UK Government rolls out AI training for 400,000 civil servants this autumn
All civil servants in England and Wales will get hands-on AI training starting this autumn. The announcement comes as part of a big civil service shakeup led by Pat McFadden, chancellor of the duchy of Lancaster, aiming to boost productivity amid a shrinking workforce.
Tens of thousands of jobs are being cut through voluntary redundancies and hiring freezes. Officials are expected to use AI tools to speed up repetitive tasks wherever possible. 
The government is already piloting an AI tool called Humphrey—named after the 1980s TV character Sir Humphrey Appleby from Yes, Minister. Early results are promising: Scotland used it to analyze responses on cosmetic procedure regulations and got human-level results much faster.
If expanded, Humphrey could save up to 75,000 days of staff time per year across 500 annual consultations, worth £20m in staffing costs.
The Department for Work and Pensions is also using AI to process high volumes of correspondence faster, flagging vulnerable benefits or pension recipients more quickly.
Cabinet secretary Sir Chris Wormald told staff he’s proud of the civil service’s “ability to continuously adapt” and urged teams to evolve with “an uncertain world” ahead.
The prime minister has set us an important task in building a productive and agile state, which will involve us preserving and championing everything that is great about the civil service while changing to meet the challenges of an uncertain world.
We need to seize the opportunities of artificial intelligence and other technological developments to continually modernise everything that we do.
Training will be delivered through the government’s One Big Thing project, focusing on key skills each year.
But worries remain about AI bias and errors. A December Guardian investigation revealed a UK government AI fraud detection system biased by age, disability, marital status, and nationality.
Related: Revealed: bias found in AI system used to detect UK benefits fraud
