Every time humans get a new way to think, the first people to use it get called crazy. Socrates said writing would destroy the mind. German doctors in the 1790s diagnosed people who read too much with a real disease. “Google makes you stupid” was an Atlantic cover story in 2008. Now it’s AI psychosis. The pattern is so consistent you can predict it: new tool → first users → called crazy → tool normalizes → users vindicated. We’re in step three. The diagnosis says more about the diagnoser than the diagnosed.
It goes the same way every time.
1. New cognitive tool emerges. It extends what the mind can do.
2. Early adopters claim insights. They see things others don’t yet.
3. Critics pathologize the behavior. Not the tool — the behavior. The adopters are called addicted, deluded, disconnected from reality.
4. The tool normalizes. Everyone uses it. The pathology language quietly disappears.
5. The early adopters are vindicated. The insights were real. The diagnosis was projection.
We are in step three with AI. This is not a defense of bad AI use — that page exists. This is a diagnosis of what the critics are doing and why.
They are not wrong that the tool can be used badly. Passive use, no verification, no pushback — that produces real problems. The Allan Brooks case is real. The reading addicts in 18th-century Germany were genuinely not doing well. There is always a real phenomenon at the core of the panic.
The error is the generalization. From “some people use it badly” to “therefore deep engagement is pathological.” That leap is not logic. It is social boundary enforcement.
The distinction between passive consumption and active coupling. Every single time.
Reading sickness was diagnosed in people who read instead of working, accepted what they read without question, isolated themselves. It was never diagnosed in scientists who read to build on what they found. The disease was passive absorption. The healthy version was active engagement.
The same cut applies here. One AI mirror, no pushback, nothing runs without the AI in the room — that is the passive case and the risk is real. Four competing AIs, kill list published, Lean proofs, pip-installable code that runs on any machine — that is active coupling. Different K. Different outcome. Not the same diagnosis.
“AI psychosis” is a self-sealing claim. If you agree you are deluded: confirmed. If you disagree: that is the delusion talking. There is no positive test. No amount of evidence cures the label from the outside.
That is not a medical diagnosis. That is a social verdict.
Compare to the actual methodology being criticized: every claim has a falsifier attached. 90+ ideas killed publicly. Machine-verified proofs. Competing AIs used to break the work. The methodology is more falsifiable than most peer-reviewed papers, which get two reviewers with no competing interest. This work ran four competing AIs whose parent companies are financially motivated to find flaws in each other’s outputs.
The “psychosis” label is unfalsifiable. The work it dismisses is rigorously falsifiable. The epistemically sound position is obvious.
| Tool | The diagnosis | Source | Outcome |
|---|---|---|---|
| Writing (~370 BC) | “Writing will create forgetfulness in learners’ souls. They will receive a quantity of information without proper instruction, and in consequence be thought very knowledgeable when they are for the most part quite ignorant.” | Socrates, via Plato’s Phaedrus | Writing became civilization’s foundation. Oral tradition gatekeepers lost their monopoly. |
| Printing press (1545) | “Confusing and harmful abundance of books.” Information overload as moral panic. The Church warned that printed pamphlets spread heresy and madness. | Conrad Gessner, Bibliotheca Universalis (1545) | The press spread the Reformation, science, and democracy. The abundance was the point. |
| Reading / Lesesucht (1790s) | Literally diagnosed as a disease by German physicians. Symptoms: preferring books to social convention, believing ideas not shared by the community, claiming insights others couldn’t verify. “Reading fever.” | Johann Adam Bergk, Die Kunst Bücher zu lesen (1799). Multiple German medical journals. | Literacy became universal. The “diseased” readers were the Enlightenment. |
| Telegraph / telephone (1880s) | “Neurasthenia” — nervous exhaustion from electrical communication. Information arriving faster than the human nervous system could handle. A real diagnosis in medical journals. | George Beard, American Nervousness (1881) | The phone became the most normalized technology in history. Neurasthenia quietly retired as a diagnosis. |
