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
DeepMind AI Controls Plasma in A Fusion Reactor: Last week, researchers at DeepMind announced they had trained an AI system to control plasma in a nuclear fusion reactor. Donut-shaped “tokamak” fusion reactors use powerful magnets to control the shape of superheated hydrogen atoms — a process that requires continuous monitoring and precise adjustments. In an attempt to replace the classical control algorithms that have been used in this part of the process, researchers from the Alphabet-owned AI lab used reinforcement learning to train a large neural network in a simulated reactor, then used a smaller, faster neural network to carry out the larger network’s predictions in the real reactor (their Nature paper is available here). The DeepMind model successfully controlled the plasma in the tokamak for two seconds — the maximum the reactor at the Swiss Plasma Center can handle before overheating. While the DeepMind system’s performance is an impressive proof-of-concept, the prospect of commercially viable nuclear fusion remains a distant goal. But experts say AI systems like DeepMind’s could help researchers experiment with new plasma configurations or optimize tokamak design, potentially accelerating fusion progress.
- More: The World’s Largest Tokamak Just Crushed the Record for Nuclear Fusion Energy | Transformative Enabling Capabilities for Efficient Advance Toward Fusion Energy
- More: A New Proposed Law Could Actually Hold Big Tech Accountable for Its Algorithms | Understanding Canada’s Algorithmic Impact Assessment Tool
- A robot autonomously performed laparoscopic gastrointestinal surgery on a pig, reportedly achieving better results than an expert surgeon. Reconnecting two ends of an intestine is a challenging procedure, and even small mistakes can lead to fatal results. As the researchers from Johns Hopkins and the Children’s National Hospital in Washington, D.C., detailed in a paper published in Science Robotics, a machine learning-based tracking algorithm guided the robot and helped it detect and adapt to unexpected changes with minimal human intervention.
- Researchers used AI to model the Omicron variant of the coronavirus, offering a potential speed and cost advantage over established methods. As Tom Simonite chronicled in Wired last month, a genomics researcher at the University of North Carolina at Charlotte used protein-folding AI tools developed by DeepMind and the University of Washington to model the structure of the Omicron variant’s spike protein. Though the AI-generated structure was not 100 percent accurate, it came just a few days after the variant’s genomic data was made public by scientists in Botswana — nearly a month before the structure was identified using established methods. That speed advantage could help researchers make an “educated guess” about new viruses while they wait for definitive answers.
- BioNTech and UK-based AI company InstaDeep created an “early warning system” for flagging dangerous coronavirus variants. According to a preprint paper published by the companies’ researchers, their system successfully flagged 12 of 13 potentially dangerous variants an average of two months earlier than they were officially designated by the WHO. That time advantage could help vaccine companies like BioNTech prioritize the development of variant-specific vaccines. The system was not perfect, however — its one miss was the Delta variant. This was likely due, the researchers said, to limited genomic data availability and which mutations the model identified as significant.
Government Updates
U.S. Navy and 59 Partners Conduct World’s Largest Unmanned Exercise: Last week, the Navy concluded a 60-nation maritime exercise that featured the largest unmanned exercise in the world, comprising 80 air, surface and underwater unmanned systems from the United States and 9 other countries. The 18-day exercise gave the Navy and its partners an opportunity to familiarize themselves with a number of unmanned and autonomous systems on the cusp of widespread deployment. Soon after the exercise concluded, the Bahrain-based 5th Fleet announced the launch of a joint fleet of 100 unmanned surface vessels, meant to patrol the area around the Persian Gulf. Vice Adm. Brad Cooper said the drones, set to be operational by the summer of 2023, would help build an “integrated, unmanned and artificial intelligence network” and improve threat detection. In the longer term, the Navy has plans to deploy even larger unmanned vessels — earlier this month, Adm. Mike Gilday, the chief of naval operations, said he would like to deploy large and medium-sized unmanned vessels alongside carrier strike groups by 2027 or “earlier if I can.”
The DOJ Ends The China Initiative: Yesterday, the Department of Justice announced that it is ending the China Initiative and replacing it with a program that is country agnostic and focuses on a wide range of threats. The controversial initiative, launched by the Trump administration in 2018, aimed to crack down on China’s targeting of U.S. technology, including early-stage research. But after a number of prominent cases involving academics at U.S. universities — including those of Gang Chen and Anming Hu — critics argued that the initiative had become overly broad despite its original mandate, was having a negative impact on collaborative research, and was engaged in racial profiling. In a speech announcing the change, Assistant Attorney General for National Security Matthew Olsen acknowledged those critiques, but defended the DOJ’s motivations, which he said “have been driven by genuine national security concerns.” To continue to address those concerns, Olsen said the department would introduce a “Strategy for Countering Nation-State Threats” that addresses illegal activity from adversarial nations — including the theft of trade secrets and malicious cyber activity — specifically naming China, Russia, Iran and North Korea.
The Algorithmic Accountability Act Is Reintroduced: Earlier this month, Sens. Ron Wyden and Cory Booker and Rep. Yvette Clarke introduced the Algorithmic Accountability Act of 2022, a bill that would require companies to assess and disclose information about how their automated systems are used. Specifically, the bill (a section-by-section summary and one-pager are also available) would require companies to conduct ongoing impact assessments of automated systems that make or help humans make “critical decisions.” The bill would create a sizeable new role for the FTC — it charges the agency with enforcing the bill’s new regulations, tasks it with developing the guidelines for assessment and reporting, requires it to publish an annual anonymized report on trends and a repository of information about automated critical decision systems, and creates a new 50-person Bureau of Technology inside the agency. The original version of the bill was introduced in 2019, but failed to gain traction.
Russia Sanctions Could Impact Emerging Tech, Semiconductors and AI: With a Russian invasion of Ukraine now underway, the Biden administration is reportedly preparing a suite of sanctions that would target Russia’s access to critical technologies. According to Reuters, these sanctions would expand the scope of the Foreign Direct Product Rule — which the government already uses to restrict some exports to China — to limit Russia’s access to key U.S. technologies, including semiconductors and other AI-related exports. In a call to semiconductor industry executives last month, White House officials warned that export controls could significantly limit their access to the Russian market. The White House also said that the invasion could have a major effect on key semiconductor imports. Both Russia and Ukraine are major exporters of many of the raw materials necessary for their production — 90 percent of the United States’ semiconductor-grade neon comes from Ukraine, and Russia accounts for 35 percent of its palladium supply. The White House urged semiconductor companies to diversify their supply chains, and prices on key commodities have risen in anticipation of supply disruptions.
In Translation
CSET’s translations of significant foreign language documents on AI
CSET’s translations of significant foreign language documents on AI
PRC S&T Progress Law: Law of the People’s Republic of China on Progress of Science and Technology. This law, amended in December 2021, regulates how Chinese technology-related industrial policies operate and how state guidance funds for S&T are run, stipulates punishments for those who embezzle S&T funds, and so on.
PRC Tech Commercialization Compendium: National- and Provincial-Level Scientific and Technological Achievement Conversion Promotion: A Policy Compendium. This document is an English-language translation of the table of contents of an extremely lengthy compendium of PRC national, provincial, and city-level laws, regulations, and policies related to the “conversion of S&T achievements into practical applications.”
PRC Smart Manufacturing Standards: Guidelines for the Construction of a National Smart Manufacturing Standards System (2021 Edition). This document details China’s plans to develop domestic standards for the smart manufacturing industry.
PRC Talent Recruitment Plans: 2020 National Foreign Expert Project Application Guide. This document briefly describes six different Chinese talent recruitment plans, all designed to entice foreign researchers, academics, administrators, or entrepreneurs to relocate to the PRC to enhance China’s strategic S&T capabilities.
