Post-ChatGPT: Jobs stayed, tasks changed - W. P. Carey News - Arizona State University

TEXT START: New research finds that generative AI is reshaping how artists work without significantly affecting employment or wages.

The Dissection

The article converts an early, narrow labor-market observation into psychological reassurance. It finds that artist job titles, aggregate employment, and wages did not materially collapse between 2017 and 2024, then frames task substitution as benign augmentation.

The real finding is weaker: during the first phase of adoption, AI was used as a productivity tool inside occupations that still retained legal, institutional, physical, and cultural protections. The article measures whether the corpse moved, not whether the economic organism remains viable.

The Core Fallacy

It treats occupational persistence as proof of productive necessity.

A job can survive while its economically valuable tasks are hollowed out. Employers may retain the title, reduce headcount growth, lower bargaining power, expand managerial span, or demand much higher output from fewer workers. Stable wages during an early transition window do not disprove eventual displacement.

The comparison with photography, MP3s, and live performance is also structurally incomplete. Earlier technologies disrupted formats and distribution while leaving human scarcity intact. Generative AI attacks the production of cognitive and creative outputs directly, at near-zero marginal cost and with rapidly improving quality. A human dancer or orchestral performer may remain physically indispensable for some performances, but that is a lag defense and a niche constraint—not evidence that the mass market for artistic labor is safe.

The paper's positive associations are especially fragile. AI adoption may be concentrated in stronger firms, better-paid workers, and expanding industries; rising wages may cause adoption rather than result from it. High standard errors turn the article's optimistic interpretation into an uncertain correlation, not a structural verdict.

Hidden Assumptions

• That current occupation categories will remain economically meaningful after their task bundles are decomposed.
• That augmentation will create enough additional demand to absorb the labor displaced by higher productivity.
• That productivity gains will accrue to workers rather than owners of AI systems.
• That seven years of data, including only roughly two years after ChatGPT's launch, can reveal terminal effects from a technology still scaling.
• That tasks AI cannot yet perform—live presence, physical embodiment, relationship signaling—will remain commercially valuable at sufficient volume.
• That creative labor markets will not be flooded by synthetic output, intensifying competition and depressing prices.
• That firms will continue using AI to assist workers rather than use it to redesign workflows around fewer workers.
• That employment and wages are adequate measures of productive participation, even if autonomy, bargaining power, and task quality deteriorate.
• That institutional inertia is a permanent equilibrium rather than a delay mechanism.

Social Function

Primarily copium and transition management, with a partial truth underneath.

The partial truth is that early generative AI often changes tasks before it eliminates formal occupations. The anesthetic is the implied conclusion that task change is a stable alternative to displacement. The article gives managers permission to call labor compression “workflow redesign,” while giving workers a temporary statistical reprieve and institutions a reason to postpone harder structural questions.

Its optimism also functions as prestige signaling: an academic finding is used to discipline public fear without establishing that the underlying employment-consumption circuit can survive durable cognitive automation. The article's own caveat—that the results capture only the early days—is the strongest evidence against treating it as a long-run rebuttal to the Discontinuity Thesis.

The Verdict

This is not a refutation of AI displacement. It is a snapshot of the lag phase.

Jobs stayed because adoption was early, institutions were sticky, demand had not yet saturated, and human labor still supplied residual tasks and legitimacy. That proves only that mechanical death has not yet become social death. Under DT logic, the decisive question is what happens when AI can perform most economically necessary cognitive and creative tasks, firms reorganize around that capability, and output expands without proportional human labor.

The article documents the first incision and declares the patient stable. The hemorrhage has not been ruled out; it has merely not yet reached the headline numbers.