AI is not a normal technology

TEXT START: The only remaining fundamental disagreement between people who think AI is a normal technology and those that don’t is the belief that the AI can do everything a human can do, but better.

THE DISSECTION

This text argues that “normal technology” is a category error. It identifies the real discontinuity: a general optimizer that can enter every cognitive labor market, generate the tasks created by its own disruption, and compete with humans at machine speed and cost.

Its strongest point is that automation does not merely eliminate existing jobs; new jobs can also become targets. It correctly identifies the deflation mechanism: once a skill becomes reproducible at scale, its scarcity and wage power collapse, even if output quality rises.

The mathematics section exposes the deeper damage. AI separates valuable answers from the human training process that once produced understanding. That is not merely cultural perversity. It is a power transfer: the owners of the optimizer inherit the output, while everyone else loses a bargaining chip.

The article correctly rejects the idea that AI is “normal” because it lacks mystical agency. Its abnormality is functional: generality, recursive capability acquisition, low marginal cost, and relentless competitive deployment.

THE CORE FALLACY

The article mistakes a conditional endpoint for a complete mechanism. If AI can do everything humans can do better, human labor loses most of its value. But that does not mean every human job disappears instantly, becomes infinitely cheap, or disappears on the same schedule.

The Discontinuity Thesis is more precise: P1, capability dominance; P2, inability to preserve stable human-only economic domains; P3, collapse of majority access to economically necessary work. The system can die while pockets of human employment remain.

The article also confuses capability with deployment. “Can do” is not identical to “is legally permitted, physically embodied, insured, trusted, powered, networked, and granted authority.” Those are lag defenses. They delay the death; they do not reverse it.

Its claim that anything AI can do becomes infinitely cheap is also too clean. AI may commoditize labor while rents remain concentrated in compute, energy, robotics, data, platforms, property, and permissions. The result is not universal free abundance. It is cheap cognition chained to expensive ownership.

The largest omission is the Sovereign question: who controls the AI capital stack? If output rises while ownership remains concentrated, transfers can preserve consumption without restoring productive participation. The worker does not become free; he becomes economically unnecessary and politically managed.

HIDDEN ASSUMPTIONS

• Superior performance automatically produces immediate substitution. Competition makes substitution likely, but liability, physical bottlenecks, trust, and institutional inertia create uneven timing.
• Human authenticity is nearly irrelevant. It is economically small, but relational trust, embodiment, legitimacy, status, and human presence can preserve niches. They are moats, not mass rescue.
• Taste is a durable seniority moat. AI can model taste, and firms controlling distribution can concentrate it. Seniority buys time and leverage, not immunity.
• New demand cannot outrun automation. Under true generality, that is a strong DT expectation, but new demand may still create temporary labor requirements.
• Global coordination will fail. This is probably the correct competitive assumption, but the article treats it as obvious rather than analyzing state coercion and strategic limits.
• Higher productivity means higher quality of life for the population. It does so only if access to output is distributed. Otherwise productivity becomes a weapon for lowering labor’s price.
• The central loss is craftsmanship or mathematical meaning. Those losses matter, but the terminal economic loss is bargaining power: the ability to sell human time for claims on social output.

SOCIAL FUNCTION

Partial truth and transition warning, with a layer of credentialed-class mourning. It is not copium. It punctures the lullaby that AI will merely create better jobs and correctly identifies entry-level cognitive labor as the first carcass.

But it converts a political-economic expropriation into a story about authenticity, taste, and the tragedy of mathematics. That makes the collapse culturally legible while leaving the ownership regime insufficiently indicted.

THE VERDICT

This is an accurate alarm with an incomplete autopsy. It correctly identifies why AI is unlike cars, factories, or the internet: it attacks the general capacity to perform economically useful cognition, including tasks created by its own disruption. It also correctly sees that “be human” is not an economy-wide defense.

Its error is treating the death of jobs as the whole event. The decisive event is the death of labor’s necessity as the distribution mechanism for social purchasing power. Human jobs may survive as lag-bound, embodied, status-bearing, legally required, or servile functions. That does not save the post-WWII order.

Once AI capital dominates, Sovereigns own the productive system, Servitors remain only where they are indispensable, and everyone else becomes a claimant, consumer, hobbyist, or managed surplus population.

Authenticity is not a lifeboat. It is a luxury niche on the deck of a sinking ship.