ICE restricts foreign students, MIT removes controversial dataset and DOD AI report highlights concerns

Sign Up
Worth Knowing

Tech Community Criticizes New Immigration Restrictions: U.S. tech companies, entrepreneurs and researchers spoke out against recently announced immigration restrictions. Amazon, Google, Facebook and Microsoft condemned the White House’s new freeze on temporary work visas, calling it short-sighted and saying it jeopardizes U.S. global competitiveness. Large tech companies recruit thousands of immigrant workers each year, and many prominent AI leaders came to the country on temporary visas. On Monday, ICE announced international students taking online-only classes would not be allowed to remain in the country; AI researchers and business leaders took to Twitter to denounce the decision, arguing it hurts students, researchers and U.S. colleges and universities.
MIT Removes Image Dataset With Racist Labels: Professors at the Massachusetts Institute of Technology and New York University have taken down the 80 Million Tiny Images dataset after being informed that it contained offensive words and images. The dataset, created in 2006, was generated automatically by selecting nouns from a list and pairing them with images from internet searches. However, new research found the dataset included racial and misogynistic slurs, as well as pornographic images. The researchers identified similar problems in other datasets and suggested steps to avoid the issues in the future. The MIT team apologized, acknowledging the dataset may have inadvertently trained biased AI systems. 
ML System For Predicting Criminality Faces Backlash: Researchers created a model they claim can predict with 80 percent accuracy whether someone is a criminal based only on photos of faces. In response, 2,400 people signed a petition against publishing this or similar articles, arguing that crime-prediction technology is harmful and inaccurate. After the outcry, publisher Springer Nature said it was not planning to print the article. Computer vision and facial recognition algorithms face increasing criticism for disproportionately failing on people of color. The New York Times also recently reported that Robert Julian-Borchak Williams, a Black man, was wrongfully arrested after being misidentified by facial recognition — the first such incident to be publicized. 
Government Updates

ICE: Nonimmigrants Taking Classes Online Only May Not Remain in US: International students taking classes entirely online will not be allowed to remain in the United States, Immigration and Customs Enforcement announced Monday. The restrictions apply to students on F-1 or M-1 visas who are currently in or entering the country. Students attending schools using a hybrid model — a mixture of online and in-person classes — will be allowed to stay, but their universities must certify that their program is not entirely online. Several U.S. universities previously announced plans to go fully online due to concerns about COVID-19; this news follows immigration restrictions announced via executive order in late June.

FY21 NDAA Progresses With AI Amendments: The House Armed Services Committee unanimously passed its version of the National Defense Authorization Act on July 1. AI-related amendments were approved, including:
AI-related amendments were also introduced for the Senate NDAA, including:
Facial Recognition Ban Introduced: Sens. Markey and Merkley introduced the Facial Recognition and Biometric Technology Moratorium Act. The legislation prohibits federal use of facial recognition and other biometric technologies unless specifically authorized by Congress. In addition, the bill withholds certain state and local law enforcement grants unless local facial recognition bans are implemented. It also forbids the federal government from using information collected from biometric technologies in judicial proceedings unless additional legislation is passed. Rep. Jayapal introduced a companion bill in the House; currently no Republicans support the legislation.

Report Finds DOD AI Systems Have Security and Coordination Flaws: A report issued by the DOD Inspector General found that the Joint AI Center has not fully developed an AI governance framework and that the DOD has AI security vulnerabilities. Intended to assess gaps in the system for AI governance and data protection, the audit instead found that a DOD-wide system had not yet been developed. The report recommends the JAIC establish a standard definition of AI, a central process to account for AI projects, and strategies for collaborating across the DOD. In addition, it identified inconsistently implemented systems for protecting AI data from cyber threats and offered a plan to correct security weaknesses. The DOD CIO said the DOD has already taken several steps outlined in the report.

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

Chinese Academy of Science Talent Recruitment Programs: 2019 Chinese Academy of Sciences Talent Program Application Guide. This document is a complete list and description of all Chinese Academy of Science talent recruitment programs. CAS has talent programs primarily targeting young scientists in strategic and emerging disciplines, including defense-related “military-civil fusion” technologies; some are aimed at domestic S&T talent, while others recruit foreign scientists.

Recruiting Outstanding Domestic and Foreign Talent: The Chinese Academy of Sciences Hefei Institutes of Physical Science: Recruiting Outstanding Domestic and Foreign Talent. This 2017 document lists technical positions that the Hefei Institutes of Physical Sciences hoped to fill through talent programs. It is an unusually detailed example of a Chinese research institution’s efforts to recruit foreign scientists.

Report on Youth Thousand Talents Program: Situation Report on the Chinese Academy of Sciences’ Work on the “Youth Thousand Talents Program.” This document details the number and affiliation of CAS-sponsored candidates for China’s “Youth Thousand Talents Program” as of 2011. Under the program, prospective employers such as CAS are responsible for identifying overseas talent they wish to recruit, and must submit applications on behalf of their candidates.

What We’re Reading

Report: Trends in U.S. Multinational Enterprise Activity in China, 2000–2017, U.S.-China Economic and Security Review Commission (July 2020)

Commentary: Balancing Standards: U.S. and Chinese Strategies for Developing Technical Standards in AI, Jeffrey Ding (July 2020)

What’s New at CSET

PUBLICATIONSRECRUITING
We’re looking to fill the following roles:
  • Research Fellow: Lead research projects, produce publications and brief policymakers. Graduate degree in relevant field and 3+ years of relevant experience required. Applications close July 17.
  • Forecaster: Predict the future of technology-security policy on CSET’s new crowd forecasting platform. Unpaid, part-time opportunity.
IN THE NEWS