China has a clear understanding of the importance of artificial intelligence (AI) to the future military balance, and a well-developed and resourced system for domestic development of AI technologies. By the judgment of seemingly most Chinese and many foreign analysts, the People’s Liberation Army (PLA) remains far from implementing revolutionary uses of defense AI and is still grappling with the institutional reforms and basic capability development needed to effectively use AI in current operations. But Chinese thinkers are working through the possibilities for AI to give the PLA both symmetric and asymmetric tools for high-end conflict with the US military and its allies, in the context of reasonably clear military strategic goals. Fear of Chinese potential was enough for the US in late 2022 to introduce severe export controls targeting China’s access to high-performance semiconductors, expressly justified by the imperative to constrain Chinese advances in defense AI.

China is ruled by a Party that follows a materialist conception of human development and wields unchallenged authority over all social institutions including the PLA, with a leader who appears entrenched for the coming decade. Official judgments about AI’s structural importance, and of the need to be equipped for long-term strategic competition with the US, are thus unlikely to change. To this end, China is leveraging its dynamic civilian economy to boost defense AI development, within the larger context of a national drive for the commanding heights of AI and other emerging technologies. Whether methods that have worked as part of an integrated global economy can still deliver results under growing “decoupling” pressures from the US remains to be seen. But China’s internal means for AI development are now sufficiently robust that analysts of military and strategic global affairs will find it imperative to watch this space, even as it becomes increasingly opaque.

1 Thinking About Defense AI

1.1 Political Context

The PLA’s subordination to the Communist Party of China (CPC) needs consideration when evaluating Chinese thinking about defense AI. The CPC exercises ideological control through the commissar system at various levels, and at the apex of the military hierarchy through the chairman of the Central Military Commission (CMC), who typically is General-Secretary of the CPC and China’s head of state (currently Xi Jinping). This means that the PLA’s development and use of AI is at least nominally governed by the CPC’s political priorities and theoretical judgments, and by the views of China’s top leader.

Under Xi Jinping, official rhetoric and policy has increasingly emphasized emerging technologies like AI. In his “work reports” to the last two CPC congresses, Xi highlighted the PLA’s need to “accelerate development of military intelligentization” and “intelligent combat capabilities,” giving top-level endorsement to a focus on defense AI. Armed conflict with the US is the PLA’s main preoccupation driving Chinese efforts to both close the defense capability gap with the US symmetrically, and to search for asymmetrical means of circumventing US military advantage. Either way, exploiting emergent information-centric technologies like AI appears critical.

1.2 Towards “Intelligentized Warfare”

China’s successive defense white papers (DWPs) show progression from conceiving warfare’s ‘form’ (形态) from being predominantly characterized by “mechanization” towards “informatization,” and since 2014 towards “intelligentization” (Fedasiuk et al. 2021: 305). This reflects the CPC’s conception of wider society’s development, in which “informatization”—the widespread application of digital information and communications technology (ICT)—is a comprehensive and transformational global trend (Lee 2022a). The PLA has paid close attention to the US “Third Offset Strategy” for evolving capability, organization and operations that emerged in the mid-2010s, with a focus on employing new technologies like AI and on China as a specific adversary.

China’s 2019 DWP cited the trend of “informatization” as a key feature of the global context for China’s defense policy, with new technologies like AI, cloud computing and big data “being applied to the military field at an accelerated pace.” However, this document still characterized “intelligentized” warfare as an emergent phenomenon, which the PLA must account for even while it strives to complete mechanization and integrate mechanized capabilities with ICT (Ministry of National Defense 2019).

By late 2020, the official assessment was that the PLA should “accelerate the integrated development of a mechanized, informatized and intelligent military, with a view to taking the initiative” in global military development (Ministry of National Defense 2020). The current (2020) edition of the doctrinally informed Science of Military Strategy includes intelligentization as a capability development requirement across the PLA’ services, stressing that the international environment now features “rapid development of military intelligentization” (Wuthnow 2021). As put by the director of the CMC’s science and technology (S&T) commission, AI’s disruptive nature offers the prospect of “overtaking (more advanced militaries) on the curve.”

