We were warned that AI would flood the internet with lies.

But what if the stranger danger is not that the web becomes more false?

What if it becomes more same?

In this episode of Apiana Essays, Elias Flores explores “The Correlated Web” — the idea that AI-generated content may damage the internet not by making every claim wrong, but by making independent-looking sources quietly repeat the same patterns.

The old web was messy, noisy, biased, and often wrong. But its errors came from many different chains of contact with the world. That messy independence made corroboration useful. A thousand sources could sometimes function like a noisy sensor array.

AI-generated pages create a different problem. A thousand pages may look like a thousand witnesses, while underneath they share the same model behaviors, training distributions, smoothing patterns, and statistical preferences. Agreement starts to look like evidence. Repetition starts to look like confirmation.

The episode pressure-tests the idea through model collapse, retrieval collapse, AI slop, provenance failures, watermark limits, and the difference between human authorship and grounded contact with reality.

The claim is not that human writing is automatically trustworthy, or that AI assistance makes work worthless. The deeper question is: where did the claim touch reality?

Did someone observe, measure, interview, test, inspect, verify, or report? Or did language merely predict more language and enter the web dressed like evidence?

The future may not split into “AI internet” versus “human internet.” The more useful boundary may be between the synthetic layer and the grounded layer: claims traceable back to reality, backed by people or institutions with something to lose.

Because the real clock is not when everyone uses AI.

The real clock is when nobody bothers to go look anymore.

Apiana Essays: strange questions, serious arguments.