The men building AI want us to fear it. We should – but not for the reasons they give

TEXT START: Amodei, Altman and Musk now back calls to slow AI down, perhaps for self-serving reasons.

The Dissection

The article tries to escape a false binary: AI executives may be manipulating public fear, and the technology may still be genuinely dangerous. It correctly identifies current harms—weaponized systems, disinformation, job displacement, and institutional incompetence—while arguing that existential-risk warnings deserve attention.

But its structure remains captive to the Terminator question. It treats mass unemployment as a nearer risk and superintelligence as the ultimate one, instead of recognizing that the destruction of the wage-employment-consumption circuit is already the central systemic event. The article notices the smoke but keeps searching for a monster in the basement.

The Core Fallacy

The core error is confusing the most spectacular possible failure with the decisive structural mechanism.

AI does not need consciousness, self-preservation, or a desire to exterminate humanity to destroy post-WWII capitalism. It only needs durable cost and performance superiority across cognitive work, followed by competitive deployment. Once institutions cannot preserve stable human-only economic domains, the majority lose access to economically necessary labor. That is sufficient for productive participation to collapse.

The article’s proposed answer—slowdown and regulation—also ignores coordination impossibility. Firms and states competing for productivity, military advantage, and capital cannot reliably maintain a global human labor sanctuary. Regulation may delay deployment or entrench incumbents, but it cannot repeal the competitive mathematics.

Its extinction argument is weaker still. “AI trained on human data may behave like humans” and “AI may compete with us for resources” are anthropomorphic speculation, not a demonstrated mechanism. The article’s strongest case is not rogue superintelligence. It is ordinary institutions deploying increasingly capable systems faster than society can absorb the economic damage.

Hidden Assumptions

• That the central danger begins when AI becomes autonomous or superintelligent.
• That human employment can be preserved through sufficient caution, regulation, or elite restraint.
• That governments can coordinate against the productivity and military advantages of faster AI deployment.
• That new jobs will meaningfully replace the labor made redundant, despite no identified mechanism for doing so.
• That preserving consumption through transfers would preserve productive participation and social legitimacy.
• That “AI safety” is a unified public interest rather than a battlefield over ownership, rents, and control.
• That the motives of AI executives are analytically central. Their incentives matter, but the competitive system can produce the same outcome without their sincerity or malice.
• That uncertainty about AGI timelines weakens the case for structural preparation. It does not. A slow transition can still be terminal if the direction is irreversible.

Social Function

Classification: partial truth, transition management, and elite self-exoneration—with a secondary ideological distraction.

The article performs useful damage control against both corporate hype and reflexive dismissal. It admits that job losses may arrive before any machine god and that human misuse is already lethal. That is the partial truth.

But by centering extinction speculation, it relocates responsibility from owners and institutions to a hypothetical future agent. The public is invited to fear an AI that might choose to kill them, while the immediate transfer of productive power to AI-capital owners remains underdeveloped. The result is transition rhetoric: regulate the speed, discuss safety, preserve legitimacy, and avoid confronting the ownership question directly.

The executives’ warnings can indeed be self-serving. Safety rules can become barriers that protect incumbents, while existential fear inflates valuations and attracts capital. Yet their incentives do not invalidate the danger. The same actors can be selling the fire alarm while owning the building.

The Verdict

This is a serious article with an incomplete autopsy. It correctly rejects the idea that self-serving motives make AI risk imaginary, and it correctly identifies labor displacement and institutional misuse as nearer threats. But it fails to follow those observations through P1, P2, and P3.

The terminal danger is not primarily that AI becomes a hateful god. It is that competitive deployment makes human labor economically unnecessary, institutions fail to preserve human participation, and consumption survives—if at all—as a ration distributed by owners of automated capital. The article calls for action without confronting that power structure. It is therefore a credible warning wrapped around a fundamental evasion: it sees the approaching collapse, but still describes it as a safety problem rather than the death of the economic order.