In a 2018 article, celebrated historian and futurist Yuval Noah Harari suggested that the democracies of the future might find themselves at a fatal disadvantage. During the 21st century, artificial intelligence is widely expected to lead to game-changing technological breakthroughs in almost every sector, from healthcare to transport to digital government. But, Harari argues, there’s a catch: while democracies have historically outperformed dictatorships because of the impossibility of efficiently processing all the information needed to govern a complex society through a centralised system, AI technology is likely to reverse this advantage.
This is because the rapid progress in AI this century – specifically, in machine learning, the use of computer programmes which can ‘learn’ from given inputs and rewrite their own algorithms accordingly – has been driven by the automated analysis of huge amounts of data. ‘If you disregard all privacy concerns and concentrate all the information relating to a billion people in one database,’ Harari suggests, ‘you’ll wind up with much better algorithms than if you respect individual privacy’. Centralised totalitarian regimes which keep their citizens under intense surveillance may leave democracies, which are held back by their concern for privacy, in the dust.
In this context, the new AI and data strategies which the European Commission released in February may appear to be swimming against the tide of history. Since the EU passed its General Data Protection Regulation (GDPR) in 2016, strengthening individual privacy rights and banning commonplace commercial practices such as selling people’s live locations without their consent, conventional wisdom has been that Europe is ‘losing the AI race’ due to its regulation of the tech sector. Humerick argues that the GDPR rights to refuse or withdraw consent for the use of one’s personal data simply ‘cannot coexist’ with AI. In his view, the EU has already lost the race: all the real innovation is happening in the US and China, and Europe’s strongest centre for AI research was the UK – which, he suggests, would be well advised to jettison the EU’s cumbersome privacy regulations after Brexit.
Yet the new policies make it clear that the EU is committed to its current approach to data privacy. In fact, the AI strategy proposes even more regulation, including a requirement for applications in high-risk areas to pass ‘conformity assessments’ before entering the market. So should we simply accept that the US and China – with their vast investments in AI research, indiscriminate collection of personal data, and concentration of information in the hands of a few tech giants – are the future, while Europe will become a technological backwater? Perhaps we should not be too pessimistic. After all, what it means to ‘win’ at AI all depends on what you are playing for.
The Commission recognises that the EU is relatively weak when it comes to consumer platforms, and that given its strict regulation of personal data, this is unlikely to change any time soon. Its answer is to focus instead on the potential for innovation based on industrial, non-personal and public data. It sets out several policy priorities which aim to promote AI development in industry and the public sector: these include a legislative framework for cross-sectoral data sharing platforms, economic incentives for companies to share their data, and increased funding for digital infrastructure. ‘Data spaces’ with common datasets are planned for nine priority sectors, including green technology, healthcare, manufacturing and transport.
It is certainly true that the EU is lagging behind when it comes to the hugely profitable exploitation of personal data to profile consumers, serve them targeted ads and manipulate their behaviour: what Shoshana Zuboff terms ‘surveillance capitalism’. But is that really a race the EU wants to win? Would advances in green technology and transport not be rather more useful?
Harari’s claim that more data means better algorithms is not wrong, but it is a generalisation that ignores the question of what data is being collected, and for what purposes. There is no single, objective definition of ‘better algorithms’: better at what? It depends on the policy goals being pursued. Being the best at surveillance is not necessarily something the EU should be aiming for. This is not just a political, but also an economic point. Surveillance capitalists not only invade our privacy and curtail our freedom; they also contribute very little of value to the economy as a whole.
In a perceptive article in The Atlantic earlier this year, Derek Thompson pointed out how little impact advances in digital technology have had in most areas of the economy: ‘We were promised an industrial revolution. What we got was a revolution in consumer convenience.’ There are huge profits to be made by exploiting personal data, and after years of low interest rates, venture capitalists starved of exciting investment opportunities have been happy to pour millions into ‘Uber, but for X’-style startups. As a result, the US tech sector is heavily biased towards consumer products, rather than deeper technological innovations.
