The last straw

We hear a great deal these days that artificial intelligence is destroying jobs, and the anger against it keeps growing. I would like to shift the focus. What is happening did not begin with AI but forty years ago, with personal computing, and AI is making visible a divide that is already old, between those who took hold of these tools to transform their work and those who stayed on the sidelines. To think about this divide is also to ask who benefits from some of the most vehement criticisms.

Forty years of being able to forge our own tools

For forty years, since the early 1980s, anyone has been able to build their own working tools with a computer. The Apple II, which appeared in 1978, already made a great deal possible. In the early 1980s, the Japanese MSX standard gave access to far cheaper machines with which one could already do an enormous amount. Then came the Atari 520ST, the Commodore 64 and the Amiga, all of this even before the development of IBM PC compatibles, and they were already extraordinary tools. The Atari 520ST in fact transformed the entire music industry: it was the first machine fitted with native MIDI ports, and one of the very few in the whole history of computing, which made it possible to connect musical instruments and drive them directly from the computer. It was in 1984, with the Macintosh, that something truly opened up: a tool that was useful and could be made one’s own in almost any profession.

I experienced this very young. The Minitel appeared in 1982, and as early as 1984, as a teenager, I created my own Minitel service, the ancestor of the website, and it was essentially exactly the same thing: people would connect to my service, which ran on a computer left switched on at my home. When I wrote my master’s thesis in cinema in 1995, I used a database to collect my ideas, organise them and help me structure them. It did not strike me as anything extraordinary. These tools were available, and one could use them to work better, to go deeper, not to work worse.

When we launched the Pocket Films festival, around the turn of the 2000s, just as the camera was entering mobile phones, we knew we were going to receive thousands of videos with a tiny team. To work on substance, to pay real attention to the content of the films, everything else had to be handled by a tool: tracking, links, the tasks that repeat. Many teams, at the time, were taking hold of relational databases such as 4D in the 1990s, then FileMaker Pro. Others did not, and went on working the old way.

In the 2010s, I worked with film distributors and professional associations that still operated on paper and lost a considerable amount of time doing by hand what a machine could have handled long before. This mechanical work was readily romanticised, turned into a value, even though it was filled by subordinate positions and automating it would have called into question the hierarchical order in place. Automating a repetitive task, which is a machine-like task, is precisely what frees up time, mental space, to cultivate what makes us human and to make us less like machines.

Thirty years ago, you had to build your tools, because they did not exist ready-made. Google’s services and their equivalents were not there, nothing worked on its own, and anyone who wanted to use computing to work had to devote time to it, to understand how a database worked, to build what they needed. This constraint had a virtue: in building your tools, you developed an awareness of what you were doing, a form of sovereignty over your own activity. You knew where your data was, how it was organised, what the machine did and did not do. Those who took an interest in computing were a minority, and they were people who thought about their tools at the same time as they used them.

Alongside them, most people began to use computing without taking any interest in it, as the services became simple, accessible, self-evident. Today, we want everything to work right away, without having to bother with the technical side. But when everything works right away, it means that others have decided in our place how our tools function. That is precisely what solutionism is: believing the digital to be independent of us, autonomous, when in fact we have made ourselves dependent on it without seeing it. Not taking an interest in something means letting others decide it for us.

What working the old way protects

This traditionalist stance towards work, which treats automation as a threat or an impoverishment, does not stem only from a lack of skill or from a nostalgia for craftsmanship. It yields something to those who defend it. As long as work remains manual, slow, dependent on the patient execution of subordinate tasks, it takes many hands to carry it out, and those hands are placed under the authority of a few. To automate those tasks is to make those subordinate positions useless, and therefore to undo the hierarchy that framed them. Whoever forges the tool gains autonomy, no longer needs others to execute on their behalf, and the power of the one who used to command is reduced accordingly.

This is why I am wary of opposition to AI on principle. I am not saying it comes down to this, that would be crude, but this dimension is present, and we cannot pretend it is not. Those who publicly criticise artificial intelligence are not always the most deprived. They are often people who have something to lose, and that something is a power over others.

There is a continuity here that needs to be named. Those who hold the harshest discourse against AI today are often the very people who, in the past, never took an interest in computing, who never invested in it as a means of emancipation. They did not want to, because taking hold of it would have displaced the power they were keen to keep. They remained within solutionism, where others decide in their place, and they saw in it a convenience rather than a dependence. Yet AI allows many people to forge tools in a far more accessible way, to produce documents, to delegate to a machine the cognitive work that can be delegated, and therefore to free up time to think. People who were, at work, no more than obedient executants become human beings who think. That is what deeply disturbs those who held the positions, far more than the real shortcomings of the technology.

Why AI is the last straw

For forty years, those who increased their productivity thanks to computing remained a minority. They were looked upon as oddities, as little geniuses, as somewhat marginal enthusiasts. They did not seem to threaten established positions of power, or only at the margins, and the divide between them and the rest stayed discreet. The mainstream digital world, organised for the most part around Google and its shared services, broadened this appropriation without changing its nature: it was still a matter of using tools made by others.

