Biden signs landmark tech competitiveness bill, DeepMind models more than 200 million proteins, and Meta’s new chatbot gets a demo (and gets in trouble)

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 Models More than 200 Million Proteins: DeepMind, the UK-based, Alphabet-owned AI lab, announced it had used AI to model more than 200 million protein structures — covering nearly every protein with available sequence data — and made the structures publicly available on a free online database. AlphaFold, DeepMind’s protein-modeling system, was hailed as “a breakthrough of the first order” when it was unveiled in 2020. Because proteins’ structures are closely related to their functions, observers speculated that AlphaFold’s ability to quickly and accurately predict protein structures (a painstaking process for previous state-of-the-art methods) could help accelerate drug discovery, biochemical research, and more. Last year, the company released AlphaFold’s source code, published a paper explaining how it worked, and, together with the European Molecular Biology Laboratory, made the predicted structures of hundreds of thousands of proteins available on its public database. Already, DeepMind says its tools have been accessed by more than half a million researchers. While the accuracy of most of its predictions has not been validated, observers say AlphaFold has already proved beneficial, and they are optimistic the massive expansion of the database will expand its usefulness further.

More: DeepMind has predicted the structure of almost every protein known to science | Drug Discovery Is About to Get Faster. Thank AI. | Google AI Blog: Towards Reliability in Deep Learning Systems

China’s SMIC Ships a 7nm Chip, Raising Concerns and Some Eyebrows: SMIC, China’s largest chipmaker, has reportedly shipped chips manufactured using a 7-nanometer process — a significant jump over the company’s earlier chips and roughly equivalent to the highest-end chips produced in the United States. While nanometer-based designations are now more marketing than a definitive measurement, SMIC’s new chips reportedly bear a strong resemblance to TSMC’s 7nm chips according to TechInsights, which conducted the initial review of the SMIC chip. TSMC’s chips represented the state of the art when they were originally shipped by the Taiwanese semiconductor giant in 2018, and are still more than powerful enough to run advanced AI applications. But as a number of observers have pointed out, there remains some reason for skepticism about China’s chipmaking prowess. As CSET’s Will Hunt noted in an interview with Protocol, evidence of limited production doesn’t mean SMIC will be capable of manufacturing 7nm at competitive scale. Indeed, as Rupert Goodwins pointed out last week, SMIC’s failure to ramp up demand during an unprecedented shortage of high-end chips likely indicates production difficulties behind the scenes. Furthermore, 7nm chips are the last generation of semiconductors that can be produced with the equipment China currently has access to. With the Dutch government blocking exports to China of ASML’s extreme ultraviolet lithography machines, Chinese chipmakers could be stuck at 7nm for the foreseeable future. Still, the news may add urgency to the U.S. government’s campaign to restrict China’s access to the semiconductor manufacturing equipment used to make 7nm chips. As we covered in the last edition of policy.ai, the U.S. government has reportedly been pushing the Dutch and Japanese governments to restrict China’s access to deep ultraviolet lithography machines, which are needed to produce chips up to the 7nm node.

More: A 70-Year-Old Taiwanese Chip Wizard Is Driving China’s Tech Ambitions | China’s SMIC Is Shipping 7nm Foundry ASICs

Meta Chatbot Gets “In the Wild” Training — But Not Without Controversy: Last week, Meta AI released a public demo of BlenderBot 3, a 175-billion-parameter, AI-powered chatbot. The company also gave a look under BlenderBot’s hood, publicly releasing the model’s weights, its code, its datasets, and model cards that provide a higher-level overview. That level of openness is in keeping with some of Meta AI’s recent releases — in May, it shared the source code and training information for several large language models it had developed, including OPT-175B (which was also used to develop BlenderBot 3). But in the week since the demo was released, the internet has been awash in examples of the chatbot’s shortcomings. Its posts have ranged from offensive, to untruthful and hallucinatory (it claimed to be a plumber), to insubordinate (it called Meta boss Mark Zuckerberg “too creepy and manipulative”), to downright ridiculous (it said Andrew Lloyd Webber’s “Cats” was its favorite musical). These issues did not come as much of a surprise to observers — many were quick to remember “Tay,” Microsoft’s 2016 chatbot that was swiftly shut down after spouting off a range of offensive opinions — but it also did not seem to surprise Meta AI. In its announcement post, the lab warned that such problems were a likely result of training the system “in the wild.” In an August 8 addendum to the launch post, the lab reiterated that the demo would help it build “truly robust conversational AI systems” and committed to “push ahead” with the project.

