UK’s AI masterplan vs reality


With the publication of its National Artificial Intelligence Strategy (NAIS), the UK joins the roll of nations jostling to assert ‘superpower’ status in the field of AI. Published in September, the government’s 10-year plan is founded on the contention that competitive pre-eminence in AI is a “top-level economic, security, health, and wellbeing priority [that is] vital to national ambitions on regional prosperity and for shared global challenges”.

Its view is backed up by a 2019 study by McKinsey that reckoned AI could deliver a 22 per cent boost to the UK’s GDP by 2030, but the country will have to show strong competitive prowess to win AI market share away from rival nations: at least 25 countries have already launched their own strategies.

These include Canada and UAE (2017), France, India, Japan, South Korea (2018), Germany and Israel (2019), Iceland and Singapore (2020). This year (2021) has already seen AI strategies from Ireland and the European Community. Long-standing AI frontrunners China and the US have had their strategies running since at least 2017.

The UK’s NAIS strives to balance nationalistic competitiveness with an openness to international co-operation, while emphasising business, economic, social and regulatory priorities. It is ambitious in its action-call delivery timetable, with some key initiatives promised to be in place before the end of 2021.

The term ‘AI superpower’ – which appears 15 times in the NAIS – makes clear that its authors see the UK taking a weighty leading role in addressing aspects of AI that must be sufficiently mature and robust to support major economic build-out at the scale it envisions. These include the development of industry-wide standards for the technology of AI, and bringing about more openness by AI vendors with regard to the inner workings of their core intellectual property (IP).

There remain other key AI considerations that the NAIS addresses tangentially, or hardly at all. They include acquisition of UK AI innovators by non-UK interests, and the necessity for better industry-standard definitions of what AI actually is – and is not.

Also, for a seamless and inclusive ‘AI economy’ to function, a way of validating AI’s value proposition would seem integral to consumer rights. In other words, how can organisations that buy AI solutions be sure they get value for money – and that what they’re buying is authentic AI?

While the NAIS acknowledges the importance of AI start-ups to the realisation of the UK’s global leadership position, it makes fewer direct references to the more established commercial AI sector, which seems to now form part of the UK’s ‘AI ecosystem’ (the NAIS refers to the ‘AI ecosystem’ 33 times throughout its 35 pages, without explanation or definition of what is meant by the term).

To date, attempts by the government to build bridges with the AI community have been mixed. Founded in 2018, government group the AI Council is a non-statutory expert committee of independent members set up to provide advice to the government and high-level leadership of the ‘AI ecosystem’. Currently, its 21 councillors comprise eight members from academic backgrounds, three appointed government advisors, three vendors (BT, Microsoft, DeepMind), three consultants, a charity (Luminate), a financial services firm (MasterCard) and an online grocery business (Ocado).

The NAIS sets out its objectives on three ‘pillars’: these include ensuring that the UK benefits in the long-term growth of AI, with focus on investment, inclusivity and governance. Six key initiatives are proposed in support of these aims.

NAIS key initiatives

• An AI Standards Hub pilot scheme to coordinate “UK engagement in AI standardisation globally”. The NAIS highlights the UK’s involvement in AI technical standards internationally (see box below: ‘UK helps lead way on AI technical standards’). An area of underrated importance, while the term ‘standard’ can be used freely in the context of AI, sometimes to denote a functionality benchmark rather than a common technical specification that AI solutions are built to and comply with.

• The development of an “AI technical standards engagement toolkit” to “guide multidisciplinary UK stakeholders to engage in the global AI standardisation landscape”; the NAIS gives no further explanation of what this will be.

• A government consultation on the options and value for a “National Cyber-Physical Infrastructure Framework”, to help identify how common interoperable digital tools, platforms and testbeds and/or cyber-physical or living labs could collaborate to form a digital and physical ‘commons’ for innovators. A Cyber-Physical System describes a computer system in which a mechanism is controlled or monitored by algorithms. “Infrastructure Framework” does not make clear whether the ultimate intention is to build an actual state-run computing infrastructure or something that specs-out a theoretical reference framework.

