With the rapid development of artificial intelligence technology, generative AIs such as ChatGPT and Sora have displayed their powerful capabilities in multiple fields. The year 2022 was dubbed the “Year of Generative AI”; in 2023, GPT-4 became popular due to its human-like intelligence; and at the beginning of 2024, Sora’s emergence once again shocked the world. Generative AI can significantly enhance the military’s capability in intelligent information acquisition, decision support, and human-machine collaborative combat, expanding its application in the military domain.
Generative AI Technology and Its Characteristics
Generative AI integrates achievements from artificial intelligence technologies such as natural language processing, computer vision, speech recognition, and deep learning. It can generate new content based on existing data and knowledge through algorithms, pushing intelligent warfare into a new stage of “autonomous combat across domains”.
Video Screenshot Generated by Sora
Basic Concept of Generative AI The National Internet Information Office’s “Management Measures for Generative Artificial Intelligence Services (Draft for Comments)” defines generative AI (AIGC) as “technology that generates text, images, sounds, videos, codes, and other content based on algorithms, models, and rules.” There are mainly five categories: First, generative text models. By learning from a large amount of text data, they can generate articles, stories, comments, and poems. Besides OpenAI’s ChatGPT, notable models include Google’s BERT and PaLM, and Microsoft’s Turing-NLG, all of which feature “strong interaction,” “strong understanding,” and “strong generation.” Second, generative image models. By learning from a large amount of image data, they can create entirely new paintings and illustrations. The most representative is OpenAI’s DALL-E2/E3 model, which can quickly generate realistic and creative images based on descriptions. Third, generative audio models. By learning from a large amount of audio data, they can generate new music and speech. The most representative is OpenAI’s Voice Engine, which can clone a person’s multilingual audio from just a 15-second voice sample. The popular video translation software HeyGen uses this engine. Fourth, generative video models. By learning from a large amount of video data, they can generate new video clips or complete video content. The most representative is OpenAI’s newly launched video generation model Sora, which can create realistic, vivid, and complex high-quality videos based on text instructions, enabling smooth transitions between multiple camera angles, producing quality close to commercial films. Fifth, generative code models. They can automatically write or assist in writing code to improve development efficiency and quality. The most representative are OpenAI’s CodeX and DeepMind’s AlphaCode.
Generative AI is significantly different from traditional AI. First, from the core goals and functions, the core goal of generative AI is innovation, capable of generating entirely new content. This creativity transcends the realm of traditional AI, which mainly establishes connections between known inputs and outputs for predictions. Second, from the learning mechanisms and technologies, generative AI widely uses generative adversarial networks, Transformers, and other technologies, relying on a large amount of sample data for training, and can evolve by learning from the content it generates, surpassing the traditional AI’s reliance solely on machine learning. Third, from the application scenarios and advantages, generative AI has broad application prospects, applicable in creative or assistive creative fields, while traditional AI is more used in scenarios requiring precise recognition and classification or predicting future trends based on historical data.
Development History of Generative AI From a technological perspective, the development of generative AI is a process of continuous breakthroughs and innovations. First, the rise of expert systems. From the 1980s to the early 1990s, expert systems emerged. They relied on rules and predefined knowledge bases to perform reasoning and decision-making in specific fields, laying the foundation for the development of generative AI in logical reasoning and knowledge expression. Second, the rise of deep learning. In the late 1990s, due to algorithmic limitations and insufficient computing power, the development of generative AI encountered a bottleneck. Entering the 21st century, the rise of deep learning technology brought revolutionary changes to generative AI. It constructs deep neural network models that automatically learn feature representations and generation rules from large-scale data, greatly enhancing the quality and diversity of generated content. Third, the development of multimodal models. Since 2022, thanks to continuous algorithm optimization and increased computing power, generative AI began to evolve from single-modal to multimodal cross-domain generation, enabling AI to simultaneously process and understand various forms of information such as text, images, audio, and video, providing broader application space for generative AI in various fields. GPT-4 is a large multimodal model that not only possesses advanced reasoning and complex instruction processing capabilities but also exhibits high creativity. This marks the entry of generative AI into a fast track of development. Fourth, the goal of general artificial intelligence. The ultimate development goal of generative AI is to achieve general artificial intelligence (AGI), which refers to machine intelligence similar to human intelligence, with broad applicability and adaptability. The founding intention of OpenAI is to ultimately achieve AGI. Currently, achieving AGI still faces many challenges, including how to enable machines to understand and process natural language, how to enable machines to learn and create like humans, and how to make reasonable decisions in the face of unknown situations. Additionally, the development of AGI requires breakthroughs in existing algorithmic and technical limitations and finding universal intelligent theories.
