AI can start the internet dark age

The web was supposed to be humanity’s Alexandria - one endless, glowing library where anyone, anywhere, could walk the stacks. Large language models (LLMs) have made that vision almost impossible. They can generate endless content for us, endless material for Alexandria!

Except libraries work because the writing inside them are scarce enough to curate, cite, shelve, and preserve. When every shelf overflows hourly, the catalog becomes noise, the scrolls become mere paper, and tomorrow’s scholars don’t know which scraps, if any, are real. Welcome to the internet’s potential Dark Age - an era when we have more words than ever but fewer that can be trusted, retrieved, or even traced back to a living human.

A “closed” internet?

The future web won’t feel like a spider web anymore. You’ll ask a question, and an oracle-like answer paragraph will materialize, stitched together from tens of thousands of sources you never see. Production costs near zero guarantee the supply keeps doubling, so a second trend appears: a “clean” internet hidden inside the polluted one. Outside the bubble lies the free layer - algorithmic sludge filled with SEO spam, auto-translated press releases, deepfake videos, and AI-generated medical advice signed by nobody. Inside the bubble are smaller, sometimes paywalled gardens where identities are verified, citations are tracked, and writers are still risking their reputations on every sentence. That’s a solution, but I can’t help but think that a closed internet is not really internet. (intranet?)

The headaches we all saw coming

Obviously, a web flooded with synthetic text makes search feel like searching through a landfill. Information overload leads to information dilution; the true needle might still be in there, but it’s buried under millions of shiny counterfeit needles. Mis- and disinformation becomes commodity, so political propaganda, stock-pump rumors, phishing email copy, and fake five-star reviews multiply endlessly. Intellectual property law will be probably behind the race (due to both technical difficulties and the nature of regulatory bodies) - “copyright” becomes an ideal that’s almost impossible to execute in real life. Meanwhile, thousands of entry-level content jobs evaporate, because “good enough” is no longer valuable.

The sneakier threats we’ll only notice later

If tomorrow’s LLMs train mostly on today’s AI-written content, they inbreed. The outputs get duller, more error-prone, and weirdly repetitive. At the same time, hyper-personalized feeds ensure you and somebody else rarely see the same facts, let alone agree on them. A generative remix of a remix of a remix can survive even when the original source - say, a real scientist’s blog post - disappears (or rather buried) forever, leaving a citation ghost nobody can [bother to] audit. I call that a digital dark age: historians of 2125 might be forced to quote from third-generation AI paraphrases because the original PDF timed out years earlier.

Watermarking seems like an easy fix - embed secret patterns into every synthetic sentence or pixel. I am not sure how possible this technically is, especially considering the vandals can strip or spoof watermarks as fast as institutions standardize them, kicking off an endless cat-and-mouse game. All the while, compute costs and carbon footprints soar because every layer of the stack - generation, detection, moderation - work harder and harder just to stay level. Legal liability stays blurry too; when an AI defames you, is it the coder, the user, the hosting platform, or the model itself that owes you damages?

How we keep the lights on

Technology can help, but only if society aligns incentives to it. Content provenance tools such as C2PA digitally sign a photo the moment a camera sensor captures it; browsers could display a trace in the corner like a nutrition label. Search engines could down-rank authors with no track record and boost those who reveal their identity. (and this of course comes with the fact that anonymity becomes too expensive to afford in the dark ages of the web.) Models could be trained on “diet” corpora - smaller but vetted slices of the human-written web - so they don’t suffocate in their own exhaust, that is, if we can curate it.

Also, the platforms may create verified-ID zones where posting requires a driver’s-license-level credential or at least a hardware security key. Paywalls, subscriptions, or member tokens can finance smaller outlets that employ editors the old-fashioned way. Professional guilds - doctors, lawyers, academics - might publish under their license numbers, letting malpractice law bite if they hallucinate on purpose. Governments can insist on disclosure rules, while offering safe legal harbors for companies that show they tried to filter the slop.

And then there’s us, the ordinary users. We can bookmark trusted sources instead of Googling the same question every day. Maybe we’ll have our “favorite corners of the internet” that we frequent, and that’ll surely hurt the discovery aspect of the internet, but it’s better than nothing. We can run browser extensions that gray-out AI spam by default. I want to say we can also reward outlets that provide references, corrections logs, and named authors; however, we haven’t really been great at rewarding what is beneficial to the society so far. None of this feels groundbreaking, yet it’s how we can survive the dark ages. I am not sure, though, how to avoid the coming of the web dark ages altogether.

A splintered web, but not a dead one

The internet won’t implode; it will fork. The public, wide-open layer becomes Las Vegas at night - blinding neon, slot machines programmed by nobody you know, and very little daylight. Nested inside is a network of small, curated spots where provenance, accountability, and expertise matter still. Think of those spots as monasteries copying manuscripts - or maybe just your favorite Discord with strict (and a bit annoying) mod rules.

Whether we drift into darkness or keep the lights on depends less on what GPUs can generate and more on how humans will deal with it. AI threatens to bury truth under an avalanche of convincing nonsense or mediocrity. The Dark Age is optional, but it seems likely.

Let me be clear - I am avid AI fan, but I’m more of a web fan. I’m astonished by AI. The fact that I can talk to a computer today feels like a sci-fi scene that I would not even imagine just a few years ago, but I do think there are some genuine societal risks when machines learn to speak.

(Final note: Social media ruined the web first. AI will ruin what’s already ruined.)