The State of Artificial Intelligence in China

From January 26, 2019

There is a really interesting concept in psychology called the Johari Window and it suggests that we rarely see ourselves as we actually are. Not only that, but we think other people see us differently than they do. Maybe you’re not as charming as you think you are. Maybe that laughter after you told a joke was nervous laughter, but you thought you were hilarious. A key takeaway is that it’s rare for people to accurately describe themselves to others as they actually are.

That’s why when we saw that the leading university in China – Tsinghua University – put out a study on the state of AI in China, we were a bit suspicious. Perhaps they might sugarcoat things a bit? Stanford University put out a paper we wrote about in our article on The Global AI Race which questioned China’s claim of filing the most AI patents. Then we thought of the Johari Window. You can learn a lot about people from what they say about themselves, regardless of whether or not it’s true. It’s a psychological foray into what the Chinese think of their AI capabilities. So, without further ado, we give you a peek into China’s Johari Window as they describe the state of artificial intelligence in China and how they’re faring in the global AI race.

China AI Development Report 2018

The paper we’re going to talk about is titled “China AI Development Report 2018,” and it was produced by the “China Institute for Science and Technology Policy at Tsinghua University.” The purpose of the report is to offer “a multi-dimensional comparison between China and developed countries in AI development” while analyzing “China’s strengths and weaknesses and its position in the international AI competition landscape.” We read every single word of that 118-page report so you don’t have to, extracting all the interesting facts we could find and trying to juxtapose them in a way that gives further insights into China’s AI progress without boring you to tears or exceeding 1,900 words. Here goes.

Artificial Intelligence in China – R&D

Academic papers are a good proxy for a country’s research and development (R&D) efforts into a particular area of research, and China’s AI papers as a percentage of the global total increased from 4.26% in 1997 to 27.68% in 2017, far ahead of other countries. It’s not corporations or government but universities that contributed the vast majority of AI papers globally, with 87 of the top 100 AI research institutions in the world being universities. Below you can see the trend of AI paper output over the years for the top-8 contributing countries (the blue line is China and the orange is the U.S.)

China is putting out an increasing amount of high-quality AI papers

China is putting out an increasing amount of high-quality AI papers – Source: Tsinghua University

We all know that quality is better than quantity, so in order to measure the quality of this output, they looked at two measures to determine how often papers were being cited.

Highly cited AI papers and hot AI papers

Highly cited AI papers and hot AI papers – Source: Tsinghua University

By both measures, China is ahead with the U.S. trailing not far behind.

When it comes to AI papers being produced by corporations, China doesn’t fare so well. Three companies on the list are way out in front: IBM, Microsoft, and Siemens (in that order) who collectively produce more papers than the next 11 companies combined. The report lists the top-20 companies with only one being Chinese (State Grid Corporation of China).

When it comes to companies that are being issued the most AI patents, we see IBM, Microsoft, and Samsung leading the pack with State Grid Corporation trailing behind in fourth place. (SGCC is a Chinese power company, and their AI-related technological inventions focus on fields such as grid control, power distribution and utilization networks, wind farms, and green energy with an upcoming interest in robotics.) Incredibly, SGCC is said to hold nearly 8x as many AI patents as Baidu, the second biggest holder of AI patents in China.

Again, not all patents are equal. In terms of patent maintenance, SGCC has the highest percentage (87%) of valid patents in its total published patents among the top ten assignees. In contrast, as many as 40% of Sony’s patents in AI have expired, says the report. It then goes on to make the controversial statement that China has more AI patents than the U.S. if we add in the 52% of all AI patents that originate from academia. China, the United States, and Japan are said to collectively hold 74% of all AI patents.

The War for AI Talent

Compared to the Stanford report, this section was a breath of fresh air as it didn’t drone on about equality of outcome based on gender, a divisive doctrine that has now led to legally mandated discrimination in the State of California, something which will only serve to destroy shareholder value. Society has thrived thus far because we have organizational hierarchies in place that – generally speaking – reward competence, something that is best measured objectively. To accomplish that, the report adopts “the h-index,” something that is widely recognized in the academic community as an indicator of research ability. This index is used to produce the “top 10% of international AI researchers” who are considered to be “top international AI talent.”

Top AI talent as a percentage of the total AI talent is low in China

Source: Tsinghua University

In the above chart, you can also see each country’s “total AI talent” which reflects researchers that have “issued patents and/or published English papers” within the last ten years. While China might have a lot of these types, only 5.4% of them would be considered “top talent.” On the other hand, just over 18% of all total AI talent in the U.S. would be considered “top talent,” something that confirms the findings of the Stanford report. China shouldn’t feel bad though, as India fared far worse. While not reflected in the above chart, India has the third biggest AI talent pool globally – almost equaling that of China – but less than 3% would be considered “top talent.” If you want to hang around the world’s top AI researchers but you don’t like Chinese food, looks like you ought to head over to Italy or France, the latter of which provides more options.