Technical Characteristics of Generative AI Generative AI is a powerful artificial intelligence technology that exhibits tremendous potential in imitation and creation. Its technical characteristics are mainly reflected in several aspects. First, data-driven. Generative AI relies on a large amount of data for training and learning to obtain sufficient information and knowledge to generate content, and it can handle large-scale, high-dimensional data to extract useful information and features. Second, scalability. The model architecture typically has modular and hierarchical characteristics, making it easy to expand and customize, allowing for widespread application in different fields and scenarios; the algorithm design usually considers computational efficiency and resource utilization, and its performance and generation capabilities can improve with the continuous development of technology and accumulation of data. Third, interactivity. Generative AI can understand and respond to users’ natural language input, dynamically adjusting and optimizing output content based on user feedback and needs, enabling more intelligent and flexible interaction with users, providing more personalized and precise services. Fourth, creativity. It no longer relies solely on rules but can imitate human creativity, drawing inspiration from multiple data sources, blending different styles, cultures, and knowledge to generate innovative and diverse content, offering infinite possibilities for its application across various fields.
Application of Generative AI in Intelligent Warfare
The relationship between military and technology has always been close, and the explosion of generative AI signals that weak artificial intelligence is evolving into strong artificial intelligence and general artificial intelligence, providing an increasing number of application scenarios for future intelligent warfare.
Huawei’s “Pangu Meteorological Model”
Intelligent Situation Analysis Generative AI has immense potential and value in battlefield situation intelligence analysis. First, multi-domain data fusion analysis. It can leverage its powerful information processing and output capabilities to break through the “fog of war” and conduct in-depth mining and fusion analysis of massive, diverse, complex, and rapidly growing battlefield intelligence data, providing comprehensive and accurate intelligence support for command decision-making. Second, joint battlefield situational awareness. In the face of rapidly changing battlefield data and information, it can achieve comprehensive awareness of the multidimensional spatial battlefield situation across land, sea, air, space, electronic, and network domains through multimodal information fusion, helping combat personnel understand the battlefield environment, identify strike targets, and provide early warnings of operational threats, enhancing the accuracy and timeliness of command decision-making. Third, situation prediction and threat assessment. By conducting deep learning on a large amount of historical and real-time battlefield data, it can identify patterns and rules of battlefield changes, construct complex battlefield models, simulate and predict the evolution trends of battlefield situations, and forecast potential enemy actions, attack methods, and targets, providing early warnings and response suggestions for our side. Huawei’s “Pangu Meteorological Model” does not use traditional numerical forecasting methods; it employs a three-dimensional neural network adapted to Earth coordinates and a hierarchical temporal aggregation strategy, completing global seven-day high-resolution numerical weather forecasts in just 10 seconds, with a computation speed increase of over ten thousand times and a reduction in computational power consumption to one hundred thousandths, which can be used for battlefield environmental support in the future.
