TL;DR

Academics at Manchester and Edinburgh have published a verdict on nearly ten years of British AI policy, and the phrase they chose is “hyperactive incrementalism” — a lot of initiatives, short horizons, little evaluation. Their argument is not that the ambition is wrong but that the machinery beneath it is missing, and that pouring AI onto that foundation will widen regional and public-sector gaps rather than close them.

The numbers behind the warning

Nathan Critch, at Manchester’s Productivity Institute, and Darcy Luke, at Edinburgh’s BRAID programme, assembled the evidence that usually gets asserted rather than counted. As of 2022, just 15% of British firms had taken up even one AI technology, mostly the larger ones, and mostly in IT, telecoms and legal services; hospitality, health and retail trailed badly. Both the companies selling AI and the companies using it cluster in the capital and the South East.

Inside government the constraint is data. In a National Audit Office survey, 62% of the public bodies that replied named data quality and availability as a serious obstacle — records stranded on legacy systems, council data siloed across platforms that cannot talk to each other. Skills shortages compound it, worst in local government, where councils cannot match private-sector salaries for scarce expertise. The paper notes the UK trains well at degree level and thinly everywhere else: in-work training, vocational and further education all under-provided relative to higher education.

The energy arithmetic

The AI Opportunities Action Plan promises a twentyfold increase in domestic compute by 2030. The researchers cost that out: AI infrastructure consuming around 3.6 terawatt-hours in 2020 would rise to roughly 72TWh — over a quarter of total UK electricity use. Britain holds roughly 3% of the world’s compute, and much of that was never tuned for AI workloads, leaving heavy dependence on hyperscaler cloud. Set against net-zero commitments, grid constraints and the water needed for cooling, the plan collides with itself.

Looking forward

This lands the same week hundreds protested outside Holyrood for a moratorium on Scottish data centres, and days after the Turing Institute warned the UK’s sovereignty window is closing. The researchers’ conclusion is bleaker than either: without policies that spread adoption beyond the already-productive, AI becomes “another problem for an already problem-laden government, rather than the solution it is hoped to be.” The paper appears in The Political Quarterly.