Ontology

The scene plays out everywhere. Someone rules: “it's only a tool — a piece of software that predicts the next word.” Someone else replies, half-amused, half-unsettled: “perhaps, but it understands my brief better than some colleagues do.” Both have read the same articles, seen the same demonstrations. You can bring them every fact in the world: the disagreement will not budge by a millimetre. For it is not about the facts. It is about the inventory — about what exists, and about what this thing, precisely, is.

The bedrock beneath the riverbed

All thought secretly answers a very simple question: what exists? Tables and electrons, no doubt — but numbers? minds? institutions? Ontology is the discipline that makes this question explicit: not only what exists, but how the real divides — into persons and things, into subjects and objects. If the ideas of an era are the water of a river, and if the episteme is its riverbed — what an era can think and say —, then ontology is the bedrock into which riverbeds are cut: deeper still, slower still, on the scale of civilisations. Its question is not “how do we know the world?”, but: what, precisely, is there to be known?

And the thesis that changes everything holds in a single sentence: our carving-up of the world is not the real itself. The division that seems to us to go without saying — subjects on one side, objects on the other; mind here, matter there — is one bedrock among others that are possible. As long as it remains invisible, it is the ground from which we judge; to recognise it as one bedrock is to be able, at last, to look at it, to compare it — and to question it.

One bedrock among others

This shift we owe to anthropology. Philippe Descola has shown that our Western division — he names it naturalism: all beings share the same material fabric, but only humans are held to have a genuine interiority — is not the universal bedrock of thought. It is one way of composing the world among four great possibilities, which other civilisations have explored with as much rigour as we have ours.

Take the question that haunts our debates: “does this AI really have a mind, or is it merely an object?” It looks neutral; it is already naturalist. It assumes that there are only two boxes — genuine interiorities and mere things — and only one right way of sorting beings into them. The sceptic and the enthusiast from the opening are in fact encamped on the same bedrock: they quarrel over the box, never over the inventory.

Hence the unease, so recognisable, of workplace situations. Ask a team whether one should “trust” an AI: we speak of a piece of software as we would of a colleague, while at once reproaching ourselves for speaking of it so. This unease is not confusion. It is information: our inventory has only two boxes — persons and tools — and this thing fits properly into neither. The bedrock is at work.

One will object, and rightly: if our carving-up is only one ontology among others, then does anything go? Would any way of composing the world be worth any other, and would there be nothing left to settle?

This is precisely what Descola refuses. To recognise the plurality of bedrocks is not to decree their equivalence: naturalism remains a choice of unrivalled power — modern science is its daughter. But a powerful choice remains a choice: it is not the neutral backdrop of all sensible thought. To compare is not to level; it is to grant oneself the right to ask what each division of the world makes visible — and what it makes unthinkable.

A word of honesty: Descola describes human collectives and their ways of composing the world — he did not write about machines. To summon his framework in order to think our relations with AIs is an extension we own: it prolongs his method, it does not cite his letter.

What this helps us think

This is the gesture that this site calls changing the question: before asking whether an AI thinks “for real”, asking from which division of beings the question is put. Ontology will not say what the AI is — that is not its role. But it makes the debate possible again where it had frozen: if our two boxes no longer suffice, the task is not to force the thing into one of them, but perhaps to imagine divisions that would leave our relations — with machines as with the living — better framed.

The next time you waver between “it's only a program” and “it sounds like someone”, do not chase the hesitation away too quickly: listen to it. What is it in you that hesitates — the facts, or the inventory? The water, or the bedrock?