FT: AI gains in public sector risk being cancelled out by public’s own AI

TL;DR:

• An FT analysis by Sarah O’Connor and John Burn-Murdoch argues that UK government AI productivity gains may be cancelled out by the public’s own use of AI when interacting with authorities — a “tug of war” rather than a one-sided efficiency win.
• The piece flags a UK government estimate of AI-related productivity savings being revised down by roughly half, from £45bn per year to £23.6bn, and draws on the Ada Lovelace Institute’s recent briefing on UK public-sector AI evidence gaps.
• Concrete examples include a UK planning tool called Extract that cut single-submission processing time from over an hour to three minutes, now being met by AI tools like Objector AI that generate planning-permission objection letters, lobbying videos and committee speeches at scale.

This is one of the more interesting reframings of UK public-sector AI productivity to date because it treats AI deployment as a two-sided market rather than a one-way efficiency lever. The FT writers are explicit that they remain optimists on AI in government — but the tug-of-war metaphor, with case-study evidence across planning, freedom of information and primary care, captures something the standard “AI will save £Xbn” narrative misses.

Three concrete tug-of-war cases

The planning system case is the cleanest. Government launched Extract, an AI tool that digitises paper planning documents and accelerates single-submission processing twentyfold. A British couple, frustrated with how hard it was to object to a development, built Objector AI — which generates objection letters, lobbying videos and committee speeches. A similar service has reportedly been used for thousands of UK planning objections. The government’s response this month was to roll out a custom Google-built AI tool to counter the AI-assisted objection flood. If AI-powered planning officers process submissions 20 times faster while AI-powered objectors submit 20 times as many objections, the planning system’s net throughput may be roughly flat.

The Freedom of Information example points the same direction. Submission volumes have grown enough that officials are considering tightening eligibility criteria to limit which requests are permissible. The FT framing is sharper than government typically allows: “AI not only making government less efficient, but potentially resulting in the public losing access to valuable information.” The GP-appointment example is the most consequential for individuals — AI-assisted booking systems are frustrating enough that some patients give up and go to A&E instead, “scarcely a less efficient use of the healthcare system”.

Looking forward

The Ada Lovelace Institute’s briefing this week argued UK policymakers need a “more robust and rigorous approach” to AI productivity claims, with service quality and equity measured alongside time-and-cost savings (see [our Ada Lovelace coverage]). The FT tug-of-war thesis is the complementary point: even rigorous measurement may understate the dynamic effects of AI being deployed simultaneously on both sides of the citizen-state relationship. For UK central and local government, the practical implication is that AI productivity business cases need a counter-AI demand assumption. For UK SMEs in govtech, the second-order opportunity is real — Objector AI is a startup, and so are the planning-counter-AI vendors.