More: Artificial Intelligence Incident Database | Chess robot grabs and breaks finger of seven-year-old opponent | 30% of Google’s Emotions Dataset is Mislabeled

Government Updates

President Biden Signs Major Tech Competitiveness Bill: President Biden signed the CHIPS and Science Act into law on Tuesday, capping a years-long legislative effort and directing billions of dollars into domestic semiconductor manufacturing while authorizing billions more for scientific research. The bill underwent several iterations as it made its way through Congress — the Senate passed the U.S. Innovation and Competition Act last summer, and the House passed its own version, the America COMPETES Act, earlier this year — and though both measures contained significant domestic chipmaking subsidies and funds for scientific research, a seemingly stalled reconciliation process raised concerns that the legislation would fail to reach the president’s desk. But a late push yielded a 1054-page, $280 billion bill (section-by-section summaries available here) that earned enough bipartisan support to pass both chambers. The final bill included a number of provisions relevant to AI and emerging technology, such as:

More than $52 billion in subsidies and a 25 percent tax credit for investments in domestic semiconductor manufacturing.
A $20 billion authorization for the National Science Foundation’s new Directorate — the Directorate for Technology, Innovation and Partnerships, which launched earlier this year — tasked with supporting domestic development of critical technologies in AI, energy, and other areas.
$13 billion authorized for NSF-funded STEM workforce development, including AI scholarship-for-service and microelectronics education programs.
$9 billion authorized for NIST, including funding for research and standards development in industries of the future.

The bill’s passage was applauded by the semiconductor industry, which had lobbied heavily for it. On Tuesday, Micron announced plans to spend $40 billion by 2030 on U.S. chipmaking, while Qualcomm and GlobalFoundries announced a $4.2 billion partnership that will expand GlobalFoundries’ New York semiconductor fabrication facility. Those companies join the likes of Intel, TSMC and Samsung, which are planning to spend billions on U.S. fabs in the coming years.

Pentagon Officials Raise Concerns About JADC2 Rollout: The Pentagon’s plan to connect and coordinate its services’ sensors — the Joint All Domain Command and Control (JADC2) strategy — may have hit some speed bumps, if recent comments are any indication. In separate events last month, Wanda Jones-Heath, the Air Force’s principal cyber advisor, and Doug Bush, the assistant secretary of the Army for acquisition, logistics and technology, suggested that the strategy lacked adequate coordination between the services’ efforts. “We are not aligned with what we need to be to be interoperable to be able to fight together,” Jones-Heath said at a Potomac Officers Club event. “Someone needs to just push us where we need to go.” Bush, meanwhile, suggested that a high-level JADC2 office — modeled on something like the Joint Counter-small Unmanned Aircraft System Office — together with a larger joint exercise, could help to bring the services’ efforts closer together. Jones-Heath and Bush don’t appear to be alone in their concerns — as part of its markup of the FY 2023 National Defense Authorization Act, the House Armed Services Committee asked both the Pentagon and the Government Accountability Office for reports on the DOD’s JADC2 efforts.

A Major Data Privacy Bill Got Through Committee — Can it Pass?: A major data privacy bill soared through the House Energy and Commerce Committee with strong bipartisan support — the first federal consumer privacy bill to make it out of committee. The American Data Privacy and Protection Act, approved by a vote of 53–2, would require individuals’ consent for collection and use of several types of “sensitive” data, including genetic and biometric information, precise geolocation details, information on sexual orientation or behavior, and more. The bill takes an expansive approach to enforcement — in addition to federal authorities, the ADPPA would allow state attorneys general or “state privacy authorities” (such as the California Privacy Protection Agency) to enforce its provisions. It would also allow individuals to bring civil cases against violators. But important sticking points remain — see Brookings’ Cameron F. Kerry’s writeup for an excellent overview — and passage through the Senate is not guaranteed. Sen. Cantwell was a vocal critic of an earlier draft of the bill, and as chair of the Senate Commerce Committee, her support will be key to its passage in that chamber. It’s not yet clear whether the changes in the most recent version of the bill will be enough to win over Cantwell and other critics.

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

PRC Academy of Engineering Budget: Chinese Academy of Engineering 2022 Annual Budget. This document is the 2022 budget of the Chinese Academy of Engineering, a state-run body of professional Chinese engineers that provides technology policy advice to the government and runs laboratories focused on applied technology research.

PRC Association of Science and Technology Budget: China Association of Science and Technology 2022 Budget. This document is the 2022 budget of the China Association of Science and Technology, a Communist Party-led professional association for Chinese scientists. CAST is funded by the Chinese government and promotes scientific ethics and science popularization. Among other projects, this year’s budget funds an effort to reform the country’s Chinese- and English-language scientific journals.

