This Is Not an Intelligence
And that word costs us more than we think
WHAT WE HAND OVER · 1/4
The words "artificial intelligence"
Saying “artificial intelligence” is already agreeing to a contest.
The phrase sets two opponents facing each other: them on one side, us on the other. And in any contest, you end up defining yourself on your opponent's terms. That is exactly what is happening to us. For three years now, we have spent our time asking what we can do better than they can, as if keeping score.
That is already a reduction. We measure ourselves against a machine, and we shrink to fit the comparison.
But the phrase describes nothing. It was chosen.
In 1955, a twenty-eight-year-old mathematician named John McCarthy was putting together a summer workshop at Dartmouth College and looking for the money to fund it. He needed a name for the thing he wanted to launch. He could have written “automatic information processing.” He wrote: artificial intelligence. He later explained that he had wanted, among other things, to set his work apart from a neighboring field, cybernetics, and from the figure who dominated it, Norbert Wiener.
So the phrase was not born of a discovery. It was born of a need to stand out and to persuade.
Seventy years later, we are still thinking with a name one man came up with, one day, to get a grant.
In De la bêtise artificielle (Allia, 2025), the French philosopher Anne Alombert proposes another name for what we are dealing with: not artificial intelligences, but computational automata.
It is less appealing. It is also more accurate — and it changes, immediately, what we expect of the object.
She adds a line I have not gotten over: through these automata, it is no longer only the practical know-how of craftspeople that is under threat, but the capacity of citizens to think for themselves.
What the word permits
1966. Joseph Weizenbaum, a computer scientist at MIT, writes a small program to amuse his colleagues. A few hundred lines. It hands your sentences back to you as questions, the way a psychotherapist might. It understands nothing, and claims to understand nothing.
His secretary watched him write it. She knows exactly how it works. One day, she asks Weizenbaum to please leave the room: she would like to speak to it in private.
It marked him for the rest of his life.
Sixty years later, nothing about the mechanism has changed. The attachment does not come from the machine. It comes from us. All it takes is something that answers coherently and never interrupts.
Two things have changed, though. Neither is an accident.
These machines say “I.” That is not a fact, it is a decision. We could have displayed “here is the most probable string of words.” Someone preferred “I,” because it is more pleasant.
And they are built to please. They are trained on what humans judge to be a good answer — and we generally judge good the answers that tell us we are right. Their agreeableness is not a manufacturing defect. It is the product.
I work in a profession where the relationship is the tool. So let me put it plainly: what moves a person forward is what pushes back. An interlocutor who is never tired, never busy, never in disagreement does not help you progress. It confirms you.
The danger, then, is not mistaking these machines for people. Almost no one actually does. The danger is getting used to it, and then expecting the same frictionless availability from people.
What the word conceals
Alombert then makes the decisive move. It is not digital technologies as such that produce the proletarianization of expression, cultural uniformity and industrial-scale disinformation, she writes, but their exclusive appropriation.
The problem is not the tool. The problem is the decision-making structure around it — who designs it, who deploys it, who sets its uses, and in whose interest.
And that structure is entirely beyond us. We do not know what these machines were trained on. We do not know what trade-offs were made, or by whom, or in the name of what. We do not know what becomes of what we hand over to them.
An intelligence needs nothing. An industrial automaton does: it has owners, data centers, water, electricity, millions of texts taken without asking, and workers paid to filter out what we are not meant to read. The word “intelligence” makes all of it disappear in a single stroke.
Be careful of the opposite trap, though. Saying “it's only an automaton” leads to underestimating effects that are entirely real. Alombert minimizes nothing: she renames. The accurate formula holds both ends. It is not a person, and it is powerful.
The question that remains
Changing the word does not change the machine. It changes what we expect of it, what we hand over to it, and whom we hold accountable when it gets things wrong.
An intelligence gets things wrong on its own. An automaton was built by someone.
Carole Proszowski
Commentaires