AI Ethics Expert Timnit Gebru Launches Research Center: Earlier this month, Timnit Gebru, a computer scientist and expert in algorithmic bias, launched a research institute that will study the impact of AI on marginalized and understudied groups. Until last year, Gebru was co-lead of Google’s Ethical AI team, a research unit the company started in 2018 as part of its public commitment to building ethical AI systems. Gebru’s controversial exit — she said she was fired for criticizing the company’s work on large language models, while Google insisted she had resigned — and the subsequent firing of her former Ethical AI team co-lead sparked concerns about the future of AI ethics research at Google and other tech giants. Gebru’s new organization, the Distributed Artificial Intelligence Research Institute (DAIR), is positioning itself as a direct response to those concerns. In a press release, the organization said it aims “to counter Big Tech’s pervasive influence on the research, development and deployment of AI” by serving as an independent space for researchers from diverse backgrounds to study the negative impacts of AI systems in their communities, with DAIR’s upcoming NeurIPS paper slated to be the first such project.
- More: Ex-Google AI ethics chief: Boost worker power to curb harmful AI | Facebook knew its algorithms were biased against people of color
- TSMC has begun risk production of chips using its 3 nm process node, according to a report from DigiTimes. Risk production generally lasts between six and 12 months, meaning TSMC looks to be on schedule to hit its roadmap target of entering 3 nm volume production by the end of 2022. TSMC says its 3 nm chips offer 10–15% performance improvements and 70% greater logic density compared to current state-of-the-art 5 nm chips, which hit the market late last year. Apple has reportedly booked all of TSMC’s early production capacity for its iPhone, iPad and Mac processors, meaning it could be until 2023 or 2024 before chips made for AI workloads make use of the new process (unless Samsung manages to woo customers with its own 3 nm chips).
- Nvidia reportedly plans to introduce its next generation, 40-series GPUs next year. The release will include a new architecture for datacenter and AI workloads using TSMC’s 5 nm process. Nvidia’s GPUs are found in most of the world’s AI supercomputers, but with customers turning to rival chips or designing their own, the new GPUs could help it thwart that challenge. The new architecture, dubbed “Hopper,” seems intended to pair with another Nvidia product named for the late American computer scientist Grace Hopper — the Grace CPU, announced earlier this year, is set to be released in early 2023 and promises 10 times the performance of today’s fastest systems when paired with Nvidia GPUs.
- More: AWS debuts new compute-intensive and AI instances powered by custom chips | Deloitte: Chip shortage will extend into 2022
The House Passes NDAA, Lining Up the Bill for a Senate Vote: On Tuesday night, the House of Representatives voted 363–70 to approve the $768 billion FY2022 National Defense Authorization Act. The annual defense bill has become law more than 60 years in a row, but protracted debate in the Senate following the House’s passage of an earlier version of the bill gave rise to concerns that a final agreement might not be reached. The new version (summary available here) was negotiated by leaders of the House and Senate Armed Services Committees, and the Senate is expected to vote on the House version of the bill without amendments. The bill contains a number of provisions related to AI and emerging technology, including:
- Sec. 217 requires the DOD to establish a national network for microelectronics R&D. Last year’s NDAA had authorized such a network as part of the CHIPS for America Act, but the updated language in the new bill mandates it.
- Sec. 226 requires the DOD to “review the potential applications of artificial intelligence and digital technology to the platforms, processes, and operations” of the military and to establish performance metrics for AI adoption efforts.
- Sec. 228 requires the DOD to establish “executive education activities on emerging technologies” for select officers and civilian staff.
- Sec. 247 requires the DOD to provide regular reports and briefings to Congress on its implementation of recommendations made by the National Security Commission on Artificial Intelligence in its final report.
DOD Reorganizes Tech Offices: In a memo published late yesterday, Deputy Secretary of Defense Kathleen Hicks announced the creation of a new position, the Chief Digital and AI Officer, who will be responsible for coordinating all of the Pentagon’s data, AI and digital efforts. Hicks’ memo also outlined a significant reorganization of three of the Pentagon’s key AI and data offices — the Joint Artificial Intelligence Center, the Defense Digital Service and the office of the Chief Data Officer. Once the office of the CDAO reaches initial operating capability (sometime between February 1 and June 1, 2022), the three offices will report to the yet-to-be-announced CDAO, who will, in turn, report directly to Hicks. The plan — which Breaking Defense’s Aaron Mehta scooped last week — was met with both praise and concern. While some observers applauded it, Mehta spoke to others who worried the reorganization would strip the offices — all of which were formed within the last five years — of their (unusual-for-the-Pentagon) independence. A Pentagon official acknowledged and tried to downplay these concerns, telling journalists ahead of the memo’s release that the changes were the next step in the young offices’ evolution and a key to the DOD’s embrace of AI and other emerging technologies as part of its Joint All-Domain Command and Control (JADC2) strategy. However, the official didn’t rule out that the three offices could be merged and subsumed by the office of the CDAO, but said that, for now, the current JAIC, DDS and CDO leaders would remain in place.
NITRD Releases Resources for Tracking Federal AI Spending: The Networking and Information Technology Research and Development program released two resources with in-depth information about the federal government’s AI spending: a Supplement to the President’s FY2022 Budget and a FY2018–FY2022 Artificial Intelligence R&D Investments dashboard. As we noted when the Biden Administration published its FY2022 budget request earlier this year, the request and its accompanying appendix were relatively light on specific references to AI. The new resources go into greater detail about the FY2022 budget request’s 8.8% increase in nondefense AI R&D, as well as the progress of recent AI initiatives, such as the 18 National Artificial Intelligence Research Institutes. At the moment, however, the FY2022 numbers remain aspirational — Congress has only passed a continuing resolution to extend FY2021 funding levels through mid-February.
