'AI Jobocalypse': Raghuram Rajan proposes AI token tax to curb job losses, boost govt coffers

TEXT START: According to details reported by The Economic Times, Rajan highlighted a structural financial distortion in modern employment markets: companies are mandated to pay social security and employment-related contributions for human staff, whereas replacing human labor with automated systems carries no equivalent fiscal obligation.

The Dissection

The text is a respectable transition-management memo wrapped around a fiscal proposal. It correctly identifies a real asymmetry: human labor carries payroll-linked social costs, while automated labor does not. It then treats that asymmetry as the central defect and proposes token taxation, retraining credits, and retention incentives as corrective machinery.

The deeper function is containment. The article converts a potential rupture in the mass employment–wage–consumption circuit into a governable policy problem: measure tokens, tax them, retrain workers, and wait for adjustment. Its language of gradual adoption, lower prices, rising demand, new firms, and augmented workers supplies institutional anesthesia. The machine is presented as disruptive, but still administratively domesticated.

The Core Fallacy

The proposal confuses slowing displacement with reversing the economic mechanism causing it.

If AI achieves durable cost and performance superiority across cognitive work, firms will be compelled to substitute it for labor. A token tax can increase the price of automation at the margin, but it does not restore the productivity, speed, scalability, or competitive advantage of human workers. It is a toll booth placed on a replacement process, not a reconstruction of the labor market.

The tax may raise government revenue and finance transfers. It may delay adoption in some cases. It may even preserve selected jobs temporarily. But under the Discontinuity Thesis, those are lag defenses. They preserve consumption or buy time; they do not preserve productive participation. The post-WWII circuit still breaks when the majority no longer possess economically necessary labor.

The argument that AI will lower prices and expand demand is incomplete. Demand requires purchasing power. If production becomes cheaper while wages and employment collapse, the system can manufacture abundance alongside mass insolvency. More output does not automatically create enough human income to buy it.

Hidden Assumptions

• That displaced workers can be retrained faster than AI expands into the next layer of work.
• That newly acquired skills will remain scarce after firms receive the same AI systems and deploy them competitively.
• That firms can retain workers without sacrificing enough cost or performance to lose against less constrained competitors.
• That governments can tax foreign AI providers effectively without driving usage offshore or through opaque intermediaries.
• That token consumption is a reliable proxy for labor displacement. It is not: a highly automated system may use fewer tokens, while a human-augmenting system may use many.
• That lower prices will generate sufficient new demand to absorb displaced labor rather than simply increasing output with fewer workers.
• That productivity gains will be distributed through wages instead of accruing primarily to owners of models, compute, energy, data, and distribution.
• That the current 20% adoption figure measures the future threat. It mainly measures the present lag before diffusion accelerates.
• That firms' acknowledgement of social responsibility can override competitive selection. It cannot, unless coordination is imposed at a scale institutions have historically failed to sustain.
• That gradual collapse is materially safer than rapid collapse. A slower death still kills the same circuit; it merely gives institutions more time to narrate it.

Social Function

Primary classification: transition management and partial truth.

Secondary function: ideological anesthetic.

The partial truth is that fiscal policy can capture some of AI's gains, fund transfers, and reduce the current subsidy favoring machine substitution. The anesthetic is the implication that this makes human economic indispensability recoverable. It does not. The proposal offers governments a manageable vocabulary for an unmanageable structural change: taxes, credits, retraining, retention, and adjustment.

Its most revealing sentence is the appeal for firms to help employees cope with uncertainty. That frames displacement as a corporate attitude problem. Under competitive pressure, it is a selection problem. Firms that preserve redundant labor indefinitely become targets for firms that automate it.

The Verdict

Rajan identifies the payroll distortion but mistakes the tax treatment for the disease. An AI token tax could become a useful extraction and transition instrument—government revenue, redistribution, and temporary braking—but it cannot reconstitute the mass labor market once AI makes human cognitive work economically nonessential.

The article is therefore not a refutation of the Discontinuity Thesis. It is evidence for it. The proposed policy is the state preparing to tax the mechanism that is severing its tax base, while calling the delay a solution. The likely endpoint is not restored employment, but managed obsolescence: ownership concentrated above, transfers below, and a shrinking class of humans retained only where they remain sovereign, indispensable, or useful as transition intermediaries.