Killer Robots Used in Libya? Not So Fast, Experts Say: A recent report from the UN Panel of Experts on Libya sparked concerns that a lethal autonomous weapons system (LAWS) — “killer robots,” colloquially — had been used in the field for the first time. The report, which detailed the fighting between the UN-recognized government and forces loyal to Khalifa Haftar, claimed that government forces had used a LAWS — namely the Turkish-made Kargu-2 drone — to track and engage Haftar-linked forces. The story was picked up by several major news outlets, with many framing it as a potential watershed in the history of autonomous weapons. But some experts — including Ulrike Franke, Jack McDonald and Michael Horowitz — pushed back, saying that the details in the report do little to distinguish the Kargu-2 system from loitering munitions, which have been in use for decades. As Zachary Kallenborn summarized in the Bulletin of the Atomic Scientists, the vague language of the UN report and the possibility that the Kargu-2’s manufacturer exaggerated its capabilities make it difficult to know how “autonomous” the lethal system was.
Chinese Researchers Announce the Largest “Large Language Model” Yet: A new natural language processing (NLP) model announced last week by the state-funded Beijing Academy of Artificial Intelligence (BAAI) is the largest ever trained. Wu Dao 2.0 has 1.75 trillion parameters — dwarfing GPT-3’s 175 billion parameters and even the 1.6 trillion parameters of Google’s Switch Transformer — and while the relationship between parameters and sophistication is not one-to-one, it is generally a good indicator of a model’s power. In addition to its high parameter count, Wu Dao 2.0 does more than just NLP — it is a multimodal system trained on 4.9 TB of text and images, meaning it can perform image recognition and generation tasks in addition to the text processing and generation tasks of traditional NLP. While BAAI has yet to publish a paper elaborating on the performance of Wu Dao 2.0, a handful of released results showed impressive performance: The model achieved state-of-the-art results on nine common benchmarks, surpassing previous juggernauts such as OpenAI’s GPT-3 and CLIP and Microsoft’s Turing-NLG.
- More: CSET: Truth, Lies, and Automation: How Language Models Could Change Disinformation | How Large Language Models Will Transform Science, Society, and AI | On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
Senate Passes Massive China Competitiveness Bill: The Senate passed the U.S. Innovation and Competition Act by a vote of 68-32 on Tuesday, advancing a bill that, if signed into law, would pour more than $200 billion into U.S. R&D and high tech manufacturing. The bill has undergone a significant transformation since its introduction as the Endless Frontier Act in April. The original 160-page proposal grew, through a series of amendments, to more than 2,400 pages, and its price tag more than doubled. The bill now includes $52 billion in emergency appropriations to support semiconductor manufacturing and research incentives and includes a host of new provisions related to the U.S. research and development ecosystem. It directs $81 billion to the National Science Foundation, $29 billion of which is meant for a new directorate focused on emerging technology. Despite its bipartisan success in the Senate, observers expect the bill to face stiffer opposition in the House. The House Science and Foreign Affairs Committees continue to work on alternative proposals to the Senate-passed bill, with the House Science Committee reportedly preparing to consider its NSF for the Future Act next week. For more on this legislation and other Congressional news, see our latest legislative roundup.
Biden Budget Includes AI Funding: President Biden released his budget request late last month, a $6 trillion proposal with substantial funding for AI and emerging technology. While the request includes a significant increase in federal R&D funding — to $176.26 billion, a 9 percent increase compared to the FY2021 budget — many observers were surprised by the relatively small increases proposed for the DOD’s R&D budget. As Patrick Tucker noted in Defense One, the DOD’s $112 billion RDT&E budget request likely means an inflation-adjusted increase of only 0.6 percent. The department’s $874 million earmarked specifically for AI is also a far cry from the $8 billion recommended in the final report of the National Security Commission on Artificial Intelligence earlier this year. As Brandi Vincent noted in NextGov, the appendix published with the budget was relatively short on references to AI and other emerging technologies compared to previous budget requests. While that doesn’t mean AI and emerging technology research has been put on the back burner, as Bloomberg Government’s Chris Cornillie told Vincent, without more details, “it’s hard to know where we stand.”
White House Expands Blacklist of Chinese Companies: Last week, President Biden issued an executive order expanding restrictions on U.S. investment in companies with ties to the Chinese military, which President Trump originally imposed last November. The new order increases the number of targeted companies from 31 to 59 and revises the criteria for listed entities. Most notably, it aims to avoid litigation concerns by focusing on certain sectors of the Chinese economy — namely those that develop surveillance platforms used to “facilitate repression or serious human rights abuse” — rather than the DOD’s list of companies via NDAA Section 1237, under which President Trump’s November and January EO’s operated. Those EOs had faced hurdles in court, with several Chinese companies successfully challenging their inclusion in the list. Senior Biden administration officials reportedly said the changes in the new order were meant to ensure it had stronger legal footing. The new order will go into effect on August 2.
