Article 7 February 4, 2023
AI’s Malicious Use and Challenges to International Psychological Security
In December 2022, the International Center for Social and Political Studies and Consulting published the second report “Experts on the Malicious Use of Artificial Intelligence and Challenges to International Psychological Security”, penned by this author. This report stems from the research project titled “Malicious Use of Artificial Intelligence and Challenges to Psychological Security in Northeast Asia” (21-514-92001) and jointly funded by the Russian Foundation for Basic Research (RFBR) and the Vietnam Academy of Social Sciences (VASS).
The responses garnered from a targeted survey of twenty-five experts from twelve countries and the subsequent analysis of their feedback aim to bring to light the most serious threats to international information-psychological security (IPS) through malicious use of artificial intelligence (MUAI), determining how dangerous these threats are, which measures should be used to neutralize them, and identifying the prospects for international cooperation in this area.
This publication attempts to define the current level of MUAI threat to the IPS as well as its possible level by 2030. The report pays special attention to the situation in Northeast Asia (NEA), where MUAI practices are based on a combination of a high level of AI technologies in leading countries and a complex of acute contradictions in the region. The results of the second survey allow us to trace the continuity and changes in expert assessments compared to the first survey conducted a year earlier, in which nineteen experts from ten countries took part.
In the year since the first expert survey, the world has become even more unstable and dangerous. NATO’s aggressive advance to the East provoked a massive military conflict in Ukraine, one that risks escalating into a world war. Gross domestic product in Ukraine is expected to decline 35 to 40 percent in 2022; Russia also expects a decline of 2,5 percent for the same period. There are clear signs of a recession in the U.S. and the EU, with this economic slowdown reaching even China. Unprecedented inflation, rising borrowing costs, and increasingly intimidating levels of indebtedness are all contributing to fears of economic collapse in many Western and non-Western countries.
However, even amid COVID-19, the ultra-rich have little to worry about. In 2022, Forbes identified more than 1,000 billionaires around the world who were richer than they were a year prior (Dolan and Peterson-Withorn, 2022). This deepening polarization in living standards inevitably intensifies the world’s most acute political, racial, and national conflicts across both developing and rich countries—including the United States, long thought immune to such turmoil. In a sick society, threats to psychological security through the malicious use of AI are much higher than in a socially healthy environment. Especially if social regression is observed against the background of the ongoing progress of high technologies.
The transition to advanced forms of AI is already happening. In 2022, DeepMind, a subsidiary of Alphabet specializing in AI, announced the creation of an intelligent agent called “Gato” that can single-handedly perform more than 600 of different tasks (MSN, 2022). The same network, with the same weights, can play Atari video games, capture images, chat, stack blocks with a real robot arm and much more. Gato decides whether to produce text, torque joints, press buttons or take other actions based on context (DeepMind, 2022). The agent could outperform human experts in 450 tasks. Gato learns several different tasks at the same time, easily switching from one skill to another without forgetting what it has learned. Previous AI models have started to combine different skills but when starting a new task, these models had to forget what was previously learned in order to move on to the next one.
However, Gato does make some mistakes that a human would not make. While Gato is still far from being an AGI creation, progress is still evident. In terms of malicious use, an agent similar to Gato repeatedly, if not by orders of magnitude, reduces the cost of preparing and implementing various actions by one agent. This makes it easier to carry out psychological operations with a combined effect on the senses—an intelligent agent like Gato can simultaneously send text messages, conduct a conversation and so on.
Modern AI capabilities already enable one to influence public consciousness. The president-elect of South Korea, Yoon Suk-yeol, used an unusual strategy during his 2022 election campaign. His campaign team used deepfakes to make an “AI avatar” that helped him win the election. This technology is helpful for appealing to younger voters and to get them more involved (Vastmindz 2022). AI Yoon’s creators believe he is the world’s first official deepfake candidate—a concept that is gaining traction in South Korea, which has the world’s fastest average internet speeds (AFP 2022). AI technology transformed Yoon Suk-yeol into a more modern candidate than his competitors from the perspective of younger voters. With neatly combed, black hair and a smart suit, the avatar looks nearly identical to the real candidate but instead, used salty language and meme-ready quips in a bid to engage younger voters who get their news online (AFPRelaxnews 2022). Some alarms were raised when the avatar politician used humor to try and deflect attention from Yoon’s past scandals (The Times of India 2022).
AI Yoon’s pronouncements made headlines in the South Korean media, and seven million people visited the “Wiki Yoon” website to question the avatar (AFP 2022). At first glance, AI Yoon could pass for an actual candidate—an apt demonstration of how far artificially generated videos have come in the last few years. “Words that are often spoken by Yoon are better reflected in AI Yoon,” said Baik Kyeong-hoon, director of the AI Yoon team (AFP 2022). However, there is a question about what should be done if the avatar of a statesman, politician or business person is a false representation, reinforcing in the public’s conscious and subconscious mind inflated qualities, creating the illusion of attributes that the real person does not possess.
The experience of South Korea’s last presidential campaign may have shown, in part, the initial form of a new method of rather dangerous political manipulation. An increasingly adaptive avatar that does not need rest creates an image that a real person will be able to compete with less and less in the public space. This prompts the question of whether “televised presidents” will soon be replaced by “deepfake presidents”?
The world’s first AI TV news anchor was unveiled in China in 2018 (Kuo, 2018), and they were immediately noticeable as the result of an algorithm. However, in December 2021, AI company Xiaoice introduced N Xiaohei and N Xiaobai—virtual replicas of two real-life news anchors—on “AI Business Daily.” According to the Chinese state-run Xinhua News Agency, Xiaoice, in collaboration with National Business Daily, has demonstrated a strong ability to develop virtual replicas that are effectively indistinguishable from real humans through advanced machine learning (ML) and rendering technologies (Lin, 2022). Unfortunately, these extraordinary achievements have the potential to be used for malicious purposes by anti-social actors aiming to destabilize the public consciousness, provoke anti-social actions, and/or deceive individuals.
