It’s important civil servants “know what ‘good’ looks like,” said Robyn Scott, founder of Apolitical, a learning platform for public officials. “Because there are a lot of great companies out there … and there are a lot of snake oil salesmen.”
Greasing the wheels of government
Right now, AI isn’t widely used across government.
The National Audit Office found last fall that 37 percent of government bodies that responded to a survey said they’d deployed AI, typically in only one or two use cases each.
The most controversial uses have drawn the most attention to date: for example, in predictive policing, the analysis of live facial recognition data and the use of algorithms to identify fraudulent benefit claims. Scant information about these use cases has been made available to the public.
But the generative AI boom of the past 18 months has drawn renewed attention to how the tech could be wielded to speed up monotonous tasks and grease the creaking wheels of Whitehall.
The previous Tory government had begun to expedite the roll-out of the tech. Former Deputy Prime Minister Oliver Dowden launched an AI “hit squad” in November aimed at cutting jobs in the civil service.
The so-called Incubator for AI set up pilot projects using generative AI to analyze responses to government consultations, power a chatbot that would interact with citizens and collate documents for ministers, replacing the traditional “red box.”
Recent reports from the U.K. government’s Central Digital and Data Office and the Ada Lovelace Institute agree with the TBI that AI holds promise, but the range of predicted productivity gains varies wildly.
An analysis of the costs of overhauling legacy digital infrastructure and implementing the new tech across government rarely features in the most optimistic forecasts. “Technology is not free,” pointed out tech consultant Rachel Coldicutt in response to Reeves’ plan to use AI to save costs.
“If something costs £500 million this year, how long will it be expected to take to recoup those costs, and could that money be better spent elsewhere?”
“We’re really at a preliminary stage of experimentation,” said Imogen Parker of Ada Lovelace. “We need to learn much more quickly about what works and what doesn’t.”
This means looking at “the messy reality” of how this tech interacts with government processes, rather than “hypothetical, perfect, idealized use cases,” she adds.
A report from think tank IPPR found that about 10 percent of tasks could be automated across jobs in the private and public sector. “There are very few jobs that you could just … plug in AI and you don’t need humans anymore,” said IPPR senior economist Carsten Jung.
Guidance issued to civil servants last year noted that generative AI tools “can, and do, make errors,” meaning officials would need to verify the outputs. It noted that the tools could be biased and pose data privacy risks too.
Another concern is how it might change the nature of the relationship between government and citizens. “When you insert a technology into a system or service, it has a ripple effect … It changes what people expect, and it changes how people behave,” said Parker.
She pointed to the example of MPs using AI to write emails to constituents or the government automating the analysis of consultation responses. “It will unquestionably change the way that residents feel about their interactions with that system … It might change people’s willingness to engage in consultations to start with.”