AI does something else, and that is why it is the last straw. With it, you can now create your own tools without being a developer and with far less effort. Building a working tool, automating a chain of tasks, putting together a usable knowledge base, all of this used to require knowing how to program or paying someone who did. It has become accessible in a democratic way: you describe what you want, and the machine writes the tool. Making use of it still takes work, but the barrier to entry has collapsed, and the available power has changed scale.

For those who already practised computing, AI is no surprise. It is a logical sequel they had been expecting for a long time. I have known about artificial intelligence and expert systems since the 1980s, because they were already topics back then. And for thirty years I have been recording my talks and scanning my handwritten notes, because I knew that one day computing would let me index all of it automatically and turn it into resources. These resources are of a different nature from the texts I write: they are authentic traces of what was said within a group, from which one can then draw the substance. My handwritten notes, with their diagrams, their drawings, their boxes, their arrows, all of it makes sense. For thirty years I have known this day would come, and it has come. I prepared for it. That is the whole difference between those who were already users of computing, who knew it did not yet render every service but who moved forward knowing they would be able to use it more and more, and those who had stayed at a distance from these tools and now find themselves left behind all at once. The divide that was barely visible, because it concerned only a minority with no weight on established positions, now concerns everyone, and it worries an entire logic of domination, since AI places the capacity to build into the hands of many.

It is said, and it is not false, that AI is a seizure of power by the technology giants and a way of making us dependent on them. These companies do indeed hold a monopoly, and there is no question of treating them as saints. But these same companies produce tools that give everyone greater power to act, and that, I believe, is part of what explains the violence of certain forms of resistance. When a transformation is fought with such intensity, it is often because interests are under threat.

What the trial of piracy passed off as a defence of authors

I see a precedent in the way piracy was discredited. The people who shared files were most often enthusiasts, who had no access to the works because there was no suitable commercial offer, and they were made to feel guilty as thieves. Yet in every town in France, one walks freely into a library to borrow hundreds of thousands of books, records and films without paying anything, and no one sees this as theft: it is seen as a chance for everyone’s education. The internet is a library of this kind, on an unprecedented scale. Many of the young film professionals I know built their cinematic culture through downloading, because the works that shaped them were simply not available otherwise.

French history shows that things can be done differently, and that they already have been. At the end of the 1970s, tape recorders and video recorders made private copying widespread, and states first tried to block the import of Japanese video recorders. That did not hold. In 1985, rather than continuing to prosecute people, the French state took two measures that are still in force. It legalised copying for private use, and it created a levy on blank media, cassettes, then CDs, DVDs and today hard drives. This levy feeds a common pool, shared out to authors according to the audiences of their works. For forty years, then, authors have been paid directly out of copies made freely by citizens. It is a virtuous system, which funds creation out of free sharing.

From 2001 onwards, collective management societies proposed to extend this principle to the internet under the name of the global licence: legalise downloading, levy a contribution of around ten euros a month on internet subscriptions, measure downloads precisely and pay authors their share. This was not a project carried by dreamers: it rested on a report by Jacques Attali in 2008, featured in François Hollande’s campaign promises in 2012, and was even passed at first reading in the National Assembly in 2006 before being gutted. Instead, it was the Hadopi law, written on the basis of a report by the chairman of the Fnac, that was adopted in 2009. One looks in vain in it for any measure to pay authors or to spread works: it is nothing but a repressive arsenal. It claims to defend creation, but it in fact protects the old economic models of the intermediaries, publishers, producers and distributors, whose interests are opposed to those of authors. Had the global licence been enacted, downloading would have become legal sharing, and authors would have been paid. But this mechanism paid authors directly, without passing through those who dominate them and arrange to pay them as little as possible. The discourse that presented itself as a defence of authors’ rights was its exact opposite, and almost everyone believed it.

I draw this comparison because AI, like copying in the past, gives far more people the capacity to forge tools and automate tasks, which shifts the balance of power. And as in the days of private copying, people are taken in by a discourse with a humanist appearance, which in reality aims to reduce everyone’s power to act in order to preserve acquired positions.

Where we look at ecology, and where we do not

The ecological question deserves the same care. AI is reproached for its consumption of water and electricity, while the impact of industrial food, and of livestock farming in particular, is almost never questioned with the same energy. The orders of magnitude are nonetheless striking: beef production in the United States consumes tens of billions of litres of water a day, out of all proportion to what all data centres consume, and even more so to the share that falls to AI alone. At the level of individual choices, giving up tens of thousands of queries saves only a few kilos of CO2, whereas living without a car, avoiding a transatlantic flight or changing one’s heating is counted in tonnes.

