Experts Answer Reporters' Questions at the Seventh Military Big Data Forum
Digital-Intelligence Empowerment: When Military Decision-Making Meets Big Data — Experts Answer Reporters' Questions at the Seventh Military Big Data Forum
■ Hua Juan, Luo Zhunchen, PLA Daily Special Correspondents; Liu Xuetao, Reporter; Li Youzhi, Reporter
The Seventh Military Big Data Forum was held in Guiyang. Photo by PLA Daily reporter Li Youzhi.
Today, as the form of warfare accelerates its evolution toward informatization and intelligentization, big data—as a new type of strategic resource—is profoundly influencing and reshaping the mode of combat-power generation.
On the future intelligentized battlefield, big data is not merely a carrier for recording information; it is the "foundational energy" driving the war machine to operate efficiently. Like the "blood" of intelligentized warfare (智能化战争), it flows through every "capillary" of the operational system, sustaining the survival and functioning of the entire intelligent combat system. It can be said that driving the efficient transformation of digital-intelligence technology (数智科技) into combat power has become the key proposition for winning intelligentized wars in the future.
On May 21, the Seventh Military Big Data Forum, hosted by the Academy of Military Sciences (AMS), was successfully held in Guiyang. The forum focused on cutting-edge issues including promoting innovative military applications of big data and intelligent technology, digital-intelligence empowerment (数智赋能) of social science research, open-source intelligence innovation in the digital-intelligence era, high-value data mining and decision support, and trusted applications of large-model intelligent agents. The forum was co-sponsored by the National University of Defense Technology (NUDT), Beihang University (Beijing University of Aeronautics and Astronautics), and Beijing Institute of Technology, and was jointly organized by the AMS Military Science Information Research Center together with the National Key Laboratory of Data Space Technology and Systems, the Information Center of the State Administration of Science, Technology and Industry for National Defense (SASTIND), the National Defense Science and Technology Industry Big Data Innovation Center, and Guizhou University.
This newspaper's reporters interviewed the following experts on topics including human-machine co-governance (人机共治), digital-intelligence security (数智安全), capability generation, and co-creation ecosystems in the era of military intelligence: Feng Yansong of Peking University's Wang Xuan Institute of Computer Science; Luo Wei of the AMS Military Science Information Research Center; Liu Shuo of the Public Security Information Technology and Intelligence Studies College at the Criminal Investigation Police University of China; Li Bo of 360 Group; Zong Chengqing of the International Committee on Computational Linguistics; Xiao Weidong of the College of Systems Engineering at NUDT; and Liu Xianglong of the School of Computer Science at Beihang University. It is hoped that this dialogue will provide intellectual support for accelerating military intelligent construction (军事智能建设).
Reporter: Compared with civilian applications, what are the unique core characteristics of digital and intelligent technology when empowering decision-making in military scenarios?
Feng Yansong: Current AI technology, represented by large language models, is fundamentally built on token-sequence generation and pursues probabilistically optimal solutions—naturally suited to civilian scenarios with high fault tolerance, such as everyday conversation. Military scenarios, however, possess attributes of high dynamism and high risk, with extremely low fault tolerance. This demands that digital-intelligence technology, when empowering military decision-making, achieve rationality, autonomy, stability, and reliability. Its core characteristics concentrate in four areas.
On data: the difficulty of acquiring and processing military data far exceeds that of civilian data. Military data processing cannot simply adopt general-purpose data frameworks; it must follow the unique laws of military affairs and be implemented through engineering practice.
On adversarial gaming (博弈对抗): digital-intelligence technology empowering military decision-making cannot directly reuse the logic of general-purpose AI applications. The conditions of use are extremely constrained. AI applications must be capable of assessing strategic intent from massive volumes of chaotic information within an extremely short time, and must be able to complete a process of active accumulation and autonomous evolution over sustained practice.
On evaluation systems: civilian static evaluation models cannot match the "dynamic gaming" characteristics of military decision-making. There is an urgent need to simulate extremely complex environments and design evaluation mechanisms for anti-interference capability.
On human-machine collaboration: in civilian scenarios, the mechanisms and division of rights and responsibilities for human-machine collaboration are mostly relatively clear, whereas in complex military scenarios, the boundary of rights and responsibilities in human-machine collaboration still needs to be further clarified.
Reporter: How can genuinely valuable and credible information be quickly and accurately screened from the massive volume of open-source military-related information?
