Is AI Taking Our Jobs? Signal Boosting Noah Smith - Economist Writing Every Day

TEXT START: Artificial intelligence is supposed to transform the labor market.

The Dissection

The article is documenting a lag and presenting it as a rebuttal. It gathers aggregate employment, wage, firm-survey, and occupational evidence showing that AI has not yet produced economy-wide displacement. It then uses the task-versus-job distinction to preserve the possibility that automation will expand demand for human labor.

The text quietly quarantines its most dangerous evidence: weaker outcomes for young workers and severe declines in digital media and creative employment. Those are treated as exceptions rather than early indicators. The article’s real function is to convert “the system has not broken yet” into “the system may not break.” That is a temporal substitution disguised as economic analysis.

The Core Fallacy

The central error is confusing short-run employment resilience with structural immunity.

The data described are snapshots of adoption, hiring, and reported employment effects. They do not establish that AI cannot sever the mass employment–wage–consumption circuit once capability, diffusion, and competitive pressure mature. New tasks and expanded software demand can temporarily offset substitution. They are not a permanent law requiring firms to retain humans after machines become cheaper and more capable.

The article also treats aggregate employment as the decisive variable. That misses the mechanism. A labor market can preserve headline employment while destroying entry-level access, bargaining power, career ladders, and the economic necessity of ordinary workers. The young-worker evidence is therefore not a footnote. It is the exposed edge of productive participation collapse.

“No statistically significant increase in unemployment” is not evidence that the underlying transition is harmless. It does not rule out hiring avoidance, occupational narrowing, wage compression, or the disappearance of the jobs through which new workers acquire experience. Statistical non-detection is not structural exoneration.

Hidden Assumptions

• Output expansion will continue to create human jobs as quickly as AI eliminates or bypasses them.
• Firms reporting “supplementation” are measuring durable employment demand rather than temporary augmentation during transition.
• New tasks will remain human-intensive instead of becoming the next targets for automation.
• Productivity gains will flow into labor demand rather than mainly increasing capital returns and reducing headcount requirements.
• Experienced workers’ complementarity with AI will protect them indefinitely.
• The collapse of entry-level work can occur without eventually damaging the broader occupational pipeline.
• Stable employment stocks reveal stable productive participation.
• Occupation-level wage growth cleanly measures demand for human labor, despite the article admitting that composition and labor-supply effects complicate the result.
• Competitive firms will voluntarily preserve human employment when doing so becomes economically unnecessary.

These assumptions smuggle permanence into evidence that only demonstrates delay.

Social Function

Primary classification: transition management and ideological anesthetic, with a legitimate partial-truth core.

The article is accurate that automation does not mechanically equal instant job elimination, that firms can expand during early adoption, and that current mass displacement has not appeared in the cited evidence. That accuracy gives the reassurance its credibility. But the reassurance narrows the time horizon until the structural threat disappears from view.

Its citations, survey percentages, and careful economic language also perform prestige signaling: the reader is encouraged to regard the popular displacement narrative as unserious because the evidence is still mixed. The result is a respectable-looking lullaby for a system whose first casualties are already visible at the entry level and in creative work.

The Verdict

This is not a refutation of the Discontinuity Thesis. It is an empirical report on the lag before the thesis becomes unmistakable.

The article proves only that AI has not yet destroyed aggregate employment at scale. It does not disprove durable cognitive substitution, institutional inability to preserve human-only economic domains, or the eventual collapse of productive participation. Its strongest counterexamples are temporary demand expansion and delayed displacement; its strongest evidence of change is the weakening of young workers and the rapid damage to creative occupations.

The text is therefore a partial truth deployed as premature reassurance: accurate about the present surface, blind to the competitive mechanism beneath it. The system is not cleared. It is still absorbing the shock.