Intelligent Command Decision-Making Generative AI can provide powerful support and assistance for command decision-making, improving the accuracy and efficiency of decisions and enhancing the flexibility and adaptability of actions. First, assisting in task planning. It can use multi-source intelligence analysis tools and decision support models to automatically extract key information, generate situational images, data charts, deployment strategies, and resource allocation documents, assisting in task planning and improving operational planning efficiency. Second, assisting in scenario simulation. It can accurately analyze battlefield situations and operational needs, automatically draft combat scenarios, generate multiple operational plans, and predict the outcomes of each plan through simulation, providing references for commanders to select optimized plans. Third, assisting in combat decision-making. It can be embedded in integrated joint combat command systems, assess battlefield situations, predict the direction of war, and continuously iterate and optimize through human-machine interaction, forming combat suggestions to assist command decision-making, achieving faster response speeds and higher decision-making levels than traditional command methods. To enhance the speed and effectiveness of command decision-making, the U.S. military has included generative AI in its “Technology Watch List,” initiated the “Artificial Intelligence Forward Program,” and the “From Data to Decision” project, reshaping the command decision-making process and constructing a capability system that transforms massive intelligence data into effective autonomous decision outputs.
U.S. Military XQ-58A Unmanned Wingman
Operational Intelligent Collaboration Generative AI can significantly enhance the collaborative efficiency between different combat units, strengthening overall combat capabilities. First, enhancing collaborative combat effectiveness. Through real-time perception and analysis of battlefield situations, it can quickly identify the actions and intentions of both enemy and friendly forces, automatically assigning and scheduling tasks based on the capabilities, positions, and mission needs of each combat unit, improving the efficiency and accuracy of collaborative combat. Second, shortening the combat process. It can decompose combat tasks, use built-in algorithms to optimize plans, and rely on its rapid response speed to accelerate the OODA (Observe-Orient-Decide-Act) combat cycle, shortening the decision-making period and creating a winning advantage through “speeding up to beat the slow.” Third, supporting human-machine collaboration. Generative AI, being a product of machines that can also understand human language, serves as an ideal human-machine interaction interface, capable of being embedded in unmanned combat platforms, supporting cross-platform and cross-domain manned/unmanned collaborative operations through intelligent algorithms and autonomous decision-making capabilities, fully supporting new styles of combat such as “loyal wingman” and “swarm” operations, greatly enhancing combat effectiveness. In future intelligent warfare, generative AI can precisely connect with unmanned combat platforms, planning combat tasks in an “order-based” manner, assigning tasks nearby in an “order-based” manner, and locking targets in a “take order” manner, potentially achieving a dimensionality reduction strike against traditional combat opponents.
Intelligent Support for Logistics Generative AI can be comprehensively integrated into various support systems, with people, equipment, and supplies interconnected, organically merging to greatly enhance the efficiency and precision of combat support. First, material management and allocation. It can analyze battlefield demand and material inventory in real-time, optimizing material allocation and scheduling; it can predict material consumption trends and replenish in advance to avoid shortages; and it can comprehensively analyze factors such as transportation routes, traffic conditions, and material demand to optimize transportation plans, reducing costs and improving efficiency. Second, equipment maintenance and repair. It can monitor and analyze equipment operation data in real-time, predict potential failures, formulate maintenance plans in advance, optimize maintenance resource allocation, and carry out timely equipment maintenance and repairs, lowering maintenance costs and extending equipment life, thereby improving equipment availability. Third, medical rescue and support. By analyzing medical data, it can quickly and accurately identify potential disease signs and injuries, predict treatment outcomes, and provide more efficient and intelligent medical support for military personnel; it can assist in remote diagnosis, offering optimal treatment plans to enhance the efficiency and accuracy of battlefield medical rescue; it can match the nearest rescue personnel and resources to quickly organize the evacuation and treatment of the injured, improving battlefield rescue efficiency. The U.S. Army’s Telemedicine and Advanced Technology Research Center collaborates with Johns Hopkins University to utilize artificial intelligence, augmented reality, and robotics to provide virtual assistants for medical personnel and soldiers on the battlefield.
Impact of Generative AI on Intelligent Warfare
Artificial intelligence is not a single discrete technology, but a general enabling technology. With its powerful learning and innovative capabilities, the wide application of generative AI in the military field will bring revolutionary changes to warfare, resulting in comprehensive and far-reaching impacts.
