Table of contents
- Further analysis
- Top 30 regions leading AI research in 2020:
- Publication Index per million population for the top 50 countries leading AI research in 2020*:
- Academia vs Industry – Share of Total Publication Index:
- A word cloud created from the article title:
- Measuring competitiveness in AI research (based on the Herfindahl index):
- AI Ranking Transition from 2019 to 2020
- Changes in the top five countries leading AI research:
- Changes in the top five global institutions leading AI research:
- Trends in the top five US universities leading AI research:
- Top 5 universities in the world leading AI research:
- Top 5 global companies leading AI research:
- Can America continue to lead China?
- data set
- read more
Top 30 regions leading AI research in 2020:
1. USA – 1677.8
2. Europe* (EEA** + Switzerland + UK***) – 556.2
3. China – 281.2
4. Canada – 114.5
5. South Korea – 76.6
6. Japan – 57.8
7. Israel – 57.7
8. Australia – 47.6
9. Singapore – 30.1
10. India – 22.7
11. Russia – 19.2
12. Saudi Arabia – 10.2 13.
Taiwan – 5.9
14. Vietnam – 2.9 15. Brazil
– 2.8 16.
South Africa – 2.5 17.
UAE – 2.2
18. Iran – 1.7
19. Chile – 1.3
20. Turkey – 1.0
21. Pakistan – 0.9
22. Egypt – 0.3
23. North Macedonia – 0.3 24.
Thailand – 0.3
25. Barbados – 0.3
26. Qatar – 0.2
27. Malaysia – 0.2
*This article groups European countries that often share a common vision for AI research, but not necessarily in clear alignment.
**EEA includes Austria, Belgium, Bulgaria, Croatia, Republic of Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Iceland, Liechtenstein, Norway ( source ).
***Bundling the UK with European countries is not a political statement, but for clarity of explanation. After Brexit is complete, the UK may seek to remain a member of the EEA ( source ).
Publication Index per million population for the top 50 countries leading AI research in 2020*:
1. Switzerland – 10.113
2. Israel – 6.378
3. Singapore – 5.269
4. USA – 5.126
5. Canada – 3.046
6. UK – 2.409 7.
Australia – 1.876
8. Denmark – 1.769
9. Finland – 1.701
10. Sweden – 1.555
11. France – 1.535
12. South Korea – 1.482
13. Austria – 1.244
14. Germany – 1.100 15. Netherlands – 0.872 16.
Barbados – 0.871
17. Belgium – 0.776 18.
Portugal – 0.647
Luxembourg – 0.538
20. Japan – 0.458
21. Italy – 0.323
22. Saudi Arabia – 0.297
23. Greece – 0.260
24. Taiwan – 0.247 25.
United Arab Emirates – 0.228
26. Cyprus – 0.209
27. Norway – 0.207
28. China – 0.198
29. Czech Republic – 0.166
30. North Macedonia – 0.160
31. Russia – 0.133
32. Spain – 0.114
33. Poland –
0.105 34. Qatar – 0.071 35. Hungary – 0.068
Chile – 0.066
37. Romania – 0.052
38. South Africa – 0.043
39. Vietnam – 0.030
40. Iran – 0.020
41. India – 0.017
42. Brazil – 0.013
43. Turkey – 0.012
44. Malaysia – 0.006
45. Pakistan – 0.004
46. Thailand – 0.004
47. Egypt – 0.003
*Publication Index divided by country population in millions published by the World Bank ( source ).
Academia vs Industry – Share of Total Publication Index:
Academia – 78.9%
Industry – 21.1%
A word cloud created from the article title:
Measuring competitiveness in AI research (based on the Herfindahl index):
The Herfindahl Index (also known as the Herfindahl-Hirschmann Index) is a measure of the size of the industry’s participants and the degree of competition among them. The calculation formula is as follows.
The interpretation of the Herfindahl exponent is as follows
- If H is less than 100, it indicates a highly competitive industry.
- An H below 1,500 indicates that the industry is not concentrated.
- H between 1,500 and 2,500 indicates moderate concentration.
- H above 2,500 indicates a high concentration state.
The Herfindahl index can be calculated in two ways: country-specific and institution-specific. The former indicates whether a particular country is “monopolyed” in AI research, and the latter indicates whether it is “monopolyed” by which institution.
