ANGELA MERKEL, Germany’s chancellor, has a reputation for being dour. But if she wants to, she can be quite funny. When asked at a recent conference organised by Ada, a new quarterly publication for technophiles, whether robots should have rights, she dead-panned: “What do you mean? The right to electric power? Or to regular maintenance?”
The interview was also striking for a different reason. Mrs Merkel showed herself preoccupied by artificial intelligence (AI) and its geopolitics. “In the US, control over personal data is privatised to a large extent. In China the opposite is true: the state has mounted a takeover,” she said, adding that it is between these two poles that Europe will have to find its place.
Such reflections are part of a wider realisation in Europe: that AI could be as important to its future as other foundational technologies, like electricity or the steam engine. Some countries, including Finland and France, have already come up with national AI strategies, and Germany is working on one. Once it is finished later this year, the European Union will condense these efforts into a co-ordinated plan on AI. Unsurprisingly, it is all highly Eurocratic: dozens of committees and other bodies are involved. But the question raised by Mrs Merkel is as vital for Europe as the ones about Brexit or immigration: can it secure a sizeable presence in between the AI superpowers of America and China?
People in Silicon Valley are sceptical. “It will screw this up, just as it has done with cloud computing,” says Jack Clark of OpenAI, a company that aims to promote human-friendly AI. Hardly anyone in the Bay Area can imagine Europe becoming a force in machine learning, the AI technique that has seen most progress in recent years. It involves feeding reams of data (pictures of faces, for instance) through algorithms so they learn to interpret other data (in this example, to recognise people in videos).
That scepticism is not only because of self-inflicted weaknesses such as Europe’s “tendency to favour incumbent business over disruptive newcomers”, in the words of Greg Allen of the Centre for a New American Security, a think-tank. The region also has a structural disadvantage: a lack of scale. Benefiting from huge, homogeneous home markets, America’s and China’s tech giants have a surfeit of the most vital resource for AI: data.
This advantage creates others. Having more data means that firms can offer better services, which attract more users and generate more profits—money that can be used to hire more data scientists. And having a lot of data creates demand for more computing power and faster processors. All the big cloud-computing providers, including Amazon and Microsoft, are developing their own specialised AI chips, an area where Europe is also behind.
Yet look beyond machine learning and consumer services, and the picture for Europe is less dire. A self-driving car cannot run on data alone but needs other AI techniques, such as machine reasoning, which is done by algorithms that are coded rather than trained—an area in which Europe has some strength. Germany has as many international patents for autonomous vehicles as America and China combined, and not only because it has a big car industry.
Nor are ever-larger pools of data and ever-more powerful chips the only way to go. More researchers are looking into what can be done with “small data”—ie, using fewer data to train algorithms—particularly in manufacturing and the internet of things. This is where Europe, home to many industrial firms, could have an advantage. “As AI gets more complex, Europe will have opportunities,” predicts Virginia Dignum of Umea University in Sweden.
Old Europe and wiser AI
Europe’s biggest opportunity, however, may be political and regulatory rather than technical. As Mrs Merkel noted, America and China represent two fairly extreme models on AI—which leaves room in the middle. “Europe could become the leader in AI governance,” says Kate Crawford, co-founder of the AI Now Institute, a research centre at New York University. Europe could pioneer rules to limit potential harm from AI systems when, for instance, algorithms are biased or run out of control. Many people hope that Europe will set global standards in AI, as it is doing with its new privacy law, the General Data Protection Regulation, whose principles have been widely copied elsewhere.
Other types of regulation offer a similar opening. Both America and China are centralised data economies, in which this resource is controlled by a few firms. Europe has a shot at developing a more decentralised alternative, in which data are traded or shared between firms. That could involve defining access rights to data (the equivalent of property rights in the digital realm) and what types of data, including commercial ones, need to be made open because of their social value—much as European banks must give fintech startups access to certain data if customers agree to this. That could make Europe the preferred home for new types of data firms.
To get there, Europe has to get a lot right. On paper, things look promising. The French and Finnish national AI strategies make for more interesting reading than the American and Chinese plans that inspired them, offering a balanced mix of measures, ranging from public spending on research and training data scientists to rules for the data economy and using AI in government. The European Commission’s communication on AI is long on such useful things as “making more data available” and “bringing AI to small businesses”.
Plenty could still go wrong in the implementation stage, however. To become more of a force in AI, Europe will have to pool its resources in research and data. But the EU tends to spread things out in order to satisfy both national and commercial interests. Expect the commission at some point to announce the launch of a loose AI research network rather than anything with a central hub.
Another reason for pessimism is institutional inertia. Much of the money earmarked for AI research will end up in existing academic institutions, which may not be the best place for it. Many European research institutes founded decades ago survived the long AI winter in the 1990s and 2000s, when reduced interest and funding killed off efforts elsewhere. The German Research Centre for Artificial Intelligence, founded in 1988, claims to be the world’s largest with more than 1,000 employees, but it is certainly not the best known. Its strengths lie in robotics and classical AI rather than in machine learning.
Most worryingly, Mrs Merkel’s interest notwithstanding, Germany does not seem ready to get fully behind AI and to team up with its European neighbours. A big reason for that is its existing economic strength. Policymakers, industry leaders and researchers tend to argue that Germany is already an AI power and that it will suffice to inject more of the technology into the continent’s industrial products and manufacturing machinery. Co-operation with France does not appear to be a priority. Plans for a joint AI research centre, mentioned in the German government’s coalition agreement, have been abandoned.
To become a powerhouse in AI, Europe will have to overcome its divisions, digital and otherwise. That seems unlikely in the current climate. But where there is a political will, there may be a way. In the 1980s, when Europe felt under economic threat from Japan, Helmut Kohl and François Mitterrand, the then German chancellor and French president respectively, pushed through an EU-wide wireless standard, which came to be called 2G and helped Europe dominate the mobile-phone industry for decades. To be sure, AI is a much more complex beast than a wireless standard. But it shows what may be possible.