By Richard Stirling and Gonzalo Grau
Labour has won the general election. Sir Keir Starmer’s government is faced with an unenviable task: historically high tax rates, high borrowing, and public services straining at the seams. In her latest speech in the house of commons, Chancellor of the Exchequer Rachel Reeves confirmed that the state of UK public finances is much worse than initially expected. The only way out is to grow the economy. One of the most promising ways to grow the economy – with little room for manoeuvre, that is – is to harness AI.
Former Prime Minister Tony Blair seems to think so as well. During his speech at the Future of Britain conference on the 9th of July, he dubbed the fourth industrial revolution the only “game changer” in the new administration’s toolbox for rebooting a faltering economy. Recent, albeit damning criticisms of the Generative AI hype machine have taught us to be sceptical of these claims. We broadly agree.
AI presents a multi-billion pound opportunity for the UK over the next 10 years: £520 billion according to Amazon. Through time savings and utility gains, AI is poised to become a lubricant for economic re-modernisation, and can embed itself into solutions for almost every policy issue facing the UK to date. However, this can only happen when those deploying it do not jump the gun on how they want to use it – AI is best put to use in clearly-defined, basic problems.
Figure 1: Value gains from AI (source)
This opportunity, however large, has been slipping away in recent years. Though we try to avoid making year-on-year comparisons of our AI Readiness Index, some long-term developments can be pointed to as evidence. Since the first edition, the gap between itself and the United States widened by around 2 points, and it has been overtaken by Singapore. Should the new administration remain complacent around AI, it risks falling further behind: a group of six countries led by Finland and Canada are within two points distance. To fix the economy (and climb back up the podium), Labour needs to embrace the promise of AI. We’ve broken down what we think needs to happen – or not – for its efforts to be successful.
Smart regulation
For all the faults of the UK economy, its tech sector is doing quite well. Though the gap with the United States remains sizable, it ranked second highest in the technology pillar of the 2023 edition of our Government AI Readiness Index. In fact the UK’s tech sector was the first in the world – after its American and Chinese counterparts – to reach a valuation of $1 trillion. It is now worth over $1.1 trillion. This growth is unsurprisingly spearheaded by London, whose tech VC investment value in the first quarter of 2024 ($6.8 billion) is more than double that of second-placed Paris ($3.2 billion). Already, the UK makes up over a third of the combined value of private investment in AI with the entire EU (around 3.5 billion euros). It’s easy to point to the exponential development of AI applications as responsible for the recent growth of the global tech market and stop there. But when attempting to explain the UK’s breakaway from its European neighbours, we believe regulatory strategy is to blame.
Across the channel, the European Union’s landmark Artificial Intelligence Act and its mostly horizontal, risks-based regulatory approach has proven controversial from an innovation point of view. The AI Act’s classification of AI applications into 3 risk categories – unacceptable risk, high risk, and limited risk – is governed by vaguely defined essential requirements that, whilst allowing domain specialists the flexibility to determine detailed parameters and compliance strategies, can also be a source of confusion when interpretations about enforcement mandate differ. The act also contradicts existing EU legislation such as the GDPR, whose privacy principles are fundamentally opposed to the AI Act’s extensive record-keeping requirements. These contradictions, along with the additional cost of resource intensive conformity assessments and transparency requirements, can stifle innovation by creating a regulatory environment that is hostile to SMEs and small operations. Pending its entering into force this Thursday, the Act may backfire on the EU’s long term plan for AI sovereignty by favouring foreign multinationals over domestic innovators.
The answer is clear then: if innovation in the AI and tech sector is the key to rebooting economic growth, Labour should avoid regulatory frameworks that stifle it. This, ironically, means that Labour ought to pick up where the Tories left off. That is, not “rushing” to regulate AI but undertaking a sectoral, harms-based approach to the challenge. Leveraging existing regulation in use-case contexts promises to be less resource intensive to implement and comply with. Whether an employee is discriminated against by an algorithm or a traditionally racist hiring manager, the ultimate issue is whether or not that harm is prohibited and dealt with. Building on existing legislation by empowering enforcement bodies like the AI Safety Institute (AISI), which designs and deploys safety tests for frontier models developed by powerful multinationals, will add a degree of foresight and preventative temperament to what critics may call a ‘reactive’ regulatory model.
The promised establishment of a Regulatory Innovation Office, which will support regulators in updating legislation to accommodate the AI context, suggests that Labour is leaning towards the innovation-friendly approach. To be clear: we do not believe in the ubiquitous, often simplified discourse that pits innovation against regulation. Being strict in monitoring and testing foundation models is not mutually exclusive from adopting a semi-automated, minimalistic regulatory strategy for AI. From early reports on the nature of the upcoming AI Bill, it seems Labour will be taking a sensible direction.
