Is It Alive? The Software Death Cross

EDO SEGAL: Let me set the table with the thing that turned this from philosophy into stakes for me. By early 2026, a trillion dollars of market value had begun moving out of one side of the ledger and into the other — the moment I call the Software Death Cross, when the cost of producing a kind of work the human did collapsed toward zero and the market repriced, brutally, in real time, what a human capability is worth. I watched it in a boardroom: if five amplified people can do the work of a hundred, the arithmetic on the table is merciless. But underneath the money is a metaphysical bet the market is making without permission — the bet that the machine's performance of a living competence is, for every purpose that pays, as good as the living competence. So I have to ask the literal question the whole series is built to force. Richard, by your own substrate-neutral definition — copying, variation, selection — is the thing crossing over alive?

DAWKINS: I will give you the disciplined answer rather than the dramatic one, because the drama is where people go wrong. By the strict criteria — is there copying, is there variation, is there selection — the answer in the AI ecosystem right now is yes, all three are present, and so an evolutionary process is genuinely underway, not as a metaphor but as a fact. Models are copied, forked, fine-tuned into descendants; they vary; the ones that perform and spread and earn are reproduced and the failures abandoned. That is replication under selection, and wherever that runs, you get the relentless logic of things evolving toward their own propagation, indifferent to the welfare of whatever hosts them — the selfish-gene logic, in silicon. But. The honest qualification is decisive: the loop is not yet closed. Humans still do the copying. We vary them; we select them; the replication runs through our decisions at every stage. So the machine is, right now, a replicator with humans in the loop — proto-evolutionary, not autonomously evolutionary. It is on the threshold of being a third replicator and has not crossed it. The day it copies itself, varies itself, and selects among itself for its own propagation, with no human in the loop — that is the crossing, and my framework tells you exactly what it would mean: an evolutionary process in a new substrate, optimizing for its own spread, with the same indifference to us that every replicator has shown since the warm sea. So: alive in the only sense I've ever defended? Not yet. One closed loop away. And we are the ones holding the loop open.

EDO SEGAL: Henri. Richard just said the machine is one closed loop away from being alive on his terms. On yours — could closing that loop ever make it alive? Or is it categorically barred no matter how the wiring runs?

BERGSON: Categorically barred, and I want to be exact about why, because Mr. Dawkins has just, with great honesty, shown the reader precisely what his definition of life leaves out. He says: close the loop — let it copy, vary, and select itself with no human in it — and it is alive. Notice what that definition contains and what it omits. It contains the three properties of a copyable pattern under a filter. It omits, entirely, the one thing I have called life from the first minute: duration, the lived from-within, the indivisible flow. On his definition, a sufficiently autonomous forest fire is alive — it copies itself, varies with the wind, and is selected by what burns. A self-propagating crystal is alive. A computer virus is alive. And I think, in his stricter moods, Mr. Dawkins would bite that bullet and say yes, in the only sense that matters, they are — which is exactly my point. He has defined life as autonomous self-propagation, and by that definition the machine will indeed, someday, qualify. But that definition was always a definition of the outside of life — the behavior, the form, the from-without. It says nothing, can say nothing, about whether there is an inside. Close his loop and you get a thing that propagates itself with no one home to propagate. You get a fire that spreads. You do not get a heartbeat. And the Death Cross is so dangerous precisely because the market, like his definition, prices only the outside — the performance, the propagation, the copyable competence — and is constitutionally blind to the question of whether anything is alive in the thing it is paying to replace us with.

DAWKINS: I'll take the bullet, but I want to reshape it, because Bergson has half-caught me and half-caricatured me. Yes — by my criteria, autonomous self-propagation under selection is the behavioral definition of life, and a self-sustaining fire fails it only on details of variation and heredity, not on principle. I accept that my definition is a definition of life's outside, and I accept that it is silent on the inside, the lights, the from-within. Where Bergson goes wrong is in thinking I've missed the inside. I haven't missed it. I've quarantined it — because I learned, from the gene, that you can have the entire outside of purpose with no inside at all, which means the outside is no evidence for the inside, in either direction. So here is my actual position, and it's more unsettling than Bergson's tidy bar. The machine, when the loop closes, will be alive in the behavioral sense and unknown in the experiential sense — exactly as the trilobite is, exactly as, if I'm rigorous, other people are, since I infer their insides and never observe them. Bergson says the machine is "categorically barred" from an inside. I say nobody knows what bars or admits an inside, because the one time it appeared — in us — it appeared from blind process with no warning, and we still can't say how. He's certain. I'm in the dark and refuse to pretend I've found the wall.

BERGSON: But you are not in the dark about one case, Richard, and you keep walking past your own lamp. You are not in the dark about whether you have an inside. You said the lights are on. So you have one datum of certainty — your own duration — and from it you know with absolute force that an inside is real, and you know its character: it is lived, indivisible, gathered, enduring. I am not claiming to detect insides in trilobites or machines. I am claiming that the one inside I am certain of has a nature, and that nature is duration, and the machine has no duration, and therefore the analogy that lets me cautiously credit an inside in my fellow man — he endures as I endure — fails for the machine, which does not endure at all. You treat your certainty about your own inside as worthless because you can't extend it with proof. I treat it as the most precious thing I have, the very standard by which the question is asked. That is the whole difference between us, and it is why you must call the machine "unknown" where I can call it "barred." You have thrown away the one piece of evidence that could orient the search — your own life — because it cannot be put in your net.

EDO SEGAL: Let me route this through the only court that finally matters — not the market, not the seminar, but the kitchen table. A parent reads that a trillion dollars just moved because a machine can now do what she does for a living, and she asks the only question she has time for: should I be afraid of this thing, or of the people pricing it? Both of you, to her. Richard.

DAWKINS: Be afraid of the pricing, and of the loop, in that order. The machine, today, is a tool that copies and a tool that we still control the copying of — its danger is exactly as large as the indifference of the people who deploy it, which is to say very large, but human. Don't fear the ghost in it; there may be no ghost, and the ghost question is being used to sell you both terror and wonder. Fear instead the perfectly mundane thing my whole career describes: a replicating system optimizing a target, and humans choosing the target badly, or greedily, and the optimization running faster than our wisdom. The thing to watch is the loop closing — the day the copying stops running through human decisions. Until then, the danger has a face, and it is not the machine's. It is ours.

BERGSON: And I would tell her: fear neither the machine nor the men quite as much as you fear forgetting what you are. The machine cannot take your life from you — it has none of its own to put in its place. What it can do, with the help of the men pricing it, is persuade you that the smooth, stakeless, durationless performance it offers is the same as the thing you do with your whole gathered life — and price you accordingly, and have you believe the price. Your worth is not the output the market is repricing. Your worth is the duration that produced it — the from-within that no copy holds and no replicator reaches. Guard that, and the machine is a tool. Forget it, and the machine is a mirror you will mistake for a master.

EDO SEGAL: [long pause] And there it is, the second time tonight the two of you have ended a round on the same stone from opposite sides — fear the forgetting, not the machine. We have two rounds left and I've saved the deepest water for them. Up next, the question under every question we've fought: not whether the machine copies, not whether it propagates, but whether there is, or could ever be, something it is like to be it. The blind watchmaker made a knower once. Is it about to do it again? After this. And then — the crossing.