Trump wants better AI. He also wants less immigration. He can’t have both.

President Donald Trump released a splashy new plan for American artificial intelligence last week. High on enthusiasm, low on details, its goal is to ramp up the rate of progress in AI research so the United States won’t get outpaced by countries like China. Experts had been warning for months that under Trump, the US hasn’t been doing enough to maintain its competitive edge.

Now, it seems, Trump has finally got the memo. His executive order, signed February 11, promises to “drive technological breakthroughs ... in order to promote scientific discovery, economic competitiveness, and national security.”

Sounds nice, but unfortunately, there’s a problem. America’s ability to achieve that goal is predicated on its ability to attract and retain top talent in AI, much of which comes from outside the US. There’s a clear tension between that priority and another one of Trump’s objectives: cutting down on immigration, of both the legal and illegal varieties.

Trump has spent the past two years of his presidency pushing away foreign-born scientists by means of restrictive visa policies. (Yes, he ad-libbed during the State of the Union that he might want more legal immigrants, but it’s really not clear how serious he was.) He’s also alienated them through his rhetoric — his decision to declare a national emergency to build a border wall is just the latest example. The result is a brain drain that academic research labs and tech companies alike have bemoaned.

If the Trump administration really wants to reverse this trend and win the global AI race, it’s going to have to relax its anti-immigrant posture.

The visa system would be a good place to start. Writing in Wired last week, Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence, called on Trump “to include a special visa program for AI students and experts.” Etzioni argued that “providing 10,000 new visas for AI specialists, and more for experts in other STEM fields, would revitalize our country’s research ecosystem, empower our country’s innovation economy, and ensure that the United States remains a world superpower in the coming decades.”

As Etzioni noted, the Trump administration has so far made it harder for foreigners to get H-1B visas, which allow highly skilled workers to perform specialized jobs. The visa process takes longer than it used to, lawyers are reporting that more applications are getting denied, and computer programmers no longer qualify as filling a specialty occupation under the H-1B program. Even those lucky ones who do get the coveted visas may have a hard time putting down roots in the US, because the administration has signaled it may nix the H-4EAD program, which allows the spouses of H-1B holders to live and work in the country.

It’s hard to overstate how anxiety-provoking all this can be for authorized immigrant families, whether or not they work in AI. Imagine not knowing, for months at a stretch, whether you’ll get to keep working in the US, or whether you and your partner will be able to live in the same country. Facing this kind of uncertainty, some would-be applicants prefer to pursue less stressful options — in countries that actually seem to want them.

Despite getting bogged down in the courts, Trump’s travel ban also impacted hundreds of scientists from Muslim-majority countries — and deprived the US of their contributions. “Science departments at American universities can no longer recruit Ph.D. students from Iran — a country that … has long been the source of some of our best talent,” MIT professor Scott Aaronson wrote on his blog in 2017. “If Canada and Australia have any brains,” he added, “they’ll snatch these students, and make the loss America’s.”

Canada has done just that. The country, which hosts some of the very best AI researchers in tech hubs like Toronto and Montreal, was the first in the world to announce a national AI plan: In March 2017, it invested $125 million in the Pan-Canadian AI Strategy. That summer, it also rolled out a fast-track visa option for highly skilled foreign workers. And as Politico reported, in a survey conducted by tech incubator MaRS, “62 percent of Canadian companies polled said they had noticed an increase in American job applications, particularly after the election.”

Canadian universities also reported a major uptick in interest from foreign students. University of Toronto professors, for example, reported seeing a 70 percent increase in international student applications in fall 2017, compared to 2016.

Taken together, Trump’s immigration policies have hurt the US when it comes to STEM in general and AI in particular. But his executive order gives no sign that the administration understands that. It features the words “immigration” and “visa” exactly zero times.

If anything, it seems to double down on Trump’s “America first” philosophy. A section titled “AI and the American Workforce” says heads of agencies that provide educational grants must prioritize AI programs and that those programs must give preference to American citizens.

This approach fits with Trump’s overall belief that immigrants, both legal and illegal, take jobs from US workers. He’s just applying that principle to the field of AI researchers.

Yet the primary stated goal of Trump’s AI strategy is not to protect the career prospects of individual American workers, but to protect Americans at large from being overtaken by other countries in the AI race. Of course, those two projects may converge to some extent. But there will also be instances where they diverge — in the context of hiring decisions, say. In those instances, it’s most effective to choose the candidate (wherever she comes from) who’ll do the best job at beefing up America’s AI strategy, so that it can ultimately benefit a vast number of Americans.

Maintaining the US advantage in AI has become an increasingly urgent project since last year, when China declared its intention to become the world leader in the field by 2030. Although Trump’s executive order avoids mentioning China by name, a US Defense Department document on AI strategy released the very next day was not so circumspect. As the public summary of a strategy that was developed last year, it offers a window into the motivations that are driving the Trump administration now. And it casts the situation in pretty dire terms:

Other nations, particularly China and Russia, are making significant investments in AI for military purposes, including in applications that raise questions regarding international norms and human rights. These investments threaten to erode our technological and operational advantages and destabilize the free and open international order. The United States, together with its allies and partners, must adopt AI to maintain its strategic position, prevail on future battlefields, and safeguard this order.

There’s good reason to be concerned about China winning a global AI race. Already, it’s using AI surveillance technologies to become one of the most repressive countries on the planet. For the US to let in fewer foreign researchers would be to risk more of them being employed by China, which could mean its military makes key advances first.

Americans who tend to worry deeply about the risks AI poses to humanity may argue that slowing AI growth means we forestall those risks. But slowing growth in the US won’t impede growth worldwide; it just means it’ll happen elsewhere.

In fact, there’s a compelling argument to be made that more immigrants working in American AI would allow firms that are concerned with AI safety to develop safer methods faster. That’s not just because more people power will yield more research and development, but because immigrants might be more attuned to racial, ethnic, and class disparities in the way AI risk gets distributed. Those disparities have been too often glossed over in Silicon Valley.

At least one firm aiming to build safe artificial general intelligence, OpenAI, seems cognizant of that. “Our goal is to build safe AGI which benefits the world, which means we want our team to be maximally representative of the world,” its website states, before specifically noting: “We sponsor visas including cap-exempt H-1Bs.”

This diversity-embracing approach is a far cry from the one laid out in the president’s executive order.

Ultimately, Trump can ensure America’s place as an AI superpower. Or he can try to keep AI jobs out of the hands of non-Americans. But he has to choose.

Sign up for the Future Perfect newsletter. Twice a week, you’ll get a roundup of ideas and solutions for tackling our biggest challenges: improving public health, decreasing human and animal suffering, easing catastrophic risks, and — to put it simply — getting better at doing good.