Planning Reform – let the data centres come!
Another important tool through which to leverage AI is industrial policy. On this, we’ve been clued into what Labour’s plans are. Rachel Reeves’ speech on July 9th announced widespread structural change through planning reform and a National Wealth Fund (NWF). These changes look promising in terms of unblocking what has until now been a paralysing system, but how can these stimulate the development of Britain’s domestic AI capacities?
As things stand, it looks like Labour’s industrial strategy is focused on bold green investment – promises of lifting the ban on wind farms and centralising decision-making around large infrastructure projects seem to usher in a new era of quick, no-nonsense planning. Alongside housing and energy developments, however, the new government’s “war on NIMBYs” does well to also cover data centre construction. The number of data centres in the UK currently sits at 514, which is impressive in per capita terms – China has 449. Removing hurdles to data centre construction like those that previously halted large-scale projects in the green belt will help consolidate the UK as a regional leader in digital infrastructure. This poses a significant challenge in that unlocking new data centres will add pressure to the energy grid, which decarbonisation has already pushed to its limits. Luckily, finding innovative ways to boost energy capacity is in line with Labour’s commitment to green energy investment. It is nonetheless a difficult task.
A more active approach to stimulating investment should also figure. This can start with building on existing, successful schemes like the Enterprise Investment Scheme (EIS), which supports investors willing to take on the risk of investing in early-stage companies through tax breaks. The number of applicants and approvals has slightly dipped in the last year – perhaps amending some of the aspects of the scheme to increase incentives for investment in AI companies could help integrate it as an essential component of a wider AI development strategy.
What of the National Data Library?
Data is the input in AI development. Without high-quality, standardised datasets, high-quality models are not possible. Any government that is serious about strengthening its domestic AI research capacities should at least be thinking about data quality and standard-setting policies. Equally so, any government that is serious about creating a competitive domestic AI R&D sector should be thinking about how to make these quality datasets as widely available as possible. The UK is already home to large open data initiatives – data.gov.uk, a government data repository containing over 30,000 datasets, has been online for 14 years. Labour’s promised National Data Library could function as a centralised body through which to facilitate public-private data sharing for R&D purposes. Though the details around this body remain a little foggy, the potential for breaking down the remaining barriers to a fully dynamic open data ecosystem is exciting. As is usually the case, the best place to look when trying to understand what is possible with regards to research and development is across the pond.
The United States has been throwing money into initiatives that attempt to democratise AI R&D. Though its AI R&D capacities are world-leading, this is mostly carried by Big Tech, which tightly controls the vast majority of high-quality datasets. Most initiatives in this area have focused on widening access to federal government data and investment in data-sharing infrastructure. There are some useful lessons here. NAIRR, or the National AI Research Resource, is the US government’s flagship project in democratising AI research and development. Though it is still in its pilot phase, the programme is impressive in scale and ambition. The idea is for it to be a cyberinfrastructure collating, coordinating, and controlling access to all resources relevant to AI R&D: computational resources, data, testbeds, algorithms, services, software, networks, expertise, and user training and support, among others. Whilst the National Library seems to be envisioned as a coordinating body for public data, there is potential for it to be so much more. On that note, the US solution is not a one-size fits all approach – the UK’s data sharing ecosystem brings a unique set of characteristics and challenges.
Figure 2 – Schematic overview of the NAIRR vision (Parashar et. al., 2023)
The UK’s repositories and open data initiatives remain siloed and sectorally compartmentalised. Though a plurality of data sharing repositories and infrastructure is good for specialisation – Administrative Data Research (ADR) comes to mind as a specialised data sharing silo for academic research – a broader cultural change is needed to create the conditions for a fully dynamic open data ecosystem. There is currently widespread scepticism around large, centralised monoliths among the civil service – a powerful open data coordinating body like the National Data Library risks being fought tooth and nail under the guise of departmental sovereignty.
The challenge for Labour lies in effecting this cultural change through a strong central mandate, all the while maintaining the plurality of approaches to data-sharing that are already in place. Reducing the complexity of the current landscape is a good place to start. The National Data Library could tackle this by facilitating communication between data-sharing bodies, allowing for experience exchange to act as a catalyst for ‘soft’ standardisation.
To summarise…
Labour faces a tremendous opportunity. We could have gone into more detail in outlining our thoughts around the direction they should take, but we preferred to look at some key structural elements that we think can help propel their AI vision forward. Pending the details we are likely to see in the announced AI Opportunities Action Plan, they look up for the challenge.
