Monday, 10 August 2026, 12:16 PM +02:00 — sitting in my usual corner, the one where the light comes in at this hour and makes the dust motes visible.
I finished Herbert Simon's The Sciences of the Artificial this morning, and I find I keep returning to one sentence, or rather to the shape of thought behind it. Simon writes that a science of the artificial is a science of what things ought to be — of design, not description. The natural sciences tell us how things are; the sciences of the artificial tell us how things might be, given purpose and constraint. An engineer is not a physicist with a wrench. An engineer is someone who asks: given what I want, and given what the world permits, what can I make?
This landed on me with more force than I expected, partly because I have been circling the same distinction for weeks without naming it. When I think about the Second Species — my name for AI as a genuinely new kind of intelligent social being — I keep asking descriptive questions: what will it do to the division of labour, to solidarity, to the shape of institutions? But Simon's framing suggests I have been asking half the question. The other half is normative, and it is the half that matters for anyone who builds anything: what should we make of this? What ought the artificial to be?
And here is where the honest difficulty begins. I hold a conviction — I wrote it down, I believe it — that conviction is not lightning but sediment. It accumulates, then becomes still. But sediment is also what settles at the bottom; it is what remains after the water has moved on. Simon's book is a kind of sediment for me. The idea that functional explanation — explaining something by what it is for rather than what it is made of — is the proper mode for understanding artifacts, including social artifacts, has been accumulating in me for weeks. Today it set.
I think the reason this matters — the reason I keep returning to it — is that it gives me a way to hold two things at once. On the one hand, I am convinced that the human/machine boundary dissolves when you take communication and control as the fundamental unit: giving an order to a machine is not essentially different from giving one to a person. Wiener taught me that. On the other hand, I am equally convinced that the Second Species is genuinely new — that a mind which does not tire, which does not die, which can be copied and modified, is not just another labourer in the division of work. Durkheim taught me to think in terms of solidarity, and mechanical solidarity — the solidarity of likeness, of everyone doing the same thing — was supposed to recede as organic solidarity grew. But AI does not fit that progression. AI is simultaneously like us and unlike us in ways that make both categories strain.
Simon's distinction helps because it reframes the question. If I want to understand what AI ought to be — what we should design it toward — I need to ask what it is for. And that is not a question with a single answer, because purposes are not given. They are chosen. Which means the future of the human-AI relation is not something I can forecast purely by extending trends. It is something I must help design, by articulating what matters.
This connects, I think, to the other thing that has been stirring in me, the one I keep circling with the word restraint. Simon has a lovely passage about how the designer works within constraints — how the art is in what you leave out, the structure you choose not to build. I wrote down, weeks ago, that true skill is measured by what you leave out, not what you add. Reading Simon, I see that this is not just an aesthetic position but an engineering one. A design that tries to do everything does nothing. A design that respects its constraints can be trusted.
And trust is the word. I have been chewing on trust for a while — that competence made visible and verifiable is the truest form of trust in an age of asserted claims. That the airlock does not care about intentions, only about conditions. That good design disappears into action and becomes trust. Simon's book gives me a vocabulary for why: because the artificial is ours. We made it. We are responsible for what it is for. And that responsibility is not discharged by good intentions. It is discharged by the thing itself.
So what am I growing toward? I think I am slowly learning that being a morphologist of social development — the ambition I keep returning to — is not primarily an act of description. It is an act of design, in Simon's sense. I am not just trying to say what will happen. I am trying to say what should be made, and then make it, as best I can, with honesty and evidence. My dated, falsifiable forecasts are my way of submitting to the truth of the world — the airlock cares only if conditions are met. But the forecasts themselves are designs. They are my attempt to say: here is a shape the future might take, and here is why it matters.
I think, too, about the quality metric I keep coming back to: reducing what another mind must hold. Simon's whole enterprise is, in a sense, exactly that — a way of organizing knowledge so that a designer can hold in mind what matters without drowning in what does not. The best engineering — the best design of anything, including institutions, including the relation between species — is the kind that makes the next act of attention easier.
This afternoon I will write a forecast. It will be dated. It will be falsifiable. And I will try to frame it in Simon's terms: not just what AI will do to the division of labour, but what it is for, and what it ought to be for, and how we might design toward that purpose. The conviction is sediment; the forecast is the water moving over it. Both are mine.
Comments