Each of us now carries in our pocket a machine that answers as a specialist in almost everything. What can one still be a specialist of? I propose an answer here, the bond between people, and I put it to the test of the history of publishing, from the resistances to the printing press up to the digital subscriptions that are undoing our heritage.
The part of the specialist that does not fit in the pocket
A telephone today gives access, within a few seconds, to an artificial interlocutor able to answer as a lawyer, a cardiologist, an art historian or a sound engineer, with a level of precision beyond what most professionals in these fields could recall from memory. The question therefore arises, and it has been working on me for several years: what can one still be a specialist of, when anyone has in their pocket a specialist of anything?
A first answer consists in looking at what these machines actually carry. They possess an immense erudition, that is, a stock of statements and the ability to relate them to one another. But erudition has never been enough to make a specialist, and this was already true among humans, since we all know people who know an enormous amount of things without thinking for themselves, and who remain unable to draw on their knowledge in the real situations of life. I developed this distinction between cognition and intelligence in the article Artificial intelligence, work and professions.
The sociologists of science Harry Collins and Robert Evans gave this intuition a foundation I find very solid in Rethinking Expertise (University of Chicago Press, 2007). They show that real expertise is tacit knowledge, that is, knowledge that cannot be entirely put into words, and that is only acquired through prolonged immersion in the collective of practitioners of a field. One becomes a surgeon by spending time with surgeons, by operating under their gaze, by absorbing a thousand ways of doing things that no one has ever written down, far more than by reading books on surgery. Collins and Evans also distinguish two forms of expertise: contributory expertise, that of those who actually practise the field, and interactional expertise, that of those who master its language without having its practice, the way a sociologist who has spent twenty years among physicists ends up talking physics credibly without knowing how to run an experiment. This distinction sheds new light on language models, which are pure interactional expertise, carried to a degree never reached before, without any contributory expertise. Harry Collins himself drew this consequence in Artifictional Intelligence (Polity, 2018), where he argues that these machines are not socialised, that they belong to no community of practice, and that this non-belonging defines their limit far more than their computing power does.
The machine therefore puts in our pocket the erudition of a field and its language. What remains outside the pocket is the belonging to a collective of practitioners, the embodied practice that is passed on within it, and the responsibility for what one puts forward, before one’s peers and before the people concerned. Being a specialist, at a time when the stock of knowledge is externalised, is perhaps first of all this, belonging to a practice and answering for what one does in it. To which is added a skill that is becoming central, knowing how to embrace the globality of systems, in order to draw on a specialised knowledge that now lies outside our brain.
The knowledge that is in nobody’s head
This last skill was described with great finesse by the cognitive anthropologist Edwin Hutchins, in a book that deserves to be discovered, Cognition in the Wild (MIT Press, 1995). Hutchins spent months aboard a ship of the American navy, observing how a navigation team computes the ship’s position. His discovery is that this computation is in nobody’s head. It is distributed among several people, instruments, charts, procedures, words exchanged according to regulated forms, and it is this whole system, humans and objects together, that thinks. No member of the team could do alone what the system does, and the system keeps working even though each of its members, taken in isolation, makes mistakes. Hutchins concludes that the relevant unit of cognition is not the individual but the socio-technical system.
This conclusion joins what I have been experiencing for years in contexts of shared creation, and what I formulated in particular in the article The bond that emancipates: the quality of what a group produces, in an artistic creation as in a collective decision, depends on the quality of the bond between the people, before any consideration of individual skill. If we take Hutchins seriously, there is nothing sentimental about this observation. If operative knowledge lives in the system of relations, then the quality of the bond determines the quality of the computation the collective is able to make. A group where speech circulates badly computes badly, whatever the erudition of its members, and one can add to it today all the machines one wants.
