Stephen Moore: What bank ATMs teach us about AI - Maui News

TEXT START: New inventions and technologies, from the lightbulb to the farm tractor to the automobile to the laptop computer to Uber drivers, cause serious job market disruptions.

THE DISSECTION

The column uses the ATM as a historical sedative. It concedes disruption, then selects a case where narrow automation reduced routine labor while expanding branches, transactions, and higher-value human interaction. From that favorable example it extrapolates a general rule: technology destroys tasks, creates demand, and rewards workers who adopt the machine first.

Its real function is normative, not analytical. The column is defending rapid AI adoption and shifting responsibility for the consequences onto workers. “Get the machine before everyone else” is presented as a labor strategy, but it ignores who owns the machine, who controls distribution, and whether workers remain necessary after the machine becomes ubiquitous.

THE CORE FALLACY

The ATM was bounded task automation. It counted cash and processed deposits; it did not replicate the full economic function of the bank teller. It left relationship banking, sales, judgment, and local service comparatively scarce.

AI is aimed directly at those supposedly higher-value cognitive functions. ChatGPT and Claude can generate text, analyze information, answer questions, produce code, and perform service work at scale. The ATM removed tedium while preserving the human interface. Advanced AI can remove the interface itself.

The column mistakes a case of complementarity for a law of technological progress. Its job-growth story required elastic demand, new physical branches, regulatory expansion, and continuing need for human workers. AI does not need a new human-staffed branch for every deployment. One system can serve millions, improve continuously, and replace entire workflows rather than merely one repetitive task.

Under the Discontinuity Thesis, this is the decisive break. P1 makes cognitive labor economically inferior; P2 prevents institutions from preserving stable human-only domains; P3 removes productive participation for the majority. Lower prices and abundant services may preserve consumption, but they do not restore the wage-to-consumption circuit.

HIDDEN ASSUMPTIONS

• Every productivity gain will create enough additional demand to absorb displaced labor.
• Human relationship work will remain scarce instead of becoming another AI target.
• Workers will meaningfully own or control AI rather than merely use capital controlled by firms and sovereigns.
• Early access to AI creates a durable labor advantage, even after competitors receive comparable systems.
• New industries will require humans in quantities comparable to the labor AI displaces.
• Markets can expand indefinitely instead of reaching saturation once machine capacity becomes abundant.
• Productivity gains will flow into wages rather than concentrating in ownership, compute, data, platforms, and distribution.
• Legal and institutional adaptation will occur quickly enough to protect workers, while also assuming those reforms can reverse structural substitution.
• More consumption automatically means more productive participation.
• ATMs, Apple, Google, SpaceX, Walmart, and AI share the same labor dynamics despite radically different levels of automation, scalability, and capital intensity.

SOCIAL FUNCTION

Classification: partial truth, copium, ideological anesthetic, elite self-exoneration, and transition management.

The partial truth is real: some technologies expand markets, create complementary work, and raise the value of tasks they cannot perform. The anesthetic is treating that contingent pattern as destiny. The elite self-exoneration is the claim that workers merely need to acquire the machine before others do. That advice conceals the central issue: if the machine becomes universal, individual adoption cannot preserve the relative scarcity of human labor.

The column turns a distributional crisis into a personal-responsibility slogan. It discusses what technology can produce, but not who captures the surplus or what happens when production no longer requires mass employment.

THE VERDICT

This is an ATM-shaped argument against learning from the ATM’s limits. It correctly describes one successful episode of narrow automation, then uses it to launder optimism about general-purpose cognitive substitution.

The analogy fails at ownership, speed, scale, and scope. ATMs made banks cheaper while preserving a human service layer. AI can make the service layer cheaper, replicable, and disposable. The article is therefore not a rebuttal to the Discontinuity Thesis. It is a polished pre-collapse lullaby: accurate about the old machine economy, structurally blind to the machine economy that is replacing it.