Government must use AI to reduce friction – not replace accountability, panel warns

Public sector leaders are being urged to focus less on the novelty of AI and more on whether it genuinely improves how citizens interact with government, following a wide-ranging panel discussion at this week’s Think Digital Government.

Speakers from government, regulators and industry explored where AI is already delivering value in government. The discussion also highlighted growing recognition that government must invest in data quality, governance and public trust if AI adoption is to succeed.

Speakers warned that public confidence in government AI systems will ultimately determine whether adoption succeeds. Alex Jones, interim director at the Incubator for AI (i.AI), argued that discussions about AI should begin by understanding the role government itself plays in society.

“Government isn’t only a service provider,” he said. “Government isn’t a business; government isn’t a startup.”

Instead, he argued, government exists because some responsibilities – such as justice, education and public safety – are “too important to leave to the private sector and to the market alone”.

Jones suggested that public trust in AI depends on two things: citizens believing government is competent enough to use the technology responsibly, and believing its incentives remain aligned with the public interest.

“The public need to see that the government’s incentives are aligned with theirs when they’re using AI,” he said.

Chris Gullick, chief data and AI officer at Ofgem, said his organisation approaches AI through the lens of its regulatory mission rather than technology experimentation.

“We start with the mission of the regulator. Protecting energy consumers, energy security, and net zero,” he said.

For Gullick, the key question is not simply whether AI can make processes cheaper or faster, but whether it improves outcomes for society.

“What are we going to use AI for to benefit society rather than just make things as cheap and as quick as possible?” he added.

Government still wrestling with experimentation vs. caution

The panel also explored the tension between encouraging innovation and maintaining sufficient safeguards around AI use. Matthew Hills, government and public sector lead for UK and Ireland at ElevenLabs (pictured), said trust is particularly critical for technologies that interact directly with citizens.

“If the trust erodes, then it erodes with the buying entity who purchased the solution,” he said.

He argued that organisations often underestimate the difference between experimenting with AI tools internally and deploying them safely into production environments.

“When it comes to actual production, that’s where the issue does arise,” said Hills.

Meanwhile, Gullick described the challenge of balancing enthusiasm from staff eager to use AI tools with concerns from others who remain sceptical or anxious about the technology.

At Ofgem, he said the organisation has focused on helping staff become “confident experimenters” through controlled opportunities to test AI tools safely.

One initiative involved bringing together technical and non-technical staff to build simple AI agents using Copilot Studio.

“That kind of activity gets something useful and gets people engaged and trying and getting excited about how it can help their day job,” said Gullick.

AI should reduce friction – not replace human judgement

Speakers stressed that AI should augment public servants rather than replace human accountability and decision-making. Jones pointed to existing government AI projects focused on reducing administrative burdens, including tools helping social workers cut time spent taking notes.

The panel also discussed where AI may not be appropriate at all. Gullick said recruitment was one area where he remained cautious because of the risks of bias and poor decision-making.

“Recruitment’s an area where I’m generally dead against the use of AI,” he said.

However, he noted that AI could still play useful supporting roles, such as helping unsuccessful candidates receive better feedback on CVs and applications.

Jones also suggested there are many “human first” processes where government should remain cautious about relying too heavily on AI, particularly where systems remain inaccurate or where decisions carry significant personal consequences.

Throughout the discussion, a consistent message was that government should not pursue AI because it is fashionable or politically attractive, but because it genuinely improves services and outcomes for citizens.