The government estimates that 62% of the work done by administrative assistants, the most junior civil service role, is routine and therefore could be automated with AI.
Government predictions suggest public sector digitisation could lead to annual productivity savings of £45 billion. Significantly, £36 billion of these savings could be made by simplifying and automating public sector delivery, according to a government response to the Science, Innovation and Technology Select Committee.
The findings are based on research by the Department of Science, Innovation and Technology (DSIT), whose methodology is not currently public. Investigators at POLITICO used Freedom of Information Requests to access this estimate.
They found that executive officers, senior executive officers and higher executive officers in the civil service spend 48, 43 and 23% of their time on routine tasks respectively, indicating plenty of scope for AI deployment in the lower ranks. Senior civil servants are estimated to spend 0% of their time on routine tasks.
Rolling out AI on a larger scale should therefore increase civil service productivity, contributing to the government's plans in the spending review to cut civil service operating costs by 15% by the end of the decade.
While outsourcing tasks to AI could lead to some job losses, the modelling is unclear as to where these losses might be, due to the grouping of civil service grades by some public bodies. Additionally, whether AI is capable of fulfilling all routine tasks - and whether it could complete some non-routine tasks as well - requires more comprehensive testing.
Giles Wilkes, senior fellow at the think tank Institute for Government, emphasised that it could also create a new, different kind of work. "There’s often some kind of rebound effect, some kind of creation of new demand," according to Wilkes, which could lead to further levels of management.
DSIT has suggested it will release a more accessible version of its methodology and open source the model's code in the future "to support transparency and understanding of the approach used." Currently, no timeframe has been set for this advancement.