The British state is currently engaged in a massive, expensive hallucination.
Every week, Whitehall press releases trumpet the arrival of a new wave of "innovators and disruptors" hired to inject artificial intelligence into the machinery of public service. We are told these elite squads of tech mercenaries will slash waiting times, optimize local councils, and automate the bureaucratic grind. Discover more on a similar topic: this related article.
It is a comforting narrative. It is also entirely wrong.
The belief that you can fix broken public infrastructure by sprinkling machine learning models on top of it is the supreme tech delusion of our time. I have spent years auditing enterprise software deployments and watching public sector tech budgets vanish into the ether. The reality is brutal: the UK public sector does not have an AI deployment problem. It has a foundational data rot problem. Hiring highly paid Silicon Valley expats to build predictive models on top of a crumbling, paper-and-Excel infrastructure is like putting a spoiler on a horse-drawn cart. It looks fast, but you are still stuck in the mud. Additional journalism by CNET explores similar views on the subject.
The Myth of the Outsourced Savior
The core flaw in the government's strategy lies in the identity of the people they are hiring.
Whitehall routinely awards multi-million pound contracts to elite consulting firms and boutique AI startups staffed by brilliant PhDs. These teams arrive with beautiful slide decks and complex neural network architectures. They assume they are entering an environment where data is clean, accessible, and structured.
Then they meet the actual British civil service.
They find legacy databases from the 1990s that cannot talk to each other. They find critical patient infrastructure split across dozens of incompatible NHS trusts. They find local council data locked inside proprietary, monolithic software systems managed by predatory third-party vendors who charge extortionate fees just to export a CSV file.
The elite disruptor quickly realizes that 90% of their time cannot be spent training sophisticated models. Instead, they spend it doing basic, grueling data engineering—shoveling digital coal just to get the data into a format the AI can read. By the time the contract expires, the government is left with an incredibly expensive, brittle prototype that works perfectly in a test environment but breaks the moment it touches the chaotic reality of live public operations.
The Brutal Truth About "Efficiency"
Let’s dismantle a common question often asked in policy circles: How can AI help civil servants process claims faster?
The premise itself is flawed. The bottleneck in public service delivery is rarely the speed of data processing; it is the complexity of the underlying legislation and the sheer volume of edge cases.
Imagine a scenario where an AI system is deployed to automate housing benefit approvals. The algorithm can process standard applications in seconds. But public sector work is defined by the non-standard. The applicant who is self-employed, working irregular hours, caring for a disabled relative, and split-renting between two properties does not fit the clean training data.
When an algorithm encounters these real-world anomalies, it does one of two things: it either flags the file for manual human review, adding a layer of bureaucratic friction, or it confidently hallucinates an incorrect decision. When a private company makes an AI error, a customer gets the wrong shoes. When a government agency makes an AI error, a family ends up on the street.
The pursuit of algorithmic speed ignores the necessity of administrative justice. You cannot automate a process that your own staff do not fully understand because the underlying policy framework is a contradictory mess of decades-old compromises.
The Hidden Cost of the Sovereign Cloud Illusion
True technical authority requires acknowledging the systemic dependency nobody wants to talk about: cloud infrastructure.
The UK government talks loudly about digital sovereignty and building native AI capabilities. Yet, the entire infrastructure relies heavily on the computing power of three American corporations: Microsoft, Amazon, and Google. Every public sector AI initiative is, under the hood, a massive wealth transfer from British taxpayers to Silicon Valley hyperscalers.
The public sector is entering a trap of compounding technical debt. The more a department integrates its workflows with proprietary machine learning APIs hosted on foreign clouds, the more locked-in it becomes. The initial vendor contracts look reasonable, but the long-term data egress fees and compute costs are fiscal time bombs.
If you do not own the iron, you do not own the innovation.
Stop Buying Models and Start Fixing Pipelines
If the state actually wants to modernize, it needs to stop chasing the glamour of generative models and commit to the unsexy, blue-collar work of data standardization.
- Declare war on proprietary data silos: Pass legislation that forces every software vendor selling to the public sector to provide open, standardized, free APIs. If a vendor refuses to let the government access its own data without a fee, blacklist them from public procurement.
- Fire the consultants, hire data plumbers: Stop paying £3,000-a-day strategy consultants to talk about transformation. Spend that money hiring permanent, in-house data engineers whose sole job is to clean, centralize, and maintain the state's data pipelines.
- Embrace boring technology: You do not need a multi-billion-parameter large language model to optimize a bin collection route or flag missed tax payments. Simple, transparent, rules-based automation and basic statistical regressions are cheaper, entirely auditable, and run on a fraction of the compute power.
The obsession with hiring external tech saviors is a symptom of a managerial class that fears technical execution. They want the magic of AI without doing the hard work of building a functional digital state. Until Whitehall fixes its broken data foundations, every penny spent on high-profile AI disruptors is just expensive theater designed to make stagnation look like progress.
Stop buying the hype. Clean the data or cancel the contracts.