Pentagon Steers Toward Responsible AI Plan: Late last month, Deputy Secretary of Defense Kathleen Hicks issued a memo outlining the steps the Pentagon will take to implement “Responsible AI” (RAI) across the DOD. Last year, the Pentagon adopted five “DOD AI Ethical Principles,” (which we covered at the time). The new memo reaffirms those principles and lists key tenets the department will adhere to when implementing RAI, including ensuring trust between warfighters and AI systems and building an “RAI-ready” workforce. The memo names the Joint Artificial Intelligence Center as the coordinating body for RAI’s implementation and directs it to form and train an “RAI Working Council,” which will develop an RAI strategy and implementation plan, a talent management plan, and an AI acquisition plan over the coming months.
CSET Job Openings
Please share with qualified candidates in your network or consider applying:
- Policy Communications Analyst will help advance our written and visual products by advising researchers on report clarity, narrative flow, cogency, presentation effectiveness and parsimony. Deadline extended! Applications now due by June 14
- Research Fellow – AI TEV&V will focus on the safety and risk of deployed AI systems by researching real-world AI incidents and use these identified incidents with other analyses of AI systems to inform policy recommendations regarding AI safety, test, evaluation, verification and validation (TEV&V) processes, standards setting and management, and the appropriate employment and operation of AI systems by businesses and the US Government (including the military). Applications due by July 1
CSET’s translations of significant foreign language documents on AI
CSET’s translations of significant foreign language documents on AI
PRC 5G Plan: “Set Sail” Action Plan for 5G Applications (2021-2023) (Draft for Comments). This document is a draft of an ambitious Chinese plan to apply 5G technology to key industries and to society at large. The plan sets many qualitative goals and a few quantitative targets for China’s 5G industry to reach by 2023.
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.
What’s New at CSET
- Security, Collaboration, and the Changing Map of Global R&D by Melissa Flagg, Autumn Toney and Paul Harris
- Comparing the United States’ and China’s Leading Roles in the Landscape of Science by Autumn Toney and Melissa Flagg
- Council on Foreign Relations — Net Politics Blog: China’s Internet Trolls Go Global by Ryan Fedasiuk
- CSET: Legislative Roundup: Concerns about China — Surveillance, STEM Competition and More — Dominate Recent S&T Legislation by Daniel Hague
CSET maintains a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:
- (Closing Soon) What percentage of software engineer job postings between July 1 and September 30, 2021, inclusive, will allow for remote work?
- (Closing Soon) How much funding will U.S. tech startups raise between July 1 and December 31, 2021, inclusive?
- Financial Times: Yesterday, FT’s #techAsia newsletter highlighted Autumn Toney and Melissa Flagg’s new data brief comparing the research output of the United States and China, noting that the paper “reveals crucial data.” In an earlier recent edition, #techAsia referred to the recent brief China’s Foreign Technology Wish List by Ryan Fedasiuk, Emily Weinstein and Anna Puglisi as “groundbreaking.”
- The Wall Street Journal: China’s Foreign Technology Wish List also earned a mention in The Wall Street Journal, which reached out to Puglisi for an article about an alleged case of Chinese economic espionage in Tennessee.
- Wired: Helen Toner spoke with Wired for a story on the AI Incident Database, to which CSET has been contributing detailed data and scholarship.
- KCBS: CSET Senior Fellow Andrew Lohn joined San Francisco radio station KCBS to discuss his recent report on disinformation with Ben Buchanan, Micah Musser and Katerina Sedova, Truth, Lies, and Automation.
- Slate: Slate’s “War Stories” columnist, Fred Kaplan, spoke to Emily Weinstein for an article about the Biden administration’s China Strategy.
- Fortune: An opinion piece in Fortune about China’s response to COVID-19 cited Weinstein’s 2020 data brief, China’s Use of AI in its COVID-19 Response.
What We’re Reading
Report: Building Resilient Supply Chains, Revitalizing American Manufacturing, and Fostering Broad-Based Growth: 100-Day Reviews under Executive Order 14017, the White House (June 2021)
Interactive: Mapping China’s Tech Giants, ASPI (June 2021)
Report: China’s Quest for Global Primacy: An Analysis of Chinese International and Defense Strategies to Outcompete the United States, Timothy R. Heath, Derek Grossman and Asha Clark, RAND Corporation (2021)
Article: The failed promise of Kenya’s smart city, Carey Baraka, Rest of World (June 2021)
- June 10: CNAS, Russia: An Assumptions Check, featuring Margarita Konaev
- June 17: Stanford HAI and CSET, The E.U. AI Act: A Risk-Based Policy Approach to AI Applications, featuring Andrew Imbrie
- June 22: CSET Foretell, Foretell in Conversation: A New Approach to Geopolitical Forecasting with Professor Michael Horowitz, moderated by Michael Page
- June 23: CNA, NSCAI and CSET, Requirements for Leveraging AI featuring Diana Gehlhaus
- June 24: CSET Webinar, Where Does India Stand in the Global AI Race? featuring Husanjot Chahal and moderated by Melissa Flagg
What else is going on? Suggest stories, documents to translate & upcoming events here.