Learning Without Fighting: New Developments in PLA Artificial Intelligence War-Gaming

The Opportunities and Challenges of Intelligentization

A lack of recent experience in combat is often characterized as a major liability and potential disadvantage for the Chinese People’s Liberation Army (PLA) in any future conflict scenario. [1] Despite notable advances in its capabilities in recent years, apparent shortcomings remain in the “software” of the PLA’s training and readiness, and perhaps even its will to fight and courage (China Brief, December 1, 2016). The People’s Republic of China (PRC) has not been at war since its 1979 conflict with Vietnam—of which several current military leaders, including members of the Central Military Commission, are veterans—and there are intense concerns today about the perils of “peace disease.” Today, the PLA’s contemporary experiences in military operations other than war (MOOTW), including counter-piracy and peacekeeping operations, offer only limited experience of direct relevance to potential high-end conflict scenarios. In future fights, the PLA could confront a range of difficulties that could include the apparent rigidity of its command structure, and the relative inexperience of its officers and enlisted personnel. Despite major reforms, the PLA could continue to struggle with joint operations, even as it seeks to leverage a new doctrinal approach that is still being formulated (Diplomat, June 6, 2017). At the same time, the PLA is redoubling its efforts in military innovation, rapidly developing and looking to operationalize emerging technologies—particularly artificial intelligence (AI)—that may require major adaptations in concepts, structures, and training.

For the PLA, Xi Jinping’s exhortation to prepare to “fight and win” future wars may thus prove a daunting endeavor. How is the PLA attempting to overcome such critical challenges? While seeking to enhance the realism and sophistication of “actual combat” (shizhan, 实战) training, the PLA is also expanding its activities in war-gaming and adopting new techniques in training, including the use of virtual reality to enhance realism and enable psychological conditioning (Xinhua, January 17, 2017). In this regard, these aspects of the PLA’s exploration of new directions in military innovation will inform its response to what it sees as a “Revolution in Military Affairs” (junshi geming, 军事革命), or RMA, which is catalyzed and deepened by today’s emerging technologies (Xinhua, August 20, 2014). In particular, AI is seen as a critical strategic technology that is transforming today’s “informatized” (xinxihua, 信息化) warfare to future “intelligentized” (zhinenghua, 智能化) warfare (CNAS, November 2017).

During his work report to the 19th Party Congress in October 2017, Xi Jinping urged the PLA to “Accelerate the development of military intelligentization [junshi zhinenghua, 军事智能化] and improve joint operations capabilities and all-domain operational capabilities based on network information systems” (Xinhua, October 27, 2017). This authoritative exhortation seems to elevate “intelligentization,” which involves leveraging AI technologies to enable and enhance a range of future military capabilities, as a guiding concept for future Chinese military modernization. Clearly, China recognizes AI as integral to future national competitiveness, with the potential to change the global balance of power. These ambitions to “lead the world” in AI were prominently highlighted in the launch of the New Generation Artificial Intelligence Development Plan (Xinyidai Rengong Zhineng Fazhan Guihua, 新一代人工智能发展规划) (PRC State Council, July 20, 2017). This plan also calls for the PRC to “Strengthen the use of new generation AI technologies as a strong support to command decision-making, military deductions [junshi tuiyuan, 军事推演, e.g., war-gaming and operations research], and defense equipment, among other applications.” The interest in the application of “military deductions” has included the use of AI in war-gaming (bingqi tuiyuan, 兵器推演), as well as complex simulations. In particular, the PLA’s apparent progress in the use of AI in war-gaming provides an initial indicator of its attempts to explore new concepts of operations for the dynamics of intelligentized operations, while also seeking to enhance the acumen and preparedness of its officers and personnel for future warfare.

PLA Experimentation with AI War-Gaming

The application of AI to war-gaming could provide an invaluable instrument as the PLA seeks to train for and explore the dynamics of future warfare, influenced by a strategic culture that concentrates on military science. In fact, some of these applications of AI have a long history within the PLA, including early initiatives to apply expert systems to military operations research that date back to the late 1980s and 1990s. [2] In recent years, the PLA’s National Defense University (NDU) has taken the lead in the development of more advanced techniques for wargames (War on the Rocks, February 17, 2015). These advances have occurred under the leadership of Major General Hu Xiaofeng (胡晓峰), who is currently exploring the integration of AI into wargaming, while also examining new directions for decision support. For instance, the level of realism from a more artificially intelligent “blue force” (lan jun, 蓝军)—the PLA’s equivalent of a “red team”—could increase the value of exercises for both training and evaluating the military balance relative to a powerful adversary. For Chinese military strategists, among the lessons learned from AlphaGo’s victories was the fact that an AI could create tactics and stratagems superior to those of a human player in a game that can be compared to a war-game. In that regard, AlphaGo appears to have served as an inspiration of sorts for this new direction in PLA war-gaming, along with related conceptual exploration of the potential of AI in future command and control. At the same time, the expansion of war-gaming within the PLA can contribute to training future commanders who may develop greater skills in strategic thinking and command decision-making, which is a current weakness for the PLA (War on the Rocks, February 2019).

