This paper is the fourth installment in a series on “AI safety,” an area of machine learning research that aims to identify causes of unintended behavior in machine learning systems and develop tools to ensure these systems work safely and reliably. The first paper in the series, “Key Concepts in AI Safety: An Overview,” outlined three categories of AI safety issues—problems of robustness, assurance, and specification—and the subsequent two papers described problems of robustness and assurance, respectively. This paper introduces specification as a key element in designing modern machine learning systems that operate as intended.

See our original translation of a 2020 document from China's standards-setting body calling for specific standards for AI technologies.

See our original translation of a Standards Development Announcement from China's state press agency, Xinhua News Agency.

See our original translation of a Chinese state-run think tank's report assessing the country's computing power.

This research agenda provides a roadmap for the next phase of CSET’s line of research on the U.S. AI workforce. Our goal is to assist policymakers and other stakeholders in the national security community to create policies that will ensure the United States maintains its competitive advantage in AI talent. We welcome comments, feedback and input on this vision at cset@georgetown.edu.

Data Snapshots are informative descriptions and quick analyses that dig into CSET’s unique data resources. Our first series of Snapshots introduced CSET’s Map of Science and explored the underlying data and analytic utility of this new tool, which enables users to interact with the Map directly.

This data brief explores how international collaboration relates to the impact and output of research publications. Focusing on the top 10 countries with the highest publication output from 2010 to 2019, the authors provide a comprehensive analysis across the major fields of science and technology.

This brief explores the development and testing of artificial intelligence system classification frameworks intended to distill AI systems into concise, comparable and policy-relevant dimensions. Comparing more than 1,800 system classifications, it points to several factors that increase the utility of a framework for human classification of AI systems and enable AI system management, risk assessment and governance.

CSET's Emily Weinstein and Ainikki Riikonen of the Center for a New American Security submitted this comment to the Office of Science and Technology Policy in response to a request for recommendations on securing U.S. government research and development against foreign government interference and exploitation.

Devices based on superconductor electronics can achieve much higher energy efficiency than standard electronics. Research in superconductor electronics could advance a range of commercial and defense priorities, with potential applications for supercomputing, artificial intelligence, sensors, signal processing, and quantum computing. This brief identifies the countries most actively contributing to superconductor electronics research and assesses their relative competitiveness in terms of both research output and funding.