8 January 2021

The US Needs More Foreign Artificial Intelligence Know-How


DURING PRESIDENT OBAMA’S two terms in the White House, Jason Furman was a top economic policy adviser and a key voice on the growing importance of artificial intelligence.

Furman served as deputy director of the National Economic Council before becoming chair of the Council of Economic Advisers. He also coauthored a report issued by the Obama administration in October 2016 that detailed the economic importance of AI to the US.

Furman, who is now a professor of the practice of economic policy at Harvard, spoke to WIRED senior writer Will Knight. This transcript has been edited for length and clarity.

WIRED: The report produced under President Obama, Preparing for the Future of Artificial Intelligence, emphasized the importance of AI to the US economy. What’s the first thing the Biden administration should do to demonstrate this?

Jason Furman: Far and away the most important AI policy change should be immigration policy. That has been a disaster for the last four years, and I hope that gets much better over the next four years.

AI is invented by humans, and if you look at the top AI engineers, they are spread all over the world. Attracting them to this country is so important.

The US has certainly lost its allure for some tech students and workers. Are there specific immigration policies you’d like to see changed?

It’s everything from discouraging students from coming to our country to making it less attractive to hire people through the H-1B program and making it harder for people's spouses to work while they're here.

If you look at the engineers in AI and the major tech companies, they're from all over the world. It’s also just sending out a “You’re not welcome in America” message. I think that's been the most harmful thing to AI innovation in the last four years. Some of the harm accumulates over time, so it’s not at all too late to deal with that and undo it.

There’s been plenty of impressive AI research over the past few years. Why can’t the US just focus on making use of that?

I think it's possible that we need more and more impressive AI just to keep up the old pace of innovation.

Look at agriculture. The most important crops there are wheat and soybeans, and there's very little room left for AI—it’s already very mechanized. There's more room for AI in advanced robotics with soft fruits like strawberries and grapes and the labor there. In some ways, that requires a much more technologically advanced input. But the output in terms of GDP just isn't as large as the stuff we've already done in agriculture, because that's a smaller part of it.

What about the Trump administration’s concerns about Chinese students or researchers coming to the US to steal ideas?

I think the paranoia around spying by Chinese students and Chinese researchers in the United States is getting in the way of our ability to attract the best scientists and make advances in many areas, of which AI is just one.

I think it's possible that we need more and more impressive AI just to keep up the old pace of innovation.

Most of what is done in universities isn't secret; in fact, we want people to hear those ideas. and take them to other countries. The larger the network of people involved in that process, the faster we’ll make progress. So the idea that China is spying on something that itself is a public good—that just doesn't make any sense.

There must be legitimate concerns, though, perhaps with companies like Huawei?

I don't have access to enough information on Huawei; I’m absent the security clearance. I’m open to that being a legitimate national security issue around protecting ourselves from espionage. But that’s different from ideas; that’s using a particular company’s technology that comes embedded with who knows what.

So I'm open to there being a trade-off between national security and economics, and I'm very open to making the trade-off on the side of national security. It’s the basic ideas, innovation, the types of things you publish in a journal, as opposed to the type of things you manufacture in a factory, that aren’t as much of a risk.

More broadly, what did the Trump administration get right about AI, and what did it get wrong?

They put AI innovation front and center, which is where I think it should be. And they did recognize that if you don't have public trust in AI, you won’t be able to advance it. My view has been, and still is, that we don’t have enough AI. So I think some additional federal funding would help; that’s something the Trump administration actually called for.

They didn't focus nearly as much on addressing issues around bias. There's an increased understanding of algorithmic bias, and racial equity is one of the four priorities that president-elect Biden has set. Clearly there's work to be done in this area. Some of it can just be reversing President Trump's executive order (regarding diversity training at entities receiving federal money), which basically made it harder to do some of the enforcement around racial bias.

I don't favor a single AI regulator, because I think you need AI expertise at the SEC, at NHTSA, at HUD, you need it everywhere.

Should the next administration be worried about the way AI expertise is being sucked into industry?

So far corporate AI has been decently open, because researchers at the companies want to share and show off. But I'm nervous about how long that will persist, and I’d like to see more university research.

The two obstacles of university research are, one, you need money to pay the researcher to compete with the private sector, and two, the data and the resources that you can work with are much better in a company. So it’s tough and expensive to compete. I remain nervous about the extent to which you can count on the private sector for basic research.

Would increasing the diversity of the AI workforce also help the US access more talent?

The United States is 20th in the OECD in the share of 3- and 4-year-olds who go to preschool. Remedying that would be an AI policy, because there are people who could make a contribution that we're missing out on, because from the very beginning of their lives they're not getting on a track that will let them express their full potential. So yes, the extreme levels of inequality of opportunity in the United States are important problems for all innovation.

What impact do you think Covid is likely to have on the spread of AI?

The biggest problem we have in our economy has been demand (for goods and services, and therefore workers). But by the end of next year, I think people will have, on balance, enough money to spend. And it will be safe to spend it because of the vaccine. But there will still be a lot of employers that don't want to hire people back, because they decided they can make do without them. This is the problem we're going to have in 2022 and beyond, and none of the emergency legislation has started to scratch that.

For all the hype around AI, it seems surprisingly few companies are using it. Should the government try to encourage greater adoption?

I don’t think the government needs to encourage that. We just need to make sure it's there, and that we’re overcoming the problems.

No comments: