news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

New Intelligence Report

Editor: Qiao Yang is so sleepy

【New Wisdom Introduction】Nature recently reported on a newly released AI industry database PARAT. Analysis showed that American and Chinese companies dominated in terms of paper citations, patent applications, etc.

In the field of AI, Silicon Valley giants Alphabet (Google's parent company) and Microsoft have the highest number of paper citations, far exceeding other companies.

However, Chinese companies Baidu and Tencent lead in patents.

These results come from the PARAT database published by the Emerging Technologies Observatory (ETO), a tool that tracks AI-related activities in the private sector, including the publication of research and patents, as well as talent indicators, to help us gain insight into AI development trends.

https://parat.eto.tech/?sort=highly_cited_ai_pubs-desc&zz_columns=name,country,ai_pubs,highly_cited_ai_pubs,ai_pubs_top_conf,ai_patents,ai_jobs,tt1_jobs

ETO is affiliated with the Center for Security and Emerging Technologies (CSET), which is headquartered at the School of Foreign Service at Georgetown University and is a think tank dedicated to policy analysis in the field of AI.

Ngor Luong, a senior analyst at CSET, said that because cutting-edge research in the field of AI is conducted as often in industry as in universities, it is important to monitor commercial activities.

AI Giants

Zachary Arnold, chief analyst at the Observatory, said that PARAT data proves that China's large companies are very competitive in the field of AI.

When sorted by the number of highly cited AI papers and preprints, three Chinese tech giants—Tencent, Alibaba, and Huawei—are all in the top ten.

Arnold said, “There is still a prejudice among many people that China is large and can produce a lot of results, but they are not truly first-class.”

However, the ETO calculates a variety of quality-adjusted metrics on which Chinese companies "excel".

According to PARAT data, the most cited paper in all AI research is "Attention Is All You Need", published in 2017.

According to Google Scholar, this paper has been cited more than 127,000 times.

The author team of this paper mainly comes from Google, and proposed the Transformer architecture to achieve "grand unification".

Chinese institutions have also produced a lot of high-quality research. For example, this paper co-authored by CUHK, Tencent, SenseTime and other institutions proposed a real-time semantic segmentation method ICNet, which has been cited more than 1.7k times.

Paper address: https://arxiv.org/pdf/1704.08545

In the past decade, only three of the top 10 companies that submitted the most AI patents were located in the United States, while the rest were located in China, Germany and South Korea.

Top Employers

The data also highlights the diversity of the AI ​​industry.

In addition to the five giants - Alphabet, Amazon, Apple, Meta, and Microsoft, there are many companies in the "long tail" that also have high citations.

Among the more well-known companies are OpenAI and Apple, as well as Disney and Japan's Mitsubishi, which are less well-known for their AI innovations.

Luong noted that the database collects paper and patent data only up to the end of 2023, so it misses more recent developments.

Arnold said other indicators in PARAT reveal AI activities that are sometimes overlooked.

For example, PARAT covers the number of AI positions in each company and is collected from the social media platform LinkedIn.

The data is most accurate for U.S. companies, which allows us to see which companies are attracting the most AI talent.

By this metric, Amazon tops the list with 14,000 AI jobs, but is closely followed by multinational consulting firm Accenture. According to Arnold, large consulting firms are now "mercenaries" for other companies and government AI projects.

This data allows us to examine the company's AI efforts from multiple perspectives.

“We see a lot of discussion about ‘who is leading in AI?’ We are data fans, and we know there are many different types of data that can be used to answer this question, and they don’t always point in the exact same direction.”

References:

https://www.nature.com/articles/d41586-024-02515-1