UK unveils ‘Humphrey’ assistant to civil servants with further AI plans to reduce bureaucracy

A week later the UK government announced a sweeping plan to make big investments in AI, it explains more about how this will be in the public sector. On the agenda: AI assistants to speed up human tasks; data sharing in closed departments; and new AI tools – named “Humphrey” after the character of the old UK TV political sitcom – to speed up the work of civil servants.

The plans will be revealed at a press conference on Tuesday led by the Department of Science, Innovation and Technology (DSIT), along with two other departments, Work and Pensions and Health / Social Care.

If you go to The UK government’s AI website to see how other projects are progressing, you will see that most of the efforts so far seem to be in the early stages, either in a few trials or an experimental phase; others are just starting out. For example, a plan to bring more AI services to the NHS that is customer-focused is “agreement” commitment to an idea.

Some also include links to the Github repository to view some of the work so far. It is unclear how many people are working on this project, and what third-party tools (such as LLMs) are being used. (We have asked these questions and will update them as we learn more.)

At their heart, all projects are about success. The government, DSIT said, currently spends £23 billion a year on technology, and the idea would be to reinvest that money into technology.

“Slow technology has disrupted our public services for a long time, costing us all time and money… Not to mention the headaches and stress we’re left with when we’re delayed or pressured to get things done.” form,” said Peter Kyle, Secretary of State for DSIT, in a statement. “My department will use AI… We will use technology to overcome the inefficient ways in which government agencies share information and work together to help the people they serve. “

The proposals include a new team within DSIT to lead the projects, as well DOGE in the US but it was designed and run by civil servants rather than technocrats.

DSIT is successful in three primary areas:

1. Duties of civil servants. Humphrey, his name wry, wise assistant played by the late Nigel Hawthorne in “Yes, Minister” and then “Yes, Prime Minister,” is a set of programs that aim to reduce the amount of daily work for civil servants, especially around the amount of material they have to read. and maintenance as part of their job.

“Consultation” is designed to read and summarize “thousands” of responses to the consultation in hours (answers, which can be long and many, are a large part of how the government takes comments from stakeholders). “Parlex” allows them to ask questions and read discussions in Parliament related to bills or other documents they are working on. “Minutes” is a secure writing service for taking notes from their many meetings. “Redbox” helps them prepare briefs and legal documents. And “Lex” allows them to search for legal requirements.

2. Another way to push efficiency will be around speeding up people-oriented projects. The idea here is to focus on the role of successors, which are many in the UK, such as the 100,000 calls that the tax authorities receive every day, or the need for people to appear in person to register a death, or (surprisingly). ) placed advertisements in local papers as one way of obtaining a driver’s license.

DSIT’s theory is that reengineering things like this with AI fuel systems could save $45 billion a year. (It is not clear if this estimate was before or after subtracting the cost of building and running AI services.)

3. The final phase will focus on collaboration between departments to facilitate data sharing to speed up service delivery and performance.

Taken together, the various projects are an indication that the government appears to mean business in its new AI push. But they also raise a number of questions.

For example, when it comes to data sharing, the DSIT currently states that the current operational concept will be “a common way of sharing information.” Central government departments, such as HMRC (tax and customs) and the Department of Trade and Industry, for example can share data with each other and local councils for fraud investigations, or to better understand how businesses are doing and what programs can help them. But what happens to people’s data protection when data is shared in ways you may not have intended?

Another possible question is around Humphrey: right now, DSIT said that some of the first programs are only in the experimental stages, but the big question will be, how far will the government go in believing in some AI principles?

There will also be human problems. As one former civil servant (who now works for an AI company) documentspast efforts to create programs that cut across multiple departments did not always work. Consistency, money and authority are ultimately the levers that can make or break any policy.