The world's top AI research institutions

Source: Tsinghua University

China’s AI Industry Development

As of June 2018, there were 4,925 AI enterprises worldwide that have AI technologies or products as their core operations. Of those, 1,011 were Chinese (20.5% of total), and 2,028 were American (41.2% of total). Beijing has the greatest number of AI enterprises in the world (395) followed by San Francisco (287) and London (274). When it comes to the type of AI enterprises the Chinese have, they seem to be focused on the senses – hearing, seeing, and speaking. 

Structure of the Chinese AI market

See no evil, hear no evil, speak no evil. At least in theory. Source: Tsinghua University

When it comes to the application of these technologies, the report says “mature AI devices are not many,” and only three AI products would be considered “fairly mature” — smart speakers, smart robots, and drones.

Smart Speakers

While many people think that the Chinese are enacting a sinister plan to spy on everyone with tiny chips, Google is going all Trojan Horse on them by simply dominating the smart speaker space with a 36.9% market share. There’s no Huawei the Chinese saw that one coming, and not far behind is Amazon who is listening in on a lot of conversations with a 27.7% market share. Only then does China make an appearance with Ali Baba coming in at an 11.8% share of the smart speaker market.

Industrial Robots

Now, this next part about industrial robots is where things get a bit confusing, as it sends mixed messages about what these numbers mean. The below chart is – we assume – meant to reflect the total number of industrial robots acquired by each country:

Industrial robot shipments in major markets

Source: Tsinghua University

We’re going to just drop the accompanying text right here so you can have a look for yourselves:

The market posted 380,000 industrial robots sold in 2017, up 29% y/y. China has been the world’s largest industrial robot market since 2013. In 2017, China posted 138,000 industrial robots in sales, followed by Korea with approximately 40,000 and Japan with approximately 38,000.  In the Americas, the United States is the largest single market, selling approximately 33,000 industrial robots in 2017.

That last sentence is what’s confusing. Are all these numbers referring to the robots each country has soldduring 2017 or purchased? While some of our MBAs probably puff the magic dragon more often than they should, we’ve always been under the assumption that China imports most of their industrial robots, and that’s what the China Daily thinks as well:

China purchased 141,000 industrial robots in 2017, up 58.1 per cent year-on-year, but foreign brands accounted for nearly three-quarters of that. Among industrial robots, 37,825 were domestically manufactured, up 29.8 per cent year-on-year.

Anyways, maybe some of this was lost in translation but it looks like in 2017, China bought more robots than Korea, Japan, ‘Murica, and zee Germans combined.

Drones

DJI is the most influential drone manufacturer in China with more than $2.5 billion in revenues for 2017 and a commanding 50% market share in the North American drone market. Other Chinese drone manufacturers like EHANG, Zero Zero Robotics, Zerotech, and XAG have also become quite influential. One example of drone applications the report cites is in infrastructure inspections. Guangdong Power Grid inspects over 111,846 miles (180,000 km) of power lines annually, a distance that’s equivalent to 4.5 times the earth’s circumference. Around 85% of that work is now performed by drones, representing the largest drone inspection workload in the world, and increasing overall inspection efficiency by 2.6 times.

AI in China – Policy and Public Perception

The last part of the report dives into AI strategies and AI policies of which the Chinese have plenty. According to the report, China has over 1,000 AI policy documents, compared to other countries and regions that haven’t done all that much – relatively speaking – like the U.S. (9 AI policies), Germany (5 AI policies), the U.K. (4 AI policies), or the entire European Union (5 AI policies). AI has also become a core part of China’s “Deep Blue” program geared to safeguard national security, though some developments like the Chinese social credit score are raising some eyebrows in the international community. Internally, nobody seems to mind much. AI is a popular science topic among ordinary people, 53% of whom support “comprehensive AI development” which is exactly what’s happening. People learn about AI from the news, with older people (30 plus) and men showing significantly more interest in the topic:

AI awareness in China

Source: Tsinghua University

Conclusions

We’ve talked before about how China is starting to kick America’s ass in some areas of technology, and it’s been interesting to see tech media progress through the phases of grief as they come to realize that China may just start eating everyone’s lunch. First, there was denial, and then one short year later, acceptance. In order to compete with China, the world needs to first understand it, and it’s a very difficult place to navigate if you’re wearing your ethnocentric sunglasses. Chinese people have an incredibly strong work ethic, a habit that’s been nicknamed 9-9-6 (9 to 6, 6 days a week), and that’s probably why 63% of self-made billionaires are Chinese women. For investors around the world looking in, it’s even harder to figure out when trying to apply traditional notions of risk vs. reward. The Chinese have a high tolerance for riskand they generally love to gamble. And it appears that they’re betting a large stack of chips on the future of artificial intelligence in China.

Collected at: https://www.nanalyze.com/2019/01/artificial-intelligence-china/