arises between what the image contains and what the system adds. I could not have known it then, but that boundary would become one of the great philosophical questions of generative artificial intelligence.

At what point does reconstruction cease to be restitution? At what point does it become invention? And when does the invention become so plausible that we forget it was not present in the original?

We were already encountering, in a primitive form, the problem of evidence.

An image is never innocent.

It can be corrected.

Transformed.

Cropped.

Interpolated.

Recolored.

Composited.

As soon as the image becomes data, it gains new power but simultaneously loses part of the innocence the twentieth century had attributed to it. For a long time people said, “I saw it in the photograph.” As though that sentence were enough. Digital technology was going to destroy that certainty.

I think I understood very early that the image would gradually cease to be evidence in itself. The more powerful the production tools became, the more we would need to know where the image came from.

Who produced it.

When.

From what source.

With what transformations.

This issue would follow me all the way to Artprice.

An image of an artwork is never merely an illustration.

It can become an element of identification.

Comparison.

Provenance.

Expertise.

But only if we know what it is and where it comes from. It would also follow me all the way to the Abode of Chaos, which would become a gigantic machine for producing, accumulating, and confronting images of contemporary history. And it follows me today with artificial intelligence, at a moment when an image that never existed can be produced in a few seconds with an extraordinary degree of plausibility.

When I look at this continuity, I understand that the Lumière Stations were much more than a technological adventure. They were a school of doubt. They taught me that an image must be viewed as a construction. That its beauty says nothing about its truth. That its precision does not guarantee its origin. That resolution is not knowledge.

And above all, that seeing more does not automatically mean understanding more. You have to know what you are looking at.

I find François again in that lesson. His knowledge of fakes and artistic techniques naturally connected with this reflection. To understand a fake, it is not enough to know an artist’s style. You have to understand how it was made.

The support.

The pigment.

The material.

The tool.

The aging.

The habits.

The accidents.

Everything that escapes the general image and belongs to the depth of the object. The eye must learn to read several layers at once. That was also what the machine did.

It decomposed.

Isolated.

Measured.

Compared.

The human being and calculation could then become complementary. I believe that this is where my relationship to what we now call artificial intelligence truly begins.

I would be very careful not to say that Lumière 999 was AI in the contemporary sense. But I was already working in a world where a machine could participate in operations of recognition, processing, interpolation, simulation, and decision-making involving an image.

That necessarily changed the way we thought. The machine was gradually ceasing to be merely an arithmetic tool. It was becoming an auxiliary to vision. And as soon as a machine begins to participate in vision, a new question immediately appears:

What does it see that I do not see?

Then its mirror image:

What do I see that it will never be able to measure in this way?

Perhaps the entire relationship between the human being and artificial intelligence is already contained in that gap.

A machine can compare millions of pixels.

A human being can be stopped by a shadow.

It can measure a continuity.

He can sense an anomaly without yet knowing how to name it.

It can detect regularities.

He can understand that the irregularity is precisely what matters.

The mistake is to ask which of the two should win. That is not the right question. The question is how their forms of perception can complement one another without either one destroying the responsibility of the other.

This principle has remained constant in everything I have built. I never wanted technology to replace the historian. I wanted it to allow the historian to see farther. I never wanted it to replace the expert. I wanted it to allow the expert to compare more traces. I never wanted it to replace the artist. I wanted it to give the artist access to forms that had previously been impossible.

The machine becomes dangerous when we delegate to it the responsibility of thinking in our place. It becomes extraordinary when it expands the territory within which we can think.

That was exactly what I felt in front of those first images.

We could see a building before it was built.

An object before it was manufactured.

A landscape after a transformation that had not yet taken place.

An old photograph at a scale it had never known.

A red, green, or blue component separated from the others.

A satellite image as a territory of data.

The visible world suddenly revealed itself as something that could be decomposed and recomposed. That could have led to a fascination with technology for its own sake. In me, it produced instead a philosophical unease. If the world can be decomposed into data and then reconstructed, what still guarantees that its reconstruction is the world?

That question has continued to grow. Today it lies at the heart of our civilization. We are entering a time when the distinction between what was captured and what was generated sometimes becomes inaccessible to the eye alone. Synthetic images are no longer recognizable by their poverty. They can now surpass photography in their power of seduction.

We will therefore have to rebuild elsewhere the trust we once placed in the eye.

In sources.

In traceability.

In metadata.

In the chain of custody.

In confrontation.

In adversarial evidence. This is exactly the same logic that led me, in documentary databases, never to separate information from its origin. The image belongs to this architecture of knowledge.

We had learned to make knowledge inhabitable. We had learned that information could become matter. With Lumière 999, that matter acquired an additional dimension. It could become visible before it existed.

I realize today how much that experience prepared the way I would build the Abode of Chaos.

For years, I was able to move mentally through spaces that had not yet been transformed.

Position masses.

Anticipate relationships.

See how one sculpture would enter into dialogue with another.

Sense that a perspective would work or that it had to be destroyed.

Calculation and professionals would then come in to verify what spatial intuition had produced. I never claim that intuition replaces an engineer. It gives the engineer something to verify. That is the entire difference. I do not ask calculation to create my vision. I ask it to tell me whether the physical world will accept it.