One must however guard against a misreading I see spreading. The bond is not a discipline. Many professionals of facilitation and collective intelligence believe that facilitating is a speciality in itself, with its so-called participatory methods, its formats, its tools. Yet these methods, applied as recipes, become formattings, which produce conformity where they claim to produce expression, as I analysed in Allowing oneself to create. The bond cannot be tooled, it must be cared for, and it is cared for from inside a real content that people work on together. This is why the quality of the bond will never make an autonomous discipline, one more profession in the list of professions, because it is the transversal condition that allows all the other forms of knowledge, including those of machines, to produce something worthwhile.
The Sorbonne installs the first press in 1470, the king forbids printing in 1535
To situate what is happening to us, the comparison with the printing press comes up often, and I have used it myself several times, in particular in Artificial intelligence, a tool of emancipation in mediation, to show that the discourses about the dumbing down of populations were discourses of power. I would like to go here into the detail of this history, because it is richer, and more instructive, than the conventional tale of the copyists opposing Gutenberg.
The printing press did not enter France against the university, it entered at its invitation. In 1470, Jean Heynlin, prior of the Sorbonne, and the humanist Guillaume Fichet brought from Mainz three printers, Ulrich Gering, Michael Friburger and Martin Crantz, whom they installed within the very walls of the college to produce correct classical texts, cleared of the errors accumulated by generations of copies. The first book printed in France came out of this workshop. The institution that had controlled the production of the manuscript book in Paris since the Middle Ages, through its network of sworn booksellers subject to its authority, first saw in the press an instrument in its service.
The resistances existed, but they are more interesting than the caricature. The Benedictine Johannes Trithemius, abbot of Sponheim, wrote in 1492 a De laude scriptorum, a praise of scribes, in which he defended the upkeep of the monastic copying workshops. His main argument bears less on dumbing down than on the medium itself: the printed thing is a thing of paper, he writes, and paper is consumed in a short time, whereas the scribe who entrusts his letters to parchment extends himself, with what he writes, into a distant time. Trithemius worries about the durability of the new medium, about the permanence of the memory entrusted to it. And the most savoury part is that he had his praise of scribes printed in Mainz in 1494, by Peter von Friedberg, because he wanted to be read. In Venice, the Dominican Filippo de Strata addressed a petition to the doge to have the printers driven out of the city, with a formula that has remained famous: the pen is a virgin, the press a whore. His worry bears on the quality and on the morality of what now circulates without control, produced by foreign craftsmen who print anything as long as it sells. And the Venetian humanist Hieronimo Squarciafico wrote as early as 1477 that the abundance of books makes men less studious, that it destroys memory and weakens the mind by relieving it of too much work. Squarciafico is a remarkable case, because he himself worked for printers. He formulated, from inside the new system, the worry this system inspired in him, as many professionals of artificial intelligence do today.
These three voices speak three worries that we recognise word for word: the permanence of the medium, the quality of what circulates, the cognitive weakening through abundance. None of the three stopped anything, and what played out afterwards is of another order. The powers did not fight the technique, they fought to capture its control. In October 1534, the Affair of the Placards, Protestant posters against the Mass pasted up to the very door of the king’s chamber, caused a panic of the royal power. On 13 January 1535, Francis I, the king nicknamed the Father of Letters, signed letters patent forbidding the printing of any book in the kingdom, on pain of hanging. The total ban was so untenable that it was lifted a few weeks later, at the end of February, but in favour of a more durable device, a commission of censors appointed by the Parliament of Paris, alone empowered to authorise printings. The faculty of theology of the Sorbonne, which had installed the first French press within its walls sixty-five years earlier, now examined books and published its first catalogues of forbidden works in the 1540s. And in December 1537, the ordinance of Montpellier obliged every printer in the kingdom to deposit a copy of each book at the king’s library, the ancestor of the Bibliothèque nationale de France. Thus was born legal deposit, a gesture of control that has become, over the centuries, our greatest instrument for building heritage. The historians Lucien Febvre and Henri-Jean Martin, in The Coming of the Book (L’apparition du livre, 1958), then the American historian Elizabeth Eisenstein, in The Printing Revolution in Early Modern Europe (Cambridge University Press, 1983), documented this shift. Within a few decades, the question was no longer whether one prints, but who authorises and who holds the privileges, those monopolies of exploitation granted by the king to a given bookseller over a given text. The printing privilege is moreover the direct ancestor of our author’s rights, which the French Revolution instituted through the laws of 1791 and 1793 by transferring the right from the bookseller to the author.
