Facial recognition pushback grows, US joins G7 AI group, and White House proclamation restricts Chinese researchers

Sign Up
Worth Knowing

Facial Recognition Pushback Grows: Demonstrations against police violence have invigorated discussion and action around law enforcement’s use of facial recognition technology. IBM announced Monday that it no longer offers facial recognition or analysis software and opposes using facial recognition for mass surveillance and racial profiling. Sens. Booker and Harris introduced the Justice in Policing Act, which includes a provision prohibiting law enforcement body cameras from using facial recognition without a warrant; Rep. Bass introduced a companion bill in the House. Additionally, the American Civil Liberties Union sued facial recognition company Clearview AI, and Sen. Markey wrote to the CEO of Clearview AI requesting information on the role it has played in supporting law enforcement over the past few weeks.
OpenAI Announces GPT-3: Researchers at OpenAI have trained a 175 billion parameter natural language processing model, by far the largest such system to date. Relative to the 1.5 billion parameter GPT-2, GPT-3 scores higher on a range of language tasks without being specifically trained for them, a capability known as “few-shot learning.” In one study, human readers correctly identified whether a text was written by a human or GPT-3 only 52 percent of the time. While the researchers say GPT-3 demonstrates the possible performance benefits from increasing model size, others argue the paper lacks new techniques and that gains from scale may soon hit limits. One estimate indicates training GPT-3 likely cost OpenAI $12 million, which would make it one of the most expensive machine learning models yet.
Semiconductor Industry Lobbies for $37B in Federal Spending: The Semiconductor Industry Association is pushing for $37 billion in federal support, the Wall Street Journal reports. The association’s draft proposal reportedly includes $5 billion in federal funding for a new chip factory jointly operated by the government and private sector, $15 billion in state grants for new chip factory incentives and $17 billion in research funding. The association estimates that China will double its share of global chip production to 28 percent by 2030, and says increased domestic funding could help to avoid such an outcome.
Machine Learning Spotlight — Some Algorithmic Progress Overstated: Some improvements in machine learning algorithms may be overstated or inconsistently measured, according to new research. Researchers at MIT evaluated 81 papers featuring a specific type of algorithm and found no clear evidence of improvement over 10 years, despite claims of progress in each paper. Other studies reached similar conclusions across different types of machine learning algorithms. One study found that small tweaks to old models allowed them to match newer models, undermining alleged improvements. The researchers suggest that tools to allow accurate model comparison may help address the problem, and they warn people to be wary of hype.
Government Updates

US Joins G7 Group on AI Ethics: The United States has joined the other G7 countries in the Global Partnership on AI, a panel for setting ethical guidelines for artificial intelligence based on democratic values. While the Trump administration initially resisted joining for fear of hampering innovation, U.S. Chief Technology Officer Michael Kratsios announced the development in an op-ed, saying America needs to stand with other democracies and counter Chinese efforts to misuse the technology. The partnership officially launched May 28. The same week, the Trump administration announced interest in expanding the G7 summit to include Russia, South Korea, Australia and India.

Proclamation Restricts Entry of Some Chinese Researchers: On May 29, the White House issued a proclamation barring entry of some Chinese graduate students or researchers affiliated with Chinese institutions that support the state’s “military-civil fusion strategy.” The order applies to Chinese nationals who are affiliated with institutions deemed problematic by the State Department, though the list of institutions has yet to be published. In addition, the Secretary of State will consider if any visas currently held by Chinese nationals should be revoked. The order is intended to limit Chinese acquisition of U.S. technologies; Sens. Portman and Carper also previewed legislation to prevent Chinese theft of U.S. research.

Legislation to Create National Cloud Computing Resource Introduced: Senate AI Caucus co-founders and co-chairs Sens. Portman and Heinrich introduced legislation establishing a task force to plan a national cloud computing system for AI research. The bipartisan legislation convenes experts from academia, government and industry to develop a detailed roadmap for implementing, deploying and developing this resource. The new system would provide researchers and students across scientific disciplines with compute, government and non-government datasets and a research environment. Reps. Eshoo, Gonzalez and Sherrill introduced a companion bill in the House.

Bill to Advance AI Research Introduced: Sens. Gardner, Peters and Wicker introduced the bipartisan Advancing AI Research Act of 2020 last Thursday. The bill allocates $250 million per year from 2021–2025 toward the creation of a federal program to advance AI research at the National Institute for Standards and Technology. In addition, it tasks the National Science Foundation with establishing at least six AI research institutes, each of which would receive up to $50 million per year from 2021–2025, creating traineeships, and launching other pilot programs. Sens. Gardner and Peters also introduced a bill to provide scholarships for technical talent. Five other bills on AI, science and technology were introduced within the past week; our latest blog post summarizes all seven bills.

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

Report on Work to Protect Overseas Chinese: Report to the State Council on Work to Protect the Rights and Interests of Overseas Chinese. The Director of the PRC Overseas Chinese Affairs Office delivered this report to the Chinese parliament in April 2018 regarding his office’s performance. According to the report, one of China’s main priorities in engaging with Chinese people outside the PRC is to recruit scientific and technical talent to serve the country’s economic development. The report also mentions a number of problems that foreign citizens of Chinese descent face in living and doing business in China, such as foreign ID documents not being accepted, difficulties securing education and healthcare for their families, and intellectual property theft.  

What We’re Reading

Report: Policy Roundtable: Artificial Intelligence and International Security, Texas National Security Review (June 2020)

Post: The Global AI Talent Tracker, MacroPolo (June 2020)

Paper: U.S. National Security Export Controls and Huawei: The Strategic Context in Three Framings, Christopher A. Ford, Office of the Under Secretary of State for Arms Control and International Security (May 2020)

Blog: Responding to the European Commission’s AI White Paper, Google (May 2020)

What’s New at CSET

REPORTSPUBLICATIONSIN THE NEWS