Can Bold Bets Keep the US Ahead of China in the AI Race?

Research shows that China’s artificial intelligence (AI) development is picking up at an unprecedented pace. Can the United States keep up with ambitious, out-there ideas?

On Track to Overtake the US

Last June, Apple and Google sponsored an academic contest at the world’s premier computer vision conference. The tech titans wanted to see if algorithms could make sense of twin camera images that varied in weather conditions from sunny to rainy. Doing so would be a lucrative achievement that could help the San Francisco-based developers with endeavors in autonomous vehicles and augmented reality.

Unfortunately for them, the contest’s winner was China’s National University of Defense Technology, a military academy with different objectives and allegiances. This event highlights how far China’s AI ambitions have come. In 2017, the country announced an AI strategy that would rival the US’s domain expertise by 2020. And the latest output data puts them right on track to do so.

Since 2005, long before the announcement of their AI strategy, Chinese researchers have published far more papers than their US counterparts. Chinese institutions have gained a reputation for low-quality (and even fraudulent) research, so naturally, questions have arisen about the work’s merit. But the numbers paint a different picture.

The Allen Institute recently analyzed data of more than 2 million AI research papers. It found that the ratio of top AI publications from China is rapidly approaching that of the US. And if the country keeps this pace up, it will be equal to the US by 2020. Even when the institute repeated the analysis to focus on research papers most cited, the US still did not come out on top very far ahead of China.

Government VS Corporate Support

Obviously, citations aren’t a perfect gauge of the quality and influence of research, so the Allen Institute plans to follow up with more analysis. But Oren Etzioni, the Allen Institute’s CEO, says one thing is readily apparent: The US government need to support American AI endeavors better.

From its military academy’s recent win in Apple’s and Google’s contest, it’s clear that the Chinese government has been very supportive of the country’s AI endeavors. And this investment is only growing. The country’s defense ministry plans to set up two new research centers in Beijing that are completely devoted to AI and unmanned systems.

Research from the Stanford-affiliated AI Index also shines a light on the Chinese government’s involvement—between 2007 and 2017, government-affiliated AI research papers grew by 400 percent. This absolutely dwarfs the output growth from Chinese corporations. In contrast, the US highly relies on companies like Alphabet for AI innovation; US corporate involvement in AI publications is seven times higher than that of China’s.

Recently, President Donald Trump signed an executive order to create a high-level AI strategy called the American AI Initiative. The order, which calls for more funding, resource allocation, and government involvement in AI, means well, but it has garnered skepticism. “It was well-intentioned but low on specifics, and it didn’t deliver the two most important things that we need—a more welcoming immigration policy to draw top research talent and significantly more research funding,” explains Etzioni.

Can Out-There AI Ideas Save America?

Putting Trump’s plans aside, there is one government entity that is focused on accelerating American AI innovation: The Defense Advanced Research Projects Agency (DARPA). When it comes to emerging technologies, the US has a long tradition of doubling down on crazy ideas. And DARPA is usually the one to see them through.

Recently, the agency showcased projects that are part of its AI Next campaign, a five-year, $2 billion plan to make the next round of radical AI concepts a reality. Initiatives under this plan include giving machines common sense, making them learn faster and need less data for training, and creating chips that can reconfigure to account for new AI capabilities.

Michael Kratsios, Deputy Assistant to President Trump for Technology Policy, says DARPA’s endeavors are essential to America staying ahead in AI: “This administration supports DARPA’s commitment, and shares its intense interest in developing and applying artificial intelligence. DARPA has a long history of making early investments in fundamental research that has had amazing benefits.”

Since its inception in 1957, DARPA has been behind some of the world’s biggest tech breakthroughs. In the ’60s, it created the networking technology that would eventually become the Internet. And more recently, it invested in the personal assistant project that would eventually become Siri, Apple’s famous AI helper.

Building a Foundation for AI’s Future

But with AI, the agency faces a different type of challenge: many of the algorithms being employed today were actually developed decades ago. As a result, they’re quite limited in capability. “We are harvesting the intellectual fruit that was planted decades ago,” explains DARPA’s Information Innovation Office Deputy Director, John Everett. “That’s why we’re looking at far forward challenges—challenges that might not come to fruition for a decade.”

So for DARPA’s AI initiatives, the agency is focused on making longlasting advancements that address these limitations. For example, giving AI common sense would fix a weakness that causes many of today’s systems to fail. Often AI systems are limited by their narrow understanding of the world. By broadening AI’s perception and reasoning abilities, using tools like AI assistants would become more intuitive and allow robots to navigate unfamiliar territory much better.

Using less data would also help circumvent another AI limitation unique to the US. Training data often determines the effectiveness of machine learning applications. But with much more stringent regulations around data, the US often falls behind China, which has access to an abundance of information that it can use in AI initiatives.

Whether the projects under DARPA’s AI Next initiative accomplish their objectives remains to be seen. But they do make one thing clear: It’s quite difficult to gauge progress and expertise in AI. Calling this competition between China and the US a race makes it seem like there’s only one path or destination to reach to achieve victory.

But this couldn’t be further from reality. In truth, which country wins will come down to smarter allocation and more innovative use of resources. And as far as this goes, each country is taking a radically different approach from the other. It will be interesting to see how things play out. What do you think will happen in the next few years? Let us know in the comments!