Changing the Nature of Warfare Disruptive technology groups centered on AI are rapidly entering the military domain, with “intelligent” elements replacing “information” elements, leading to significant changes in the nature of warfare. First, the combat system is transforming into human-machine mixed formations. Generative AI will completely overturn the main combat forces on the battlefield, with intelligent payloads, intelligent platforms, and intelligent systems becoming new trends in equipment development, and human-machine mixed formations and unmanned autonomous formations becoming new forms of combat force composition. Second, the combat style is transitioning to intelligent unmanned operations. Under the guidance of generative AI, human-led human-machine collaborative combat, algorithm-dependent intelligent cognitive combat, cloud-centered open-source data combat, and knowledge-driven all-domain distributed combat will become the main combat styles in the future, transforming the “kill chain” into a “kill network” with characteristics of low risk, low cost, and low threshold. Third, the winning mechanism is evolving to a victory through intelligence. The development of war control rights is consistent with the evolution of the nature of warfare, with “intelligence control rights” becoming the core of the contest for control in intelligent warfare, concentrated in the comprehensive competition of “algorithms + data + computing power,” highlighting victory through intelligence, speed, coordination, and calculation, making intelligence the primary factor, autonomous capability mastery, and wisdom-driven victory the basic principles of warfare.
Unmanned X-62A Engaging in Air Combat with Manned F-16
Lowering the Threshold for Combat Generative AI can effectively lower the threshold for actions such as intelligence gathering, cyberattacks, and cognitive warfare, significantly increasing the range of combat participants. First, it lowers the threshold for intelligence gathering. Without requiring high-level professional skills and experience, it can seamlessly integrate and process massive amounts of multi-source heterogeneous battlefield data in real time, providing combat personnel with real-time battlefield situations and strike targets, as well as early warning threat information. During the Ukraine crisis, the Ukrainian military used “crowdsourced intelligence” apps to call on the public to upload real-time information on Russian military movements, utilizing generative AI technology to analyze and process collected data, images, videos, etc., enhancing their situational awareness and data processing capabilities in a complex environment. Second, it lowers the threshold for cyberattacks. It can quickly realize functions such as discovering network vulnerabilities, writing malicious code, forging phishing emails, and spreading and executing malware, significantly reducing the entry requirements for cyberattacks, potentially even achieving zero thresholds. A former SpaceX software engineer used GPT-3 to audit a Git repository and discovered 213 security vulnerabilities, while the best commercial tools could only find 99. Recent studies have found that GPT-4 can generate malicious code exploiting vulnerabilities by reading vulnerability security announcements. Third, it lowers the threshold for cognitive warfare. Cognitive domain warfare involves multiple areas such as religion, language, culture, psychology, and thinking, requiring a large number of specialized talents to support it. Generative AI can innovate and generate high-quality public opinion guidance and psychological warfare attack information based on combat needs, allowing for bulk replication and deployment without interruption, reducing the production costs of “cognitive ammunition” and the thresholds for cognitive domain warfare. In 2017, a Harvard University report suggested that AI technology is more accessible compared to other military technologies (such as nuclear technology) and has lower usage thresholds.