- The Herfindahl index by country is H=3,366, indicating a high level of industrial concentration . Since H=3,434 in 2019, AI research has become more competitive at the national level than last year.
- The Herfindahl index by institution is H=142, indicating that the industry is decentralized . With H=146 in 2019, AI research has become slightly more competitive at the institutional level than last year.
AI Ranking Transition from 2019 to 2020
Changes in the top five countries leading AI research:
1. United States – 1260.2
2. China – 184.5
3. United Kingdom – 126.1
4. France – 94.3
5. Canada – 80.3
1. United States – 1677.8
2. China – 281.2
3. United Kingdom – 161.0
4. Canada – 114.5
5. France – 102.9
Changes in the top five global institutions leading AI research:
1. Google (USA) – 167.3
2. Stanford University (USA) – 82.3
3. MIT (USA) – 69.8
4. Carnegie Mellon University (USA) – 67.7
5. UC Berkeley (USA) – 54.0
1. Google (USA) – 220.1
2. Stanford University (USA) – 106.1
3. MIT (USA) – 99.6
4. UC Berkeley (USA) – 86.7
5. Carnegie Mellon University (USA) – 71.3
Trends in the top five US universities leading AI research:
1. Stanford University – 82.3
2. MIT – 69.8
3. Carnegie Mellon University – 67.7
4. UC Berkeley – 54.0
5. Princeton University – 31.5
1. Stanford University – 106.1
2. MIT – 99.6
3. UC Berkeley – 86.7
4. Carnegie Mellon University – 71.3
5. Princeton University – 45.0
Top 5 universities in the world leading AI research:
1. Stanford University (USA) – 82.3
2. MIT (USA) – 69.8
3. Carnegie Mellon University (USA) – 67.7 4.
UC Berkeley (USA) – 54.0
5. University of Oxford (UK) – 37.7
1. Stanford University (USA) – 106.1
2. MIT (USA) – 99.6 3.
UC Berkeley (USA) – 86.7
4. Carnegie Mellon University (USA) – 71.3
5. University of Oxford (UK) – 51.9
Top 5 global companies leading AI research:
1. Google (US) – 167.3
2. Microsoft (US) – 51.9
3. Facebook (US) – 33.1 4.
IBM (US) – 25.8
5. Amazon (US) – 14.3
1. Google (US) – 220.1
2. Microsoft (US) – 66.5
3. Facebook (US) – 48.5
4. IBM (US) – 29.7
5. Huawei (China) – 14.3
As you can see, apart from Huawei overtaking Amazon and UC Berkeley overtaking Carnegie Mellon University in 2020, the top five list is fairly stable. But like Lewis Carroll’s Red Queen race, top institutions need to publish significantly more papers each year to stay on top.
To stay in this place, you have to run as hard as you can. And if you want to go somewhere else, you have to run twice as fast as you are now (Lewis Carroll)
Can America continue to lead China?
Today, a heated debate continues over the current state of the US-China strategic competition for AI dominance. As we strive to put a more balanced perspective on these debates, before we begin our analysis, let’s step through a bit of the history of the US-China AI competition (some of the historical accounts below will look familiar to regular readers of our AI research rankings).
In 2016, two important events happened in AI history. First, Google’s AlphaGo became the first computer program to beat Lee Sedol, a ninth-dan professional Go player without a handicap. Second, the Obama administration “prepared for the future of artificial intelligence . ” This is the announcement of a strategy regarding the future direction and considerations of AI. In China, these two events created a “Sputnik moment” that helped persuade the Chinese government to make artificial intelligence research and development a priority and dramatically increase funding. See AI Superpowers by Kai Fu Lee ).
In response, the Chinese Communist Party has set 2017 and 2030 as deadlines for its ambitious AI goals. The goals called for China to become the top AI economy by 2020, achieve large-scale new breakthroughs by 2025, and become the world leader in AI by 2030. This strategy has become known as the ” New Generation Artificial Intelligence Development Plan ” and has spurred billions of dollars of investment in AI research and development and AI policy promotion from ministries, provincial governments, private companies and others.
Certain think tanks like CNAS argued that China’s AI strategy merely reflected key principles of the Obama administration’s report. However, now that the Obama administration has ended, it is China, not the United States, that has adopted a large-scale AI strategy. This Chinese copying strategy is not new. Quoting Peter Thiel’s ” Zero to One ” (Japanese title ” What can you create from zero? ” NHK Publishing), “The Chinese are 19th-century railroads, 20th-century air conditioners, and even Even entire cities have obediently copied everything that worked in the developed world (just like wireless telephony prevailed without installing landlines), they may have skipped a few steps along the way. I don’t know, but I copied them all the same anyway.”
