中國軍事論壇 | 人工智慧賦能合成訓練，提升品質與效率
現代英語：
The form of warfare determines the form of training. Currently, the widespread application of artificial intelligence technology will reshape the form of warfare and combat patterns, and trigger profound changes in military training. As an important part of the organizational structure of the new military training system, combined arms training urgently needs to be infused with an “intelligent core” of artificial intelligence, so as to better play its pivotal role in the new military training system, realize the transformation from “formal integration” to “spiritual integration,” and from “elemental coordination” to “intelligent leadership,” and promote the continuous advancement of combined arms training in the new era to higher quality and higher level.
Breaking the deadlock: Driving a change in training logic
Artificial intelligence empowers synthetic training not only as an “efficiency enhancement tool” to improve training effectiveness, but also leads to changes in the connotation, extension, mechanism, and standard requirements of synthetic training.
Achieving intelligent coupling involves a shift in the logic of convergence. Overcoming division through unity and disunity through cohesion are crucial battlefield principles. The key to combined arms training is “unity.” Artificial intelligence empowers combined arms training to better adapt to the collaborative needs of intelligent warfare, making it crucial for creating a “chemical reaction” in operational coordination. The training focuses on deeply integrating human creativity and value judgment with the computing power and intelligence of machines, forming a cognitive advantage at a higher dimension, and achieving a highly integrated, flexible, and intelligently coupled training system. Manned-unmanned collaborative training is a typical example of deeply integrating manned combat forces with unmanned combat systems possessing “intelligent brains,” pursuing minimal casualties and maximum operational efficiency.
Achieving an iterative logical transformation into a closed-loop system. Traditional training is limited by physical conditions, resulting in high trial-and-error costs and long iteration cycles. By leveraging artificial intelligence to create a “digital twin” training environment, through virtual-real interaction and iterative feedback in parallel systems, synthetic training can shift towards a process of continuous exploration, trial-and-error optimization, forming a new training closed loop. Training at different levels can be implemented simultaneously, and thousands of tactics can be tested and optimized in parallel in virtual space at low cost and high speed. The various elements of overall combat capability can be generated almost independently without regard to sequence. At the same time, the generation of combat capabilities exhibits certain characteristics of distribution, synchronicity, integration, and nonlinearity, significantly compressing the traditional training cycle, accelerating the synchronous generation of combat capabilities across levels, and further expanding the iteration of combat capabilities to “intra-domain foundation, cross-domain collaboration, and full-domain integration”.
Extending the value logic of intelligent emergence. Traditional training cannot pre-plan all possible interactions, nor can it easily generate new tactics and collaborative modes that go beyond pre-set plans. This dilemma is difficult to overcome when facing the demands of intelligent warfare. However, artificial intelligence is quietly changing this model, transforming the value of the training ground from simulating past wars to exploring the possibilities of future wars. Artificial intelligence empowers synthetic training, injecting it with the underlying driving force to generate “intelligent emergence.” For example, game-like confrontations with intelligent opposing forces force trainees to break out of conventional thinking frameworks, potentially leading to previously unthinkable, counterintuitive tactical combinations. The purpose of synthetic training is not only to execute known tactics, but also to hone the ability to innovate methods and update strategies in adversarial environments.
Reconstruction: Shaping Synthetic Training Patterns
Synthetic training incorporating artificial intelligence is gradually evolving into a new training model that emphasizes combat-oriented organization, focuses on enhancing intelligence and integration, shifts towards distributed autonomy, and is geared towards dynamic battlefields.
The training focuses on combat-oriented grouping. Today’s combined arms training features more diverse training subjects, more varied force compositions, and higher capability requirements. The training emphasizes combat-oriented grouping, focusing on mission-driven consistency between training and combat, and is characterized by modularity, innovation, and scalability. Artificial intelligence, acting as a “dispatch center,” can assess the status of combat units based on the battlefield situation, quickly generate optimal force grouping plans, allocate relevant elements as needed, integrate new domain and new quality forces, and practice how to quickly aggregate and disperse forces to form flexible “mission-customized” combined arms groups. This provides the system with plug-and-play capability modules that can be dynamically reconstructed, efficiently linked, and adaptively adjusted like building blocks.
