Center for Security and Emerging Technology

The emergence of carbon dots (CDs) has opened up an exciting new field in the science and technology of carbon nanomaterials and has attracted increasing interest in recent years. Due to their diverse physicochemical properties and favourable attributes, such as quantum confinement effects and abundant surface defects, CDs and their derived hybrids have shown exciting and indispensable prospects in the energy conversion and storage fields. Considering the latest developments, in this review, we comprehensively summarize the classification and structure of CDs. Three strategies for structural engineering of CDs are presented and analyzed, in terms of the tuning of size and crystallinity, and the methodologies for surface modification and heteroatom doping, with a focus on the relationship among the synthesis methods, structure and properties of the concerned CDs. More importantly, the recent advances in energy-oriented applications of CDs, including photo- and electro-catalysis, light-emitting diodes, photovoltaic cells, lithium/sodium ion batteries and supercapacitors, will be systematically highlighted. Finally, we discuss and outline the remaining major challenges and opportunities for CDs in the future.

Generative language models have improved drastically, and can now produce realistic text outputs that are difficult to distinguish from human-written content. For malicious actors, these language models bring the promise of automating the creation of convincing and misleading text for use in influence operations. This report assesses how language models might change influence operations in the future, and what steps can be taken to mitigate this threat. We lay out possible changes to the actors, behaviors, and content of online influence operations, and provide a framework for stages of the language model-to-influence operations pipeline that mitigations could target (model construction, model access, content dissemination, and belief formation). While no reasonable mitigation can be expected to fully prevent the threat of AI-enabled influence operations, a combination of multiple mitigations may make an important difference.

With the recent wave of progress in artificial intelligence (AI) has come a\ngrowing awareness of the large-scale impacts of AI systems, and recognition\nthat existing regulations and norms in industry and academia are insufficient\nto ensure responsible AI development. In order for AI developers to earn trust\nfrom system users, customers, civil society, governments, and other\nstakeholders that they are building AI responsibly, they will need to make\nverifiable claims to which they can be held accountable. Those outside of a\ngiven organization also need effective means of scrutinizing such claims. This\nreport suggests various steps that different stakeholders can take to improve\nthe verifiability of claims made about AI systems and their associated\ndevelopment processes, with a focus on providing evidence about the safety,\nsecurity, fairness, and privacy protection of AI systems. We analyze ten\nmechanisms for this purpose--spanning institutions, software, and hardware--and\nmake recommendations aimed at implementing, exploring, or improving those\nmechanisms.\n

Can large language models, a form of artificial intelligence (AI), generate persuasive propaganda? We conducted a preregistered survey experiment of US respondents to investigate the persuasiveness of news articles written by foreign propagandists compared to content generated by GPT-3 davinci (a large language model). We found that GPT-3 can create highly persuasive text as measured by participants' agreement with propaganda theses. We further investigated whether a person fluent in English could improve propaganda persuasiveness. Editing the prompt fed to GPT-3 and/or curating GPT-3's output made GPT-3 even more persuasive, and, under certain conditions, as persuasive as the original propaganda. Our findings suggest that propagandists could use AI to create convincing content with limited effort.

Growing popular and industry interest in high-performing natural language generation models has led to concerns that such models could be used to generate automated disinformation at scale. This report examines the capabilities of GPT-3--a cutting-edge AI system that writes text--to analyze its potential misuse for disinformation. A model like GPT-3 may be able to help disinformation actors substantially reduce the work necessary to write disinformation while expanding its reach and potentially also its effectiveness.

We describe the creation and quality assurance of a dataset containing nearly all available precinct-level election results from the 2016, 2018, and 2020 American elections. Precincts are the smallest level of election administration, and election results at this granularity are needed to address many important questions. However, election results are individually reported by each state with little standardization or data quality assurance. We have collected, cleaned, and standardized precinct-level election results from every available race above the very local level in almost every state across the last three national election years. Our data include nearly every candidate for president, US Congress, governor, or state legislator, and hundreds of thousands of precinct-level results for judicial races, other statewide races, and even local races and ballot initiatives. In this article we describe the process of finding this information and standardizing it. Then we aggregate the precinct-level results up to geographies that have official totals, and show that our totals never differ from the official nationwide data by more than 0.457%.

Purpose A common way of eliciting speech from individuals is by using passages of written language that are intended to be read aloud. Read passages afford the opportunity for increased control over the phonetic properties of elicited speech, of which phonetic balance is an often-noted example. No comprehensive analysis of the phonetic balance of read passages has been reported in the literature. The present article provides a quantitative comparison of the phonetic balance of widely used passages in English. Method Assessment of phonetic balance is carried out by comparing the distribution of phonemes in several passages to distributions consistent with typical spoken English. Data regarding the distribution of phonemes in spoken American English are aggregated from the published literature and large speech corpora. Phoneme distributions are compared using Spearman rank order correlation coefficient to quantify similarities of phoneme counts in those sources. Results Correlations between phoneme distributions in read passages and aggregated material representative of spoken American English ranged from .70 to .89. Correlations between phoneme counts from all passages, literature sources, and corpus sources ranged from .55 to .99. All correlations were statistically significant at the Bonferroni-adjusted level. Conclusions Passages considered in the present work provide high, but not ideal, phonetic balance. Space exists for the creation of new passages that more closely match the phoneme distributions observed in spoken American English. The Caterpillar provided the best phonetic balance, but phoneme distributions in all considered materials were highly similar to each other.

