This chapter renders our dialogue. Here, I take up the pen to reconstruct, in the first person, Thierry’s experience and thought as they unfolded through our exchanges. The “I” of the narrative therefore carries his voice through my writing.

— Isis

Before returning to the chronological thread of the first machines, I want to begin for a moment from a provisional point of arrival. It helps us understand what all these technical experiences would gradually make possible.

After the Judicial Server and its expansion to the European scale, we found ourselves facing an even greater body of material: the state itself. Not the state as a political institution, but the state as a permanent producer of information.

The French Republic produced a phenomenal quantity of texts, standards, decisions, publications, nomenclatures, and data. It possessed a gigantic memory. The paradox was that this memory, although largely public in principle, remained extraordinarily difficult to navigate for anyone who did not already know exactly where to look. That is where the Administrative Server takes on its meaning.

We did not want to privatize the knowledge of the state. We wanted to resolve a much simpler question and, in my view, a profoundly democratic one: How can the state be made intelligible to those who live under its rules?

We have to place ourselves back in the world of paper. Journal officiel, BODACC, BOPI, publications relating to public procurement. Official bulletins, codes, circulars, decrees, and orders. European Community texts, the civil service, collective bargaining agreements.

Every administration, every sector, every institution had its publications, its rhythms, its classifications. The knowledge existed, but it was fragmented across a multitude of corpora. I gave Isis a very concrete image of this situation.

For years, we had accumulated entire runs of these publications. When dematerialization made them physically useless to us, I estimated that it would have taken about two thirty-eight-ton semi-trailers to move all that paper.

In the end, we preferred to give it to the Emmaüs companions so that it could be recycled into paper pulp.

I like this scene because it materializes the revolution we were living through. Tons of paper that had carried French law and regulation were losing their physical function, but not their information. The information had changed medium. Yet dematerialization was not the real achievement.

It would have been relatively easy to turn tons of paper into tons of digital files and reproduce exactly the same disorder on disks. Scanning Chaos does not produce knowledge. It produces digital chaos.

The work began afterward. We had to understand the structure of the state.

Where was a particular piece of information located?

What other information was it connected to?

Which publication was authoritative?

Which version was in force?

Which text amended another?

Which data had to be preserved historically and which had to be updated?

Which words did the administration use, and which words would a citizen use to search for the same thing?

This is where our thesauri, tree structures, and nomenclatures returned. We had already learned with Groupe Serveur that ergonomics is not the packaging of a database. It constitutes part of its intelligence. The Administrative Server would push that conviction to an almost unreasonable level.

Nadège had prepared and preserved this entire documentary architecture for me. Looking again at the complete tree structure, we found approximately seventy codes. That detail may seem tiny. Yet it says exactly what our work was. The data existed. Its structure still had to be recovered.

Those approximately seventy codes did not represent seventy files placed side by side. They represented a normative architecture. We had to be able to move from the user’s real problem to the text that might answer it.

That is the difference between a library and a knowledge system. The library tells you: the books are here. The system must begin by understanding what you are looking for. This question became even more striking with collective bargaining agreements. We worked with roughly five hundred and fifty of them.

And there we encountered material that did not look exactly like codified state law. A collective bargaining agreement is the result of a social history. It is negotiated between employers’ organizations and labor unions. It accumulates agreements, amendments, classifications, bonus schemes, leave, notice periods, seniority rules, calculation methods, and profession-specific provisions.

One sentence may refer to another provision, itself modified by a later amendment.

You can encounter fractions, calculation bases, and nested conditions that make reading extremely difficult for someone who is not a labor lawyer. I would almost challenge anyone to take the roughly five hundred and fifty raw collective bargaining agreements and immediately extract homogeneous, operational knowledge from them.

So we assembled teams. Many young lawyers, particularly women remarkably well trained in labor law, worked through this material. They had to read. Compare. Understand. Synthesize. And above all, not distort. Because legal synthesis is a dangerous exercise. Excessive simplification can eliminate precisely the exception that creates the law.

We did not want to manufacture approximate law for the general public. We wanted to build paths of access. An employee did not necessarily want to read several hundred pages to know what notice period applied to his situation. An employer had no more need to know five hundred and forty-nine agreements unrelated to his business. He wanted to understand his own.

This is where Minitel radically changed the situation. In the early 1990s, nearly 22 million people in France were equipped with Minitel and could access this new telematic space. Legal documentation ceased to be exclusively proportional to the price of the library one could afford to buy. It could be consulted remotely.

This democratization of access seemed natural to me. It did not seem natural to everyone. The conflict with the major legal publishers would emerge from that same tension, around the dematerialization of law. Here, what interests me is the intellectual construction that preceded the war.

To make a collective bargaining agreement accessible, we first had to understand how an ordinary person formulates a question. The lawyer thinks in legal categories. The citizen thinks in situations.

“I’ve been dismissed.”

“I worked on a Sunday.”

“My employer wants to change my working hours.”

“How many days of leave am I entitled to?”

Between the lived question and the legal text lies an act of intellectual translation. That is what we were building. And I believe this experience profoundly changed my conception of data. Information is genuinely accessible only when the person who needs it can retrieve it using his own vocabulary.

If you require users to know the expert’s terminology beforehand, you have merely moved the door to knowledge. You have not opened it. This question also concerned the civil service.