These products do not even necessarily respond to consumer needs. As Zuboff outlines, the much-hyped ‘Internet of Things’ and the profusion of internet-connected home appliances have less to do with our desperate desire for smart doorbells and smart fridges than with the economic imperative for companies whose business model is surveillance and manipulation to gather ever more data about our lives. She quotes the unnamed marketing director of an IoT company saying that it’s ‘all push, not pull. Most consumers do not feel a need for these devices…[Silicon] Valley has decided that this has to be the next big thing’. Collecting all this data and concentrating it in private hands may produce technically brilliant AI systems, but there’s no reason to assume that will translate into meaningful improvements in collective wellbeing; that’s not the goal with which these systems will be designed.
The US may be the world leader in AI and home to many of the world’s most successful tech companies, but technological advances have had very little impact on its overall economic productivity, which has been rising by just 1.3% per year since the financial crisis and actually fell in one quarter of 2019. Perhaps one reason is that so many of the best minds of our generation are too busy finding innovative new ways to get people to click on ads and to track their locations at all times. If strict data regulation leads the EU to focus its resources, research funding, and industrial strategy towards transport, green energy, and manufacturing instead, maybe it will not be such a crippling disadvantage after all.
This illustrates how inappropriate it is to conceive of AI development as a race, in which everyone is going in the same direction and the US and China are getting there faster. As Virginia Dignum argues, races are linear and towards a specific end goal; in AI development there is no fixed end point at which we declare someone has won and go home. It will not be possible or desirable for the EU to ‘win’ at every possible aspect of AI development, since there are a vast number of possible applications and technological approaches, some of which are more socially valuable than others, and some of which are actively harmful. The strategy recognises that the EU is investing much less in AI development than the US and China, and calls for an increase, but expectations should be realistic; it cannot pour infinite resources into trying to keep up with them on all fronts, so focusing its energies on a few priority sectors is the most sensible way to move forward.
Moreover, winning is not an end in itself. AI advancement is useful insofar as it helps the EU reach its other policy goals, which are unlikely to be exactly the same as the US and China’s. In this regard, it should not be forgotten that what happens in Europe affects the rest of the world, and vice versa. Since the GDPR was passed, a number of other jurisdictions (including California, hardly a backwater when it comes to AI) have adopted similar privacy legislation. Since companies from all countries have to comply with the GDPR at least for their operations within the EU, it is influencing business practices around the world, including for US and Chinese tech giants. As the biggest market in the world, the EU has significant economic leverage over other countries: last year Japan finished implementing stringent new privacy safeguards in order to obtain an ‘adequacy decision’ allowing it to exchange more data with the EU. This all rather complicates the idea that EU regulations are holding back European innovators while the rest of the world freely races ahead. In fact, regulation does not just help the EU to uphold its values within its borders, but also to promote its values and policy priorities worldwide.
Finally, the point that there is not a single race to be won but a multiplicity of goals to be pursued applies not only to the problems we try to solve using AI, but also to the technological solutions themselves. Machine learning has produced most of the major American and Chinese breakthroughs in AI, but that does not mean that it is the only form of artificial intelligence worth pursuing. ‘True intelligence,’ as Dignum points out, is not just about perceiving patterns in data but also ‘the capability to reason, interact and decide based on little, incomplete and contradictory information’. In her view, the Commission’s strategy is still too narrowly focused on data as the basis for AI development, and is not capitalising enough on Europe’s academic strengths in other types of AI – although it is still open to consultation, so this could change.
What is clear, however, is that winning the ‘AI race’ is not the goal the EU should be trying to achieve – and although it also adopts the ‘race’ rhetoric rather uncritically, the new strategy shows that the Commission has its priorities straight. It is clearly concerned with Europe’s competitive position, but accepts that it lags behind when it comes to consumer technology and focuses instead on areas like industrial AI, environmental technology and transport – the fields where technological progress is most urgently needed. Ultimately, the goal is to promote the creation of new AI applications which genuinely improve people’s lives. If it succeeds in this, whether the EU is ‘behind’ or ‘ahead’ on some unspecified metric is beside the point.