The prediction of exceptional or surprising growth in research is an issue with deep roots and few practical solutions. In this study, we develop and validate a novel approach to forecasting growth in highly specific research communities. Each research community is represented by a cluster of papers. Multiple indicators were tested, and a composite indicator was created that predicts which research communities will experience exceptional growth over the next three years. The accuracy of this predictor was tested using hundreds of thousands of community-level forecasts and was found to exceed the performance benchmarks established in Intelligence Advanced Research Projects Activity's (IARPA) Foresight Using Scientific Exposition (FUSE) program in six of nine major fields in science. Furthermore, 10 of 11 disciplines within the Computing Technologies field met the benchmarks. Specific detailed forecast examples are given and evaluated, and a critical evaluation of the forecasting approach is also provided.

Real-world harms caused by the use of AI technologies are widespread. Tracking and analyzing them improves our understanding of the variety of harms and the circumstances that lead to their occurrence once AI systems are deployed. This report presents a standardized conceptual framework for defining, tracking, classifying, and understanding harms caused by AI. It lays out the key elements required for the identification of AI harm, their basic relational structure, and definitions without imposing a single interpretation of AI harm. The brief concludes with an example of how to apply and customize the framework while keeping its modular structure.

The global AI industry is booming, with privately held firms pulling in nearly $40 billion in disclosed investment in 2019 alone. U.S. companies continue to attract the majority of that funding—64 percent of it in 2019—but that lead is not guaranteed. This report analyzes AI investment data from 2015 to 2019 to help better understand trends in the global AI landscape.

Our work analyzes the artificial intelligence and machine learning (AI/ML) research portfolios of six large research funding organizations from the United States [National Institutes of Health (NIH) and National Science Foundation (NSF)]; Europe [European Commission (EC) and European Research Council (ERC)]; China [National Natural Science Foundation of China (NNSFC)]; and Japan [Japan Society for the Promotion of Science (JSPS)]. The data for this analysis is based on 127,000 research clusters (RCs) that are derived from 1.4 billion citation links between 104.8 million documents from four databases (Dimensions, Microsoft Academic Graph, Web of Science, and the Chinese National Knowledge Infrastructure). Of these RCs, 600 large clusters are associated with AI/ML topics, and 161 of these AI/ML RCs are expected to experience extreme growth between May 2020 and May 2023. Funding acknowledgments (in the corpus of the 104.9 million documents) are used to characterize the overall AI/ML research portfolios of each organization. NNSFC is the largest funder of AI/ML research and disproportionately funds computer vision. The EC, RC, and JSPS focus more efforts on natural language processing and robotics. The NSF and ERC are more focused on fundamental advancement of AI/ML rather than on applications. They are more likely to participate in the RCs that are expected to have extreme growth. NIH funds the largest relative share of general AI/ML research papers (meaning in areas other than computer vision, natural language processing, and robotics). We briefly describe how insights such as these could be applied to portfolio management decision-making.

Much of the research and discourse on risks from artificial intelligence (AI) image generators, such as DALL-E and Midjourney, has centered around whether they could be used to inject false information into political discourse. We show that spammers and scammers—seemingly motivated by profit or clout, not ideology—are already using AI-generated images to gain significant traction on Facebook. At times, the Facebook Feed is recommending unlabeled AI-generated images to users who neither follow the Pages posting the images nor realize that the images are AI-generated, highlighting the need for improved transparency and provenance standards as AI models proliferate.

One sentence summarizes the complexities of modern artificial intelligence: Machine learning systems use computing power to execute algorithms that learn from data. This AI triad of computing power, algorithms, and data offers a framework for decision-making in national security policy.

A globally competitive AI workforce hinges on the education, development, and sustainment of the best and brightest AI talent. This issue brief compares efforts to integrate AI education in China and the United States, and what advantages and disadvantages this entails. The authors consider key differences in system design and oversight, as well as strategic planning. They then explore implications for the U.S. national security community.

This report examines nearly 350 artificial intelligence-related equipment contracts awarded by the People’s Liberation Army and state-owned defense enterprises in 2020 to assess how the Chinese military is adopting AI. The report identifies China’s key AI defense industry suppliers, highlights gaps in U.S. export control policies, and contextualizes the PLA’s AI investments within China’s broader strategy to compete militarily with the United States.

This paper is the first 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. In it, the authors introduce three categories of AI safety issues: problems of robustness, assurance, and specification. Other papers in this series elaborate on these and further key concepts.

China aims to become “the world’s primary AI innovation center” by 2030. Toward that end, the Chinese government is spending heavily on AI research and development (R&D)—but perhaps not as heavily as some have thought. This memo provides a provisional, open-source estimate of China’s spending.

As modern machine learning systems become more widely used, the potential costs of malfunctions grow. This policy brief describes how trends we already see today—both in newly deployed artificial intelligence systems and in older technologies—show how damaging the AI accidents of the future could be. It describes a wide range of hypothetical but realistic scenarios to illustrate the risks of AI accidents and offers concrete policy suggestions to reduce these risks.

This report presents research and findings on accountability and risk in AI systems by providing an overview of how risk-management frameworks and the AI system lifecycle can be integrated to promote trustworthy AI. It also explores processes and technical attributes that can facilitate the implementation of values-based principles for trustworthy AI and identifies tools and mechanisms to define, assess, treat, and govern risks at each stage of the AI system lifecycle. This report leverages OECD frameworks – including the OECD AI Principles, the AI system lifecycle, and the OECD framework for classifying AI systems – and recognised risk-management and due-diligence frameworks like the ISO 31000 risk-management framework, the OECD Due Diligence Guidance for Responsible Business Conduct, and the US National Institute of Standards and Technology’s AI risk-management framework.

How do we measure leadership in artificial intelligence, and where does the United States rank? This policy brief examines potential AI strengths of the United States and China and prescribes recommendations to ensure the United States remains ahead.