A report studying global artificial intelligence (AI) trends points to progress for AI research in China in the last year, as well as persistent roadblocks.
The study, published by academic journal and research firm Elsevier, analyzed over 600,000 scholarly publications from 1998 to 2017 and found that Chinese publications are increasing in volume and show enhanced performance in some markers of quality.
Between 1998 and 2002, Chinese researchers wrote only 9% of academic publications, compared to 24% in the 2013 to 2017 time period. Europe lost 5 percentage points and the US 8 percentage points in the same time period but combined, accounted for more than half of the AI research worldwide.
Chinese research has mostly grown in the area of computer vision. In 2011, this topic overtook neural networks as the most popular among Chinese academics. That year, Chinese researchers wrote 3,000 papers on computer vision. Six year later, they wrote approximately 6,500, more than double than on the second most-popular topic, neural networks.
Europe follows a similar trend on computer vision research, but the consistent growth of this field is matched by that of planning and decision-making. In absolute numbers, the latter category maintains a lead in European research over computer vision, with approximately 750 more papers being published in 2017.
Another source of growth for Chinese research are conference papers. China’s AI-related academic publications increased by 13.8% between 2008 and 2018, compared with a 7.7% increase in Europe and 5.3% in the US.
The US may be lacking in volume of papers, but it is winning in research impact. Elsevier used the field-weighted citation impact (FWCI) to measure how often a paper is cited in other publications, adjusted for the average of the field.
Papers published from American institutions are cited 1.5 times more than the mean of the related field, a figure that has held and even increased since 1998. By contrast, European institutions started at the mean in 1998, and have progressed to about 1.25 in 2017.
China’s growth in this respect is “tremendous,” the study finds. China’s FWCI in AI research has galloped from half the world average in 1998 to reaching the mean in 2017.
This trend held true in the years from 2013 to 2017, when the top Chinese universities in terms of impact are, in order, the Chinese Academy of Sciences, Tsinghua University, Harbin Institute of Technology, Shanghai Jiao Tong University, and Zhejiang University.
Professor Chuan Tang of the Chinese Academy of Sciences (CAS) was interviewed for the paper. He finds three main obstacles in China’s contribution to global AI research. First, it is lacking the chip technology to support AI technology.
Second, “China lacks long-term efforts in AI basic research,” and scholars tend to follow Western trends, he told Elsevier. Third, it lacks experts of high quality, as only 38.7% of researchers working in China with more than 10 years of experience, he said.
Globally and in all academic disciplines, papers have higher impact, as measured by the FWCI, when they are published in partnership with industry professionals. Only 3.4% of AI-related papers worldwide involve academic-corporate collaboration, but they achieve, on average, a 2.53 FWCI score.
The US is leading in cross-sector collaboration; it is responsible for 8.9% of papers involving industrial partners worldwide. This share of American papers has an astounding academic impact, with an FCWI score of 3.41.
Europe and China have yet to work with corporate partners in AI research to this extent, with shares of 3.6% and 2.3% of global academic-corporate papers published, respectively, involving academic-corporate collaboration.
Chinese studies that involved corporate partners achieved an FWCI score of 2.64, slightly ahead of their European counterparts at 2.46.
China is also lagging behind in international collaboration. It holds the highest percentage of researchers who never leave the region, while the US has the largest number of researchers who migrate out of or into the country. Researchers who tend to stay within their region have the lowest impact and productivity on the field, compared with their migratory counterparts.
Slightly more researchers migrated to China for around two years between 1998 and 2017 to work on AI academia. China gained 0.1% more researchers in this period, close to the US’s net inflow of 0.3%.
However, researchers who stayed in the US in these two decades have the highest impact on the field, which “might indicate a reason for international inflow into the country,” the paper concludes.
Finally, the paper includes a case study of graduates from the Chinese Institute of Automation and the Chinese Academy of Sciences. The research indicates that graduates from AI-related fields are far more likely to end their education with a dispatch, meaning they are employed in jobs that the university or research institute helped them find.