AI, captain: The ‘new space race’ to steer the tech ship is on
The question is no longer whether AI is shaping our world, but who is shaping AI
If you think artificial intelligence is an entirely theoretical pursuit, a quick trip to Kaggle is illuminating.
Kaggle is a website that runs competitions, sometimes with serious cash prizes, between teams vying to solve real-world problems by designing algorithms that trawl through big data sets.
There are currently 17 competitions up and running. The biggest prize, $100,000, is being offered by an investment company for the team that reveals how news reports affect stock prices.
The second and third biggest – both offering about $50,000 – seek to predict when earthquakes will occur and how loyal Brazilian shoppers are.
There’s even $25,000 to quantify “cuteness” in Malaysian pets, so that shelters for homeless animals there can understand which get adopted fastest, and improve the online profiles of those no one wants to take home.
In each case, a worldwide community of experts is being harnessed to build computer processes – largely a branch of AI known as machine learning – which have a very practical impact.
These are not academic exercises, like beating the world’s best player of the game Go, as an algorithm built by the former British company Deepmind did in 2016.
They are about saving lives – human and hamster – or helping businesses grow. The real similarity is that, like once-British Deepmind, which Google bought in 2014, Kaggle is also owned by the US giant, which snapped it up in 2017.
Indeed, while artificial general intelligence – computers able to rival mankind’s astonishingly adaptable cognition – is still far off, specific AI – computers brilliant at learning to solve certain tasks – is becoming ever more embedded into the world around us.
So in 2019, the question is no longer whether AI is shaping our world, but who is shaping AI.
In most accounts, Google and America’s other titans sit in one corner of the ring, the Chinese state in the other – both mining vast troves of data and training their algorithms to reach ever greater sophistication as they go.
The two countries’ AI strategies are profoundly different, one driven by the state, one by the private sector. The highly respected AI index report, released last December, shows that in 2017, for example, China’s government and its affiliates produced four times as many AI-related papers as its corporations.
In America, by contrast, it was the private sector that produced most by far, almost twice the papers produced by the government.
“The proportion of corporate papers in the US is 6.6 times greater than that in China,” the report noted.
And the private versus state approach adopted has huge implications for investment. While the Chinese government seeks to stimulate its AI industry with a seemingly neverending stream of billion-dollar-plus initiatives, Washington relies on its businesses to do the job.
For example in 2017, Amazon alone spent $16.1bn on research and development – more than five times the $2.9bn spent by the most celebrated research wing of the US government, the Defence Advanced Research Projects Agency (Darpa).
The stats reveal nothing short of an ideological gulf between the US and China, echoing that between the US and the USSR in the 1960s. No wonder many are already calling AI the new space race.
Ultimately the identity of the biggest winners will matter to all of us, because as AI seeps into everyday life, most companies will have to buy the technology. It’s too complex to develop in-house. Deloitte, for example, predicts that this year, 70% of companies that adopt AI technology will do so through cloud-based enterprise software.
AI is spreading to the many, but it will still be supplied by the few. So China and America can seem like the Symplegades of AI – clashing rocks that smash all between them.
Certainly, that is the view espoused by Kai-Fu Lee, once an executive at Apple, former president of Google China and author of an influential recent book – AI Superpowers: China, Silicon Valley, and the New World Order.
He puts Europe so far behind China and the US that it “wouldn’t even claim a bronze medal in this AI competition”.
Such defeatism is commonplace among European politicians and academics too, many of whom suggest clubbing together across the continent to compete.
One initiative is to build an AI equivalent to Cern, the giant European particle smasher, which its backers would call the European Laboratory for Learning and Intelligent Systems, or Ellis.
A second proposal is called the Confederation of Laboratories for Artificial Intelligence Research in Europe, or Claire.
Both deploy the language of desperation. “Europe is not keeping up,” warn Ellis’s backers in an open letter.
AI developed elsewhere “is not well-aligned with European values”, say those behind Claire.
Yet the truth is that in some respects Europe is still a global AI leader.
Analysis compiled at the end of last year reveals that Europe accounts for 28% of research papers published on AI, ahead of China on 25% and the US on 17%.
And in Europe the UK is, by any metric, the biggest player by far. Britain matters.
The UK far outstrips France – Europe’s second power in AI – by the number of PhD students, research papers published, venture capital investment in new AI companies, the valuations of those companies, leading entrepreneurs in the sector, how much they are paid – you name it.
As a study commissioned by the European parliament reported last year: “The UK is currently the only leading country for AI that has succeeded, thanks to a favourable AI ecosystem, to stand in international competition for AI funding, research and talent.”
Brexit risks, it notes, will primarily be to the EU, which “will be deprived of its most vibrant player”.
The European Commission went even further a month ago in its own report, Artificial Intelligence, A European Perspective.
It noted the UK was home to more than 6% of the 35,000 workers in the global AI sector, more than twice the number in France.
Reflecting on centralised state efforts to stimulate national AI sectors, such as President Emmanuel Macron’s £1.3bn AI plan unveiled last March, or the UK’s own £1bn AI sector deal announced a month later, the commission acknowledged: “An AI race for leadership has just started and sees the US, China and Europe [mostly the UK] as the largest players.” Mostly the UK.
“When it comes to AI talent, there is nothing in the world to compare with the triangle of London, Oxford and Cambridge,” says Murray Shanahan, professor of cognitive robotics at Imperial College London (who also happens to be a senior research scientist at Deepmind).
And in one respect, the AI sector in the UK and continental Europe is characterised by the same ideological rivalry as that between the US and China. What is the best way to become globally competitive? Is it through centralising, state-led, top-down initiatives, or by bolting world-leading research on to deregulated free markets, then letting enterprise and entrepreneurship do the rest?
Across the Channel, the inclination towards the former is clear. German economy minister Peter Altmaier stated his preference at the end of last year. “We need a kind of Airbus for AI,” he said at a summit in Nuremberg.
But some thinkers and investors believe bureaucratic commands from on high are highly unlikely to succeed.
Nicolas Colin is a former French bureaucrat who is now the founder of The Family, an investment firm specialising in European startups.
He says European technologies that seriously rival the US and China can only be driven from the bottom up.
He suggests that such companies, forced to overcome Europe’s national barriers of different languages and administrations, may even come to have a competitive edge.
“We need to make use of our disadvantage – our fragmentation – and turn it to our advantage.”
What is clear is that for world-leading AI companies to be born and thrive in the UK and continental Europe, it will not be possible to simply to replicate China or America. Rather, a different model must be established, a model to let them leapfrog rivals, a model whose founding philosophy remains up for grabs.
Will it be top-down along EU/Airbus lines, or deregulated, with a hive of interconnecting support services, following the recipe that has made the City so successful?
As with financial services, Britain’s significant European lead in AI is one way of ensuring it is the latter. That would be a fitting success.
For from the 1960s to the late 1980s AI research itself was dominated by a top-down approach, Classical AI, in which machines navigated the world through an ordered, often complex set of symbols framed by a designer.
It promised much, but in the end ran into the sand. Instead, the great AI advances of today have come through biologically inspired processes that start off very simply, but that over time, layer upon layer, deliver extraordinary results with insight Classical AI never demonstrated.
In other words, creating effective AI networks once depended on a radical bottom-up approach.
Today, creating an effective network of AI companies in Europe may depend on British leadership delivering the same.
– © The Sunday Telegraph