As the PLA seeks to advance theoretical innovations in order to prepare for future warfare, the introduction of AI to war-gaming activities can provide not only training, but also a means of studying and even seeking to “design” future intelligentized warfare. In September 2017, the China Institute of Command and Control (CICC) co-sponsored the first national Artificial Intelligence and Wargaming Forum convened at NDU’s Joint Operations Academy, which debuted an AI system called “Prophet 1.0” (Xianzhi 1.0, 先知1.0), developed by the Chinese Academy of Sciences Institute of Automation (Science Network, September 29, 2017). Prophet 1.0 proved victorious over the human teams, seven to one, with no human intervention in the process (Global Times, September 28, 2017). In December 2017, in another round of competition, Prophet 1.0 again beat top teams in this human-machine confrontation (renji duikang, 人机对抗)with a three to one record (Science Network, December 28, 2017). Evidently, this initial “AI commander” served as a proof of concept for future advances in AI in war-gaming, including through generating data that could be used in machine learning to enhance their sophistication. Such activities are starting to extend across a range of Chinese military and civilian research and educational institutions, with plans to evolve from simpler scenarios of land warfare to the complexities of naval, aerial, air-sea, and cyber battlefields. Between September and December 2018, there was another round of war-gaming competitions titled “Prophet • Battle Sage” (Xianzhi • Bingzheng, 先知•兵圣) in which participants from a number of universities, research institutes, and technology companies competed to test their AI agents in war-gaming (CICC, September 4, 2018). This competition, involving “human-machine confrontation” and “machine-machine confrontation,” was intended as an open test platform to improve the capabilities of the AI agents involved and the human players.

As the PLA appears to prioritize leveraging AI in decision support for commanders on the future battlefield, initial efforts to develop an “AI commander” in war-gaming may contribute to future progress. For instance, in April 2018 the Chinese Academy of Launch Vehicle Technology (CALT) convened an intelligent human-machine competition (zhineng ren-ji duikang sai, 智能人机对抗赛) called “Decisive Victory” (Juesheng Qianli, 决胜千里). This competition involved an “AI commander program” (rengong zhineng zhihuiguan chengxu, 人工智能指挥官程序) known as “Xiao Yi” (小奕), which confronted human players from the Chinese Academy of Sciences and Tsinghua University—and reportedly defeated them six to two (CALT, April 23, 2018). This platform for human-machine intelligent confrontation involves games that can “simulate applied decision-making processes,” enabling sophisticated confrontations, including a competition against the AI agent developed by the Chinese Academy of Sciences Institute of Automation. Decisive Victory has been described as an open platform to support the research, development, and training of intelligent decision-making models for aerospace domain systems, thus contributing to the continued development of intelligent game technologies. However, any attempts to transition from these scenarios of ‘game confrontation’ to the actual battlefield will inevitably prove challenging, given the much greater complexity inherent in actual combat.

Conclusion and Implications

Going forward, the PLA’s evident interest in the application of AI to war-gaming constitutes a notable direction of development. Although debates persist regarding the extent to which the PLA can be a “learning organization,” such experimentation and competitions could contribute to training and conceptual improvements. To date, the PLA appears to be undertaking these activities at greater scope and scale relative to the United States, seemingly motivated by a hope of ‘designing’ the shape of future warfare. These activities can produce data that is valuable to training AI systems for advances in war-gaming and novel techniques for decision-making. Concurrently, the PLA appears to be expanding its use of simulations, as well as virtual reality, to support more realistic training, and this experimentation could contribute to the refinement of initial concepts and theories.

What lessons will the PLA learn from these war-gaming activities? Could the PLA develop unique theories of victory and concepts of operations for future “intelligentized” operations? Might the PLA overestimate the applicability of lessons learned in these activities to the complexities of real-world warfare? Could the uncertainties of a new era of intelligentized warfare increase the likelihood of miscalculation of the military balance? At present, the PLA remains in the process of speculation and experimentation, ahead of full implementation. [4] In the process, the tendency to leverage AI in efforts to compensate for self-diagnosed weaknesses could remain a significant influence on the direction of developments. In the absence of opportunities to “learn from fighting” (zai zhanzheng zhong xuexi zhanzheng, 在战争中学习战争), innovations in AI in war-gaming could prove to be leading factors influencing future conceptual innovations by the PLA.

Elsa B. Kania is an Adjunct Senior Fellow in the Technology and National Security Program at the Center for a New American Security, and she is also a PhD student in Harvard University’s Department of Government.

Notes

[1] For a thoughtful discussion of the ways in which combat experience can matter but is contingent upon other factors, see: Timothy R. Heath, “China’s Military Has No Combat Experience: Does It Matter?”, RAND, November 27, 2018. https://www.rand.org/blog/2018/11/chinas-military-has-no-combat-experience-does-it-matter.html

[2] Deborah R. Harvey and Barbara R. Felton, “Military Operations Research in China: A Defense S&T Intelligence Study,” March 1994, DST-1820S-187-94. This document was regraded unclassified and released in March 1998.

[3] AlphaGo is a computer program designed by Deep Mind Technologies, an AI development company in London. In a series of high-profile matches from 2015-2017, AlphaGo decisively defeated a series of world-class human opponents in the complex board game go. See: “The Story of AlphaGo So Far,” Deep Mind Technologies, undated. https://deepmind.com/research/alphago/.

[4] For reference on the theoretical framework that informs my thinking on these dynamics, see: Thomas G. Mahnken, “Uncovering Foreign Military Innovation.” The Journal of Strategic Studies 22, no. 4 (1999): 26-54.