That is why “seeing before calculating” has never meant contempt for science. Quite the opposite. To truly respect a measurement, you have to know what you are asking it to measure. Vision formulates the problem. Science tests its possibility. Technique builds the passage.

This alliance would accompany me everywhere.

In databases.

In Groupe Serveur.

In Artprice.

In sculpture.

In architecture.

In the Abode of Chaos.

And today in my relationship with artificial intelligence.

The Lumière experience also taught me something else: the most advanced technologies become fascinating when they encounter memory. We could have confined ourselves to producing images of the future. We also used these tools to return to the earliest images of cinema.

That gesture probably contains my entire trajectory. I do not believe in progress that erases what came before it. I believe in technology that increases our ability to question what preceded us. The future interests me only if it has a memory. Otherwise, it is merely acceleration.

Perhaps that is why the name Lumière stayed with me.

It brought science and art together.

Chemistry and image.

Technique and memory.

Invention and transmission.

I also remember that period as a time when we went to meet those who had devoted their lives to the image. My time at Studio Harcourt Paris belonged to the same attitude. I did not go there as someone seeking an external mark of distinction.

I saw it as a place where a science of light, the face, and portraiture had accumulated. A professional memory. Knowledge polished over generations. I have always felt an almost instinctive form of respect for those who truly master a technique. True mastery rarely makes people arrogant. It makes them precise.

You recognize someone who knows a material deeply by the way he knows it can resist him.

François had that.

Bernard Chardère did as well, in his field.

So did the great artisans of the image.

I sought their proximity because their knowledge prevented me from becoming a prisoner of the technology itself. A new machine can very quickly create the illusion that everything begins with it. That is false.

Behind image-processing software lie centuries of perspective.

Behind a digital palette, centuries of color theory.

Behind a camera, optics.

Behind optics, physics.

Behind portraiture, an entire history of human representation.

Technology is never year zero. It is an additional layer.

That is precisely why the word transmission is so important to me. François Moillo did not simply transmit knowledge to me that I could later have replaced with other knowledge. He transmitted a way of looking. That is much more difficult to quantify.

A master does not merely give you answers. He changes the structure of the questions you will ask yourself for the rest of your life. I believe that is what happened.

When I look today at an image produced by artificial intelligence, part of me still observes it through eyes formed during those years.

I look at the matter that does not exist.

The shadows.

The continuity.

The inconsistencies.

The relationship between the detail and the whole.

I look for the fabrication behind the appearance. The technology has changed by several orders of magnitude. The inner gesture has remained. Never allow yourself to be hypnotized by the result. Look for the process.

That principle will take on a new significance in the world that is coming. For a long time, humanity has learned to distrust text. We know that a text can lie. We know that a witness can be mistaken. We know that propaganda can manufacture a narrative. We had retained a much deeper trust in the image.

It seemed to bear the physical trace of what had stood before the lens. That period is over. The image is becoming, in turn, a language capable of inventing with the realism of evidence. We will have to become more intelligent than our own eyes.

I do not find that tragic. I find it demanding. Every major technology takes away one of our forms of naivety. Then it forces us to construct a deeper knowledge.

Writing externalized memory.

Photography transformed representation.

Cinema transformed time into image.

Computing transformed the image into data.

Artificial intelligence is now transforming data into possibility.

At every stage, something is lost. At every stage, something else becomes thinkable.

We simply have to reject two symmetrical attitudes: fear of technology and worship of technology. I believe in neither. A technology is never a destiny. It is a power. And every power immediately raises the question of what we are going to do with it.

In 1987, standing before the Lumière 999 Stations, I obviously could not know the world that would be ours forty years later. I could not imagine that one day a machine would instantly produce a photograph of a scene that had never taken place. I could not foresee that billions of images would be produced every day, copied, transformed, analyzed, and distributed across an almost instantaneously connected planet.

But I could already see something. The image had just changed in nature. It was no longer merely captured. It was becoming calculated.

From that point on, everything became possible.

Wonder.

Knowledge.

Simulation.

Restoration.

Creation.

But also the fake.

Manipulation.

The disappearance of immediate evidence.

The same tool opened in every direction.

Perhaps that is what I mean by seeing before calculating. Seeing does not consist merely in anticipating the form of an object. It means trying to glimpse the consequences of a tool before its power forces us to endure them.

I obviously have not always succeeded. No one does. But I have tried to live that way. Look at an emerging system. Ask what it might become. And when an intuition seemed strong enough, begin building without waiting for the world to decide that the time had come.

Lumière 999 belongs to that history. François Moillo does as well. Not as a technological footnote. As one of the men who taught me that the eye can be educated, that matter speaks, that a fake always has a genealogy, and that a technology has meaning only when it increases the depth of our gaze.

The machines have changed. Monochrome screens have disappeared. Three hundred megabytes frighten no one anymore. RGB passes have become an abstraction ignored by most users. Calculations that once required patience have become instantaneous.

But the question has not aged. Perhaps it has become more important than ever.

What do we actually see when we look at an image?

And behind it:

Who constructed what we believe we are seeing?

All my life, I have tried to look behind the surface. The Lumière Stations were one of the first places where I understood that from then on, behind the image, there would also be algorithms.