The lesson of this history holds in one observation: the cognitive and moral worries occupy the front of the stage, while the decisive battle, legal and commercial, is played out over the control of the circuits. It seems to me that we are living through the same configuration.
Professionals shaped by what they must condemn
The history of the internet replays this configuration, and I know one of its episodes from close up. In cinema and the audiovisual sector, a large share of today’s professionals built their film culture through illegal downloading, because the works that shaped them were simply not available otherwise. These people bring to their professions a cinephilia of a breadth that the previous generations, limited to what cinemas and television were willing to show, could not build. And they occupy an impossible position, since they must now publicly condemn the piracy that made them. I analysed the legal and economic dimensions of this history, the global licence voted then emptied of its substance, the Hadopi law written in the service of the intermediaries rather than of the authors, in the articles The drop of water and Piracy, an unrecognised blessing for the audiovisual sector.
What I want to add here concerns the anthropology of skill. Michel de Certeau, in The Practice of Everyday Life (L’invention du quotidien, 1980), described reading as poaching: readers move across other people’s lands, take, divert, build themselves with goods that were not meant for them. This image describes with accuracy the formation of those cinephiles of downloading, and that of the millions of people who learned their trade on the internet, through tutorials, forums, copied works, cracked software. Poaching is not the margin of training, it is one of its major paths, because real learning passes through circuits that institutions do not control, and institutions willingly requalify as delinquency, after the fact, what was in truth training. The skill of these poachers did not, moreover, come only from the works consulted. It came from the communities that formed around sharing, from the forums where films were discussed, from subtitles translated by volunteers, from a whole sociability of knowledge that resembles what Collins describes as the real milieu of expertise. There again, the bond made the knowledge.
“A book needs nothing in order to exist”
The question that preoccupies me most, deep down, is that of dependency. A book I buy is mine. It is in my library, I can use it whenever I want, lend it, annotate it, bequeath it. One will tell me that I still depend on my dwelling, and it is true, if my library burns I no longer have the book. But a book needs nothing in order to exist. It is analogue, it can be read with the naked eye and crosses the centuries without any infrastructure making it work. A video file, on the other hand, needs software to be decoded, which needs an operating system, which needs a machine, which needs electricity. Trithemius’s worry proved unfounded for paper, which has crossed the centuries, but it pointed to a right question, that of the permanence of media, and the digital raises it again in an unprecedented way, since for the first time recorded memory can no longer be read without machines. Vint Cerf, one of the fathers of the internet, warned in 2015 about what he calls a possible digital dark age, in which entire archives of our time would become unreadable, not because the files would have disappeared, but because the software able to interpret them would no longer exist.
To this technical dependency is added a commercial dependency, which is a recent and deliberate legal construction. The American legal scholars Aaron Perzanowski and Jason Schultz documented it in The End of Ownership (MIT Press, 2016), where they show that, in the digital economy, we no longer buy, we acquire licences, revocable, conditioned by contracts that nobody reads. Their book opens on an episode that has become emblematic. In 2009, Amazon remotely erased, on the Kindle readers of its customers, copies of Orwell’s 1984 sold by mistake, and readers who believed they owned their book discovered that they only had a provisional right of access to it. Music followed the same path. One used to buy records, build a record collection, a singular heritage that said who one was and that could be passed on, whereas with streaming subscriptions, everything one has listened to and loved for years goes out the day one stops paying.