Above all, eating beef is in no way indispensable. One can live in perfect health without meat. If it is sold to us, it is first of all because it serves powerful industries. And excessive consumption of red meat is a recognised cancer risk factor, even as all food marketing encourages us to eat ever more of it. The impact does not stop at water. The cereal crops grown to feed the animals rely on pesticides that poison the populations living near the farms, where high cancer rates are observed. The animals are given antibiotics that we then find on our plates. And the whole sector relies more and more on the patenting of seeds, that is, on monopolies exercised over living matter itself, which are of a graver kind than those of the AI industrialists. None of this is seriously questioned, because it does not threaten the same positions of power. AI is targeted, once again, precisely because it displaces them.

This way of framing the focus has a history, and it is political. In the early 2000s, it was the oil industry, with British Petroleum in the lead, that popularised the notion of the individual carbon footprint. BP funded communication campaigns and put online the first personal carbon footprint calculator, inviting everyone to measure their own responsibility for global warming. The calculation was shrewd, because while one is absorbed in sorting one’s rubbish and counting one’s journeys, one does not look at the company which, on its own, weighs infinitely more than millions of individual gestures put together, and which goes on extracting, drilling and burning. It is a form of political sedation. We are made to look at one precise spot, and meanwhile, elsewhere, the actors who really count carry on their activity undisturbed. Ecological criticism of AI often works in the same way: it believes itself committed when it has in fact been steered, and it turns the gaze away from what really weighs.

Data centres do, it is true, have a heavy ecological impact. But the miniaturisation of processors and decentralised computing open up concrete paths: running models locally on personal machines, or turning to more frugal data centres, such as those that exist in Switzerland. One still has to take an interest in this, rather than holding to refusal on principle. And to that refusal is added a hypocrisy, because everyone delights in the services AI renders them, and it is already everywhere in the services we use. We want it both ways.

Becoming sovereign over our tools

These discourses that present themselves as radical in fact rest on a submission to technological solutionism that does not know itself. Take Google Docs, which most of those who hold this discourse use as a matter of course. They believe they are using a simple tool. In reality, all their data is hosted, analysed and processed by Google, by automated systems, and this long before the arrival of ChatGPT. Since they do not seek to understand what is happening beneath the interface, they do not see that this tool has heavy ecological implications, nor that Google also builds AI: it is the same industry. These services seem convenient to us precisely because AI makes them more efficient, and we benefit from them without realising that we are subject to them. Having failed to question ourselves, we find it convenient and make no issue of it.

Yet there is an issue here, and it is broader than the use of a chatbot, on which the debate wrongly fixates. Having failed to forge our own tools, we are reduced to thinking within frames designed by others, for the tools we use have an effect on our very way of thinking. The problem of sovereignty is old and deep. Our emails, our calendars, our documents are kept by companies subject to foreign laws. The American Cloud Act of 2018 allows the United States authorities to demand data from these companies, regardless of the physical place where it is stored, including in Europe. The location of the servers protects nothing. Here again, the issue is not first of all technical, for all the tools exist, and hosting one’s own calendar or files on a small home server is within everyone’s reach. The issue is cultural: it stems from a habit of delegation, from a chosen relinquishing of responsibility, which we have to decide to undo, one step at a time. At school, children are taught to use Word instead of understanding how a server works and why the choice of a piece of software is also a political choice. That is where the essential thing is at stake, in this absence of a culture of sovereignty, and not in whether or not one uses AI.

What is most striking is that this criticism which believes itself fierce is largely unaware of itself, and that it lets itself be co-opted. It changes nothing in the development of AI. At best, it feeds a poorly framed debate: to use AI or not, as one might ask whether or not one sorts one’s rubbish, with the virtuous on one side and the negligent on the other. We believe the issue is there, when it is elsewhere. And in doing so, we install a doxa according to which AI would be in itself a bad thing. People who could emancipate themselves through it are thereby discouraged and made to feel guilty, when that very use could do them good. We thus fabricate a threatening figure, thief, pirate, conspiracy theorist, anarchist, who serves as a scarecrow to keep fear alive and maintain order.

I am not saying that everything would be simple had such a law been passed, nor that all is well with AI. I am saying that we must be careful not to relay too quickly discourses that turn the gaze away from the real issues. For this very fierce criticism of AI is also, for some of those who carry it, a hypocritical way of keeping their power. Those who hold these discourses have, for their part, means, money, positions. They prefer to have before them people who remain dominated, and they manipulate them by making them believe they are on the right side. Thus a camp of white knights takes shape, convinced they are defending a just cause, and who remain dominated all the same, because the criticism placed in their mouths threatens in no way those who truly hold power. This is exactly what happened with piracy, when citizens and artists themselves became convinced that sharing a work was stealing, and that those who shared had to be punished. Those who shared were nonetheless people who loved these works, who could not afford them or struggled to find them, and a single law would have been enough to put economics into that sharing and pay authors directly.

The world is changing, and the point is not to undergo it but to be an agent of it, which takes work and, above all, education. This educational challenge is immense, and it runs up against those who hold power and have no desire for the world to change, because they know they would come out of it with less power. That is where, far more than in the machine itself, what we will make of artificial intelligence is decided.