The US and Europe might be top dogs in machine learning at the moment, but the East is catching up fast, helped by massive government spending.
That's one of the key findings the second annual AI Index, a compendium that empirically tracks how AI research is progressing over time. The organization tracks research papers, industry statistics and other quantifiable metrics.
East, West, which is best?
In terms of research Western countries across the US and Europe still lead in research impact, but China is catching up. Meanwhile, South Korea and Japan rank second and third for number of AI patents filed after the US.
Last year, 28 per cent of AI papers on Scopus, a large database covering over 30,000 journal titles, were written by European authors, followed by Chinese researchers at 25 per cent and American authors at 17 per cent.
Although the US seems to have published less papers, they’re more influential and cited more often than those written by European and Chinese researchers. Citations are calculated using the “field-weighted citation impact (FWCI)”, which measures the average number of citations received by AI authors in a particular region divided by the average number of citations by all AI authors.
The US had a score just under 2, whilst Europe hovered below 1.5 and China has been steadily rising going from under 0.5 to almost 1 over 1998 to 2016.
The most recent figures reveals that Chinese AI researchers are more interested in engineering and technology and agricultural sciences. The US and Europe, however, have shifted their focus to humanities and medical and health sciences.
“Overall, we see a continuation of last year’s main takeaway: AI activity is increasing nearly everywhere and technological performance is improving across the board,” the report states.
Public verses private
The national interest in the technology is often framed around a competitive spirit between the East versus the West but there's another factor. In the East it's governments that drive the research.
“In 2017, the proportion of corporate AI papers in the U.S. was 6.6x greater than the proportion of corporate AI papers in China,” according to the report. “The number of Chinese government-affiliated AI papers has more than doubled since 2009. There were 1.7x as many corporate AI papers in 2017 as there were in 2009.”
This is due to funding. In the US, development in AI is spearheaded by private companies that invest tens of billions of dollars to research and development. Amazon splashed out on $16.1bn and Alphabet - which includes companies like Google, Waymo and DeepMind, invested $13.6bn.
These amounts are larger in comparison to the amount spent by the US government. “To put this in perspective, the total budget for the National Science Foundation, together with DARPA and Department of Transport's investment in autonomous and unmanned systems totals just $5.3bn in the 2019 budget.”
The research coming out of China is increasingly being accepted into the top AI academic conferences like AAAI, NeurIPS, and ICML. There has been a surge in the number of students studying AI at Chinese universities - especially Tsinghua University.
The combined enrollment in its AI and machine learning course increased by 16x in 2017 compared to 2010. It’s the biggest jump compared to other international universities specialising in computer science in Israel, Canada, or Switzerland.
AI is being used across the world in pretty similar ways. North America, Europe, and other developing markets - including China are adopting for robotics and automation processes. The scores were also level for other areas of AI such as chatbots, speech understanding, and machine learning.
But countries from the developing markets stood out ahead for computer vision and autonomous vehicles. China is well known for aggressively rolling out AI cameras on its streets to catch jaywalkers (sometimes mistakenly) or in restaurants and airports.
Bit of a boy's club
Now for some slightly shocking statistics. Diversity in AI has always been pretty poor and the field has attempted to fix this and lots of different support groups like Women in Machine Learning (WiML), AI4ALL, and Black in AI have popped up to increase outreach. But it’s still heavily dominated by men.
On average, a whopping 80 per cent of AI professors are male in top universities across US and Europe. Applications for AI-related jobs in America were more likely to be made by men, as they made up about 71 per cent of the application pool.
It appears speech recognition roles are the most heavily skewed towards men - only just over a quarter of those applications were from women. The figures are a tiny bit more balanced for other areas such as computer vision, robotics, and deep learning.
It’s not all bad news, however. Workshops help by WiML aimed at supporting women in AI saw a 600 per cent increase in attendance from 2014 to 2018. AI4ALL, a non-profit focused on helping young people from less represented backgrounds break into AI, had an increase of 900 per cent for its educational program compared to 2015. ®