The UK government has leapt into the AI hype with a raft of "Exemplar" programs it claims will deliver billions in value – including a Clippy-style assistant to help citizens navigate complex forms and legal jargon, rather than simply making them clearer in the first place.
"We can entirely rethink and reshape how public services help people through crucial life moments using the power of emerging AI technology," technology secretary Peter Kyle MP claimed in a statement on the plan. "Using agentic AI to its full potential, we could provide a level of service to citizens across the country that was previously unimaginable – helping people to find better career opportunities, avoid wasting their time on government admin, and more.
"We are asking the world's brightest AI developers to work in collaboration with our own brilliant AI teams as we test how valuable their latest tech can be in helping people in their day-to-day lives. At each step, we'll only progress if the technology can be used in a safe and reliable way – but if it works, we could be the first country in the world to use AI agents at scale."
Kyle did not respond to The Reg's questions about which of the "world's brightest AI developers" had been invited to participate, nor whether legally binding guarantees and exemptions would be provided should a government-backed large language model do what large language models do best and "hallucinate" an answer-shaped object entirely at odds with reality – which, if applied to an official government form, could land the citizen in question in extremely hot legal water.
At the same time, the government is also planning to give doctors access to large language model technology to "draft discharge documents faster by extracting key details from medical records, such as diagnoses and test results" – a commonly cited potential application of LLM technology that, again, overlooks issues with context window length and the "hallucination" problem, which is likely to result in at least some patients' medical records being entirely incorrect.
Unlike the plan for "AI helpers" for the citizenry at large, this medical application isn't theoretical. Development is already underway at Chelsea and Westminster NHS Trust, with funding from the AI Exemplars program. Another project, dubbed "Justice Transcribe," will see an unspecified machine learning model used for live note-taking and transcription by probation officers – again, already in active use and due to be rolled out to the nation's 12,000-strong probation officer workforce "following the pilot phase outcome."
These and other AI Exemplar efforts will, Kyle claims, unlock "£45 billion in productivity gains." Other projects announced include an "AI Content Store" for schools, a tool dubbed "Extract" for more rapid extraction of data from old, handwritten planning documents and maps, and the previously announced "Humphrey" assistant for civil servants – named, with no sense of irony, after the self-serving, truth-twisting, Machiavellian lead in classic political satire Yes, Minister.
The Register asked the Department for Science, Innovation and Technology (DSIT) to clarify how it plans to deal with the "hallucination" problem inherent in LLM technology and whether or not users would be insulated from the legal impact of any errors made by such models, but it did not answer by the time of publication. It has, however, stated that the citizen-facing program, to be developed into a prototype over the next six to twelve months, will be "entirely optional to use." Should the prototype pass evaluation, it would then be rolled out across the nation "from the end of 2027."
Helping the government with all of this will be Jade Leung, who has been picked to split her work as the chief technology officer at the AI Security Institute with acting as the "AI Adviser" to the Prime Minister – and is, at the least, an actual human advising on the topic of AI, rather than the kind of "AI Adviser" the government is hoping to foist on the public.
A statement from DSIT on the appointment claimed that Leung, who graduated from the University of Oxford with a PhD in AI governance after a change of direction from a master's in environmental policy at the University of Cambridge, will "work to position the UK as the leading nation to help unlock the benefits and prepare for the impacts of transformative AI, working closely with the Prime Minister to harness the technology as it delivers the strong foundations and economic growth which are central to the government's Plan for Change."
Prior to the new role, Leung spent a little over two-and-a-half years at OpenAI, first as governance and policy adviser then governance lead. Prior to that, she was head of research and partnership at the Centre for Governance of Artificial Intelligence (GovAI) – founded in 2018 as part of Oxford's Future of Humanity Institute before setting out on its own in 2021.
Leung will remain at the AI Security Institute, the government's announcement confirms, "splitting her time" between her full-time role there and bending the Prime Minister's ear on the topic of AI. Leung had not responded to a request for comment by the time of publication.
One thing is clear, though. The government is very much on the AI bandwagon, and Leung is unlikely to slow things down. ®
Source: The register