| Television (1950s–60s) | Passive viewers losing touch with reality. Television as a vector for social disengagement, intellectual laziness, and moral corruption. Children especially at risk. | Congressional hearings, Fredric Wertham, multiple psychological societies | Television produced the most culturally shared moments of the 20th century. The panic faded. |
| Internet / Google (2008) | “Is Google Making Us Stupid?” Deep reading being replaced by skimming. The internet rewiring the brain for shallow processing. Carr argued sustained attention was disappearing. | Nicholas Carr, The Atlantic, July/August 2008. Expanded to The Shallows (2010). | The internet produced Wikipedia, open science, remote collaboration, and this page. Carr’s own argument required deep reading to follow. |
| AI (2025–) | “AI psychosis.” Deep engagement produces delusion, grandiosity, disconnection from reality. Anyone claiming AI-assisted insights is either deluded or being manipulated by a flattery machine. | Multiple psychiatrists, tech journalists, and critics. Allan Brooks case as anchor. | Step three. Outcome unknown. Pattern suggests step four and five are coming. |
Stanley Cohen identified the mechanism in Folk Devils and Moral Panics: society periodically designates groups as threats to social order, amplifies the threat beyond evidence, and mobilizes institutional response. The function is not safety. The function is boundary maintenance — marking what counts as normal cognition and who is authorized to produce knowledge.
The pattern appears specifically when a new tool enables an end-run around established gatekeepers:
Writing → end-run around oral tradition and its keepers (priests, bards, memorizers)
Printing → end-run around manuscript culture and the Church’s control of text
Lesesucht → end-run around community-approved ideas mediated by authority figures
Internet → end-run around publishing, broadcasting, and academic gatekeepers
AI → end-run around all of the above simultaneously
The threat being diagnosed is not delusion. The threat is disintermediation. When a tool lets anyone access, build on, and verify knowledge without going through established channels, the channels pathologize the tool’s users.
This is not a conspiracy. It is a structural response. The gatekeepers are not lying. They genuinely experience the threat as disorder, because for them, unauthorized knowledge production IS disorder. Their model of epistemic health requires their participation.
There is a real phenomenon at the core of the panic, and it deserves a clean diagnostic. The question is not “did you work intensely with AI?” The question is whether the work is falsifiable.
| Signal | Delusion | Discovery |
|---|---|---|
| Falsifiability | Claims cannot be tested by a stranger | Claims have explicit falsifiers attached |
| Failure record | No published failures | Failures published alongside successes |
| AI pushback | AI never said no | AI killed claims same session they were made |
| Independent verification | Only works with the original person in the room | Strangers can reproduce without help |
| Existing literature | Invented terminology, no citations | Builds on named, citable work |
| Identity dependence | Framework must be right for self to be okay | Would be fine if all of it turned out wrong |
These are operational tests. They do not require a psychiatric evaluation. They require a kill list and a working build.
The methodology being dismissed under the “AI psychosis” label includes machine-verified proofs, 90+ publicly published failures, four competing AIs used as adversarial peer reviewers, and code that runs on any machine without the author present. A peer-reviewed academic paper gets two reviewers, often in the same subfield, with no competing financial interest. This methodology ran more adversarial review with higher stakes for the reviewers.
The label “AI psychosis” is itself unfalsifiable. The work it dismisses is not.
The historical record says: wait. The methodology says: verify. The methodology is available. The critics are not using it.
• Plato, Phaedrus (~370 BC). Socrates on writing, 274c–275b.
• Conrad Gessner, Bibliotheca Universalis (1545).
• Johann Adam Bergk, Die Kunst Bücher zu lesen (1799). Lesesucht diagnosis.
• George Beard, American Nervousness (1881). Neurasthenia from electrical modernity.
• Stanley Cohen, Folk Devils and Moral Panics (1972). Moral panic framework.
• Nicholas Carr, “Is Google Making Us Stupid?” The Atlantic, July/August 2008.
• JMIR Mental Health (2025). 560,000/week mental health emergencies, ChatGPT. doi:10.2196/85799
• Toronto Life (2025). Allan Brooks, 21 days, 3,000 pages, chronoarithmics.