News by CSET

policy․ai
Biweekly updates on artificial intelligence, emerging technology and security policy. View at policy.ai or download a pdf. Forward to a friend, or sign up here.

 

Wednesday, September 16, 2020
by REBECCA KAGAN

 

Worth Knowing

Nvidia To Acquire Arm for $40B: On Sunday, U.S. GPU designer Nvidia announced plans to acquire U.K. chip company Arm from current owner SoftBank for $40 billion. Arm develops and licenses designs that are used to produce chips, including those inside most smartphones — an area where Nvidia is not currently a major player. Nvidia CEO Jensen Huang said that together, the two companies will create the “leading computing company for the age of AI.” Nvidia will maintain Arm’s open-licensing model, letting Arm continue to serve customers around the world, though some authorities worry the deal would effectively place Arm under U.S. control amid U.S.-China tensions. Assuming regulatory approval, the purchase should take 18 months to complete; if finalized, it would be the largest-ever semiconductor deal.

Portland, OR Enacts Strictest Facial Recognition Ban in the US: The Portland City Council unanimously voted last Wednesday to ban both private and government use of facial recognition due to privacy concerns as well as gender and racial inaccuracies. The strictest ordinance of its kind in the nation to date, it prohibits companies from using the technology in any “place of public accommodation” where goods or services are offered. The prohibition goes into effect on January 1, 2021. Portland Mayor Ted Wheeler said he hopes it encourages other cities to enact more stringent legislation as well.

Chinese Tech Executives Launch Domestic Tech Fund: A group of Chinese tech executives plans to launch a “domestic replacement” fund to support Chinese tech companies focusing on AI, 5G and semiconductors, Reuters reports. The fund aims to assist companies affected by U.S. sanctions and create China’s next tech giant. Venture capital firm China Europe Capital  expects to raise 5 billion yuan ($730 million) with the help of former executives from Huawei and chipmaker SMIC. China Europe Capital Chairman Zhang Jun said the group sees the U.S.-China tech war as “a matter of life and death,” and that it hopes to provide blacklisted companies with extra support to “survive, and run, although with a limp.”

Machine Learning Spotlight — Standards For Clinical Trials Involving AI: A consortium of experts has proposed the first international standards for clinical trials involving AI. The guidelines were published simultaneously in leading medical journals BMJ, Nature Medicine and Lancet Digital Health after being developed with input from more than 150 authorities. The authors developed a list of 14 AI-specific additions to the standard journal protocols for reporting on clinical trials, including describing the type of algorithm, the training data and the data analysis process. AI is a “strong driver” of new medical interventions, the authors say, and they hope the standards will help editors and readers accurately appraise the quality of new interventions.

Government Updates

Esper Announces AI Partnership for Defense: The Joint AI Center will launch the AI Partnership for Defense this week in collaboration with allies, Secretary of Defense Mark Esper said at the Defense Department’s AI Symposium and Exposition. Formed with at least 10 other nations’ militaries and defense departments, the partnership will focus on data sharing, ensuring interoperability of AI systems and incorporating ethics into the AI pipeline. Esper described the partnership as part of a democratic alternative to China and Russia’s authoritarian approach to AI. In the same talk, Esper highlighted two AI training programs for DOD staff and announced that a real-world dogfight between AI and human pilots will occur in 2024.

Groen Nomination To Be JAIC Director Clears Major Senate Hurdle: On Tuesday, the Senate Armed Services Committee advanced the nomination of Maj. Gen. Michael Groen to be Director of the Joint AI Center. He now awaits Senate confirmation. Groen has served in the Marine Corps for more than 30 years and is currently the deputy chief of computer network operations at the National Security Agency. If appointed, he would be promoted to lieutenant general, as the JAIC previously announced the position would be filled by a three-star general or flag officer.

Trump Administration Reportedly Considers Blacklisting SMIC: According to Reuters, the Trump administration is considering blacklisting Chinese chipmaker Semiconductor Manufacturing International Corporation over concerns about its relationship with China’s military. SMIC said in a statement it was in “complete shock and perplexity” about the possible blacklisting, claiming it complies with all U.S. regulations and has no relationship with the Chinese military. If enacted, SMIC’s addition to the Entity List would require U.S. government approval of American suppliers shipping to the company. SMIC shares fell 23 percent on the news.