As a result of China’s collective efforts, the US’ superiority in AI is rapidly disappearing. In 2017, the US had an 11x lead over China ( source ), but in 2019 the US dropped to a 7x lead ( source ), and in 2020 the US has a 6x lead. (See the first part of this article). Moreover, according to this analysis from the Allen Institute for Artificial Intelligence , China is steadily increasing its share of authors in the top 10% of most-cited papers. all right.
Changes in the shares of the United States and China in the top 10% of AI papers in the world
It may be argued that America’s competitiveness in AI over the next decade will not be very good. However, the outcome of the competition depends on the interplay of the evolution of three key elements of modern AI: algorithms, hardware, and training data. considered necessary.
Based on decades of advances in computer science at world-class universities such as MIT, Stanford, Carnegie Mellon, and UC Berkeley, we believe America will hold a strong lead in AI algorithms for years to come. there is Additionally, companies such as Google and Facebook have published their internal research at AI conferences, creating a mature ecosystem where top AI researchers now move seamlessly between academia and industry.
In addition, the United States, home to Silicon Valley in the original definition of the word silicon, has been at the forefront of hardware innovation. Over the next five to 10 years, it will be extremely difficult for China to catch up with the United States in advanced microprocessor technology, especially given the huge patent portfolios protected by Intel, AMD, and NVIDIA. .
But US superiority has been questioned when it comes to the availability of training data. Access to data forms part of a broader debate about the conflict between privacy and public interest, with the United States tending to the former and China to the latter. In China, AI currently scans faces from hundreds of millions of street cameras , reads billions of WeChat messages , and analyzes millions of health records . all follow the “data-as-a-public-good” argument. The availability of this training data, combined with China’s population of 1.4 billion, creates a huge strategic advantage for the country.
Censoring text messages is relatively easy, but censoring images presumably uses image recognition . If it is not possible to immediately determine whether an image is subject to censorship by image recognition, the similarity with past images that have been subject to censorship is evaluated. In one instance of information censored in this way , Chinese researchers gave birth to genetically engineered twin girls .
Chinese tech giant Tencent, which develops and operates WeChat, is under “huge pressure” from the Chinese government to enforce censorship on the app. Despite being censored, daily life in modern China is impossible without the app .
It’s hard to draw any conclusions, but we believe that the first two factors (algorithms and hardware) will eclipse the last factor (data availability) and that the US will maintain its lead in AI for the next few years. there is However, such a slight advantage of the United States is the big harvest from the above analysis . , we need to allocate significant public and private funding to boost AI research at America’s top research universities.
In August 2020, the White House worried that the United States would lag behind rivals like China in AI and quantum computing research, saying these technologies would not only help economic development but also national security. In response to a number of policy advisers who warned, it announced a $1 billion investment to advance research into AI and quantum computing. However, we believe this number should be closer to the $1 trillion proposed by Mark Cuban in 2016 for investments in AI and robotics . Otherwise, the US risks losing its strategic advantage over China in AI, just as it lost it in high-speed rail and space exploration .
Investing in robotics is important, he said, because advances in robotics will automate jobs and put people out of work. In other words, if you become a robot maker, you can avoid the risk of unemployment, and by extension, you will have an industrial advantage over other countries. By the way, it is no secret that robotics is an important application field for AI .
In response to the above situation, the United States should continue to secure its superiority in outer space. It’s a step forward, said Zivitski. He also argues that while the coronavirus pandemic will make America’s finances tougher, cuts to the defense budget won’t help the economy recover .
It should be noted that even data science societies have yet to publish paper data in a Python-friendly format, so the 🤷♂️analysis this time was rather manual (i.e., first parse the HTML, then Correcting typos in first names, standardizing, splitting lines by authors in multiple institutions, summarizing in pivot tables, etc.). If you find a bug, please email us . If you want to download the dataset, you can find it here .
If you liked this post, you might also be interested in our analysis of ICML 2020 and NeurlLPS 2020 . Their analysis showed AI research rankings at each conference. If you want to check last year’s AI research ranking 2019, please click here .