Luo Wei: Open-source military-related information online is enormous in volume and mixed with true and false content. The key to quickly identifying authentic, credible, and valuable content lies in doing a good job of information discrimination and verification, which can be approached from two main directions. First, assess source credibility. Prioritize tracing and verifying sources; give precedence to accounts whose publishing entities are clearly identified and whose writing style is measured, and classify accounts by credibility level based on the accuracy of past published content, update frequency, and similar factors. Second, conduct cross-verification of information from multiple parties. Information from a single channel is inevitably one-sided; we must integrate multiple independent, non-interest-affiliated sources for cross-comparison and corroboration. If multiple sources express consistent accounts, the information can generally be accepted; if the information circulates only within a narrow range with no other corroborating evidence, it must be accepted with great caution.
Beyond this, two core supporting points must also be addressed. First, leverage intelligent technology. We can use big data, artificial intelligence, and other technologies to automatically scan and filter massive volumes of military-related information, automatically filter out invalid junk information and duplicate content, sort out the internal connections among information, and substantially improve information screening efficiency. Second, leverage professional personnel for oversight. Machines struggle to comprehend complex backgrounds and contexts, and find it difficult to see through the true intent behind information. The work of verifying the authenticity and assessing the value of important information still requires the participation of professional personnel. In brief: machines are responsible for efficiently screening massive volumes of information, while professional personnel are responsible for providing guidance and professional judgment at critical nodes.
Reporter: In the era of military intelligence, what new core capabilities do intelligence analysts most urgently need to develop?
Liu Shuo: Intelligence analysis is currently transitioning from "information mining" to "intelligent assessment" (智能研判). For intelligence analysts, they urgently need to equip themselves with three "new keys" to become composite-type personnel (复合型人才) who wield AI tools and possess strategic depth.
The first key is "the ability to define the problem"—that is, the strategic insight to find the right direction in the fog. The starting point for future intelligence analysts will no longer be "what has been collected," but rather "beneath the complex and confusing surface, where is the enemy, what are the risks, and what are the key variables?" The second key is "the translation ability for human-machine dialogue." Intelligence analysts must be proficient at translating strategic intent into instructions that machines can understand, while simultaneously translating machine-generated outputs into language that decision-makers can comprehend. The third key is "the ability to question and verify." In daily life, we frequently encounter AI that solemnly talks nonsense or produces systematic bias. The core value of future intelligence analysts also lies in their ability to scrutinize AI-generated "seemingly perfect conclusions" with critical thinking—identifying the traps, assumptions, and blind spots within them—and genuinely converting machine assessment results into reliable intelligence products.
Reporter: From the perspective of military command-and-control logic, how should we define the permission boundaries and security red lines for high-authority autonomous AI agents such as "OpenClaw"?
Li Bo: For autonomous AI agents like "OpenClaw," high authority is the foundation for their operational efficiency, but the balance between authority and security is the core challenge in their deployment. We must conduct advance assessment of military network and national defense critical information infrastructure security, focus on building "native security" (原生安全) command-and-control logic, and strictly observe four ironclad security rules.
First, uphold the ironclad rule of "least privilege": do not grant agents super-administrator privileges, restrict agents to their designated access scope, strictly prohibit them from accessing classified and sensitive data, and prevent risks at the source. Second, implement the ironclad rule of "operational isolation": through containerization and sandboxing isolation approaches, delineate an independent operating environment for core systems. In this way, even if an agent is maliciously exploited, it cannot penetrate core systems and internal networks. Third, practice the ironclad rule of "full-process auditability": all configuration changes and operational behaviors of agents must be fully logged, and any anomalous behavior must be addressed in a timely manner. Fourth, adhere to the ironclad rule of "continuous patch updates": for agent version vulnerabilities and third-party plugin supply chain risks, we must establish a normalized security inspection and patch update mechanism to prevent vulnerabilities from being exploited.
Reporter: Against the backdrop of accelerating digital-intelligence military strengthening (数智强军) and intelligentized operations, how should humans and intelligent machines scientifically divide labor to achieve a "1+1>2" multiplier effect on operational effectiveness?