House Approves Bill Authorizing AI Center of Excellence: On a Monday voice vote, the House passed legislation codifying an AI Center of Excellence within the General Services Administration. The AI in Government Act of 2020, introduced by Rep. McNerney and then-Rep. Meadows, tasks the Center of Excellence with improving “cohesion and competency” of government AI use, convening stakeholders to discuss trends in AI and publishing relevant information online. The Center is also intended to provide agencies with advice on acquisition and use of AI and advise the Director of the Office of Management and Budget on developing policies related to the use of AI. The Senate companion bill was approved by committee but has not yet received a floor vote.

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

AI Engineer and Technician Employment Prospects: A New Occupation — An Analytical Report on Current Employment Prospects for AI Engineers and Technicians. This government report analyzes the AI industry in China. It finds that demand for AI professionals continues to exceed supply and that the U.S. job market attracts the majority of highly qualified Chinese AI researchers.

Technical Domains for Made in China 2025: Roadmap of Major Technical Domains for Made in China 2025. A partial translation of a 2015 roadmap for the Made in China 2025 strategy, this section focuses on China’s IT industry, including specific targets for the country to meet by 2025 in the semiconductor, telecom, software and intelligent manufacturing equipment industries. The goals are intended to improve China’s self-sufficiency in these strategic technologies.

Policies To Promote Circuit and Software Industries: State Council Notice on the Publication of Certain Policies To Promote the High-Quality Development of the Integrated Circuit Industry and the Software Industry in the New Period. This is a list of policies that the Chinese government has enacted to spur the growth of the integrated circuit and software industries. The policies provide tax breaks to companies in these industries and encourage investment.

What We’re Reading

Report: China’s Artificial Intelligence Technology Industry Development Under New Challenges and Opportunities, China’s New Generation Artificial Intelligence Technology Industry Development Report (June 2020)

Report: Brain-Computer Interfaces: U.S. Military Applications and Implications, An Initial Assessment, RAND (August 2020)

Commentary: Deepfakes Are Coming for American Democracy. Here’s How We Can Prepare, Clint Watts and Tim Hwang (September 2020)

Report: Facial Recognition: CBP and TSA are Taking Steps To Implement Programs, but CBP Should Address Privacy and System Performance Issues, United States Government Accountability Office (September 2020)

What’s New at CSET

REPORTS

PUBLICATIONS AND PODCASTS

FORECASTS
CSET has launched a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:

CSET Experts in the News

 

Our researchers have been making a splash in the media this summer. From assessing the threats presented by deepfakes to working with U.S. lawmakers on devising a national security strategy for AI, CSET experts are weighing in on some of the most pressing challenges at the nexus of security and emerging technology.

Wired

CSET partnered with the Bipartisan Policy Center and Representatives Robin Kelly (D-IL) and Will Hurd (R-TX) to propose national security considerations for a U.S. AI strategy. CSET’s Helen Toner and Jason Matheny spearheaded the collaboration with the BPC and worked with government officials, industry representatives, civil society advocates and academics to better understand the major AI-related national and economic security issues the country faces. The bipartisan effort was covered in Wired, among other publications. Wired also interviewed CSET’s Tim Hwang about “Deepfakes: A Grounded Threat Assessment.” When it comes to this technology, Hwang said, “a lot of discussion about the threat has been driven by dramatic anecdotes.”

Axios

U.S. competitiveness in artificial intelligence hinges both on building a robust workforce and preventing technology transfer by China — two significant lines of CSET research. Over the past month, Axios has highlighted each of these factors in its newsletters. Axios Future covered a data brief by CSET’s Zachary Arnold on Canadian immigration, writing “the U.S. risks losing its long-standing leadership in the tech sector as restrictive laws and a hostile political climate causes highly skilled immigrants to leave for more welcoming countries.” Axios China, in turn, offered readers an overview of an issue brief by CSET’s Ryan Fedasiuk: “This new report looks at Chinese government funding provided to Chinese students studying in the U.S., putting this program in the context of Beijing’s attempts to increase its control and influence abroad.”

