North Korea’s International Network for Artificial Intelligence Research
In recent years, North Korea has undertaken substantial efforts to bolster its artificial intelligence (AI) capabilities. These endeavors include reforming legal and institutional frameworks and promoting specialized AI education programs within academia. However, the extent to which the nation is cultivating talent and acquiring necessary technology for AI development beyond its borders remains uncertain. Because AI is a dual-use technology with potential military applications, exploring North Korea’s international collaboration in AI development is crucial, particularly given United Nations Security Council Resolution 2321, adopted in November 2016, which prohibits scientific and technical cooperation with North Korea.
As part of 38 North’s AI-focused series, this report investigates North Korea’s foreign collaborations in AI development, drawing from analysis of co-authorships of DPRK scientific journal articles on AI, and select case studies to get a better sense of the country’s position within the global AI landscape. The findings are significant, especially considering North Korea’s continued engagement in collaborative AI research with foreign universities and individuals—especially those in China, the US and South Korea—where stringent sanctions and export controls are implemented.
Methodology
Capturing a comprehensive view of a country’s international collaborations for AI development at the national level is challenging. AI technology is comprised of a broad set of techniques akin to a field of study that lacks a single and universally adopted definition. It also serves as a horizontal technology, applicable across various platforms to enhance functionality rather than as a standalone product. Consequently, this study employs a keyword-based analysis focusing on AI-related technical terms rather than conceptual or end-use-based words to acquire more direct insight into North Korea’s research activities.
This study focuses on open-source publication data, including scientific journal articles and conference papers on AI that involve North Korean researchers, extracted from a database of peer-reviewed literature. A list of search keywords was generated using two types of information sources: North Korea’s legal national standards for AI and technical terms in the index sections of AI-related educational materials.
North Korea established legally binding national standards, called KPS (국규), for AI-related terms in the early 2000s. The KPS for AI is aligned with International Organization for Standardization (ISO)/International Electrotechnical Commission (IEC) standards 2382-28, 29, 31 and 34.[1] However, many terms defined in the respective ISO/IEC standards are conceptual and applicable across a broad range of academic fields, such as domain knowledge, knowledge base, and knowledge acquisition. Some terms more related to end-uses of AI, such as speech analysis and speech synthesis, were not considered in this analysis.
Furthermore, the ISO/IEC standards do not encompass many AI-specific technical terms. Because of this, the study sourced technical terms from the index of an AI-dedicated educational book to ensure comprehensive coverage. The keyword list derived from this index also includes terms for statistical methods commonly employed in AI research but not exclusive to AI, such as principal component analysis (PCA) and hierarchical clustering analysis (HCA). The project included such words to capture the widest possible scope of technical efforts by North Korea, albeit with the caveat of capturing scientific studies that are not strictly focused on AI.
Utilizing the keyword list of approximately 800 terms, this project gathered data on research studies published in North Korea from 2017 to 2023, a period subject to international sanctions. The data cleaning process included standardizing the names of universities and research institutes, as well as correcting misclassifications where South Korean entities were erroneously identified as North Korean.
North Korea’s Global Standing in Global AI Development
From a global perspective, when looking at the country’s publication volume, North Korea is not among the leading countries in AI development. From 2017 to 2023, 160 countries published over 2.5 million articles potentially involving AI, with China at the forefront, having published approximately 860,000 articles. China is followed by the United States, India, the United Kingdom, Germany, Japan, Canada, and South Korea (Figure 1). Meanwhile, North Korea ranks 145th, with only 161 publications, comparable to the output of African countries like Togo and Swaziland.
However, using publication numbers as a quantitative measure may not accurately reflect the country’s AI capabilities, as the publication output is not necessarily proportional to technical capability. Furthermore, the database omits articles from North Korean domestic journals like the Journal of Kim Il Sung University and Information Science, which do not focus exclusively on AI but frequently feature AI-related content. Lastly, the project’s keyword list is a living document and may not encompass all AI and machine learning (ML)-related terms that North Korea might be focusing on.
Nevertheless, the significantly lower volume of publications from North Korea indicates that the country lags behind in academic research compared to other leading countries. In general, academia is crucial for supporting the AI industry by cultivating talent and conducting foundational studies, underpinning commercial AI development. Promoting AI-focused domestic educational programs and seeking international collaborations could be one of the major tasks that North Korea may need to focus on to keep abreast of progress made by other leading countries.
