AI Could Make Governments Slower by Sami Mahroum - Project Syndicate

TEXT START: Governments are increasingly turning to AI tools to process vast amounts of information and make public services more efficient.

The Dissection

This is a bottleneck memo disguised as an efficiency warning. It correctly identifies the immediate choke point: AI can produce filings, arguments, and administrative material faster than humans can verify, adjudicate, or authorize them. But it treats that friction as a governance-management problem rather than evidence that human oversight is becoming the scarce and obsolete layer.

The article is really asking how institutions can keep human decision-makers attached to an accelerating machine pipeline. It does not examine whether those decision-makers remain economically necessary, or whether AI will force triage, automated review, restricted access, and machine-mediated adjudication.

The Core Fallacy

The core error is confusing institutional slowness with a counterforce to AI displacement.

AI does not need governments to become faster. It only needs to make cognitive production cheaper and more abundant than human institutions can process. The resulting overload does not restore the wage-to-consumption circuit or preserve mass productive participation. It exposes the human bottleneck.

Under the Discontinuity Thesis, overloaded courts are evidence for P2: human institutions cannot preserve human-only oversight at scale. The likely responses are automation, filtering, rationing, delegation, and standardized machine decisions. Those responses further reduce the number of humans required, even if the institution itself becomes slower, harsher, or less legitimate.

The article identifies a real lag effect, then mistakes the lag for system survival.

Hidden Assumptions

• Human review will remain mandatory rather than being automated, narrowed, or bypassed.
• Every AI-generated submission must enter a human queue instead of being filtered by other systems.
• Institutions can absorb the volume through more staff, funding, and procedural reform.
• More filings represent useful demand rather than cheap procedural noise.
• Legal and administrative legitimacy will continue to require human judgment at its current scale.
• A slower state means weaker AI rather than a state increasingly unable to govern its own machine-generated workload.
• Bottlenecks will be shared socially rather than captured by owners of AI infrastructure, who will control access, verification, and prioritization.

Social Function

Classification: partial truth, transition management, and ideological anesthetic.

Its partial truth is substantial: finite human attention can turn machine efficiency into institutional overload. Its transition-management function is to prepare governments for procedural redesign. Its anesthetic function is to frame the crisis as a solvable administrative mismatch instead of confronting the larger consequence: AI is eroding the necessity of human cognitive labor and making existing institutions structurally unfit for the volume they generate.

The Verdict

The article is accurate at the surface and evasive at the structural level. AI may make governments slower because it floods human institutions with machine-produced output. That is not a rebuttal to the Discontinuity Thesis; it is one of its symptoms. The machine accelerates, the human gatekeeper jams, and the institution responds by eliminating or subordinating the gatekeeper. Governmental slowness is lag-hospice behavior, not economic survival.