China sets new R&D targets, Biden’s interim strategic guidance touts AI, and a bipartisan Senate proposal for an international tech partnership

Sign up
Worth Knowing

Beijing’s Plans for R&D, AI and Chips: China’s annual legislative sessions — the Chinese People’s Political Consultative Conference and the gathering of the National People’s Congress — convened earlier this month, bringing with them important implications for AI and the semiconductor industry. Headlining the sessions was China’s 14th Five Year Plan, which was released earlier this month and approved last week. The plan (a draft of which CSET translated in December) lists AI, quantum computing and integrated circuits among the areas critical to China’s national security. On March 5, Chinese Premier Li Keqiang announced a target of 7% annual growth in R&D spending over the next five years, higher than China’s GDP growth target and its budget increase for the military — 6% and 6.8%, respectively — but a significant scaling-back from the years of double-digit growth the country had maintained. The China Development Bank, meanwhile, announced its intention to offer $60 billion in loans for firms working in strategic technologies, plus plans to invest an additional $30 billion in microchip development. Some observers noted that this year’s proposals hinted at the failure of the 13th Five Year Plan’s “Made in China 2025” campaign, which had set ambitious targets for domestic semiconductor manufacturing. With the country still producing only 16% of its chips domestically, modest increases in R&D spending and investment are unlikely to springboard China to its 70% target by 2025.
Stanford Publishes Yearly AI Report: Earlier this month, Stanford University’s Institute for Human-Centered Artificial Intelligence released its 2021 AI Index. Now in its fourth edition, the yearly report aims to provide a comprehensive account of the state of AI research and development, education, policy, investment and more. Among its findings:
  • Chinese AI research continues to grow. After overtaking the United States in total AI journal publications in 2017, China took the top spot for total number of AI article citations in 2020. While this finding has received some public attention, observers noted that total citations is a poor proxy for quality (see also here) and U.S. and European publications maintain consistent leads in weighted measures of citation impact.  
  • AI scholars are flocking to private industry. The percentages of new North American AI PhD graduates joining academia and industry were roughly equal in 2010, but have diverged dramatically over the past decade. In 2019 — the most recent year for which data was available — 65.7% of graduates went into private industry, compared to 23.7% who remained in academia.  
  • The field of AI still struggles with diversity. In 2019, only 2.4% of U.S. AI PhD graduates were Black, and only 3.2% were Hispanic. Meanwhile, women made up 22.1% of North American AI PhD graduates in 2019, a proportion that has changed little over the last decade.
  • Despite increased investment overall — up to $40 billion in 2020, an increase of 9.3% compared to 2019 — the number of newly funded companies fell for the third year in a row. After peaking above 4,000 in 2017, fewer than 1,000 new AI companies received funding in 2020.
  • The United States continues to receive the lion’s share of private AI investment, drawing $23.6 billion compared to second-place China’s $9.9 billion and the third-place U.K.’s $1.9 billion.
  • More: Global AI Vibrancy Tool | Measurement in AI Policy: Opportunities and Challenges 
OpenAI Finds Artificial Neurons That Work Like the Real Thing: Researchers at OpenAI announced they had discovered human-like “multimodal neurons” — neurons associated with an abstract concept rather a specific visual cue — in their general-purpose computer vision system “CLIP.” Fifteen years ago, researchers discovered that the human brain would fire the same brain cell or cluster of brain cells in response to related inputs — seeing a picture of a famous actress or reading her name would fire the same cells, despite being visually dissimilar. Earlier this month, OpenAI — the AI lab behind the natural language generator GPT-3 —  revealed that its CLIP system works similarly, resulting in an ability to associate photos, drawings or text descriptions that relate to a concept, even if it wasn’t trained on those specific examples. While the system’s ability to “think” multimodally is intriguing, it is not without hiccups. CLIP was easily fooled by simple adversarial attacks — putting a note reading “iPod” on a piece of fruit, for example, led the system to confidently identify the image as a picture of the MP3 player. 
Government Updates

