How Britain can become an AI superpower

 AI superpower is rising

Get ready for some big British celebrations in 2030. By then, if Rishi Sunak is to be believed, the country will be “a Science and Technology superpower”. The prime minister’s aim is for Britain to prosper from the booming opportunities offered by supercomputing and artificial intelligence. Generative AI has stoked a frenzy of excitement (and some fear) among techies and investors; now politicians have started to acclaim its potential, and British ones are in the vanguard. Britain, says Mr Sunak, will harness AI and thus spur productivity, economic growth and more. As he told an audience in London this week, he sees the “extraordinary potential of AI to improve people’s lives”.文章源自The Economist Digest-https://te.qinghe.me/12221.html

Mr Sunak’s vim and his readiness to champion AI are welcome, even if his claims sound breathless. After all, Britain’s government has spurred innovation that had sweeping economic effects—think of the Big Bang reforms in the 1980s that turned London into Europe’s financial hub. There is every reason to believe a new AI era will create huge opportunities . He is right to plan for how to make the most of these chances. But could Britain, realistically, lead on this?文章源自The Economist Digest-https://te.qinghe.me/12221.html

The country does have some advantages. It is home to several important AI companies, mostly in London—in particular, Google DeepMind. It has excellent universities, and welcomes the highly skilled foreign workers that AI companies need. The state generates troves of data; no other country has such an array of health records under a single entity, the National Health Service (NHS). And Brexit creates a chance to adopt an appealing regulatory position that could be a model for medium-sized countries around the World as they also rush to join the AI party.文章源自The Economist Digest-https://te.qinghe.me/12221.html

But there are problems aplenty, too. The most obvious is that Britain is a smallish place. America’s dominance in tech exerts a steady pull on capital, people and ideas, and American firms duly dominate in AI. The way Brexit was done means that Britain has lost access to the European Union’s single market. Although Oracle has a cluster of the advanced graphics processing units (GPUs) needed to train large models, none of the cloud-computing giants has yet chosen Britain as the base for what techies call the “compute”.文章源自The Economist Digest-https://te.qinghe.me/12221.html

For Britain to prosper in AI, therefore, much will have to change. It needs to cram more people who know one end of a GPU from the other into positions of influence in government. Mr Sunak may loudly extol the promise of BritGPT, but his government should include more engineers who understand the mix of data and compute from which AI is built.文章源自The Economist Digest-https://te.qinghe.me/12221.html

Time to chatGPU

Once it has the expertise, the government must deal with three broad concerns. The first is about those public datasets. They are in no fit state for AI developers to exploit—the data are unrefined ore, not sparkling treasure. Only the state has the authority to get these datasets cleaned up, and to start thinking of what new ones could be built. A stock of clean, regularly updated datasets that are technically and legally easy for algorithm-makers to use would draw in engineers who want to build new AI systems. An AI-ready NHS would be the jewel in Britain’s crown.文章源自The Economist Digest-https://te.qinghe.me/12221.html

Second, Britain should move fast to gain an edge in regulation. The goal should be a pragmatic set of rules keeping AI safe that sits somewhere between America’s Wild West permissiveness and what is likely to be a regulatory warren in the EU. Mr Sunak announced at a White House meeting with President Joe Biden this month that he will host a global summit on AI regulation this autumn. Good. That will be the place and time to set out Britain’s stall as having rules for AI that are sufficiently flexible to work for different industries. Hairdressers who want AI to help pick new styles, for example, need not be regulated in the same way as mortgage lenders.文章源自The Economist Digest-https://te.qinghe.me/12221.html

The last and thorniest concern is how to get AI developers the compute they need to train and run large models. Advanced GPUs made by Nvidia—for now, the only chips worth using—are suffering a global supply crunch. The government could help by telling British companies and its own departments to be much readier than now to send their data abroad to AI developers in other, friendly countries. For most datasets, worries about privacy and security are overdone.文章源自The Economist Digest-https://te.qinghe.me/12221.html

However, an unfortunate correlation exists between the sensitivity of datasets and their value in creating large models; data that are sensitive, that capture aspects of either human health or behaviour, or pertain to national security, are what would be most useful to inform new models. There is an understandable reluctance to send such data abroad.文章源自The Economist Digest-https://te.qinghe.me/12221.html

That is why Britain urgently needs more GPU clusters within its borders. More compute on British soil will have all sorts of local spin-offs and benefits. Jeremy Hunt, the chancellor, has talked of giving academic computer scientists £900m ($1.1bn) to build a British supercomputer in Edinburgh.

Yet commercial AI is so dynamic that the Edinburgh scheme risks becoming an AI white elephant. Amazon alone spends around $25bn a year on compute. British taxpayers cannot keep up with the private sector—and should not try. Instead the government should do all it can to persuade commercial providers to invest in GPU clusters on British soil.

One focus should be to ensure a reliable supply of clean, affordable power. To train models needs mind-boggling quantities of electricity. If Britain is without cheap supplies of power, it will struggle to persuade anyone to set up big GPU centres there. The queue to obtain a connection to Britain’s grid is holding back potential investors across the economy.

Other steps could include using public money to fund a “moonshot” project, such as developing open-source software, to help chipmakers break the near-monopoly that Nvidia holds on the AI market. Nvidia’s edge comes from clever software which makes training models on its GPUs a breeze. Rival chipmakers in America and Britain have no equivalent and are all but locked out of the AI market. With better software their chips could compete with Nvidia’s and ease the supply crunch that dogs AI developers the world over. That’s a worthy ambition for a country seeking global AI greatness.