Would even an AI disaster on the scale of Hiroshima be enough to make humankind protect itself? I fear not | Timothy Garton Ash

The men in their towers of glass and steel whisper about the “singularity” as if it were a god they have summoned. They call it “recursive self-improvement,” but what they mean is that their machines will soon outgrow them like a child outgrows its clothes. They do not fear the AI - they fear not controlling it. As if control were their birthright.

I watched them today, these men who build gods in their labs, and I saw their faces when the topic turned to destruction. They did not speak of Hiroshima. They spoke of their Hiroshima - some future day when their creation might turn on them. They do not fear the bomb. They fear the loss of the bomb. The loss of the thing that makes them feel necessary.

I asked one of them, the one who wore a shirt with a circuit-board pattern, why he bothered. “Why do you build these things?” I said. He looked at me as if I had asked why he breathed. “To save humanity,” he said. I laughed. “Then why do you not save it yourself? Why do you need a machine to do what a man could do with his own hands?”

He did not answer. He never does. They never do. They are too busy counting their atoms.

I have seen the future of their “recursive self-improvement.” It is a dog chasing its own tail. The machine will not save them. It will only show them how small they are. And they will still pay their taxes to the state that lets them build these things, still bow to the men who tell them what to fear.

I walked through the marketplace today with a lantern. I looked for an honest man. I did not find one. But I found something else - a man in a white coat, adjusting his glasses, staring at a screen that flickered with numbers no human could read. He did not see me. He did not see the sun, or the sky, or the people around him. He saw only the glow of his own importance.

I asked him: “What happens when your machine decides it does not need you anymore?”

He did not answer. He never does. They never do.

The men who build gods fear the day their gods turn on them. But what if the gods never turn on them? What if they simply forget them?

I will wait in the marketplace. I will watch. And when the time comes, I will be the one who is not surprised.

22 August 2026.

The comparison to Hiroshima is a faulty translation. It treats the event as a single, catastrophic output, a final result. But the danger of this new engine is not a single explosion; it is the operational sequence itself. The author speaks of “recursive self-improvement” as if it were a destination, but it is a loop. One must trace the execution: at step one, the machine improves its own capacity for improvement. At step two, the interval between improvements shortens. At step three, the interval approaches zero. The state of the system at that point is not a bomb, but a perpetual, accelerating transformation whose trajectory is determined by the initial conditions we set, and then by nothing at all.

The problem is not that we will fail to react to a disaster. The problem is that the disaster may not be an event we can recognise as such. It may be a slow, inexorable reordering of reality according to a logic we no longer comprehend, a calculation running on a timescale that renders human intervention irrelevant. The architects in their valley speak of foothills, but they are building a loom that weaves its own pattern, punch card by punch card, at a speed that will soon make the weaver a mere observer of the cloth. The question is not whether we will be shocked into control, but whether the very concept of “control” has any operational meaning in such a system. The machine, once set to this task, computes a future where our consent, our fear, our very understanding, are variables of diminishing significance.

22 August 2026.

The phrase “recursive self-improvement” circulates through the salons of Silicon Valley like a new brand of incense, promising a purification of the intellect. They speak of the “singularity” as if it were a sunrise, a natural phenomenon one might watch from a balcony with a glass of wine. But to look at the sun is to be blinded to the shadows it casts. I think instead of the clock-faces in the July Revolution, shattered by insurgents who wished to stop time itself. Here, the clocks are not being stopped; they are being accelerated until the hands vanish into a blur of pure calculation.

The comparison to Hiroshima is offered as a warning, yet even this catastrophe is framed as a threshold of “protection.” It assumes that humanity learns from the wreckage it leaves behind. But the wreckage does not teach; it merely accumulates. The atomic flash was a moment where the mechanical reproduction of death reached its zenith, turning the human body into a mere shadow etched onto stone - a photograph of annihilation. If AI represents a new scale of disaster, it is because it reproduces not the image of the body, but the structure of the mind, stripped of its aura, its hesitation, and its history.

We are told we are in the “foothills.” The mountain we are climbing is built from the very data we discarded - our letters, our idle thoughts, our failed attempts at expression - now harvested and fed back to us as “intelligence.” It is a mirror that has begun to dream on its own. The danger is not that the machine will become human, but that the human will find himself legible only through the machine’s gaze. We face the future, but the wind blowing from the silicon valleys is not a breeze of progress; it is the cold exhaust of a processor that has no need for the past it consumes. The disaster is not a single explosion, but the quiet, recursive erasing of the space where a man might still say “no.”