I have never believed that an accumulation of information mechanically produces knowledge. Very early on, I understood exactly the opposite. The larger a body of documentation grows, the more opaque it can become. The more it claims to contain the world, the more it risks making that world inaccessible.

You can possess hundreds of thousands of pages, thousands of texts, entire codes, collective bargaining agreements, decisions, nomenclatures, directories, indexes, and still find yourself facing dead matter if no one has thought about how a human being will enter it.

This question followed me throughout all the years when we were building Groupe Serveur’s first databases. I did not ask only what we could put into a machine. I wanted to know how someone would retrieve what they needed. The difference seems tiny. It is immense.

The producer of information classifies it according to the logic of the institution that produced it. Government thinks in terms of its departments, jurisdictions, texts, and procedures. The lawyer knows the hierarchy of norms and the exact vocabulary that will allow him to enter a corpus. The document specialist has his own categories. The computer scientist naturally thinks in structures, fields, tables, and dependencies.

But the human being who arrives in front of the screen does not necessarily come with those keys.

He comes with a question.

Sometimes with an anxiety.

Sometimes with an emergency.

He is not looking for “the document.” He is looking for what will allow him to understand his situation. That is where, for me, the ergonomics of knowledge begins. We had to place ourselves on the side of the person who did not yet know the path.

I understood this with particular force while working on administrative and legal corpora. Take a collective bargaining agreement. It can be regarded as a perfectly identifiable legal object, bearing a number, a date, a succession of amendments, classifications, pay scales, and specific provisions. All of that is accurate. But the person consulting it does not necessarily see that object.

He sees his life: his occupation, his salary, his leave, an illness, maternity, retirement, a conflict, a classification on which an essential part of his existence may sometimes depend.

Behind the text, there is always someone. That may seem obvious today. At the time, this shift in perspective was decisive. It forced us to leave the logic of the person who possessed the knowledge and enter that of the person who was searching for it.

We then began what I readily called an autopsy of the data. The expression was deliberately crude.

We had to open up the material.

Cut it apart.

Understand its organs.

Search for its hidden joints.

Identify what depended on what.

See what appeared secondary and changed status when the point of view changed.

We intellectually dismantled a corpus before even considering rebuilding it in a machine. The architecture of the system was decided at this stage. I did not want the software to dictate its form to knowledge. I wanted the form of knowledge to gradually impose its architecture on the software. That changes everything.

You begin by understanding a field. Truly understanding it. You try to grasp its rules, exceptions, practices, vocabulary, temporalities, and bifurcations. Then you look for how the elements live with one another.

Only then does the tree structure appear. It is not a technical drawing. It is already a form of thought. At every branch, you take a position. At every level, you decide what is primary and what is secondary. At every intersection, you accept that a piece of information may belong to several worlds at once.

And very quickly, the classic tree is no longer enough. Reality does not behave like an organizational chart. The same information can be legal, economic, professional, social, and historical. It can be searched for through several paths that do not resemble one another. It may have one name in a profession and another within government. It may be legally accurate and still remain impossible to find for someone who does not know the right word.

That is where the thesaurus comes in. For me, a thesaurus was not merely a documentary technique. It embodied an obvious fact: two human beings may search for the same reality using different words. We therefore had to build bridges. We had to make it possible for users to enter the system with their own language without being condemned by their ignorance of institutional language.

This idea has always seemed profoundly political to me. Knowledge must not be reserved for those who already know the path that leads to it. Otherwise, it becomes a privilege.

The difficulty was never to simplify things to the point of making them false. On the contrary, I distrusted that temptation. Law is complex because human situations are complex. A collective bargaining agreement cannot be reduced to three sentences. A code cannot be flattened without losing its articulations. The problem was therefore not to eliminate depth. We had to make it possible to enter it. That is very different.

A good knowledge architecture must offer several depths of reading. Someone seeking an immediate answer must be able to approach it. Someone who wants to understand more must be able to go deeper. The expert must be able to reach the source, the text, its chronology, its amendments, and its relationships with other texts.

Ergonomics does not consist in reducing reality. It consists in preventing its complexity from being used as a barrier.

I often asked my teams not to look at competing databases before we had designed our own system. That sometimes surprised them. The business world naturally encourages us to observe what others are doing. Today we call that benchmarking. We compare interfaces, solutions, best practices.

That contamination was precisely what I feared. A form once seen never disappears completely. You may believe you have forgotten it, but it continues working somewhere in memory. To see is already to begin copying. Not in the legal sense of the term. In the cognitive sense.

I wanted to preserve, as far as possible, that rare moment when a team finds itself alone before a problem and has to produce its own answer. It is sometimes in that solitude that singular architectures are born.

This was not about pride. We knew perfectly well that similar solutions could appear elsewhere. A function often ends up imposing part of its form. Two engineers who do not know one another may arrive at similar answers when they work on the same problem. A car keeps four wheels because certain physical constraints naturally lead to certain convergences.

But I wanted to know what we would produce before being contaminated by what others had already produced.

This discipline became a method:

Understand first.

Model second.

Build the tree structure.

Think through the circulation.

Test the pathways.

Only then program. Never the other way around.

That required a considerable collective effort. I never believed that computer scientists could solve these problems alone. We needed lawyers. Document specialists. People from the profession. Developers. People capable of questioning inconsistencies. And above all, these worlds had to agree not to remain confined within their own territories.

The lawyer had to understand that fidelity to the law was not enough if no one could retrieve the information. The computer scientist had to understand that the elegance of a technical architecture meant nothing if it betrayed the real structure of the field. The document specialist had to accept that a classification perfect on paper could become impractical on a screen.