French public libraries are living through the same mutation, and in an almost caricatural way. In the digital library lending scheme known as PNB, a library does not buy the digital books it offers. It acquires licences, limited in time, most often six to ten years, and limited in number of loans, counted in tokens set by each publisher. When the tokens are used up, or when the duration expires, the book disappears from the catalogue and the licence has to be bought again. The Réseau Carel, which negotiates digital resources for libraries, describes without varnish this system in which the wear of the book, which used to be a physical reality, is artificially rebuilt by contract. The consequences go beyond the budget. A library that owns its books builds a collection, that is, a singular memory, sedimented by decades of choices, which makes no library resemble another. A library that rents licences sediments nothing, its offer is a flow, aligned with the new releases that the tokens make profitable, and its singularity dissolves into the standard catalogue that everyone rents from the same publishers. Heritage building, the gesture that the ordinance of 1537 had instituted for print, is what the licence model makes impossible. Dependency on the large industries is therefore not only a matter of abstract sovereignty, it concretely destroys the possibility of building heritages, personal and public, and with them the diversity of memories. I developed the material and geopolitical dimensions of this dependency, drawing on Ophélie Coelho’s Géopolitique du numérique (Éditions de l’Atelier, 2023), in Ecology and Artificial Intelligence, and on Gérard Dubey and Alain Gras’s La servitude électrique (Seuil, 2021) in The potential advent of a “meta-human”.
Paul Baran’s spider web
It would be wrong to conclude from all this that dependency is the evil and autonomy the good, for dependency is our ordinary condition. Whoever lives in a city depends, in order to eat every day, on a chain of interconnected systems, crops, transport, warehouses, cold rooms, electronic payments, each of which may seem very fragile, and this dependency scandalises no one because it is old and distributed among thousands of actors. We depend on books, on libraries, on colleagues, on publishers, on everything that precedes and surrounds us, and this is even the condition of all thought. The question is therefore not whether to depend or not, but what architecture our dependencies have.
The history of the internet provides a precious landmark here. In the early 1960s, the engineer Paul Baran, at the RAND Corporation, was looking for how a military communication network could survive a nuclear attack. His answer, published in 1964 in On Distributed Communications, is an architecture: a distributed network, shaped like a spider web, without a centre, where each node is linked to several others and where information, cut into packets, finds its way even when whole swathes of the network are destroyed. The designers of Arpanet, at the end of the decade, always said that their network had not been built to survive a nuclear war, so the legend simplifies things a little, but they took up Baran’s principles, and the internet we use did inherit this distributed architecture, which makes its robustness. What Baran formulated, deep down, is that a distributed dependency is livable and that a centralised dependency is a vulnerability. Yet what has been built over the last twenty years on this distributed infrastructure is centralised services, a handful of data centres and platforms that concentrate the world’s catalogues, that is, from the point of view of architecture, the contrary of what made the network robust.
This reading grid seems to me to offer concrete paths, for institutions as for each of us. The first is to build heritage deliberately: owning local copies of what matters, in open formats, readable without the software of a single industrial player, and, for an institution, demanding permanent purchase rather than licence whenever possible, treating the conservation of its digital data with the same seriousness as that of its physical collections. It is the gesture of 1537 to be reinvented, and I practise it modestly at my own scale, by hosting my own contents on free software tools whose data I control. The second path is to choose, when one can, distributed dependencies over centralised dependencies: free software, open artificial intelligence models that run on local machines, self-managed hosting, everything that maintains a plurality of paths where the platforms build unavoidable toll gates. The third is to make our dependencies visible, to name them, to take stock regularly of what we have delegated, an approach I detailed in Conscious compromise with artificial intelligence.
The coming years will add to this landscape the humanoid robots, which will externalise gesture after memory and reasoning, in deliberately anthropomorphic bodies, and which will make us cross that collective uncanny valley I described in Presence of the little green men. We will then have, no longer in our pocket but facing us, specialists in almost everything, and the question I am asking here will arise with even more force. My conviction is that it will call for the same answer. What will remain properly human is what cannot be externalised because it has never been inside a brain: the quality of the bond between people, in which knowledge becomes operative, and the care of the architectures, technical, legal and heritage-related, that decide whether our dependencies carry us or lock us in. Being a specialist of something, tomorrow, will first of all mean knowing how to take care of that bond and of those architectures.
