China wants to become the world leader in artificial intelligence by 2030 — but a new analysis by Seattle’s Allen Institute for Artificial Intelligence, or AI2, suggests that Chinese researchers are on track to take the lead well before that.
The analysis is based on a tally of the most impactful research papers in the AI field, as measured by AI2’s Semantic Scholar academic search engine.
“If current trends continue, within five years, China will surpass us in terms of the top, highest-impact papers,” the institute’s CEO, Oren Etzioni, told GeekWire. “The other thing to realize is that citations are what you might call a lagging indicator, because the paper has to be published, people have to read it, and they have to write their own paper and cite it.”
Thus, the analysis is likely to understate China’s current influence in AI research, Etzioni said. “The bottom line is, Chinese AI research is startling in quantity and quality,” he said.
AI2’s findings are consistent with what tech analysts have been saying over the past year or two. Last year, an analysis by CB Insights found that 48 percent of the $15.2 billion invested in AI startups globally in 2017 went to China, with just 38 percent going to U.S. startups.
Etzioni said the AI2 analysis shows that research in artificial intelligence has grown dramatically over the past three decades, from 5,000 published papers in 1985 to 140,000 in 2018. Over that time, there have been many studies tracking the progress of AI research, but Etzioni said Semantic Scholar provides new perspective.
“First of all, this is the most up-to-date result, because we’ve analyzed papers through 2018,” he said. “Secondly, what’s unique is we looked at this notion of most-cited papers, because we’re after impact.”
The analysis shows that, in terms of sheer volume of research papers, China surpassed the U.S. back in 2006. Since then, China’s trend line has gone through ups and downs (and ups), but never fell below the U.S. totals.
Semantic Scholar told a different story when it came to the top 50 percent, the 10 percent and the top 1 percent of academic studies, as measured by citation counts. Those charts show a gradual decline in the percentage of papers attributed primarily to U.S. authors, and an accelerating rise in the Chinese percentage.
The Chinese Academy of Sciences led the list of China’s research institutions when it came to citations, followed by Tsinghua University, the Chinese University of Hong Kong and the Hong Kong University of Science and Technology.
If the trend lines are extended, China should surpass the United States this year for the top 50 percent, next year for the top 10 percent, and by 2025 for the top 1 percent.
China’s AI rise has already sparked concerns in Washington, D.C., leading to the establishment of a national security advisory commission as well as a Joint Artificial Intelligence Center at the Pentagon. The White House budget proposal for fiscal year 2020, released this week, would set aside $208 million for the AI center.
Etzioni argued that the federal government’s AI strategy should put more emphasis on basic research.
“We need to stop what the Trump administration has been doing, which is using various ways to discourage immigration of students and scholars into this country,” he said. “We need more of those talented people, like we always have. AI2 is highly international, and that’s been a huge boon for us.”
Setting aside more money for basic research in AI will also be essential, Etzioni said. Last month, President Donald Trump signed an executive order calling for a boost in AI research, and this week’s budget proposal made several references to AI’s importance. But those documents didn’t provide specifics.
“We need those specifics,” Etzioni said. “And we need them even sooner than we had thought.”
Authors of the AI2 analysis, “China to Overtake U.S. in AI Research,” are Field Cady and Oren Etzioni. The researchers used Microsoft Academic Graph to classify AI papers for the purpose of the study. Check the full analysis for details about the methodology. An earlier version of this report incorrectly included Singapore-based Nanyang Technological University as a Chinese research institution.