The White House weighs more targeted chip restrictions for China, a warning about ML research reproducibility, and Beijing gets access to dozens of popular algorithms

Sign up
Work at CSET

We’re hiring! If you’re interested in joining our team, check out the positions in the “Job Openings” section below or consult our careers page.

Worth Knowing

Princeton Researchers Sound the Alarm on ML & Research Reproducibility: The widespread adoption of machine learning–based research methods has created a “reproducibility crisis” across a number of research fields, according to two computer science researchers. In their preprint paper, Princeton Professor Arvind Narayanan and PhD student Sayash Kapoor reviewed papers across 18 fields, including political science, medicine, satellite imaging, and computer security, and identified issues in more than 300 of them. Machine learning has been increasingly popular among military planners, investors and law enforcement because of its potential to predict future outcomes from past results, but Narayanan and Kapoor warn that common issues stemming from the way training data is collected, sampled and processed can produce models that are significantly overconfident in their predictions. Because these “data leakage” issues can be difficult (if not impossible) to identify with current peer-review practices, Narayanan and Kapoor warn that the problems are likely to persist unless changes are made. To help researchers and reviewers identify machine learning–related issues, they proposed adopting “model info sheets” (inspired by and similar to model cards) that researchers could fill out and publish alongside their papers.  
Russia Announces Two New AI Organizations: Last week, the Russian Ministry of Defense said it had launched a new department for AI development, and the government announced plans to open a National AI Center in September. The announcements came during “Army 2022,” an international military expo held near Moscow, during which Russian companies (as well as vendors from China and Iran) showcased new military hardware. While little public information is available about the military’s new AI development office, observers said it likely will serve a similar purpose as the U.S. military’s Chief Digital and Artificial Intelligence Office, which oversees the Pentagon’s AI efforts. Meanwhile, according to Deputy Prime Minister Dmitry Chernyshenko, the National AI Center will help coordinate Russian AI development across industry, academia and government, and will work to bolster the country’s AI workforce, which Chernyshenko said is short 5,000 AI specialists. Moscow has set ambitious goals for the country’s AI industry — most notably in its 2019 national AI development strategy — but the government’s ability to deliver on these goals is constrained by systemic problems and the more recent sanctions, big-tech withdrawals, and brain drain challenges that have arisen in response to Russia’s invasion of Ukraine.
Chinese Companies Disclose Information About Their Algorithms: Earlier this month, the Cyberspace Administration of China published a list describing the algorithms used by some of China’s tech giants, including Alibaba, Xiaomi, Baidu, Tencent, and ByteDance. As part of new regulations on algorithmic governance that went into effect this past March, Chinese companies are required to register their algorithms with the CAC. The list (translation available here) includes information about the algorithm’s type (e.g., algorithms for personalizing content or for filtering searches) and function. Some observers raised concerns that the CAC had been granted access to the source code of the algorithms, prompting fears that Beijing may have inside knowledge of the “secret sauce” of popular apps like ByteDance-owned TikTok. But it appears the information companies were asked to provide was much more surface-level, such as 200-character descriptions of their algorithms and information about the data used to train them. While the list currently comprises 30 algorithms, it is set to expand as more companies register their algorithms with the CAC.  
Government Updates

White House Plans to Restrict China’s Access to FinFet Tech: The Biden administration is reportedly working to cut off China’s access to the technology necessary to manufacture semiconductors that use a particular type of high-end transistor. According to Protocol’s Max Cherney, the administration plans to block exports of the manufacturing equipment necessary to make chips with fin field-effect transistors (FinFETs), which would also prevent U.S. companies from servicing equipment they had already sold to Chinese companies. The Commerce Department already moved earlier this month to block the software needed to design gate-all-around transistors, a more advanced type of transistor design, but the reported restrictions would be more expansive and would target chip manufacturers rather than designers. The shift in policy would represent a refinement of the government’s current approach, which aims to prevent China from manufacturing chips at the 14 nm process node and below. But as we covered in the most recent newsletter, that has proved difficult to carry out. Not only are nanometer-based descriptions more of a marketing designation than anything else, Chinese companies have been able to repurpose equipment used for less-advanced chips to make chips beyond the 14 nm level. While the administration’s pivot could be more effective, it also seems likely to face some pushback. U.S. companies such as Applied Materials, Lam Research and KLA Corp have a significant stake in China and could stand to lose up to a third of their revenue if shut out of the Chinese market. 

