The internet's collective memory is disappearing — and AI is both the culprit and the casualty. A provocative essay from The Walrus argues that the infrastructure that once made the web a reliable repository of human knowledge is breaking down in real time, and the implications go far beyond bad search results.
You've probably noticed it: Google searches increasingly return AI summaries that are confidently wrong. One Colorado resident set up an outdoor projector to watch a sunset, only to find that Google's AI had told them the sunset had already happened. It hadn't. The AI simply invented a time.
This isn't a minor bug. It's a symptom of a deeper problem: the web's knowledge infrastructure is degrading. Link rot erases pages daily. The Library of Congress briefly lost key sections of the U.S. Constitution due to a coding error. And the systems we've built to find and retrieve information are increasingly replaced by systems that generate plausible-sounding answers instead.
Wikipedia is one of the most significant volunteer-run knowledge resources in human history. For years, search engines sent billions of viewers to its pages. Now, AI systems scrape and ingest Wikipedia's content directly, presenting it in their own results — eliminating the need for users to visit Wikipedia at all.
This creates a vicious cycle. AI systems consume Wikipedia's content without contributing back. As Wikipedia's traffic drops, so does its volunteer base, which means fewer editors maintaining and expanding articles. The content that AI systems were trained on starts to degrade because the source is being starved of the attention and contribution that sustained it.
The Internet Archive, home of the Wayback Machine, is buckling under multiple pressures. The engineering strain of indexing and storing an ever-growing repository is enormous. The organization faces cyberattacks and costly litigation — after publishers successfully sued over its digital lending program, the Archive's ability to preserve the web has been compromised.
If the Wayback Machine goes down, link rot becomes irreversible. Pages that disappear from the live web will have no backup. The internet's memory — already fragile — would be genuinely lost. The problem isn't just that content is disappearing. It's that much of it is never being preserved in the first place. Instagram Stories, WhatsApp status updates, TikTok videos, and Discord conversations — enormous portions of cultural, social, and political communication exist only in ephemeral formats that vanish within 24 hours.
Previous generations left letters, newspapers, books — physical artifacts that survive in attics and archives. Our generation leaves Stories that disappear before the sun sets. The historical record of our era may be thinner than any since the invention of the printing press.
The dissolution of FiveThirtyEight illustrates the problem. After Disney acquired it through subsidiaries, the blog continued for over a decade. Its founder, Nate Silver, left in 2023. By March 2025, the site was effectively gone — its archive of data-driven political and sports analysis, built over years, scattered or lost.
When a major knowledge resource disappears, the links that pointed to it break. Citations in academic papers, references in news articles, bookmarks in browsers — all of them now lead to 404 pages. And the knowledge that was once accessible through those links is effectively erased from the retrievable web.
The standard framing of this problem is technical: "search is getting worse" or "AI hallucinates." But the deeper issue is cultural sovereignty and institutional memory.
When knowledge infrastructure degrades, the communities and institutions that depend on it suffer. Researchers can't find prior work. Journalists can't verify facts. Citizens can't hold institutions accountable. The web was supposed to democratize access to knowledge. Instead, it's becoming a medium where knowledge appears and disappears like mist — present in the moment, gone shortly after.
Some countries are already pushing back. France has developed Tchap, a homegrown messaging app for civil servants, replacing foreign platforms. European governments are moving away from foreign tech infrastructure, treating it as a threat to technological sovereignty and national cybersecurity.
Several approaches could help preserve the web's knowledge:
Decentralized archiving: Support the Internet Archive and similar organizations. The more copies of the web that exist, the harder it is to lose.
Persistent identifiers: Use systems like DOI, ORCID, and ArXiv IDs that remain stable even when URLs change.
Self-hosting knowledge: Organizations should maintain their own archives rather than relying solely on third-party platforms that may disappear.
Supporting Wikipedia: Wikipedia's volunteer model is under threat from AI scraping. Supporting the Wikimedia Foundation and contributing as editors helps maintain this critical resource.
Regulating AI scraping: Some form of fair-use framework for AI training on public web content needs to exist — one that doesn't starve the sources that make AI possible.
The irony of the AI era is that the technology that promises to make all human knowledge accessible may be destroying the very infrastructure that makes that knowledge possible. If the web's memory disappears, AI models trained on that memory will degrade too. We're not just losing the web — we're losing the training data for the future.