THE DEAD INTERNET THEORY
| ORIGIN | Around 2016 to 2021, in forum posts, notably on Agora Road's Macintosh Cafe |
| STRONG CLAIM | The internet died around 2016; most content and users are now artificial |
| MEASURABLE | Automated traffic exceeded human traffic in several industry measurements from 2024 |
| STATUS | Unverified as stated. The weak version is increasingly supported |
THE CLAIM
That the overwhelming majority of internet content and activity is now generated by automated systems rather than people, and that this transition was deliberate.
THE THEORY, AS ITS PROPONENTS TELL IT
The theory has a strong form and a weak form, and they have moved in opposite directions since it was first stated.
The strong form holds that the internet effectively died around 2016, that most accounts and content since are artificial, and that this was orchestrated by governments or corporations to manage the population. As stated, it is unfalsifiable: any evidence of human activity can be reinterpreted as sophisticated simulation, which is the property this archive treats as disqualifying.
The weak form makes a measurable claim: that an increasing majority of internet traffic and content is machine-generated. Security industry measurement has repeatedly found automated traffic at or above half of all web traffic, and reported it crossing that line and continuing upward in recent years. Generative text and image systems have made bulk content production effectively free.
The interesting part for this cabinet is epistemic rather than technical. Every other file here was resolved, or left open, on the basis of documents, physical evidence and testimony, and increasingly all three can be manufactured convincingly at scale. This file is about the reliability of the method the other 199 depend on.
WHAT IS KNOWN
Automated traffic is a large and growing share of internet activity, and this is measured by commercial security firms with a commercial interest in the figure being alarming, which is a genuine caveat.
Coordinated inauthentic behaviour is documented and admitted. Platforms remove billions of fake accounts, publish the numbers, and have done so for years.
Generative systems can now produce text, images, audio and video that most people cannot reliably distinguish from human work, which is a change in kind rather than degree.
None of this establishes the orchestration claim. Bulk automated content is well explained by ordinary commercial incentives: advertising fraud, engagement farming and search manipulation are profitable and require no coordination.
EVIDENCE FOR
- Industry measurement showing automated traffic at or above half of all web traffic.
- Platforms' own published figures on removal of billions of inauthentic accounts.
- The collapse in the cost of generating plausible text and images to near zero.
- Documented state information operations, which establish capability and intent at smaller scale.
EVIDENCE AGAINST (THE STRONG VERSION)
- The orchestration claim is unfalsifiable as usually stated, which is a property of the claim rather than of the evidence.
- Commercial incentives explain bulk automated content entirely, without requiring any coordinating body.
- Traffic percentages are measured by firms selling bot mitigation, and the definition of automated traffic includes search crawlers and monitoring tools that nobody considers sinister.
- Human activity online remains enormous and verifiable, which the strong version can only accommodate by making itself untestable.
ASSESSMENT
Unverified as stated, increasingly supported in its weak form, and placed last deliberately. Every file before this one rests on a document that could be found, an object that could be examined, or a person who could be asked. This archive's method has been the same throughout: name the specific thing that would settle it. What this file records is that the supply of specific things is being diluted, that manufactured evidence is now cheap, and that the discipline of asking what would change the grade is about to matter considerably more than it did when the OSA opened its first file. Read the other 199 with that in mind.
WHAT WOULD CHANGE THE GRADE
- The strong version, that the transition was orchestrated, is unfalsifiable as usually stated: any evidence of human activity can be reinterpreted as sophisticated simulation. That is a property of the claim, not of the evidence, and it is disqualifying.
- The weak version is increasingly supported and is measurable. Industry measurement puts automated traffic at or above half of all web traffic, though it is produced by firms selling bot mitigation and counts search crawlers and monitoring tools.
- Commercial incentives explain bulk automated content entirely, without any coordinating body. Advertising fraud, engagement farming and search manipulation are profitable and require no conspiracy.
- What would change the grade for the whole archive, not just this file: manufactured documents, images and testimony are now cheap. Every other file here rests on evidence that could be found and examined, and the supply of checkable things is being diluted.
FURTHER READING
- Dead Internet theory · Wikipedia
- Imperva, Bad Bot Report · annual automated traffic measurement
- Platform transparency reports on removal of coordinated inauthentic behaviour
RELATED FILES IN THIS ARCHIVE
- OSA-181 · CAMBRIDGE ANALYTICACONFIRMEDHarvested Facebook data used for political targeting
- OSA-075 · QANONDEBUNKEDThousands of predictions, zero hits
- OSA-074 · PIZZAGATEDEBUNKEDA misread email, a nonexistent basement, an armed man
- OSA-199 · SLENDER MANDEBUNKEDA fictional creature with a documented author, and a real attempted murder