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:
- Research Analyst (multiple): CSET RAs are vital to our work across a range of lines of research. Research Analysts collaborate with Research and Senior Fellows to execute CSET’s research. Applications close TOMORROW, February 25 and be sure to list your areas of research interest in your cover letter.
- Data Research Analyst (multiple): DRAs work alongside our analysis and data teams to produce data-driven research products and policy analysis. This role combines knowledge of research methods and data analysis skills. Those with experience in common data visualization, programming languages, and/or statistical analysis tools may find this position of particular interest. Applications close TOMORROW, February 25.
- Business Operations and Management Specialist: Reporting to CSET’s Director of Operations, the management specialist will have responsibility for sub-grant processing, contracts management and grants management for the entirety of CSET. Excel/gsheets skills are a must. Apply by March 4.
- Research Fellow – Cyber/AI: CSET’s CyberAI project is currently seeking Research Fellow candidates to focus on machine learning applications for cybersecurity to assess their potential and identify recommendations for policymakers. Apply by March 14.
What’s New at CSET
CSET MAP OF SCIENCE UPDATE
- The CSET Map of Science has undergone an update, with new features including the ability to filter research clusters by counts of papers cited by patents, to view papers from a cluster in the Dimensions web portal, and to explore cluster country collaboration counts over time.
- CSET: Data Snapshot: Exploring CSET’s PARAT by Sara Abdulla
- Defense One: To Get Better at AI, Get Better at Finding AI Talent by Diana Gehlhaus
- OECD — The AI Wonk blog: The OECD Framework for Classifying AI Systems to assess policy challenges and ensure international standards in AI by Jack Clark, Dewey Murdick, Karine Perset and Mark Grobelnik
- The Inoculation Podcast: Will Super-Powerful Artificial Intelligence Turbocharge Disinformation? featuring Micah Musser
- The John Batchelor Show: The AI Race with the PRC featuring Ryan Fedasiuk
- U.S.-China Economic and Security Review Commission: Dakota Cary testified on “China’s Cyber Capabilities: Warfare, Espionage and Implications for the United States.” Read his testimony and watch the full commission hearing.
- On February 16, CSET’s webinar More than Deepfakes: AI and the Future of Disinformation Campaigns featured a conversation between CSET Research Fellow Katerina Sedova and CSET Senior Fellow and Director of the CyberAI Project John Bansemer about countering the threat of automated disinformation.
- On February 22, CSET Director Dewey Murdick joined Karine Perset, Marko Grobelnik, Olivia Erdelyi, Sebastian Hallensleben and Viknesh Sounderajah for the launch of the OECD’s Framework for Classifying AI Systems.
- Munich Security Conference 2022: The annual Munich Security Report named the report by Ryan Fedasiuk, Jennifer Melot and Ben Murphy, Harnessed Lightning: How the Chinese Military is Adopting Artificial Intelligence, one of its “Food for Thought” recommended readings.
- NPR: For a story about the potential impacts of a Russian invasion of Ukraine, Becky Sullivan reached out to Research Fellow Katerina Sedova, who discussed the prospect of coordinated influence campaigns and attacks on critical infrastructure.
- National Defense Magazine: A Stew Magnuson article about information operations cited Sedova’s recent brief — co-authored with Christine McNeill, Aurora Johnson, Aditi Joshi and Ido Wulkan — AI and the Future of Disinformation Campaigns.
What We’re Reading
Report: Assessing Systemic Strengths and Vulnerabilities of China’s Defense Industrial Base: With a Repeatable Methodology for Other Countries, Cortney Weinbaum et al., The RAND Corporation (2022)
Article: TimeLMs: Diachronic Language Models from Twitter, Daniel Loureiro, Francesco Barbieri, Leonardo Neves, Luis Espinosa Anke and Jose Camacho-Collados, arXiv (February 2022)
Upcoming Events
- March 10: CSET Webinar, Bringing the Chipmakers Home: Attracting Manufacturers and the Talent to Sustain Them, featuring Will Hunt and John VerWey
What else is going on? Suggest stories, documents to translate & upcoming events here.