Civil servants say AI-drafted MSP questions swamp staff - Resultsense

TEXT START: Scottish Government civil servants say they are struggling under a rising volume of parliamentary questions and freedom of information (FOI) requests, some apparently written with AI.

The Dissection

The text documents an early-stage failure of institutional throughput. Generative AI has made producing parliamentary questions and FOI requests nearly costless, while government must still verify, route, research, redact, approve, and answer them through human-heavy procedures.

The result is an asymmetrical arms race: one side industrializes demand; the other absorbs the workload through scarce bureaucratic labor. AI is not yet replacing these civil servants. It is first exposing them as the bottleneck.

The article’s proposed endpoint—AI answering requests that AI helped write—is directionally correct but too narrow. The deeper outcome is an automated contest of generated demands and generated responses, with human officials retained mainly for accountability, exception handling, and liability. That is not restored productivity. It is bureaucratic carcass management.

The Core Fallacy

The text treats the crisis primarily as a problem of request volume, disclosure rules, and inadequate government tooling. Under the Discontinuity Thesis, the underlying problem is structural: institutions built for costly human cognition cannot remain stable when cognition becomes cheap and scalable.

Investment in AI may reduce response costs, but it also reduces the cost of generating more requests, counter-requests, challenges, and verification demands. Efficiency therefore feeds demand. The system does not return to equilibrium; it enters an automation arms race.

The implicit hope that better technology will make government “leaner” while preserving meaningful scrutiny is the central error. Once routine drafting, searching, summarizing, and checking are automated, the remaining human roles contract toward supervision and liability. The technology solves the workload problem by making much of the human work unnecessary—not by preserving broad productive participation.

Hidden Assumptions

• AI-generated requests will remain identifiable and regulable.
• Volume limits can be imposed without weakening legally protected scrutiny or becoming a political weapon.
• Automated answers will preserve accuracy, accountability, and public trust.
• Government systems can be integrated securely and cheaply enough to match the speed of external request generation.
• The current staffing shortage is temporary rather than evidence that routine administrative cognition is becoming structurally obsolete.
• More technology produces a leaner state without producing mass displacement inside the state.
• Human officials remain economically necessary once AI can search records, draft briefings, classify requests, and prepare responses.

These assumptions conceal the real transition. The state is not merely overloaded. Its information-access architecture is being attacked by zero-marginal-cost cognition, and its human staffing model cannot scale against it.

Social Function

This is a partial truth wrapped in transition management and ideological anesthetic.

It correctly records the mechanical pressure: AI makes demand cheap and overwhelms institutions whose response processes remain expensive. But it frames the event as an operational nuisance that can be handled through better rules and investment. That framing postpones the harder conclusion: administrative work is entering the same compression process as other cognitive labor, and the eventual solution will be filtering, rationing, automation, and headcount reduction.

The article also performs elite self-exoneration. It presents the burden as an unfortunate collision between new technology and old procedures, rather than asking who controls the systems, who bears the displacement, and whether public scrutiny can survive when both questions and answers become machine-produced.

The Verdict

This is not yet proof that civil servants have been made obsolete. It is proof that the institutional lag defenses are failing.

P1 is visible: AI can mass-produce cognitively plausible requests at negligible cost. P2 is emerging: government cannot preserve a stable human-only scrutiny regime once machine-generated demand scales beyond human administrative capacity. P3 is not complete, but the direction is clear: the civil servant’s role is being compressed from producer of analysis into verifier, gatekeeper, and liability shield.

The likely sequence is mechanical displacement first, social resistance later. Filters, disclosure rules, quotas, automated triage, and AI-written replies will slow the collapse. They will not reverse it. The article is watching the administrative state discover that its defenders can be buried without anyone needing to defeat them.