Defining AI Psychosis. Part 2: "Prolific AI Psychosis"

TEXT START: Last week, I suggested that the term AI psychosis is applied to at least three different situations:

THE DISSECTION

The text identifies a real transitional pathology: AI multiplies output faster than humans can verify, understand, or connect it to actual value. It correctly distinguishes production from productivity and exposes the slot-machine reinforcement built into agentic tools.

But its deeper function is containment. It converts the destabilizing arrival of cognitive automation into a problem of individual discipline: supervise better, cultivate taste, sleep, and maintain boundaries. The article treats AI primarily as an intoxicating instrument used by developers, not as capital that can replace developers and concentrate productive power in its owners.

THE CORE FALLACY

The central error is scale blindness. The article assumes that human judgment, craft, and supervision remain durable economic advantages if applied carefully. Under the Discontinuity Thesis, they are temporary bottlenecks.

P1 makes cognitive work increasingly automatable. P2 prevents institutions from preserving large human-only domains once AI is cheaper and more capable. P3 follows: fewer people are required for economically necessary production. Human review may improve today’s AI output, but it does not preserve mass employment when one sovereign-controlled system can eventually generate, test, revise, and evaluate more work than entire teams.

The article correctly sees that bad AI output creates chaos. It misses the more consequential endpoint: sufficiently good AI output makes the supervising human scarce only temporarily, then makes much of that supervision cheaper, narrower, and more concentrated. “Human judgment” is a workflow requirement, not a salvation mechanism.

The author also treats fear of replacement as a contributor to irrational behavior. That is psychologically plausible but structurally incomplete. The fear is not merely hype-induced distortion. Under DT logic, it is an accurate perception of approaching dispossession. Workers racing to produce more may be maladaptive at the personal level while still responding rationally to a collapsing labor market.

HIDDEN ASSUMPTIONS

• Human taste and craft remain irreducibly superior to machine-generated judgment.
• Human supervision will continue to cost less than the errors it prevents.
• Markets will keep rewarding individual usefulness rather than rewarding ownership of AI infrastructure.
• Better workflows can preserve the developer’s role instead of merely increasing the output controlled by a smaller number of owners.
• Quality will remain scarce in a way that creates broad employment rather than a narrow verification elite.
• The current agentic bottleneck is a permanent feature, not a lag in the automation curve.
• Personal boundaries can counter competitive pressure from firms and workers using increasingly autonomous systems.
• The labor system remains intact enough for “productive AI use” to translate into continued livelihood.

The claim that subjective qualities such as taste cannot be found within an LLM is asserted as a cultural preference, not established as a structural limit. Even if human taste remains valuable, AI can model, imitate, rank, and operationalize enough of it to eliminate most human roles built around its routine application.

SOCIAL FUNCTION

Primary classification: partial truth, transition management, and ideological anesthetic.

The partial truth is substantial: AI-generated volume without verification is slop, and indiscriminate delegation can destroy value. The transition-management function is to teach surviving knowledge workers how to operate inside the early phase of automation. The ideological anesthetic is the relocation of a systemic labor crisis into personal habits and cognitive hygiene. If the worker is drowning, the proposed remedy is better swimming technique; ownership of the machine and control of the water remain unexamined.

THE VERDICT

This is a sharp micro-level autopsy attached to a strategically inadequate macro-level diagnosis. It correctly identifies one symptom of cognitive automation: output inflation outrunning judgment. It mislabels the broader condition as individual “psychosis” when much of it is a predictable response to competitive displacement and collapsing scarcity.

The article can help one developer avoid building an unmaintainable pile of machine-produced code. It cannot preserve the post-WWII wage circuit. Once AI becomes capable enough, the decisive question is not whether humans can supervise it skillfully. It is who owns the systems that produce, verify, and distribute value. The text diagnoses the smoke. The fire is the removal of human productive necessity.