• The Central Digital and Data Office (part of the Cabinet Office, responsible for the government digital, data and technology function) is conducting research with a view to developing a cross-government standard for algorithmic transparency to help build confidence and trust in how UK citizens’ data is being processed and analysed – a standard which presumably would cover concerns over algorithmic bias. This could prove to be a contentious initiative if its outcomes mandate that suppliers of AI solutions to UK government projects must reveal more about how their products actually operate.

• Establishment of National AI Research and Innovation Programme (a sort of co-ordinating agency that would have oversight of public engagement and funding).

• Working with the “thriving AI ecosystem” the Office for AI (a joint BEIS-DCMS unit responsible for overseeing implementation of the NAIS) will develop the UK’s national position on governing and regulating AI, to be set out in a White Paper in early 2022.

AI standards

UK helps lead way on AI technical standards

The National AI Strategy points out that the UK’s global approach to AI standardisation is exemplified by its involvement in the International Organisation for Standardisation and International Electrotechnical Commission (ISO/IEC) on four AI projects, as well as the UK’s initiation of and engagement in the Industry Specification Group on Securing AI at the European Telecommunications Standards Institute (ETSI).

At ISO/IEC, the UK, through the BSI, is engaged in the development of AI international standards in concepts and terminology, data, bias, governance implications, and data life cycles. At ETSI it has published, along with other submissions, ETSI GR SAI 002 on Data Supply Chain Security, which was led by the National Cyber Security Centre.

The ISO/IEC work programme includes the development of an AI Management System Standard (MSS), which intends to help solve some of the implementation challenges of AI. This standard will be known as ISO/IEC 42001 and will help organisations to develop or use AI responsibly in pursuing their objectives.

The NAIS also interconnects with government’s National Data Strategy (2020), which sets out its vision to tackle barriers to data availability and use across the UK economy, which the government calls “a vital enabler” for successful AI. In relation to the NAIS’s emphasis on the vital importance of an AI to the UK’s future economic wellbeing, this looks toward turning “unlocking” the value of data into a matter of national criticality.

The NAIS outlines several important issues for the development and deployment of AI in the UK in the decade to come, but there remain three aspects of strategic planning for AI that have not been brought within its scope. These could be summed up as validation of AI’s value proposition, the danger of AI start-ups becoming AI ‘bought-ups’ and the need for tighter definitions.

On the first point, the UK NAIS maintains that AI represents “tremendous potential” to drive “substantial” economic growth and prosperity for the UK. DCMS Minister Chris Philp, for one, is sure that “AI technologies generate billions for the economy”.

Assertions about the value AI adoption will release into future economies and societies often refer to AI as a key that will “unlock the power of data” or drive growth and efficiency. Certainly, the NAIS is founded and predicated on the view that whether by introducing more efficient systems and working practices, or by “unlocking” the power of data across the economy, AI perforce would release value that will drive economic growth – the UK mainstream economy itself will indeed evolve to become “AI enabled”.

In the global context, PwC has suggested that AI could contribute up to $15.7tn (£11.4tn) to the global economy by 2030. Of this, $6.6tn (£4.8tn) would probably come from increased productivity, and $9.1tn (£6.6tn) is likely to come from “consumption-side effects”. However, some commentators have suggested that such expectations should be further validated before large-scale investments are made into new AI initiatives.

According to Wim Naudé, professor of economics at University College Cork, “AI is not making advanced economies more productive. Even […] firms [that] are among the top investors in AI and whose business models depend on it – such as Google, Facebook and Amazon – have not become more productive.”

This contradicts claims that AI will inevitably enhance productivity, says Naudé. “Related to this is the fact that most AI applications just are not that innovative,” he has said. “Firms do not adopt it, not because they do not trust it, but because it makes little business sense.”