Notably, this risk is not purely theoretical, with deepfakes having been used on several occasions to trick companies into sending money by them (Statt, 2019; Stupp, 2019; Veldkamp, 2022). Of course, this practice is not yet common, meaning that the risk is fairly minor—for now. In June 2022, the FBI warned that scammers are beginning to use deepfakes to apple for sensitive jobs; this type of fraud could be used to gain access to company networks and obtain company data (Ferraro and Chipman, 2022).
AI technologies are being tested in military conflicts, including psychological warfare. The use of Western-produced AI technologies in the ongoing military conflict in Ukraine is significant. US facial recognition startup Clearview AI has provided technical support to Ukraine. Clearview AI’s tools can identify faces in videos, comparing them to a company database of 20 billion images from public networks, identifying potential spies and people who have been killed. AI tools also play an important role in Ukraine’s propaganda war and in processing critical conflict information.
A program from the US company Primer can perform speech recognition, transcription and translation that captures and analyzes Russian information, including conversations between Russian soldiers in Ukraine. The Swiss encrypted chat service called Threema allows Ukrainian users to send this data to the military without revealing their identities. According to former Google chief executive officer Eric Schmidt, now an AI adviser to the US government, the Ukrainian military receives thousands of such reports every day, which are then filtered by an AI program (Global Times, 2022).
Russia has not been slow to respond. The Russian Federation’s Ministry of Defense Office for the Development of Artificial Intelligence conducted an examination of the innovative projects to be presented at the Army 2022 Forum for AI technology use (IMTF “Army–2022,” 2022). Vasily Elistratov, head of the Russian Defense Ministry’s Department for the Development of AI Technologies, said that special military operation experience in Ukraine is used to improve weapons systems (Ixbt.com, 2022). Unfortunately, the arms race that is growing in terms of both size and participation, not least in the field of AI, is making international security, including its psychological component, less and less durable.
The potential for MUAI today is astounding. Noting this, researchers from Collaborations Pharmaceuticals in cooperation with European scientific institutions conducted a conceptual experiment. Instead of synthesizing new drugs, they asked the opposite of the MegaSyn AI neural network: to identify substances that are the most toxic to the human body. The neural network correctly understood the task and, in under six hours, generated a list of 40,000 substances that are optimal components of chemical and biological weapons.
The AI independently designed not only many known chemical warfare agents, but also many new ones that are more toxic. This simple inversion of the ML model turned a harmless generative model from a useful tool into an enabler of mass murder (Urbina et al., 2022). It is reasonable to suspect that this inversion approach could be applied to other areas, such as finding optimal ways to have negative psychological impacts on the public consciousness.
The most worrisome aspect of these mounting concerns is simple: legislative responses are lagging behind. The sluggish speed of the legislative process (especially relative to that of the MUAI-development process) is clear in the fact that, for example, the EU Artificial Intelligence Act, at the time of writing this text, has not yet passed the first reading in the European Parliament (EUR-Lex, 2022).
In analyzing the current international situation, most studies focus on economic, military, and geopolitical changes without sufficiently linking them to the growing impact of qualitative technological changes on international dynamics, including those in the sphere of psychological security. This underestimation is clear in the near-complete lack in the literature of a systemic analysis of the possible risks of MUAI to psychological security at individual, group, national, and international levels. The absence of such an analysis is unacceptable for numerous reasons.
First, the penetration of modern AI into countless spheres of life makes it a critical component of continued development and progress. Investment in AI may consist of trillions of dollars within the next two decades. According to a PricewaterhouseCoopers Middle East (PwC) report released at the World Government Summit in Dubai, 14 per cent (U.S. $15.7 trillion) of global economic growth will stem from the use of AI by 2030. PwC believes that the greatest gains in 2030 will be in China, where AI will be responsible for up to 26 percent of economic growth (Rao and Verweij, 2018, p. 3). However, as already established, alongside AI’s development will come tremendous risks, including in the field of psychological security.
Second, there is already some research into international practices of MUAI in the psychological field. It is short-sighted not to continue existing research on a systemic interdisciplinary basis, taking into account the counter progress of AI technologies and the regression of existing public structures, national and international institutions, and simply wait until the consequences of AI become an even greater threat to international security. At the same time, strong support from state and non-state actors is required for such research that goes beyond such areas as cybersecurity and information security, as well as psychological security before the introduction of AI technologies, because we are talking about qualitatively different possibilities of causing irreparable damage to the development of human civilization than simply quantitative expansion of propaganda capabilities.
Third, the international situation is continually worsening in several respects. This is not because of particular politicians being in power but because of the severity of the problems faced during the transition to a qualitatively new stage in human development. This transition has been accompanied by an increase in economic problems, socio-political tensions, and geopolitical competition. AI can become both the technological foundation of a new, more socially oriented world order, and the tool of high-tech undemocratic dictatorships, actively used for antisocial purposes by various non-state actors.
The conducted expert survey and the subsequent analysis of its results are precisely intended to fill the gap in the awareness of the nature and scope of threats to society associated with modern and future MUAI practices in the context of the global crisis, as well as an assessment of possible risks in this area in the near future. Further research in this direction obviously follows from the nature and dynamics of the changes taking place in the world.
December 20, 2022
DSc., Professor, Leading Researcher, Institute of Contemporary International Studies, Diplomatic Academy at the Ministry of Foreign Affairs of the Russian Federation