US export controls are no guarantee against China’s AI advances

In February, Michael Kratsios, one of Donald Trump’s top advisers on technology, gave a short speech at a Washington think-tank that revealed the White House’s thinking on artificial intelligence. The US had not yet reached its “potential” in the field, despite being the global leader, he said. This meant it had to “embrace new discoveries so that future is made in America, by America, and for the benefit of our great American people”.

But Mr Kratsios’s vision was also about defending America’s gains against global competitors such as China. “We will not allow adversarial nations and bad actors to steal our ideas, copy our technology and cheat their way to leadership, in a field central to our nation’s security,” he told the audience at the Center for a New American Security.

Yet while the awareness of the challenges facing the US on artificial intelligence has reached the upper ranks of the American government, it is far from clear that Donald Trump’s administration is embarking on the right policy path to match the stakes. “The US still retains a lead in some areas, but that lead is diminishing over time, and more than that, it is clear that China is accelerating its investments in AI while the US has no concrete plan on how it is going to handle this,” says Daniel Castro, vice-president of the Information Technology and Innovation Foundation in Washington.

“There is really no expectation that the US will come out on top in this battle for global dominance,” he says. Analysts in the US capital say the problem is not inaction: Mr Trump held a big AI summit at the White House last year and signed an executive order in February designed to boost America’s capabilities in the sector.

Instead, the concern is that some of the Trump administration’s measures, including cuts to discretionary spending in the budget, and curbs on immigration including for high-skilled workers, could blunt America’s edge.

On top of that, in mid-November, the Trump administration had proposed to include AI and robotics in a list of emerging technologies that were being considered for export controls, a treatment normally reserved for weapons and other products with possible military applications. Within artificial intelligence, the US administration suggested eight categories that might be subject to trade restrictions, including neural networks and deep learning, genetic computation and computer vision.

Among the categories of robotics that could face restrictions are micro drone and robotic systems and swarming technology.

The new export controls are expected to be narrowed significantly when the policy is finalised in the coming months. But technology groups — and many experts — are nervous that they could undercut America’s ability to innovate by discouraging research across borders that is pivotal to the AI sector in the US.

“The US needs to be really careful that it doesn’t implement knee-jerk responses that made sense in the closed off government-held technology model but will strangle innovation and hurt the US industry in the long term,” says Helen Toner, director of Strategy at the Center for Security and Emerging Technology at Georgetown University.

Part of the difficulty for US policymakers is that they may have missed the window to defend their indigenous artificial intelligence capabilities more aggressively, since China has already made so much progress in the field.

“China is already developing a very strong AI sector, so using export controls and other regulations to forestall progress? I’m not sure that’s really going to get you where you want to go,” says Steven Feldstein, non-resident fellow at the Carnegie Endowment for International Peace. The debate has become part of the broader tensions in the US relationship with China, including the trade war, which makes it harder to take a more pragmatic approach.

But experts say a more positive agenda will be crucial for the US not to end up shooting itself in the foot. One of the key areas of focus could be to ensure that the free flow of data can continue — since it is a bedrock of AI — but it could bump up against efforts to tighten privacy laws.

“Access to data, particularly certain kinds of data, will be critical for training AI algorithms,” says Samm Sacks, a cyber security and China digital economy fellow at the New America Foundation, a think-tank. “The rules for what companies can do with data have yet to be written,” she adds.

The wide-ranging policy debate on AI in the US shows how far the topic has evolved from discussions about its impact on the labour market. But the risk, perhaps, is that it has moved too far into the terrain of national security, which could make America clam up.

“The US is exceptional in being a country where the best and brightest from all around the globe really want to come here and work and research here and contribute their talents to US companies and US labs,” said Ms Toner. “Supporting that will help the US retain its advantage over the longer term.”