EU Leaders Reach Agreement on AI Act — Ratification Still in the Works: In December, EU leaders reached an agreement on the bloc’s landmark AI Act. The sweeping regulatory package will set important new standards for AI development and deployment within the EU, and — much like earlier EU regulatory efforts such as the General Data Protection Regulation (GDPR) — will likely have a significant impact on AI development abroad. While the official compromise text has not yet been released, an EU press release laid out the key pillars of the agreed-upon regulation. The AI Act will:
Prohibit AI systems for untargeted facial image scraping, emotion recognition in workplaces and schools, social scoring, manipulation of human behavior, and sensitive biometric categorization (though the act includes law enforcement exemptions for some uses of biometric identification systems).
Include obligations for systems classified as “high risk,” such as those used in healthcare, environmental applications, and elections. Such systems will be subjected to mandatory fundamental rights impact assessment among other requirements.
Impose transparency and compliance requirements on “general-purpose AI” systems (also known as foundation models). These requirements include publishing technical documentation, publishing information about content used for training, and compliance with EU copyright law.
Using AI to Speed Up Scientific Research: A handful of recent papers show how AI can be used to accelerate scientific discovery:
A research team from MIT and Harvard used AI to identify a new class of antibiotics that is effective against two of the most dangerous types of drug-resistant bacteria: methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci (VRE). According to their paper published in Nature, the researchers used a deep learning-based system to screen millions of compounds for antibiotic properties. Their use of graph-based search algorithms meant that the AI predictions were more “human-understandable” than alternative “black box” AI approaches — an important factor in helping researchers understand the system’s predictions and potentially identify related antibiotics. Two of the AI-identified candidates successfully killed both MRSA and VRE and proved effective in mice, but they will still need to go through the long process of clinical trials before they can be approved for human use.
AI and Emerging Tech in the FY2024 NDAA: Before breaking for the holidays, Congress wrapped up its consideration of the Fiscal Year 2024 National Defense Authorization Act, passing the critical authorization bill for the 63rd year running. The bill contains a number of AI provisions:
It includes provisions supporting the use of AI for optimizing aerial refueling in contested environments (Sec. 346) and maintenance operations at shipyards serving the U.S. Navy (Sec. 350).
It authorizes the Chief Digital and Artificial Intelligence Officer to access and control any DOD data and establishes a CDAO Government Council to oversee the ethical collection and use of such data (Sec. 1521).
It directs CDAO to develop and maintain data assets that support cyber operations preparations (Sec. 1523).
It directs the DOD to develop a bug bounty program for foundational models used by the department (Sect. 1542) and a prize competition to develop watermarking and other detection technologies for identifying content created by generative AI (Sec. 1543).
It tasks the DOD with defining guidance for near- and long-term strategies for the adoption of AI, as well as policies to support its ethical and secure use (Sec. 1544).
On December 13, the Senate approved the final conference report negotiated between the two chambers by a vote of 87 to 13. The House of Representatives followed a day later, approving by a vote of 310 to 118 and the President signed it into law on December 22.
NSF to Stand Up NAIRR Pilot Next Week: Next week, the National Science Foundation will launch a pilot program for the National AI Research Resource, as directed by President Biden’s November executive order on AI. The NAIRR — envisioned as “a shared research infrastructure” that would provide computing power, access to open government and non-government datasets, and training resources to students and AI researchers — has been in the works since 2020, when the National AI Initiative Act of 2020 (passed as part of the FY2021 NDAA) established a task force to research its feasibility. The task force published its final report early last year, in which it recommended a $2.6 billion investment over six years to build out the NAIRR in full, as well as a pilot program to make computing resources available sooner. The president’s November EO directed the head of the NSF to identify and enlist computational, data, software, and training resources from across the Federal government and private sector for the pilot program. NSF officials told reporters that the initial effort will be a “modest… proof-of-concept effort” comprising “resources that we have in hand” and “in-kind contributions from different technology companies.” It’s not clear exactly how robust the resourcing will be, but it will likely be a far cry from the computing power recommended by the task force for the full NAIRR. NSF officials said the goal of the pilot is to prove the value of the NAIRR concept. Should it prove a success, it will be up to Congress to fund the full project.
GAO Reports Explore DOD’s AI Workforce and Federal Agencies’ AI Adoption: Two recent Government Accountability Office reports highlight the steps the federal government still needs to take to fully realize its AI ambitions:
A study on the Pentagon’s AI workforce found that the DOD had not sufficiently defined and identified its AI workforce, hindering its ability to effectively meet its strategic goals and objectives. As the GAO report noted, the DOD has previously identified the cultivation of AI expertise as a strategic focus area, but without consistent definitions, the DOD can’t effectively assess progress or confidently set future goals. GAO recommended that the CDAO should be tasked with sufficiently defining and identifying the DOD’s AI workforce.
A review of 23 federal agencies’ AI adoption found that the agencies had identified approximately 1,200 AI use cases, most of which were in the planning phase. While agencies like NASA and the Department of Commerce reported the highest number of AI use cases, GAO found that overall AI implementation remains inconsistent. Many agencies have not fully met federal AI requirements, with incomplete or inaccurate data in their AI use case inventories and a lack of comprehensive implementation planning. GAO recommends that agencies update their AI inventories, align them with guidelines, and implement AI requirements in federal law and policy to enhance management and oversight of AI applications.
In Translation CSET’s translations of significant foreign language documents on AI
PRC Computing Power Policy:Action Plan for the High-Quality Development of Computing Power Infrastructure. This document is a Chinese government policy for the near-term development of computing power. The plan urges the expansion of compute in China, particularly of supercomputing and “intelligent compute” optimized for AI applications. But the policy also emphasizes improving the energy efficiency and lowering the carbon footprint of computing power infrastructure such as data centers. An appendix includes various compute-related metrics for China to strive for in 2023, 2024, and 2025.
Beijing Municipal Plan:Beijing Municipal Implementation Plan for Promoting the Innovative Development of Future Industries. This Beijing municipal government plan identifies 20 “future industries” that Beijing is targeting with favorable industrial policies, so as to build up world-class companies in these industries by 2035. The industries that Beijing — home of China’s best universities and a disproportionately high number of its leading tech companies — is boosting are in the AI, healthcare, manufacturing, energy, materials, and space sectors.
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.
Necessary cookies are absolutely essential for the website to function properly. This category only includes cookies that ensures basic functionalities and security features of the website. These cookies do not store any personal information.
Any cookies that may not be particularly necessary for the website to function and is used specifically to collect user personal data via analytics, ads, other embedded contents are termed as non-necessary cookies. It is mandatory to procure user consent prior to running these cookies on your website.