Training content leans towards enhancing intelligence and integration. Traditional training focuses more on assessing whether coordinated actions are completed according to plan, time limits, and standards. In intelligent warfare, humans and intelligent systems together form the basic combat components, exerting combat effectiveness through their functional division and deep integration. Therefore, the focus of new-era integrated training should also pay more attention to improving human-machine integration capabilities. In the past, training content based on human-to-human collaboration—including technology upgrades, experience-based training, and self-awareness training—has become less effective. Training content that enhances intelligence and integration is gradually becoming the key to integrated training. In tactical coordination training, trainees need to master how to collaborate and interact efficiently with artificial intelligence systems, how to use artificial intelligence to reorganize collaborative relationships, close the kill chain, coordinate joint troop actions, and achieve “combined punches”.
Training methods are shifting towards distributed and autonomous approaches. The changes brought about by artificial intelligence to combined arms training are primarily reflected in training methods. This involves not only mastering coordinated operations and solidifying the foundation of collaboration, but also in how to innovatively lead the evolution of combat systems. Distributed training, relying on AI technology, supports simultaneous, remote joint training between different combat units under the same combat background, scenario, and battlefield situation, improving training effectiveness. Autonomous training, employing a “human-outside-the-loop” approach, hones trainees’ ability to handle contingencies and act autonomously. Through feedback and self-adjustment, it promotes autonomous iterative upgrades. Conducting adversarial training breaks through the limitations of learning to fight from experience in the past. It introduces an AI-powered “blue team” to “learn” to fight in a simulated complex battlefield environment, adding random, extreme, and highly harassed scenarios.
Training scenarios are geared towards dynamic battlefields. Traditional training scenarios are mostly “pre-set scripts” designed around “established capabilities” and “known threats,” unable to break free from limited cognition and established thinking patterns. Artificial intelligence empowers synthetic training, transforming it into a “dynamic game system” targeting “unknown capabilities” and “emerging threats,” making it more “imaginative.” Based on training objectives, artificial intelligence autonomously generates logical, multi-domain, and multi-dimensional virtual combat scenarios. Through repeated practice in such highly complex and uncertain environments, trainees are more likely to develop new understandings of the future battlefield.
Exploration: Prospective Synthetic Training Path
Artificial intelligence-enabled synthetic training is an iterative evolutionary process. Looking ahead at its development path, the aim is to transcend developmental limitations and narrow-minded thinking, directly addressing “multi-agent game theory” and “digital twin training grounds,” thereby achieving multi-dimensional and systematic advancement.
Build a comprehensive training foundation. Based on digital twins and intelligent technologies, create a comprehensive training environment to achieve intelligent interaction between people, equipment, and environment. This will enable all training combat units to become dynamically adjustable “intelligent agents,” conduct cross-domain training, improve the command, decision-making, and adaptive coordination capabilities of human-machine hybrid intelligence, and incubate new tactics and formation patterns in a realistic battlefield environment.
Deploy an intelligent blue force system. Build an algorithmic adversary with autonomous evolution capabilities and dynamic game theory thinking, shifting training from “adapting to the known” to “coping with the unknown.” Through deep reinforcement learning and game theory models, the intelligent blue force can not only learn known tactical experiences but also autonomously generate diverse tactics based on real-time situations. Furthermore, it can gain insights into the opponent’s behavioral patterns during interactions, prompting the development of real and effective strategies in dynamic confrontations, and honing the unit’s tactical innovation and human-machine collaboration capabilities through continuous high-intelligence confrontations.
Innovate integrated training models. New-era combined arms training demands innovation-driven, technology-enabled approaches, requiring bold exploration and willingness to experiment. This necessitates seamlessly integrating testing grounds, training grounds, and battlefields, and innovating an integrated training model encompassing operational testing institutions, training institutions, and troops. Trainers are not merely simple technology providers and supporters, but rather embedded as training designers, process analysts, and evaluators within the training process. This allows for a better understanding and methodological revolution in training, validating new technologies, tactics, and formations in combined arms training, exploring future combat winning mechanisms, and simultaneously using data from real-world training to optimize artificial intelligence models, forming an integrated and interactive closed loop that truly integrates training with real-world application.