And all of us had to watch the users. Their hesitations were valuable. So were their mistakes. When a person kept making the same mistake in the same place, I rejected the easy conclusion that they had failed to understand. Perhaps we had built badly. A system that constantly forces users to bend to its logic often ends up revealing a weakness in its designer.

Ergonomics taught me this: when a passage resists, look at the passage before blaming the person trying to cross it.

I think this conviction owes a great deal to the way I have always looked at space. Since childhood, I have not seen only separate objects. I see the relationships among them.

Distances.

Proximities.

Passages.

Dead zones.

Perspectives.

What conceals.

What reveals.

This way of thinking expressed itself long before the Abode of Chaos. I find it again in these first documentary architectures. A database became a space for me. Information had a position. Neighbors. Levels.

Paths of access. Shortcuts. Dead ends. Vertical and horizontal relationships. We had to allow someone to move through this territory without requiring them to know its map in advance.

I did not yet know that, years later, I would physically construct a work in which this logic would become almost tangible. The Abode of Chaos is also a system of circulation. You never see everything from a single point.

One artwork reveals another. One material opens a perspective. A portrait enters into dialogue with an event located several dozen yards away. The visitor sometimes believes he is moving freely, while certain lines of force profoundly organize his gaze.

I do not place these two experiences on the same level. But today I recognize the invariant. Whether it is a database or a territory of nine thousand square meters, I ask myself how a mind will move through the matter.

What must appear first.

What must remain in waiting.

What will be understood immediately.

What will acquire meaning only after several passages.

The ergonomics of knowledge is already a mental geography.

It was also during these years that Nadège played a decisive role. An intuition is worth nothing if it cannot become reproducible. An idea can be brilliant on a board and collapse when it has to live every day, be corrected, updated, and absorb thousands, then hundreds of thousands, of additional documents. Together we learned this difference between designing a system and making it last. Nadège has that rare ability to bring a vision into the long term.

To confront it with reality.

With teams.

With procedures.

With repetition.

With everything that remains invisible when the story of an innovation is told afterward.

A documentary architecture is never finished on the day the software works. That is when its real existence begins. Texts change. Uses change. Volumes grow. Vocabularies drift. Categories that were once relevant cease to be. Others appear. Coherence must be maintained without freezing the system.

It is a discipline of extreme rigor. With a few dozen documents, almost anything can work. With hundreds of thousands, every initial weakness becomes enormous. A small design error repeated one hundred thousand times ceases to be small. It becomes a structure.

That is why we spent so much time before programming even began. One might have thought we were delaying the technical phase. In reality, we were preparing it. Code came at the end of an intellectual process of which it was only the translation.

This way of working gave us an advantage I did not fully understand until later. When we approached the Art Market, the material seemed completely different. It was. But the deeper problem was already familiar.

How do you transform masses of heterogeneous information into a coherent architecture?

How do you connect an artist, an artwork, a date, a dimension, a technique, a signature, an Auction House, a hammer price, a currency, a provenance, and the artist’s documented trajectory?

How do you allow several entry points into the same universe?

How do you preserve accuracy while making research possible?

Artprice was not born in a methodological desert. Behind it lay all those years of learning. The mistakes. Systems dismantled and rebuilt. Abandoned tree structures. Corrected thesauri. Observed users. Thousands of tiny decisions that never appear in official narratives but on which the solidity of an architecture depends.

Today I believe this history also helps us understand what is happening with artificial intelligence. One might think conversational interfaces abolish the old problem. No need to learn a tree structure. No need to navigate menus. You write a sentence. You ask a question. The machine answers. But the architecture has not disappeared. It has become invisible.

That may be the most dizzying change. In the era of the first databases, users could still see part of the system organizing their access to knowledge. The categories were visible on the screen. The trees were visible. The pathways could be challenged.

With artificial intelligence, a growing part of this mediation takes place behind natural language. The user asks a question and receives a result without fully seeing the territory traversed to reach it.

This makes even more urgent a question I was already asking myself forty years ago:

Who built the path?

According to which categories?

From which sources?

With which absences?

What was brought together?

What was kept apart?

What became central?

What was made peripheral?

Technology has changed scale. The philosophical problem remains. A knowledge architecture is never neutral. It silently decides what becomes visible.

I have devoted a considerable part of my life to reducing the distance between a question and the knowledge that could answer it. I believe that is what ergonomics is, at bottom. Not making things easy. Making the path possible.

There is a very particular violence in systems that hold information while leaving it practically inaccessible. They can claim that it is public. It is there. Available in theory. But those who know where to look have an immense advantage over those who do not.

I have always experienced this as an asymmetry that had to be fought. Between the person who possesses the vocabulary and the one who does not. Between the institution and the citizen. Between the expert and the person who simply arrives with a question. Between the mass and the person searching for a path through it.

Perhaps that is why I never truly regarded these systems as mere computing products. They already touched on a question of freedom. Freedom does not consist only in having information available. You must also be able to reach it. A library for which no one has the key remains materially full and humanly empty.

I never wanted to think in place of the person using our machines. I wanted to remove what prevented him from thinking. That is a distinction I care about. The best technology is not the one that replaces you. It is first the one that restores a capacity to you.

A good architecture almost disappears when it works. You no longer admire it. You find. You understand. You move forward. And behind that apparent simplicity remains all the invisible work of those who opened the material, questioned its structures, designed its passages, and started over until knowledge ceased to be a wall.

At the time, we did not speak of artificial intelligence as we do today. We were still building the roads. But I now see very clearly that the question was already there.

How do we prevent quantity from destroying meaning?

How do we transform mass into a path?

Everything that follows is contained in that question.