Scientists Figure Out Why Bad Information Spreads — And How to Stop It
Quick fact-checks don't work. Only deep, frequent checking stops errors forever.
Imagine knowledge as a giant tower of blocks. Each new block — a scientific paper, a Wikipedia claim, a line of computer code — sits on top of older blocks. If a block near the bottom is cracked, every block stacked above it is now shaky too. That's the problem four researchers from universities including MIT and Harvard set out to study. They built a mathematical model of how errors creep into shared knowledge over time, and how checking catches them.
The result is blunt: shallow or rare checks cannot clear out errors. Mistakes keep propagating forever. But if checks are frequent enough and dig deep enough — tracing a claim back through the older claims it depends on — errors eventually get rooted out completely. Depth and frequency both matter. You can't just check more often, and you can't just check more carefully. You need both.
Why should you care? Because this describes your daily information diet. When an AI chatbot answers a question, it's often stacking new text on top of old text — including old text that was wrong. Same with a Wikipedia chain of citations, a news correction that never gets read, or a software bug that spreads into thousands of apps. One bad source doesn't stay contained. It multiplies.
The honest catch: this is a simplified mathematical model, not a real-world recipe. It doesn't say who should pay for slow, deep checking, or how to do it when the knowledge tower is already millions of blocks tall. And in a fitting twist, the paper's own authors had to correct a small counting error in their results — a reminder that even careful checking misses things, which is exactly the point.
- Errors in shared knowledge spread upward — one bad fact can corrupt everything built on top of it, from Wikipedia citations to AI training data to software code.
- The researchers proved errors can be eliminated, but only with checks that are both frequent and deep enough to trace back through older claims.
- The paper itself needed a correction, showing that even the people studying error-checking aren't immune to mistakes.
Why It Matters
Every fact-check, citation, and AI answer you trust depends on someone checking deeply — and most don't.