Beyond the hype: thoughts on digital, data, and AI and the first 100 days

By Richard Stirling and André Petheram

The 2024 UK general election already feels like a distant memory. PM Starmer and Labour  are in power and they are moving quickly. If there was perhaps not too much for digital-government-types to get excited about during the campaign, you’d expect the new government to reveal its hand soon.

Already, Labour have announced that at least some, and possibly all, of the functions of the Government Digital Service (GDS), the Central Digital and Data Office (CDDO), and the Incubator for AI will be moved into the Department for Science, Innovation and Technology (DSIT) from the Cabinet Office. The King’s Speech will contain a bill that regulates generative AI. Finally, during Labour’s first full week of government, the heavily-Labour-aligned Tony Blair Institute hosted the Future of Britain conference, with a concerted focus on how the government can transform to allow it to handle and exploit AI. It will be very intriguing to see whether the ideas discussed make their way into actual policy in the next few weeks and months.

What might new ministers and the civil servants tasked with implementing their priorities need to think about when they think about digital, data and AI, then?

There are a couple of critical priorities they’ll need to meet, which will be in occasional conflict with each other. Firstly, productivity and efficiency. The UK has little fiscal room for manoeuvre. Public sector productivity is insipid. Digital, data and AI projects will be judged on whether they can deliver clear outcomes from comparatively low investments. 

Secondly, Labour will quickly need to demonstrate tangible improvements in the experience of public services. Sir Keir Starmer’s promise of a ‘government of service’ that ‘treat[s] every single person in this country with respect’ will mean little to voters if their interactions with the state do not reverse their perceptions of decline. A reinvigorated commitment to user-centred design and the rapid delivery of transformative new services will need to be at the heart of this. 

Given this context, there are (at least) three more specific things that new ministers and civil servants might consider.

1. Review and prioritise digital projects on the basis of their contribution to productivity and quality.

There are very many digital projects across government with genuinely laudable aims and committed people behind them (we’re biased, but we’ve been lucky enough to have been part of a few of these teams). These efforts, though, need to be embedded in a consistent cycle of measurement, evaluation, learning and prioritisation. 

Across government, with DSIT’s support, departments should focus on, for example, making 100 small digital bets, finding the 10 that work and start scaling them up. They should demand that, at the beginning of projects, digital teams are absolutely clear about a) their expected Return on Investment and b) the improvements in service quality they want to drive. Departments then need to be ruthless about prioritising projects based on how they are delivering against these outcomes. (Of course, there will be tradeoffs between productivity and service quality; how departments balance these could define Labour’s term, given the priorities outlined above).

This approach should offer a means of breaking out of the loop of perpetual discovery. Departments, service owners, project sponsors, and so on, should demand that evaluation capabilities are consistently included in multidisciplinary digital delivery teams. It should then be those teams’ responsibility to clarify, measure and report on their desired outcomes.

2. Get a grip on AI in government – how should it be delivered?

The first thing to say here is, well, beware the hype. AI is not a silver bullet. It is expensive – possibly prohibitively so for fiscally-challenged governments like Britain’s. Generative AI, especially, needs a lot of energy – which could arguably be put to better use elsewhere or simply saved. Automated decision-making, when deployed by government, has created real suffering, with crucial questions of accountability and redress unresolved. 

But, AI systems remain powerful possible solutions to certain problems (pattern-finding with vast amounts of data; optimisation for efficiency in highly complicated systems, for example).

What can new ministers and civil servants do to navigate these complexities?

The first thing to consider is how people – especially frontline staff – will actually react to new AI systems. Has a given system been designed with their needs at the forefront? How will unions react (a critical but often-neglected point)? We tend to argue that governments should look for where AI can complement public servants’ professional judgement and support their job satisfaction. Rapid (and too-narrowly-measured) productivity benefits  risk being outweighed by making people feel like their jobs don’t matter, leading to higher staff turnover, ultimately hollowing out institutional knowledge and slowing down an organisation’s decision-making and information-processing abilities. 

It’s also useful to remember that AI’s widely varying use cases and capabilities are often bundled together under terms like ‘AI in government’ (we are not innocent here!). This elides the differing ways in which AI systems could be put to use as, for example, information-provision-machines (e.g. chatbots or automated brief-writers); decision-makers (as above, probably the most contentious in government – things like fraud-detection algorithms that may lack human oversight); or, as a kind of super-policymaker which ingests vast amount of data from across departmental domains and suggests interventions perfectly calibrated to created the desired outcomes (this remains both futuristic and deeply problematic). As we find ourselves asking over and over again, what is the problem you’re trying to solve? 

To get a grip on these kinds of issue, ministers/civil servants really need… 

• a) a strategy that clarifies the outcomes and impacts they want to see from AI systems, the principles/values that must underpin its use (especially transparency, explainability and accountability), and the skills needed in the department; 
• b) a roadmap for delivery that, for example, sets out how to scrutinise the available data; is realistic about costs; maps likely suppliers; and, crucially, sets out timeframes and stage gates at which the risks of further development are reviewed and decisions made on continuation. 

To further emphasise the point, it is better to make many smaller bets and to measure, review and iterate than to make large bets on highly-risky monoliths.

3. Remember that simpler is often better.

The answer to that fundamental question – what is the problem you’re trying to solve? – really might not mean pursuing things that, on the surface at least, seem especially ‘innovative’ as a way to delivering better public services. The really hard work is often that of simplification: taking the time to understand how users behave within complex environments; identifying where tactical solution after tactical solution has been layered on top of each other to solve short-term problems, creating unnecessary complication and duplication; designing easily-maintainable products that reduce the cognitive load on users and just give people a bit more time back in their days. 

Fortunately, this is pretty much the GDS playbook; whatever the outcome of shuffling units and departments around, the new Labour government will be in a position of strength if it empowers the people who defend these principles.