Zong Chengqing: In the context of digital-intelligence military strengthening and intelligentized operations, the core of human-machine collaborative division of labor is to give full play to human strengths, make full use of machine capabilities, and achieve complementary human-machine advantages. Leveraging their advantages in high-speed computation, massive storage, and sustained operation, intelligent machines can undertake low-to-medium-complexity tasks such as open-source information acquisition, simple processing, and rapid translation. Humans, meanwhile, should focus on key links in operations such as complex situational assessment, and strategic and campaign-level decision-making, with particular emphasis on handling ambiguous and deceptive problems that machines cannot understand or correctly judge. When necessary, intelligent machines can assist in generating decision-making hypotheses that exceed conventional cognition, providing diverse references for operational assessment and command, further amplifying the effectiveness of human-machine collaboration, and ensuring that the operation of intelligent machines never departs from the track of human will.
At the present stage, we must firmly uphold the fundamental principle of "human primacy, machine empowerment" (主导在人，赋能在机). On one hand, personnel must avoid over-reliance on intelligent systems, lest independent battlefield assessment, improvised command, and actual combat handling capabilities be weakened; simultaneously, personnel must cultivate critical thinking and the ability to operate intelligent equipment, using machines rationally and trusting machines prudently, without blindly following machine conclusions. On the other hand, we must avoid the disconnect between training and use, and idle waste, after intelligent equipment is fielded. It can be said that intelligent technology provides technical support for personnel command and decision-making, while the comprehensive quality of personnel is the fundamental guarantee for mastering intelligent equipment and seizing the initiative in war—neither can be dispensed with.
Reporter: Faced with rapidly iterating large-model technology, how do we establish a long-term mechanism for the continuous evolution and dynamic updating of military large models, ensuring that large-model technology upgrades in step with tactics (战法) and training methods (训法)?
Xiao Weidong: Solving this problem requires us to treat military large models as an ecosystem capable of adapting to environmental changes and continuously self-evolving. Specifically, we need to focus on the following several aspects. First, strengthen the digital distillation of the combat training process. The continuous evolution of large models depends on a supply of large quantities of high-quality, dynamically updated data and knowledge. In each mission, the system must be able to automatically collect data such as model decision-making processes, commander correction opinions, and actual combat outcome assessments, and inject this data into knowledge bases and datasets. Second, pursue both periodic iterative upgrades and on-demand "targeted treatment" (靶向治疗). On one hand, we must formulate iterative cycle standards for various types of large models and scientifically set differentiated iteration frequencies for different large models. On the other hand, for failure cases of large models, we must apply advanced technology to conduct "targeted treatment" of large models' actual-combat shortcomings. Third, leverage intelligentized simulation training to drive large-model self-evolution. Simulation training is the most effective pathway for testing and optimizing large models. We can deeply embed large models into command-and-control systems and unmanned equipment, allowing large models to self-learn and optimize through trial and error in adversarial gaming against an "intelligent OPFOR" (智能蓝军), ensuring that large models remain synchronized with the latest tactical thinking. In sum, efficiently converting feedback generated in actual combat and exercise-training missions into nourishment for optimizing models and innovating tactics and training methods is the key to the long-term development of military large models.
Reporter: At present, how should we break through the communication barriers between military digital-intelligence technology developers and operational requirement stakeholders, and build a healthy and sustainable military digital-intelligence technology application ecosystem?
Liu Xianglong: The key to resolving the disconnect between military digital-intelligence technology research and development and operational application, and to breaking through the communication barriers between supply and demand sides, lies in driving developers and operational requirement stakeholders to move from shallow supply-demand interface to deep symbiotic co-creation (共生共创), building a healthy and sustainable military digital-intelligence application ecosystem.
First, anchor on core objectives, deeply bind technological innovation to actual combat scenarios, build a collaborative community with aligned objectives and shared value, and eliminate the misalignment and disconnect between research and application at the source. Second, uphold complementary strengths and scientific coordination: developers primarily leverage their technological innovation advantages, while operational requirement stakeholders primarily provide core resources such as actual combat scenarios and operational data, consolidating the foundation for joint research and ensuring the fit between technology and actual combat scenarios. Third, establish a normalized two-way deep linkage mechanism, with scientific research personnel embedded at the front lines and operational requirement stakeholders participating upstream, jointly identifying genuine pain points and requirements. Fourth, the two sides should build a long-term, diversified co-creation cooperation model: using short-term project cooperation as a foundation, form an iterative closed loop through joint research, co-built platforms, and collaborative project initiation, and continuously drive the implementation of military digital-intelligence technology.
Layout design: Wu Huaijiang