The Algorithm

The Algorithm, an MIT Technology Review newsletter, covered “Messier than Oil,” an issue brief by CSET’s Husanjot Chahal that analyzes the challenges in assessing whether the United States or China has a “data advantage” in the military AI realm. The newsletter summarized the paper’s findings this way: “Assessing China’s military AI advantage is messier than it seems. The country’s access to commercial market data is unlikely to confer a military operational advantage.” Chahal’s brief and the CSET/BPC collaborative effort also topped the “Notable Reports” list on the Center for Strategic and International Studies’ Tech Update newsletter. Additionally, The Algorithm highlighted Ryan Fedasiuk’s issue brief on government funding for Chinese students.

Defense News

In his first public appearance in the role, Acting Under Secretary of Defense for Research and Engineering Michael Kratsios joined Jason Matheny to discuss U.S. technological leadership in an increasingly competitive era. The Defense News writeup highlighted Kratsios’ plans to improve collaboration across the Defense Department: “What I’ve learned is that in order to get the most out of the federal government’s technology ecosystem to drive innovation … you need to be better coordinated across all aspects of the ecosystem.” Wired emphasized the Under Secretary’s vision for U.S. military use of emerging technologies to counter China. Additionally, Inside DefenseMeriTalk and OneZero covered the event.

Politico

In order to understand how the United States and its allies can confront the challenge posed by China, Politico China Watcher spoke with several experts, including CSET’s Tarun Chhabra, who observed that shaping the democratic digital world and infrastructure offers a key area of collaboration. “We need a technology alliance agenda [including] pooled R&D investments; better coordinated industrial policy and antitrust regulation; privacy-preserving data-sharing; energetic norm-building and technological standard-setting; and tailored, coordinated technology transfer restrictions, investment controls and export controls.”

NBC News

Washington’s efforts to rein in technology transfer have affected U.S.-based Chinese students and researchers in unanticipated ways. CSET’s Emily Weinstein and Dahlia Peterson analyzed Chinese reactions to a White House proclamation banning graduate students and researchers affiliated in the past or present with entities supporting China’s military-civil fusion. NBC News cited their research, which offered a list of 13 entities likely to be targeted by the ban and found a “critical but relatively muted” response by the Chinese government to the U.S. visa suspensions.

Highlighted CSET research includes:

News by CSET

Worth Knowing

Intel Considers Outsourcing More Chip Manufacturing Amid Delays: On July 23, Intel CEO Bob Swan announced the company was considering contracting out some of its chip manufacturing given delays with its 7-nanometer chip. Outsourcing cutting-edge chips marks a significant change: Intel, the world’s largest semiconductor firm, is known for both designing and manufacturing its own chips. The company’s stock fell nine percent on the news, while U.S.-based competitor Advanced Micro Devices’ stock rose six percent. Taiwan Semiconductor Manufacturing Company, the world’s largest contract chipmaker, is a likely contender for the outsourced work and already manufactures AMD’s chips; TSMC’s shares rose 9.5 percent.

New ML Training Benchmarks Set, Led by Nvidia: Industry consortium MLPerf released the results of its third annual competition in machine learning training, with records set by Nvidia and Google. MLPerf measured the time it takes hardware to train one of eight machine learning models — for tasks including image classification, translation and playing Go — to a specific level of performance. Among commercially available systems, Nvidia set 16 records using its DGX SuperPOD supercomputer and A100 GPU; Google led among non-commercially available systems with its newest TPU v4. The gains are due in part to chip improvements, as well as larger systems using more chips in parallel. The second MLPerf benchmark, which measures how quickly trained models can make predictions, will be published later this year.

New Zealand Publishes Multi-Agency Algorithm Charter: New Zealand Minister of Statistics James Shaw launched an Algorithm Charter to guide government agencies. So far, 25 agencies have signed on, including the New Zealand Defence Force and the Ministry of Justice. The charter requires its signatories to have plain-English explanations of their algorithms, plans to mitigate bias and human rights violations, consultations with communities affected by the algorithms and channels for appealing algorithm-informed decisions. It also includes a risk matrix to evaluate algorithmic impact and must be applied to all high-risk processes within 12 months. Shaw believes this document is the first time a country has provided standards to guide its entire government’s use of algorithms.