現代國語：
戰爭形式決定訓練形式。目前，人工智慧技術的廣泛應用將重塑戰爭形式和作戰模式，並引發軍事訓練的深刻變革。作為新軍事訓練體系組織結構的重要組成部分，諸兵種合成訓練亟需注入人工智能的“智能核心”，以更好地發揮其在新軍事訓練體系中的關鍵作用，實現從“形式融合”到“精神融合”、“要素協調”到“智能領導”的轉變，推動新時代諸兵種合成訓練不斷邁向更高水平、更高質量的發展。
打破僵局：驅動訓練邏輯的變革
人工智慧賦予合成訓練的權力不僅在於將其作為提升訓練效果的“效率增強工具”，更在於引發合成訓練在內涵、延伸、機制和標準要求等方面的變革。
實現智慧耦合意味著融合邏輯的轉變。以團結化解分裂，以凝聚力化解紛爭，是戰場上至關重要的原則。聯合兵種訓練的關鍵在於「團結」。人工智慧賦能聯合兵種訓練，使其更適應智慧戰爭的協同作戰需求，從而在作戰協調中產生「化學反應」。該訓練著重於將人類的創造力和價值判斷與機器的運算能力和智慧深度融合，形成更高維度的認知優勢，並建構高度整合、靈活且智慧耦合的訓練體系。有人-無人協同訓練是將有人作戰部隊與擁有「智慧大腦」的無人作戰系統深度融合的典型例證，旨在最大限度地減少傷亡並提高作戰效率。
實現迭代邏輯轉換，形成閉環系統。傳統訓練受限於物理條件，導致試誤成本高且迭代週期長。透過利用人工智慧創造「數位孿生」訓練環境，在平行系統中實現虛擬實境互動和迭代回饋，合成訓練可以轉向持續探索、試誤優化的過程，形成新的訓練閉環。不同層級的訓練可以同時進行，數千種戰術可以在虛擬空間中以低成本、高速度並行測試和最佳化。整體作戰能力的各要素幾乎可以獨立生成，無需考慮順序。同時，作戰能力的生成呈現出一定的分佈性、同步性、整合性和非線性特徵，顯著壓縮了傳統訓練週期，加速了跨層級作戰能力的同步生成，並將作戰能力的迭代進一步擴展至「域內基礎、跨域協同、全局融合」。
拓展智能湧現的價值邏輯。傳統訓練無法預先規劃所有可能的交互，也難以產生超越預設計劃的新戰術和協同模式。面對智慧戰爭的需求，這一困境難以克服。然而，人工智慧正在悄悄改變這個模式，將訓練場的價值從模擬過去的戰爭轉變為探索未來戰爭的可能性。人工智慧賦能合成訓練，為其註入了產生「智慧湧現」的內在驅動力。例如，與智慧敵軍進行遊戲式的對抗，迫使受訓人員打破傳統的思維框架，可能催生出以前難以想像、違反直覺的戰術組合。合成訓練的目的不僅在於執行已知的戰術，更在於磨練在對抗環境中創新方法和更新策略的能力。
重構：塑造合成訓練模式
融合人工智慧的合成訓練正逐步演變為一種新的訓練模式，強調以作戰為導向的組織，專注於提升情報和協同作戰能力，轉向分散式自主作戰，並適應動態戰場環境。
訓練重點在於以作戰為導向的編隊。如今的聯合兵種訓練具有更多樣化的訓練科目、更豐富的兵力構成以及更高的能力要求。此訓練強調以戰鬥為導向的分組，專注於訓練與實戰之間任務驅動的一致性，並以模組化、創新性和可擴展性為特點。人工智慧作為「調度器」發揮作用。「指揮中心」能夠根據戰場態勢評估作戰單位的狀態，快速生成最優兵力編組方案，根據需要調配相關要素，整合新領域和新素質的部隊，並演練如何快速集結和分散兵力，形成靈活的「任務定制」合成兵種群。這為系統提供了即插即用的能力模組，可以像積木一樣動態重構、高效連接和自適應調整。