White House Issues Interim National Security Strategic Guidance: The Biden administration has issued its interim national security strategic guidance, naming artificial intelligence as part of “a revolution in technology that poses both peril and promise” and emphasizing the necessity of maintaining the United States’ technological advantage. The interim guidance — meant to help federal agencies align their priorities with the White House before it releases its full national security strategy later this year — promises a shift in both emphasis and funding from “unneeded legacy platforms” to new technologies; a streamlined development, testing and acquisition process; changes in training; and rules and normative frameworks to ensure the new technologies are “used responsibly.” Both the guidance and Secretary of State Antony Blinken’s speech introducing it emphasized plans to work with allies and partners to shape the rules around emerging technologies. While the White House has not yet given an intended publication date for its full National Security Strategy, Secretary Blinken said it would be finalized “over the next several months.”

U.S. Court Reverses Xiaomi Ban: Last Friday, a U.S. District Court granted a preliminary injunction against restrictions imposed on the Chinese consumer electronics maker Xiaomi that would have prevented all U.S. investment in the company. In the final days of the Trump administration, the Department of Defense added a number of companies, including Xiaomi, to a list of “Communist Chinese military companies” (CCMC), subjecting them to President Trump’s November executive order that banned U.S. investment in Chinese military-linked entities. Xiaomi’s restrictions were set to take effect Monday. In granting the injunction, Judge Rudolph Contreras of the U.S. District Court for the District of Columbia criticized the Pentagon’s explanation of its decision, writing that the department had failed to provide the “required rational connection—or any connection” between “the facts found and the choice made.” There is speculation that the decision could open the door for other CCMC-designated companies, such as Huawei Technologies Co. and Semiconductor Manufacturing International Corporation (SMIC), to challenge their own Trump administration-imposed restrictions.  

Senators Introduce Bill to Create Tech Partnership Among Democracies: On March 4, a bipartisan group of Senators introduced the Democracy Technology Partnership Act. The bill aims to confront China’s technological influence by creating a partnership of democratic countries that would set international standards and coordinate export controls on critical technologies, including artificial intelligence. Sens. Bennet, Cornyn, Menendez, Rubio, Sasse, Schumer and Young joined lead sponsor Sen. Warner in co-sponsoring the bill, which would establish an International Technology Partnership Office at the State Department — headed by a Special Ambassador for Technology and staffed by officials from various agencies — tasked with building the “technology-based partnership of democratic countries.” The act would also establish a $5 billion fund for supporting joint research projects and investing in technology in third-country markets.

In Translation
CSET’s translations of significant foreign language documents on AI

2018 PRC Central Committee Reorganization Plan: CCP Central Committee Publishes Plan for Deepening the Reform of Party and State Agencies. This document is the final, authoritative version of China’s 2018 Party and state agency reorganization plan. It continues and expands upon the Xi Jinping-era trend of merging Party and government agencies with similar functions and creates a number of new ministries. This final reorganization plan is much more modest in scope than the plan the PRC government originally proposed, indicating that the PRC leadership decided the original plan was too ambitious.

What We’re Reading

Article: He got Facebook hooked on AI. Now he can’t fix its misinformation addiction, Karen Hao, MIT Technology Review (March 2021)

Report: Where are the women? Mapping the gender job gap in AI, Erin Young, Judy Wajcman and Laila Sprejer, The Alan Turing Institute (March 2021)

Report: Military-Civil Fusion Terminology: A Reference Guide, Alex Stone, China Aerospace Studies Institute (March 2021)

What’s New at CSET

REPORTSPUBLICATIONS AND EXPERT TESTIMONYFORECASTS
CSET has launched a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:
Interested in supporting Foretell? Apply to be a Foretell Ambassador (Remote, Part-Time Position): CSET is looking for an Ambassador to engage a virtual team of at least 10 active forecasters per month on Foretell.

IN THE NEWS
Upcoming Events


What else is going on? Suggest stories, documents to translate & upcoming events here.