FTC Sues to Block Nvidia-Arm Deal: Last week, the Federal Trade Commission sued to block Nvidia’s acquisition of British chip designer Arm. The news serves as the latest blow for the beleaguered deal, originally reached in 2020 for $40 billion (though Nvidia’s surging stock price has now driven the value of the deal up to $80 billion) — the UK government, which completed an initial investigation into the deal’s national security implications earlier this year, launched a “Phase Two” investigation last month, soon after the European Commission opened its own probe. The FTC’s complaint argues that the deal would significantly undermine competition across the computing industry — it singled out driver assistance systems, DPU-based SmartNICs, and cloud computing CPUs as the markets most likely to be undermined by the merger — and incentivize Nvidia to “harm its Arm-reliant rivals.” Often called “the Switzerland of chips” because of its relatively neutral approach to licensing, Arm sells its chip designs to many of Nvidia’s direct competitors. While Nvidia originally planned to finalize the deal by March 2022, the FTC suit and the ongoing investigations have made that timeline all but impossible. An administrative trial for the FTC’s case is scheduled for May 10.
CSET’s translations of significant foreign language documents on AI
CSET’s translations of significant foreign language documents on AI
PRC Biodiversity Protection: Opinions on Further Strengthening Biodiversity Protection. This official PRC document offers guidelines for the protection of biodiversity in China. It sets specific targets for the years 2025 and 2035 for environmental indicators such as forest cover and wetland protection. The document calls for more severe punishments for individuals and businesses that damage ecosystems, and recommends the inclusion of biodiversity protection language in Chinese officials’ performance evaluations.
Russian AI Think Tank Report: Artificial Intelligence Almanac: 2020 Index Report. This report by a Russian AI think tank provides an overview of Russia’s AI landscape as of the end of 2020. The authors argue that the Russian AI industry is lagging relative to other countries, and recommend a three- to five-fold increase in Russian AI research funding.
If you have a foreign-language document related to security and emerging technologies that you’d like translated into English, CSET may be able to help! Click here for details.
Please apply or share the role below with candidates in your network:
- AI Research Subgrant (AIRS) Program Director: CSET’s AIRS program will promote the exploration of foundational technical topics that relate to the potential national security implications of AI over the long term via research subgrants. The Director of AIRS will manage all technical, programmatic, and financial aspects of the new AIRS program.
What’s New at CSET
- Key Concepts in AI Safety: Specification in Machine Learning by Tim G. J. Rudner and Helen Toner
- Measuring AI Development: A Prototype Methodology to Inform Policy by Jack Clark, Kyle Augustus Miller and Rebecca Gelles
- AI for Judges: A Framework by James E. Baker, Laurie N. Hobart and Matthew G. Mittelsteadt
- AI and the Future of Disinformation Campaigns: Part 1: The RICHDATA Framework by Katerina Sedova, Christine McNeill, Aurora Johnson, Aditi Joshi and Ido Wulkan
- ChinaTalk Podcast: CSET: How to Break the Think Tank Mold featuring Dewey Murdick
- Korea on Point: Strengthening the U.S.-ROK Technology Alliance by Ryan Fedasiuk
- Foreign Policy: Beijing’s Strategic Blueprint Is Changing as Tensions Grow by Ryan Fedasiuk and Emily Weinstein
- CSET: Data Snapshot: Exploring protein-folding AI research with the Map of Science by Sara Abdulla
- CSET: Classifying AI Systems by Catherine Aiken, Brian Dunn and Jennifer Melot
Foretell has launched a new project that combines expert and crowd judgment. You can read more about the experts’ views, including how they think trends like China’s military aggression, political polarization, and the strength of the tech sector affect the DOD-Silicon Valley relationship. See all 20 forecast questions associated with this project here.
IN THE NEWS
- The New York Times: For an article about the effect on universities of the U.S. Department of Justice’s China Initiative, Amy Qin reached out to Senior Fellow Anna Puglisi.
- ProPublica: Puglisi also spoke to Sebastian Rotella for a ProPublica investigation about Beijing’s efforts to retaliate against dissidents studying in the United States.
- EE Times: George Leopold picked up on the release of Tim G. J. Rudner and Helen Toner’s new brief, Key Concepts in AI Safety: Specification in Machine Learning, interviewing Toner and recapping the brief’s findings for an EE Times article.
- VentureBeat: For a piece about a new multilingual AI language model, VentureBeat’s Kyle Wiggers cited the 2020 brief by Roxanne Heston and Remco Zwetsloot, Mapping U.S. Multinationals’ Global AI R&D Activity.
What We’re Reading
Report: How to Win the Future: An Agenda for the Third Generation of the Internet, a16z (October 2021)
Report: China’s rise in semiconductors and Europe, Jan-Peter Kleinhans and John Lee, Stiftung Neue Verantwortung (December 2021)
- December 16: CSET Webinar, Deconstructing China’s Vision for the Future of Warfare, featuring CSET Research Analyst Ryan Fedasiuk and Lieutenant General (retired) Jack Shanahan
What else is going on? Suggest stories, documents to translate & upcoming events here.