Enhancing Combat Effectiveness Generative AI can automatically process large amounts of data and information, aiding command decision-making to be more scientific and reasonable, strengthening human-machine collaboration in completing reconnaissance, attack, and defense tasks, and enhancing overall combat effectiveness. First, it enhances battlefield situational awareness. It can conduct automatic recognition and fusion processing based on multi-source data, efficiently identifying abnormal behaviors and potential threats based on command associations, quickly mining action patterns from a large amount of enemy data, and autonomously judging target value, combat intentions, action plans, and force comparisons, providing basis and plans for battlefield situation assessments. Second, it aids in intelligent decision-making for command. In the face of increasingly complex battlefield situations and exponentially growing battlefield information, it can efficiently analyze the operational needs of commanders, providing risk assessments and decision suggestions through large model processing and combat simulations, helping commanders reduce subjective judgment interference, objectively assess situations, and make sound decisions, transitioning from computer-assisted decision-making to intelligent decision-making integrated with human-machine collaboration, moving from “humans in the loop” to “humans above the loop.” Third, it promotes the autonomy of combat actions. The “open interface” mode and platform attributes of generative AI provide possibilities for the autonomy and intelligence of traditional weapons or automated weapons. The U.S. military has retrofitted F-16 fighter jets with data tracking computing systems, flight automation control systems, and other intelligent modules, transforming traditional fighter jets into AI-controlled X-62A unmanned wingmen. In April 2024, this type of aircraft conducted the world’s first air combat test with a manned aircraft, engaging in aerial combat at speeds of up to 1.58 Mach, with the closest distance being about 610 meters. Fourth, it diversifies innovative combat styles. The creativity of generative AI offers it infinite possibilities and vast application prospects in warfare. The Ukraine crisis is considered the first AI-enabled war, with various parties exploring the use of AI technology and continuously innovating combat actions. The Ukrainian military used AI facial recognition technology to search over 2 billion photos on Russian social media, establishing a facial recognition database for Russian soldiers, quickly confirming target identities through information comparisons, and verifying the identities of Russian agents infiltrating Ukraine. The Russian military utilized AI self-learning algorithms to rapidly adapt to changes in network defense strategies, achieving intelligent and automated cyberattacks, with the frequency of large-scale cyberattacks rising from 15 times a month to 125 times.
Expanding the Risks of War The participation of generative AI in warfare presents characteristics that are comprehensive, three-dimensional, and multi-faceted, evolving at a surprising speed, potentially becoming a “terrifying fusion” that changes the situation of warfare and even impacts the course of human civilization. First, it can easily lead to the proliferation of false information. Generative AI can rapidly produce false information, fake news, and fog of war on a large scale, targeting specific topics and audiences. The knowledge of generative AI is defined by its training datasets; when questions fall outside the training set, it may “speak nonsense seriously.” Academically, this is referred to as “hallucination.” If operational command heavily relies on generative AI, it can easily lead to false intelligence, guiding erroneous decisions and triggering “fatal decision-making dehumanization,” challenging human dominance over war. Second, it may impact the fundamental principles of warfare. Generative AI possesses “human-like awareness,” potentially fundamentally altering modes of warfare and rules of engagement, causing disruptive impacts on the foundational principles of the laws of war. Generative AI’s deadly reliance on massive data training, coupled with the complex and variable battlefield environment and the enemy’s “deception tactics,” may make it difficult to apply the principles of distinction; generative AI can only make “purely rational” judgments that are most favorable to achieving military objectives, lacking the ability for value judgments, making it difficult to adhere to the principle of proportionality; generative AI lacks pain perception, a natural flaw that makes it difficult to comply with the principle of avoiding unnecessary suffering. Third, the inability to predict its capabilities is frightening. Generative AI accelerates the development of general artificial intelligence, intensifying human fears and concerns about the weaponization of AI. Currently, human fears about AI largely stem from imagination, which is very different from nuclear weapons. While the destructive power of nuclear weapons is sufficient to annihilate humanity, humans can “find ways” to control them. However, in controlling AI, humans cannot “think” faster than machines, leading to fear. People worry that general artificial intelligence may become a nightmare for human society. Elon Musk has compared AI development to “summoning a demon,” stating, “Every sorcerer claims they can control the demon, but none have ever succeeded.”
Copyright Statement: This article was published in the October 2024 issue of Military Digest,Authors:Wang Xin, Li Lingjie, et al.,If reprinted, please be sure to indicate “Reprinted from Military Digest.” 
Follow our public account for more information
For membership applications, please reply “Individual Member” or “Unit Member” in the public account
Welcome to follow the media matrix of the China Command and Control Society
CICC official Douyin
CICC Toutiao
CICC Weibo
CICC official website
CICC official WeChat account
Journal of Command and Control official website
International Unmanned Systems Conference official website
China Command and Control Conference official website
National Wargaming Competition
National Aerial Intelligent Game Competition