North Korea’s Research Collaborators for AI Development
During the specified period, North Korea co-authored with institutions from at least 12 countries across Asia, Europe, Africa and America. These countries include China, South Korea, Japan, Germany, Lithuania, Sweden, Switzerland, the United Kingdom, Egypt, Uganda, Canada and the United States. China was the most frequent collaborator, participating in roughly 70 publications, including studies directly involving AI. Many publications with other countries did not directly involve AI or were only nominally involved in co-authorship, such as writing a review paper summarizing an event in which North Korean scholars participated. Meanwhile, researchers affiliated with universities in South Korea and the United States each jointly published a paper on AI with North Korean scholars during the subject period, as covered in the select case studies below.
A total of 45 universities and research institutes share co-authorships with North Korean entities. Several Chinese universities in geographical proximity to North Korea, including Northeast Normal University, Northeastern University, Northeast Forestry University, Harbin Institute of Technology, and Harbin Engineering University, have been particularly active in potential AI research engagements with North Korea. Other notable collaborators include the University of Detroit Mercy in the United States and George Mason University’s campus in South Korea. These foreign universities primarily cooperated with three North Korean institutes: Kim Il-Sung University, National Academy of Science and Kim Chaek University of Technology.
This suggests that North Korea’s international scientific collaborations on AI and related subjects have persisted despite the 2016 ban on such activities, including individuals affiliated with universities from the United States and South Korea. Furthermore, several Chinese universities that have collaborated with North Korea are currently on the US trade deny list called the Entity List due to their linkage with the People’s Liberation Army (PLA). While there is no indication that North Korea’s international AI collaborations involve direct military applications, it remains critical to closely monitor these partnerships since technical know-how and tacit knowledge gained through these collaborations could potentially be diverted toward military AI development.
North Korea’s Research Collaborations on AI Development: Select Case Studies
North Korea’s research on AI between 2017 and 2023 spanned a wide array of industries, applications and academic fields. The relevant AI techniques were applied to textiles, robotics, telecommunications, aerospace and cybersecurity sectors. In addition, North Korea’s AI applications ranged from object detection to visual tracking, text-to-speech synthesis to remote sensing, and cryptocurrency. AI research topics extend into environmental, educational, medical, and geological studies, demonstrating North Korea’s broad scope of interest and investment in AI research.
Case Study 1: A Joint Publication With Chinese and American Universities
In 2017, North Korea published “Adaptive robust speed control based on recurrent elman neural network for sensorless PMSM servo drives.” The study aims to develop adaptive control schemes for sensorless permanent magnet synchronous motors (PMSM) by employing a recurrent Elman neural network (RENN), suitable for processing sequential data.[2] PMSM is utilized in automation controls for industrial robots due to its high-power density, reliability, efficiency and controllability.[3] Furthermore, the removal of sensors from PMSMs can decrease their size and cost, enhancing their efficiency. However, this “sensorless” approach requires precise control to maintain the performance of PMSM, and North Korea applied AI techniques to tackle this technical hurdle. The study explicitly specified the intended end use of sensorless PMSM as joint motors in industrial robots. Possible dual-use applications of North Korea’s PMSM could include computer numerically controlled (CNC) machine tools, given the country’s heavy emphasis on CNC in its economic development. In addition, PMSM can also be used for a wide range of applications, such as aerospace science, the shipbuilding industry and unmanned aerial vehicles (UAVs).
The study involves North Korean and Chinese universities that have most actively collaborated on potential AI research, as well as individuals affiliated with a university in the United States. North Korean author Myongguk Jong is a professor at Kim Chaek University of Technology. Another North Korean scholar, Ryongho Jon, earned his PhD from the School of Information Science and Engineering at Northeastern University in China, where Chinese scientist Zhanshan Wang is a professor. As a correspondence author, Wang might have played a crucial role in the project and brought extensive experience to the research given his extensive publications on PMSM and AI.[4] Moreover, the study involves Chaomin Luo, a scholar affiliated with the University of Detroit Mercy in the United States at the time of publication.
Wang and Luo’s involvement in the project does not necessarily constitute a violation of the sanctions. For instance, in 2016, the four scholars had already published a paper on sensorless PMSM under grants identical to the study published in 2017, provided by Chinese foundations such as the National Natural Science Foundation of China and the Fundamental Research Funds for the Central Universities.[5] This indicates the publications could have been deliverables for a multi-year project that commenced and had possibly concluded before the adoption of the sanctions measures on technical cooperation in 2016.
