How the dream of AI boosting Whitehall inefficiency is sure to unravel

Digital tech has failed to do much to make public administration better – it just generates waffle

Remember how Big Data was going to revolutionise the state? Or how Agile and web thinking would transform the civil service? How about some magic apps?

On each occasion, ministers had hoped that new technology would solve the seemingly intractable problems of bureaucratic inefficiency and dysfunction. Now it’s Oliver Dowden’s turn to rub the bottle and pray that a genie comes out.

Artificial Intelligence is “the closest thing you have to a silver bullet in terms of driving efficiency for the taxpayer”, the deputy prime minister told us last November. Last week, he urged Whitehall to rub the bottle harder, as the Cabinet Office unveiled plans to spend £110m on AI tools and technical staff. Even more AI investment is expected in the Budget.

But it’s quite reasonable to be pessimistic about the outcome – largely because whatever the buzzword of the day, technology has failed to do much to make public administration leaner and meaner.

The utopian belief that technology gives us improvements in national economies, in organisations, or by making individuals happier and more productive at work, is falling apart at every level. Let’s start at the top.

The “productivity paradox” describes how we have more technology than ever but see less gain. As Mckinsey consultants wrote in 2018, the productivity boom of the 1990s from the introduction of the PC and databases was running out of puff by 2005. The financial crash caused a further fall. The benefits of digitisation, Mckinsey intoned, “have not materialised at scale”.

For over a decade, businesses have been urged to undertake “digital transformations” and are now exhausted by it all. Technology seems to generate its own impediments for staff. Instead of being freed up to work, they’re swatting away swarms of emails and Slack messages. Research for HP found that these technology related interruptions reduced workers’ IQS by an average of 10 points.

By comparison, a separate study found that smoking marijuana led to only a four-point reduction in IQ. No one would suggest getting employees stoned to make them more productive, but managers and ministers keep believing in the digital drug.

Alongside these trends, the British state has waned and then waxed. The 2010 coalition government cut the size of the civil service by 15pc, and by 2016 it was smaller than it had been since World War II, with 384,000 employees. But since then it’s followed a U-shaped curve, and in 2023 was larger than it was in 2010. That’s partly because it was given more to do, with resources thrown at leaving the EU, and then the pandemic. Core competencies were repatriated from Brussels, while Covid created new administrative schemes.

Alex Thomas, who spent 17 years in Whitehall, and is now programme director at the Institute of Government explains that most of the extra staff have been taken on for policy and digital work. Consultants have also returned in large numbers. Just as Silicon Valley is laying off hundreds of thousands of tech workers, the Government is boasting it’s hired more than ever. “The Government digital and data profession has grown from 4pc to 5.4pc of total Civil Service headcount, close to our target of 6pc”, said Alex Burghart, parliamentary secretary to the Cabinet Office, in a statement on Friday.

The Government is encouraging staff to adopt Microsoft’s Co-pilots, or “Co-parrots” as civil servants are calling them. But listening to Dowden, he appears to be describing a new technology that hasn’t been invented yet. It’s certainly more capable and more reliable than anything on offer today. But we can’t trust them, as the Civil Service’s own advice cautions: “Output from generative AI is susceptible to bias and misinformation, they need to be checked and cited appropriately,” the official guidance published in January explains.

A Co-parrot burping up incorrect answers to Parliamentary Questions, or inaccurately condensing the Ministerial Red Box is a disaster waiting to happen. “This sounds like a JR [Judicial Review] nightmare,” observed Politico’s Joe Bambridge, a former Treasury official. “If you can’t rely on it, I don’t see where the big productivity gains come from?” Quite.

But the biggest problem with throwing AI into an organisation is much more subtle. It simply generates lots of what we don’t need: waffle.

AI becomes a prop that allows the idle to look busy, which in turn, encourages the most mediocre staff and demoralises the most able. I call this the Fishburne Effect, after the cartoonist who first identified it.

In Fishburne’s cartoon, one employee boasts that AI created a three-page email for him he didn’t need to send, while another boasts that AI read a three-page email she didn’t need to receive. So it has turned out, as studies show that CHATGPT is good news for people who perform poorly in their jobs. The use of AI embeds inefficient processes and lacklustre workers. If this doesn’t change, don’t expect any productivity gains in the office – or Whitehall.