A controversial departure from Google AI, Huawei implicated in Uyghur facial recognition scandal and Congress passes an NDAA with AI implications

Sign up
Worth Knowing

Google’s Split With AI Ethics Researcher Sparks Controversy: Timnit Gebru, former co-lead of Google’s Ethical AI team, said early this month she’d been fired over an email criticizing Google’s management. Gebru — an expert on algorithmic bias, co-founder of Black In AI, and one of the few Black women leaders in the industry — emailed the internal group “Google Brain Women and Allies” critiquing the company’s treatment of women and minority employees and its handling of a paper she had co-written. The research paper reportedly identified problems with AI models used by Google and others to understand human language. After the paper was submitted to an academic conference, Google management reportedly asked her to retract it or remove her name and the names of other Google employees. Google AI chief Jeff Dean wrote in an internal email that the paper “didn’t meet our bar for publication” and “ignored too much relevant research,” but Gebru argued Google’s demands amounted to censorship. While Gebru’s 2018 hiring helped bolster Google’s public commitment to “ethical AI,” observers say her departure highlights the inherent difficulties of producing independent research at for-profit companies like Google.
Huawei Developed Facial Recognition Tech With “Uyghur Alarm”: According to a document shared with The Washington Post, the Chinese company Huawei developed and tested facial recognition technology that could identify an individual’s ethnicity and alert authorities when it registered a member of the Uyghur population. Huawei, the world’s largest telecommunications manufacturer, confirmed the document’s authenticity. A spokesman for the company said the system was only a test and had not been deployed for real-world use, but The Post reported that Chinese-language marketing materials on a Huawei site advertised a number of products’ ethnicity-tracking capabilities. Huawei has reportedly supplied several surveillance tools to authorities in Xinjiang, the northwestern region of China where a majority of Uyghurs live and where millions have been interned in state-run camps, but it was not clear whether these tools were equipped with the ethnicity-tracking capability.
Alphabet Researchers Develop AI-Controlled Stratospheric Balloons: Loon, an Alphabet subsidiary that deploys high-altitude balloons to provide internet access in remote areas, is using artificial intelligence to steer them. In a peer-reviewed article published in Nature earlier this month, Loon researchers described the reinforcement learning process that trains the new navigation system. Loon’s AI controller reportedly outperforms the company’s previous system and is already being used to provide internet over Kenya. Autonomous balloons could have applications in a wide range of commercial and scientific endeavors, including monitoring weather, observing volcanoes, tracking seismic activity and exploring extraterrestrial environments.
Fewer Bits Could Change Deep Learning: During the annual NeurIPS conference last week, IBM researchers proposed a new process that could make deep learning much more efficient. In their paper, the researchers explain that 4-bit computers could perform the same AI processes as 16-bit computers (the current standard) with only a minor loss in accuracy. Training efficiency could rise by more than 700 percent through 4-bit deep learning, opening the door for deep learning on smaller devices like smartphones. Speaking with MIT Technology Review, some observers advised caution: because 4-bit training would require new hardware, its real-world application is still several years away, and its actual performance may not measure up to the theoretical levels proposed in last week’s paper. One expert noted that promising 2016 research on 5-bit training failed to deliver on lofty expectations. 
Government Updates

Congress Passes NDAA: Last Friday, the Senate approved the National Defense Authorization Act for Fiscal Year 2021. The $741 billion defense policy bill garnered bipartisan support in both the House and Senate: the votes were 335 to 78 and 84 to 13, respectively. The bill includes a number of provisions related to AI and emerging technology, including:
  • The National AI Initiative Act, which directs the president to establish a National Artificial Intelligence Initiative to support U.S. leadership in AI. The measure authorizes approximately $5 billion for National Science Foundation AI research and education grants, as well as more than $1 billion for Department of Energy AI research.
  • Provisions from the CHIPS for America Act to support the development and production of semiconductors in the United States. The included provisions direct DOD to establish a public-private consortium for the development of chips. They also authorize a fund to advance microelectronics research with allies and partners, provided those partners enact controls on chip exports to China.
While the bill passed both chambers of Congress with more than the two-thirds majority needed to override a presidential veto, President Trump has continued to threaten one. The president has until December 23 to veto, sign, or allow the bill to become law without his signature.

Courts Rule On Immigration Suits: U.S. District Court rulings struck down two different attempts to limit high-skilled immigration, one by the Trump administration, the other by a U.S. technology workers union. U.S. District Judge Jeffrey S. White of the Northern District of California ruled against the Departments of Labor and Homeland Security in a lawsuit by the U.S. Chamber of Commerce over changes to the H-1B visa program. Those changes would have excluded an estimated third of current recipients by reducing the list of qualifying occupations, tightening degree requirements, and significantly raising minimum salaries. Earlier this week, the U.S. District Court for the District of Columbia also ruled against the Department of Labor in a related case filed by Purdue University and several other schools and organizations. Also in the D.C. District Court, Judge Reggie B. Walton ruled against the Washington Alliance of Technology Workers, which had sued the Department of Homeland Security in an attempt to end the Optional Practical Training. OPT allows foreign students to work in the United States for up to three years after graduating while still on their student visas. Getting rid of the program would have cut off one of the country’s key sources of AI talent; roughly half of all graduate students who enroll in OPT hold AI-related degrees.

President Trump Signs Executive Order On AI Development: On December 3, the president issued an executive order that sets guiding principles for the federal government’s AI use. It instructs agencies to adhere to nine principles when designing, developing, acquiring or using AI, including that it be “Understandable,” “Responsible and traceable,” “Accountable” and “Regularly monitored.” The order directs OMB to develop a roadmap for policy guidance within 180 days, a deadline more than four months into the Biden administration. Although the Biden campaign advanced positions on AI policy, the transition has not yet released an AI strategy.

House Unanimously Passes Resolution On National AI Strategy: Last week, the House passed a nonbinding resolution outlining a national AI strategy. The resolution, co-authored by Reps. Hurd and Kelly, drew on four white papers by the Bipartisan Policy Center written in consultation with Hurd and Kelly, including one co-authored by CSET. The measure encourages collaboration with allies, increased R&D funding and expanded support for technology education programs. The Senate has yet to consider a similar resolution.

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

PRC Cybersecurity Report: Research Report on the Status of China’s Information Security Professionals (2018-2019). This study by a PRC government cybersecurity center analyzes China’s complement of cybersecurity and IT security professionals. The study finds that PRC cybersecurity practitioners are still too few in number and are often burdened with non-security-related IT tasks, though China is making progress solving these problems. The authors recommend that all Party leaders take courses on cybersecurity to raise awareness of the importance of the topic.

What We’re Reading

Report: The Global AI Index, Tortoise Intelligence (December 2020)

Data Visualization: Stateoftheart AI (December 2020)

Article: A Narrowing of AI Research?, Joel Klinger, Juan Mateos-Garcia, Konstantinos Stathoulopoulos (November 2020)

What’s New at CSET

REPORTSJOB OPENINGS
CSET is hiring! Please share with qualified candidates in your network or consider applying.
  • Research Analyst: We’re looking for a Research Analyst to focus on biotechnology efforts. Collaborate with Research Fellows to develop and execute research projects. BA in relevant area and comfort with data analysis/viz required (MA preferred).
PUBLICATIONS, PODCASTS AND WEBCASTSFORECASTS
CSET has launched a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:
EVENTSIN THE NEWS
Upcoming Events


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