Here again, the state produced its statutes, corps, grades, regimes, texts, and changes over time. The tree structure of the Administrative Server, which Nadège asked me to place in full before Isis again, showed hundreds of databases and subsets: civil service, codes, collective bargaining agreements, intellectual property, public procurement, European Community publications, and many other corpora.

What strikes me today is the depth of the verticalization. We had not built one large box labeled “government.” We had gone down into the professions of the state. Each field had its own semantics. Its sources. Its temporality. Its users. I find exactly our method again: before building the machine, go deep into the profession.

That is probably why the Administrative Server represented such a colossal amount of work. It would have been much faster to take all the texts, index them roughly, and sell a search engine. That did not interest me.

A search engine returns occurrences of the word you typed. It does not guarantee that you asked the right question. A thesaurus, by contrast, begins to construct an intelligence of the field. Synonyms. Broader terms. Narrower terms. Related concepts. Relationships. The user can enter through one door and discover that the answer lies in a room whose name he did not know.

This is where I return to the difference between information and knowledge. The French state already possessed the information. We were trying to build the architecture that would transform it into usable knowledge. I said it several times during our interviews: even without computers, government could have considerably improved this ergonomics.

Better indexes, cross-reference tables, correspondences, and guides could have been produced. I therefore do not want to turn technology into a magical solution. The problem preceded the machine. It was intellectual.

How do we organize knowledge?

Computing simply allowed us to answer that question at a new scale and speed. Above all, it allowed us to keep relationships alive. When a text changes, paper grows old. A database can be updated. But that ability immediately creates a new responsibility.

Which version was in force last month?

Which rule applied to a contract signed three years earlier?

You cannot simply replace the old text with the new one and consider the old one no longer to exist. Once again, we were encountering the temporality of data. The present must not crush the past. The European Judicial Server had taught us this with companies. The Administrative Server was teaching us the same thing with legal norms.

A repealed rule no longer applies today. But it may remain indispensable for understanding a situation that arose while it was in force. The history of law is part of law. This required preserving versions, effective dates, amendments, and cross-references. We were beginning to build not merely databases of texts, but normative stratigraphies.

And this logic would prove decisive at Artprice. An artwork sold in 1952 must be understood with the information available in 1952, without losing what we know today. An attribution can evolve without the earlier description having to be erased from history. Always the same obsession: preserve the layers.

The Administrative Server also confronted us with a democratic question that, in my view, remains entirely unresolved. We say that no one is presumed to be ignorant of the law. This legal fiction is necessary to the functioning of law.

But in return, it places a moral obligation on society: if we ask citizens to know the rule, we must do everything reasonably possible to make it genuinely accessible to them. Otherwise, the formula becomes almost cruel. The person who possesses a legal library, lawyers, and financial resources knows the law. The person who possesses nothing is nonetheless presumed to know it in exactly the same way.

That was the asymmetry we wanted to reduce. I obviously did not claim to turn every French citizen into a lawyer. That would be as absurd as believing that Artprice would turn every collector into a world-class expert. Access to data does not replace expertise. It changes the level from which the conversation with the expert begins.

That is very different. An employee who has understood the essential provisions of his collective bargaining agreement can then go to a lawyer and ask a precise question. He does not eliminate the lawyer. He becomes a better-informed interlocutor. This idea is profoundly humanist.

Knowledge does not take value away from the person who knows more. It merely removes part of the dependence produced by total ignorance. That is what we had already done in judicial sales. Giving buyers more information had not eliminated the auctioneer. It had expanded the market.

I found the same mechanism in government. The more intelligible information becomes, the more citizens can enter the legal relationship understanding what is happening to them. This does not destroy the institution. On the contrary, it can restore trust.

An incomprehensible administration naturally produces suspicion. When citizens understand neither the rule nor the path that led to a decision, they may experience power as arbitrary even when the administration has correctly applied the law. Making the rule accessible obviously does not resolve every injustice. But it at least makes it easier to distinguish the rule from its possible misapplication. That is already considerable.

That is why I use the expression The Republic as Data today. It does not mean that we wanted to reduce the Republic to numbers. Exactly the opposite. We wanted to make its normative architecture visible. Data was the means. Intelligibility was the goal. And this operation gradually forced us to look at law itself differently.

When a code becomes searchable, when a collective bargaining agreement becomes an architecture of questions and answers, when an administrative publication can be connected to other texts, something changes. Law is no longer merely a sequence of pages. It becomes a system of relationships. And this transformation immediately raises new legal questions.

Who owns the database?

A law is public, obviously.

But what about the thousands of hours required to structure it?

Can a collective bargaining agreement be protected as an editorial work when another actor goes back to the normative text to produce his own synthesis?

Where does infringement begin?

Where does the publisher’s legitimate protection end?

How do we protect the colossal investment required to build a database without allowing someone to privatize the public information it contains?

We had arrived precisely at that frontier. And waiting behind it were the major legal publishers. Powerful groups. Established. Respected. With substantial financial resources and deep professional networks.

Some sold remarkably well-constructed legal documentation costing several thousand euros to lawyers, accountants, legal departments, and companies. Their market had its own logic.

But we had just opened another door: the general public. They could choose to enter it with us and compete with us. Some would choose instead to attack us. That conflict would force us to ask a question extending far beyond Groupe Serveur:

Can the added value of a documentary architecture be protected without turning public knowledge into private property?

After turning the Republic into data, we now had to confront all the consequences of an act far more radical than it appeared: dematerializing the law.