訓練內容傾向於增強智慧化和一體化能力。傳統訓練更著重於評估協同行動是否按計劃、按時、按標準完成。在智慧戰中，人和智慧系統共同構成基本的作戰要素，透過功能分工和深度融合發揮作戰效能。因此，新時代一體化訓練的重點也應更重視提升人機融合能力。過去基於人際協作的訓練內容——包括技術升級、經驗訓練和自我意識訓練——效果已下降。增強智慧化和一體化能力的訓練內容正逐漸成為一體化訓練的關鍵。在戰術協調方面，在訓練中，受訓人員需要掌握如何與人工智慧系統高效協作和互動，如何利用人工智慧重組協作關係，完善殺傷鏈，協調聯合部隊行動，並實現「組合打擊」。
訓練方法正朝著分散式和自主化方向發展。人工智慧為聯合兵種訓練帶來的變革主要體現在訓練方法上。這不僅包括掌握協同作戰和鞏固協作基礎，還包括如何創新地引領作戰系統演進。分散式訓練依賴人工智慧技術，支援不同作戰單位在相同作戰背景、場景和戰場情勢下進行同步遠程聯合訓練，進而提高訓練效率。自主訓練採用「人外環」的方式，磨練受訓人員處理突發事件和自主行動的能力。透過回饋和自我調整，促進自主迭代升級。對抗訓練突破了以往從經驗中學習作戰的局限性，引入人工智慧驅動的「藍隊」進行「學習」。在模擬的複雜戰場環境中作戰，並加入隨機、極端和高度騷擾的場景。
訓練場景面向動態戰場。傳統的訓練場景大多是圍繞著“既有能力”和“已知威脅”設計的“預設腳本”，無法突破認知限制和既定思維模式的束縛。人工智慧賦能合成訓練，將其轉變為針對“未知能力”和“新興威脅”的“動態博弈系統”，使其更具“想像力”。基於訓練目標，人工智慧自主產生邏輯嚴密、多域、多維度的虛擬作戰場景。透過在高度複雜和不確定的環境中反覆練習，受訓人員更有可能對未來的戰場形成新的理解。
探索：合成訓練的未來路徑
人工智慧賦能的合成訓練是一個迭代演進的過程。展望其發展路徑，目標是超越發展局限和狹隘思維，直接面向“多智能體博弈論”和“數位孿生訓練場”，從而實現…多維度、系統性推進。
建構綜合訓練基礎。基於數位孿生與智慧技術，創造綜合訓練環境，實現人、裝備、環境的智慧互動。這將使所有訓練作戰單位成為動態可調的“智能體”，開展跨域訓練，提升人機混合智能的指揮、決策和自適應協調能力，並在真實戰場環境下孵化新的戰術和陣型。
部署智慧藍軍系統。建構具備自主演化能力和動態博弈論思維的演算法對手，將訓練重心從「適應已知」轉向「應對未知」。透過深度強化學習和賽局理論模型，智慧藍軍不僅能夠學習已知的戰術經驗，還能根據即時情況自主生成多樣化的戰術。此外，它還能洞察對手在互動中的行為模式，進而促進戰術的演進。在動態對抗中製定切實有效的戰略，並透過持續的高智慧對抗來磨練部隊的戰術創新能力和人機協作能力。
創新一體化訓練模式。新時代的聯合兵種訓練需要創新驅動、技術賦能的方法，需要大膽探索和勇於嘗試。這就要求無縫整合試驗場、訓練場和戰場，並創新涵蓋作戰測試機構、訓練機構和部隊的一體化訓練模式。教官不再只是技術提供者和支持者，而是作為訓練設計者、流程分析師和評估者融入訓練過程中。這有助於更好地理解訓練方法並進行方法論上的革新，驗證聯合兵種訓練中的新技術、戰術和陣型，探索未來作戰的製勝機制，並同時利用來自真實世界訓練的數據來優化人工智慧模型，從而形成一個真正將訓練與實際應用相結合的整合式互動式閉環。
來源：解放軍報 作者：聶曉麗 趙澤夏 責任編輯：王一亙 2026-01-13 07:xx:xx
聶曉麗 趙澤夏