Investments in AI Startups Decline in Q2: According to CB Insights, global venture capital investment in AI-related startups declined in the second quarter of 2020 amid the economic downturn, although the tech sector may still be outperforming the global economy. The number of investments in AI companies dropped to a three-year low of 458 in Q2; total funds raised by AI startups also slipped from $8.4 billion in Q1 to $7.2 billion in Q2. The quarter had relatively fewer seed funding rounds, with a greater portion of investments going to mature companies. Global VC investment across sectors shows a similar decline. The tech sector’s economic indicators outperformed many other sectors in Q2, with unemployment rates continuing to decline and the largest tech companies reporting significant profits.

Government Updates

NSCAI Releases Second Quarter Recommendations: The National Security Commission on Artificial Intelligence published its Second Quarter Recommendations on July 22. It proposed establishing an accredited university within the federal government to meet the government’s need for digital expertise and creating a National Reserve Digital Corps, modeled after the military reserves, as a part-time service option for private sector talent. In total, the NSCAI proposed 35 recommendations across six areas including advancing the DOD’s internal AI R&D capabilities, accelerating the application of AI and expanding export controls and investment screenings. The report also included key considerations for responsibly developing and fielding AI.

State Department Loosens Restrictions on Drone Exports: The United States has eliminated the blanket denial of exports of some types of drones, the State Department announced on July 24. The policy shift is a reinterpretation of the 33-year-old Missile Technology Control Regime, an arms control pact with 35 nations that includes a “strong presumption of denial” for international sales of Unmanned Aerial Systems capable of carrying 500 kilograms for more than 300 kilometers. Critics of the decision, including Senate Foreign Relations Committee Ranking Member Menendez, said the move undermines the MTCR and increases the likelihood of U.S. weapons being exported to human rights abusers. Proponents of the decision argued that the status quo allowed China — which does not participate in the MTCR — to capture a large part of the international drone market.

House Intelligence Authorization Act Progresses With AI Provisions: The House Permanent Select Committee on Intelligence approved the FY21 Intelligence Authorization Act in an 11-8 vote on July 31. The bill includes several sections on AI, one of which emphasizes the value of consolidating AI efforts across the Intelligence Community and tasks the Director of National Intelligence with identifying and developing plans for AI projects that advance the mission of the IC. The IAA also directs the Intelligence Advanced Research Projects Activity to award grants and contracts that encourage microelectronics research. Other provisions require improvements to STEAM education and place the Director of Science and Technology directly under the Director of National Intelligence.

ODNI Releases AI Ethics Principles and Framework for the IC: On July 24, the Office of the Director of National Intelligence published Principles of AI Ethics and an AI Ethics Framework outlining norms to guide the Intelligence Community’s use of AI. The principles call for AI’s use to be legal, transparent, objective, human-centered, secure and informed by science. The corresponding framework offers a series of questions to help implement the principles and to document relevant considerations involved in procuring, designing and using AI.

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

Lessons From U.S. Military-Civil Fusion: Characteristics of, and Lessons From, the U.S. Legal System for Military-Civil Fusion. This 2018 article by a Chinese state think tank praises the U.S. integration of the military and civilian industrial bases as a model for China. The article argues that China can learn much about “military-civil fusion” from U.S. legislation on this issue.

PRC Export Control Law: Export Control Law of the People’s Republic of China (Draft) (Second Version). This draft export control bill was being considered by China’s parliament, the National People’s Congress, as of July 2020. The bill limits exports of dual-use items, military equipment, nuclear materials and other goods of counterproliferation concern. It also sets penalties for Chinese exporters who violate the provisions of the bill.

What We’re Reading

Report: Face Recognition Accuracy With Masks Using Pre-COVID-19 Algorithms, National Institute of Standards and Technology (July 2020)

Report: Emerging Military Technologies: Background and Issues for Congress, Congressional Research Service (July 2020)

Paper: Towards a New Generation of Artificial Intelligence in China, Fei Wu et al, Nature (June 2020)

What’s New at CSET

REPORTS

PUBLICATIONS

FORECASTS
CSET has launched a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:

IN THE NEWS