Case Study 2: A Joint Publication With South Korean Universities
In June 2019, South Korea and North Korea jointly published “A Study on Features for Improving Performance of Chinese OCR by Machine Learning.” Specifically, the study involves Chul Kim, a scholar affiliated with George Mason University in Incheon, South Korea, and Jangsu Kim and U Ju Kim from Pyongyang University of Science and Technology. Three scholars applied ML techniques, including k-Nearest-Neighbor (KNN) and tree classifier, to enhance the efficiency and accuracy of an optical character recognition system (OCR) designed to convert printed Chinese characters into digital information. The paper states that the authors planned to focus on behavioral authentication as a future endeavor by developing a signature verification system applicable to both off- and online domains. It is uncertain if the collaboration has continued for the authentication system as there has been no other publication by the authors added to the database since 2019.
Conclusion
Amid global advancements in AI, North Korea lags behind leading countries like China and the United States. However, despite sanctions, North Korea persists in academic partnerships for potential AI research, often relying heavily on China. Given China’s prominent role in the global AI landscape, it is crucial to consider how its collaborations with North Korea may influence North Korea’s AI capabilities. Specifically, monitoring cooperation between Chinese universities with known relations with North Korea and its institutions is crucial in assessing the North’s AI research direction, as well as monitoring sanctions compliance. Universities could also establish internal compliance programs (ICP) to ensure that all students’ and faculty members’ activities meet sanctions and nonproliferation regulations. Furthermore, other countries’ academia could be exploited by North Korea, highlighting the need for enhancing due diligence in international collaborations.
- [1]
Iso/IEC 2382-28 was eventually replaced by ISO/ICE 2382:2015. However, each term is marked with the original document and can be traced to what was written in 2382-28. See: Kang Yeong-sil, “2000년대 초 북한의 인공지능 개념과 범위 [The Concept the Scope of Artificial Intelligence in North Korea in the Early 2000s],” NK Tech, June 10, 2020; and “‘북한의 인공지능 뛰어나다’? [‘North Korea’s artificial intelligence exceptional’?],” NKTechTV, https://www.youtube.com/watch?v=JBbWo-rkzJg.
- [2]
- [3]
See Pierre-Yves Brulin, Fouad Khenfri and Nassim Rizoug, “Generating Fault Databases Through Simulated and Experimental Multi-Rotor UAV Propulsion Systems,” IEEE Transactions on Vehicular Technology 73, no. 4 (2024): 4671-4682, https://ieeexplore.ieee.org/document/10387779; and “Permanent Magnet Motors for Drones,” Unmanned Systems Technology, October 20, 2023, https://www.unmannedsystemstechnology.com/expo/permanent-magnet-motors-for-drones/.
- [4]
See Ryongho Jon, Zhanshan Wang, Chaeomin Luo and Myongguk Jong, “Adaptive robust speed control based on recurrent elman neural network for sensorless PMSM servo drives,” Neurocomputing 222, (2017): 131-141, https://www.sciencedirect.com/science/article/abs/pii/S0925231216313455; Zhanshan Wang, Longhu Quan and Ziuchong Liu, “Sensorless SPMSM Position Estimation Using Position Estimation Error Suppression Control and EKF in Wide Speed Range,” Mathematical Problems in Engineering 2014, no. 1 (2014): 1-11, https://www.hindawi.com/journals/mpe/2014/480640/; Longhu Quan, Zhanshan Wang and Xiuchong Liu, “Sensorless control of SPMSM using complex number model based position estimation and EKF,” IEEE, (2014): 2663-2668, https://ieeexplore.ieee.org/document/6852623; 1. Longhu Quan, Zhanshan Wang, Xiuchong Liu and Mingguo Zheng, “Sensorless PMSM Speed Control Based on NN Adaptive Observer,” Lecture Notes in Computer Science 8866 (2014): 100–109, https://doi.org/10.1007/978-3-319-12436-0_12; and 1. Chao Cai, Fufei Chu, Zhanshan Wang and Kaili Jia “Identification and Control of PMSM Using Adaptive BP-PID Neural Network,” Lecture Notes in Computer Science, 2013, 155–62, https://doi.org/10.1007/978-3-642-39068-5_19.
- [5]
The 2017 study states, “This work was supported by the National Natural Science Foundation of China (Grant Nos. 61473070 and 61433004), the Fundamental Research Funds for the Central Universities (Grant No. N130104001), and SAPI Fundamental Research Funds (Grant No. 2013ZCX01).” The 2016 study states, “This work was supported by the National Natural Science Foundation of China (Grant Nos. 61473070 and 61433004), the Fundamental Research Funds for the Central Universities (Grant Nos. N130504002 and N130104001), and SAPI Fundamental Research Funds (Grant No. 2013ZCX01).”