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 Fellow — AI Assessment: This Research Fellow will focus on standards, testing, evaluation, safety and national security issues associated with AI systems. To do this, they will examine how the limitations, risk, and society and security impacts of AI can be understood and managed. Apply by August 15.
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. Apply by August 15.
Research Fellow — CyberAI: This Research Fellow will focus on exploring topics at the intersection of AI and cyber. AI techniques, specifically machine learning, offer an opportunity to uplift cyber defenses, but they also may create new threats to cyber infrastructure. This Research Fellow will examine potential machine learning (ML) applications to identify associated challenges and make relevant policy recommendations. A key to success will be the ability to explain the implications of these emergent capabilities to a non-technical audience. Apply by August 15.
Visual Communications Specialist: The Visual Communications Specialist will support the External Affairs Team in raising the profile of CSET researchers and our research through a variety of outreach activities, with a focus on graphics, multimedia, and publications. Apply by September 16.

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

REPORTS

Counting AI Research: Exploring AI Research Output in English- and Chinese-Language Sources by Daniel Chou
Decoupling in Strategic Technologies: From Satellites to Artificial Intelligence by Tim Hwang and Emily Weinstein
U.S. High School Cybersecurity Competitions: Building Cyber Talent Through Extracurricular Activities by Kayla Goode, Ali Crawford and Christopher Back
Will AI Make Cyber Swords or Shields? by Andrew Lohn and Krystal Jackson

PUBLICATIONS AND PODCASTS

Foreign Affairs: Making War More Difficult to Wage: How Ukraine Is Changing the West’s Use of Export Controls by Emily Weinstein
Council on Foreign Relations: Why States Need an AI Education Agenda–Now! by Diana Gehlhaus
Center for New American Security: Regenerate: Biotechnology and U.S. Industrial Policy by Ryan Fedasiuk
Scientific American: The Biden Administration Must Designate Civilian Satellites Critical Infrastructure by Dakota Cary
On AiR Podcast: AI and the Quad, featuring Husanjot Chahal and Ngor Luong

TESTIMONY

U.S.-China Economic and Security Review Commission: On August 3, CSET Research Fellow Emily Weinstein testified before the U.S.-China Economic and Security Review Commission at a hearing on “Challenges from Chinese Policy in 2022: Zero-COVID, Ukraine, and Pacific Diplomacy.” Read her testimony or watch it here.

EVENT RECAPS

On July 28, the CSET webinar Creating an Innovation Workforce for Uncertain Times featured a conversation between Department of Labor Chief Innovation Officer Chike Aguh and CSET’s Anna Puglisi and Diana Gehlhaus about how the United States can prepare an innovative workforce for the future.

IN THE NEWS

The Washington Post: Jeanne Whalen interviewed Research Analyst Will Hunt about hiring challenges as the U.S. semiconductor industry ramps up production.
The New York Times: Farah Stockman cited Hunt’s February brief Reshoring Chipmaking Capacity Requires High-Skilled Foreign Talent multiple times in an opinion piece about global semiconductor competition.
Politico: Brendan Bordelon and Eleanor Mueller also cited Hunt’s February brief in an article about the semiconductor industry’s need for immigration reform.
The Wire China: An Eliot Chen overview of the Chips and Science Act rounded out the hat trick with another citation of Hunt’s brief.
FedScoop: Jon Harper recapped Ryan Fedasiuk, Karson Elmgren and Ellen Lu’s brief Silicon Twist: Managing the Chinese Military’s Access to AI Chips in a recent article.
NBC News: Fedasiuk spoke with Marc Caputo about the Chinese government’s access to TikTok’s data.
Defense One: Fedasiuk commented on how an invasion of Taiwan would affect global semiconductor supplies for a Patrick Tucker article that also cited Fedasiuk, Elmgren and Lu’s brief.
CNN: Director of Biotechnology Programs and Senior Fellow Anna Puglisi discussed the federal government’s response to PRC espionage with Katie Bo Lillis.
The Diplomat: A Justin Feng article about Chinese dependence on foreign semiconductor manufacturing equipment cited Saif Khan’s 2021 brief, Securing Semiconductor Supply Chains.
Nature: In a piece about PhD attainment rates, Virginia Gewin recapped the findings of Remco Zwetsloot, Jack Corrigan, Emily Weinstein, Dahlia Peterson, Diana Gehlhaus and Ryan Fedasiuk’s 2021 brief, China is Fast Outpacing U.S. STEM PhD Growth.

What We’re Reading

Report: Semiconductor Supply Chain: Policy Considerations from Selected Experts for Reducing Risks and Mitigating Shortages, United States Government Accountability Office (July 2022)

Article: Artificial Intelligence with American Values and Chinese Characteristics: A Comparative Analysis of American and Chinese Governmental AI Policies, Emmie Hine and Luciano Floridi, AI & Society (June 2022)

Old Meets New in Online Influence

This website uses cookies.