NIST Publishes A New Draft of Its AI Risk Management Framework: Last week, the National Institute of Standards and Technology published the second draft of its AI Risk Management Framework, a document meant to help AI developers anticipate and manage the risks unique to AI systems. In the draft, NIST clarifies that the AI RMF is not meant to serve as a checklist — it is intended to be consulted voluntarily by organizations throughout their AI products’ lifecycles, from early stage planning through deployment. Alongside the framework, NIST also released an early version of the accompanying AI Risk Management Framework Playbook, which includes specific actions organizations can take to ensure the trustworthiness of their systems. The playbook is only partially complete — a NIST official told FedScoop that the rest of the playbook will be completed after stakeholder feedback is received. Both the draft and the playbook are open for public comment until September 29, 2022. NIST will then host a third and final workshop on October 18–19, and aims to officially publish the AI RMF in January 2023.

Director of National AI Office Steps Down — What’s Next for AI in the OSTP?: Lynne Parker, the director of the White House National Artificial Intelligence Initiative Office, stepped down earlier this month, having served in the post since the office launched in January 2021. Her departure comes during a time of broader transition for the Office of Science and Technology Policy (where the NAII office is housed). President Biden’s first OSTP director, Eric Lander, resigned earlier this year after facing allegations of workplace bullying. In June, the president nominated Arati Prabhakar to succeed Lander in the post, but she has yet to be confirmed by the Senate (while the Senate Committee on Commerce, Science, and Transportation approved her nomination in July, a date for a full Senate vote has not yet been set). Many of the initiatives overseen by the NAII office are mandated by the National AI Initiative Act of 2020 and are set to continue — an update to the national AI strategy and a final report from the National AI Research Resource task force are both expected later this year, for example. But other initiatives, such as the proposed “AI Bill of Rights,” have murkier futures. Last year, Lander and OSTP Deputy Director for Science & Society Alondra Nelson announced plans to develop a “bill of rights for an AI-powered world.” Earlier this year, Nelson said the document would be published in May 2022, but it remains unreleased. It’s not yet clear whether the bill of rights will still be a priority with Prabhakar at the helm.    

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

PRC AI Training White Paper: China Artificial Intelligence Talent Training Report. This white paper, co-authored by China’s prestigious Zhejiang University and Chinese tech company Baidu, examines China’s system for AI training in depth. Pointing to the United States’ dominance in AI talent, the authors urge closer and more rational cooperation between Chinese universities and tech companies in training AI talent to improve China’s self-sufficiency in AI. The authors also call for China to establish an international standards body for AI talent training.  

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.

Job Openings

We’re hiring! Please apply or share the roles below with candidates in your network:
  • Visual Communications Specialist: The Visual Communications Specialist will support the work of the External Affairs Team to raise the profile of CSET’s research through a variety of outreach activities such as graphics, multimedia, and publications. Apply by September 16.
  • Research Fellow — AI Applications: This Research Fellow will focus on helping decision makers evaluate and translate new and emerging technologies, particularly in the field of AI, into novel capabilities by separating real trends and strategic opportunities from technological hope and hype. Rolling application — Apply today.
  • Senior Fellow: This Senior Fellow will provide mentorship and intellectual leadership to CSET researchers; shape analysis that is aligned with our research priorities; and facilitate engagements with government, military, academic, and industry leaders.  Rolling application — Apply today.
  • Georgetown Student Opportunities: We are currently seeking applications for Emerging Technology Observatory Research Assistants (JobID: 29149), an External Affairs Assistant (JobID: 26749), and an Editorial Assistant (JobID: 27389). Please note that applications must be submitted via the SEO website. Rolling application — Apply today.
Please bookmark our careers page to stay up to date on all active job postings. You can also subscribe to receive job announcements by updating your subscription preferences in the footer of this email.

What’s New at CSET

REPORTSPUBLICATIONS AND DATA VISUALIZATIONSIN THE NEWS
What We’re Reading

Book: Knowledge Regulation and National Security in Postwar America, Mario Daniels and John Krige (2022)  

Report: The Brussels Effect and Artificial Intelligence: How EU regulation will impact the global AI market, Charlotte Siegmann and Markus Anderljung, Centre for the Governance of AI (August 2022)