AI is cheaper than workers. Bill Gates wants to change that | PCWorld

TEXT START: AI is cheaper than workers.

The Dissection

The article is a soft policy intervention disguised as AI commentary. It accepts the central economic fact—that firms will replace labor when machine cognition is cheaper—and then searches for a regulatory brake strong enough to preserve the old employment bargain.

Gates’s “Human Reserved” category is mostly ceremonial. Jobs are bundles of tasks, not sacred human territories. Firms can automate the cognitive core while retaining a thin human shell for liability, legitimacy, or customer preference. The label therefore protects appearances more easily than livelihoods.

The token tax is more serious, but it remains transition management. It could correct a tax asymmetry, slow marginal deployments, and finance retraining or transfers. It cannot make human labor structurally competitive once AI delivers superior cost, speed, scalability, and consistency.

The Core Fallacy

The article treats replacement as a tax-policy distortion rather than a competitive convergence. If AI is materially cheaper and capable enough, taxation can change the speed and route of adoption; it cannot repeal the incentive to automate.

Under the Discontinuity Thesis, the tax may delay P1 and cushion P3, but it cannot defeat P2. Businesses will arbitrage the rules through cheaper models, private infrastructure, offshore entities, local inference, bundled software, robotics, or automation shifted into untaxed capital and workflows. A token tax can move the carcass. It cannot resurrect the wage-consumption circuit.

The proposed retraining programs also smuggle in a false solution: that displaced workers can be moved into a sufficiently large supply of new economically necessary roles. If automation is broad, retraining changes occupational labels while the scarcity of human-required labor remains.

Hidden Assumptions

• Governments can define and measure an AI token consistently across rapidly changing models and architectures.
• Major economies can coordinate enforcement despite global competitive pressure.
• Firms will absorb the tax rather than pass it to consumers, cut other labor, relocate activity, or evade it.
• Human labor remains substitutable enough that a tax can preserve mass employment rather than merely postpone displacement.
• Retraining creates durable demand instead of producing credentialed surplus labor.
• “Human” work is economically distinct from AI-assisted work and can be protected through job labels.
• The political system will tax AI capital without allowing owners to capture the proceeds or rewrite the rules.
• Preserving consumption through transfers is equivalent to preserving productive participation.

Social Function

Primarily transition management and ideological anesthetic, with a substantial partial-truth component. The article correctly identifies a tax code that may favor automation and correctly notes that implementation would be messy. But it frames the crisis as an adjustable incentive problem, allowing readers to imagine that competent taxation can preserve the familiar order.

That is the lullaby: the machine may take the job, but a clever enough committee can keep the system intact. The article never reaches the decisive question—who owns and controls the productive AI capital after labor loses bargaining power. Its policy proposal manages distribution after displacement; it does not solve sovereignty.

The Verdict

The article identifies a real lever and mistakes it for an escape hatch. An AI token tax could buy time, fund transfers, and slow the first wave of substitution. It cannot stop the Discontinuity Thesis: once machine cognition becomes cheaper and institutionally deployable, mass employment becomes a protected remnant, not the engine of capitalism. “Best bad idea” is accurate—but only if “bad” means hospice care for a dying economic order.