‘If AI and automation were a force to reckon with, we would have seen skyrocketing labour-productivity growth and rising unemployment.’

Wim Naudé, University College, Cork

In his foreword to the NAIS, Business Secretary Kwasi Kwarteng says that UK start-ups are “already leading the world in building the tools for the new economy”. His point is echoed later in the strategy: the government wants AI start-ups to have “access to the people, knowledge, and infrastructure you need to get your business ahead of the transformational change AI will bring, making the UK a globally competitive, AI-first economy”.

However, the prickly issue of acquisition and venture capital financing presents the big challenge to this ideal. If ownership of a start-up remains in the UK, it consolidates the median value of its national IP stock. The snag is, in a free market, UK start-ups have become something of an internationally traded commodity, and recent history is of some of the best of them being snapped up by foreign interests before they have reached the point where they can be financially self-sustaining.

The UK’s foremost AI exemplar, DeepMind, is lauded seven times in the NAIS, and while it’s rightly referred to as UK-based, the company was sold to Google for $400m (£290m) in 2014. In 2015, Apple bought UK AI start-up VocalIQ for between $50-$100m (£36-£73m). The next year, Twitter bought Imperial College AI spinout start-up Magic Pony for $150m (£109m); and Microsoft bought London-based AI start-up SwiftKey for $250m (£182m). Facebook bought Bloomsbury AI for $23m-$30m (£17m-£22m) in 2018 and Atlas ML for around $40m (£29m) in 2019.

These are headline lump-sum acquisitions; there are many UK start-ups now funded by venture capital from around the world. This is going to be a dilemma for the government and the NAIS, because VCs are focused on return on investment, not nationalistic continuity.

‘‘‘Artificial Intelligence’ as a term can mean a lot of things,” the NAIS acknowledges, “and the government recognises that no single definition is going to be suitable for every scenario. In general, the following definition [of AI] is sufficient for our purposes: ‘Machines that perform tasks normally requiring human intelligence, especially when the machines learn from data how to do those tasks’.”

This definition seems like something of a composite of various other definitions (the government has set out a different legal definition of AI in the National Security and Investment Act). If AI is going to move from specialised communities and enter the mainstream, as the government would like to see happen, jargon and terminology will play a significant part in its acceptance by non-technological people. For instance, words like ‘machines’ in the context of AI are liable to evoke mental images of dodgy robots to the lay person.

The issue of a standard definition is important if, as the government maintains, AI is to become mainstream over the next 10 years. A situation where multiple AI definitions are bandied about will hinder wider acceptance as well as potentially create legal tangles when it comes to the buying and selling of AI-based products and services.

The question of definitions would also clarify any ambiguity over the extent to which a software product or service sold as being ‘AI-powered’ or ‘AI-enabled’ really does deploy standard-strength AI. Vendors of AI solutions can be reticent about disclosing what actually makes their products ‘work’. Too often the terms ‘AI’ and ‘Machine Learning’ are used interchangeably. A UK enterprise about to buy into an AI solution needs some kind of independent assurance that the Trades Description Act (1968) is not being contravened.
The NAIS, meanwhile, does succeed in highlighting how the future of AI is likely to be directed as much toward its potential for driving revenue generation as understanding of the secrets of the human brain.

With the London Stock Exchange contributing tens of billions of pounds to the UK GDP, the UK’s strategy could perhaps have gained from referencing the fact that AI already exerts a major controlling interest on the UK economy: and that is through the trading platform algorithms that detect market trends and execute trades fastest. It will continue to shape stock markets through to 2030 – and probably well beyond.

Policy

National and regional AI strategies

To guide and foster the development of AI, countries and regions around the world are establishing strategies and initiatives to coordinate governmental and intergovernmental efforts. Since Canada published the world’s first national AI strategy in 2017, more than 30 other countries and regions have published similar documents as of December 2020.


UK AI Strategy inline

Image credit: .


UK AI Strategy infographic

Image credit: E&T