However, this optimistic judgment about AI’s potential is matched with pessimistic assessments of the PLA’s state of development with defense AI compared to the US military (Dahm 2020). As one example of this lag in “intelligentized” capabilities despite the PLA’s rapid inventory expansion, a 2018 Chinese assessment identified a deficit of AI-enabled tools to process the volume of information provided by the PLA’s then-already extensive ISR (intelligence, surveillance and reconnaissance) assets (Qiao et al. 2018). By one US expert’s judgment in 2019, PLA modernization was still at the stage of strengthening the separate services’ basic capabilities rather than of developing a sophisticated joint operations capability, let alone effectively integrating AI for “intelligentized” warfare (Fravel and Carson 2019).

This is unsurprising given the general assessment by Chinese commentators, including Xi Jinping himself, that the PLA still lacks the organizational and human capital to conduct system-centric warfare as described below. PLA modernization and reform has been a methodical rather than revolutionary process, with the official goals remaining “full modernization” by 2035 and becoming a “world class military”—by implication, on par with that of the US—by 2050.

In this context, Chinese thinking about AI’s potential for more advanced and radical military uses can be assumed to be still largely theoretical. Describing the “end state” for PLA organization, capability and doctrine that AI applications will fit into apparently remains a work in progress.

While this doctrinal and organizational evolution is playing out, the PLA is steadily deploying a range of increasingly capable semi-autonomous platforms (Lee 2022b). Discussion in China (as elsewhere) about defense AI is grappling with the gap between current capabilities and the implications of full machine autonomy (Fedasiuk 2020). One 2021 survey of Chinese state media outlets concluded that the PLA has yet to reach an official consensus on AI’s fundamental attributes, absent which operational doctrine cannot be developed (Pollpeter and Kerrigan 2021).

1.3 “System-Versus-System” Operations

Nonetheless, the PLA does seem to have established an orthodox view of current and future warfare as essentially a contest between operational systems, in which more effective employment of ICT will prevail (Cozad et al. 2023). In its description of “basic operational doctrine,” China’s 2015 DWP refers to “prevailing in system-versus-system operations” (Ministry of National Defense 2015). In this context, AI’s information processing capacity was identified at least a decade ago by Chinese theorists as useful to achieve the “information dominance” needed to win wars under “informatized” conditions, the official description of the PLA’s basic mission.

One aspect of information dominance is enhanced command and control (C2). One recent Chinese publication highlights AI’s utility in providing strategic early warning; assisting operational decision-making; integrating operational command (within a unified information environment); and optimizing resource allocation and mission management (Zhou and Chen 2022). These capacities overlap with those attributed by Chinese defense media commentary to next-generation (6G) telecommunications, identified as an enabler for AI (Lee et al. 2022a). This combination of AI with other emerging technologies into an operationally and strategically superior “system of systems” is the key theme in published Chinese descriptions of “intelligentized” warfare.

The advantages of enhanced C2 are often expressed by Chinese writers in terms of the OODA Loop concept, enabling the PLA to get inside the adversary’s decision-making cycle and so deploy resources faster and more effectively (Zhou and Chen 2022). In the context of “system-versus-system” operations, this would allow simultaneous rather than sequential attacks on an adversary system’s elements, bringing rapid paralysis of the adversary’s war making capacity. Long-range precision fires, as exhibited by the US military over the past three decades and increasingly accumulated in the PLA’s inventories, remain central to this operational conception. But kinetic strikes on disaggregated and thereby vulnerable enemy force elements are just part of the objective to achieve cascading and compounding failure in the enemy’s warfighting system (Dahm 2020).

With information acquisition and processing seen as the glue for a superior warfighting “system of systems”, AI stands out as a technology enhancing the PLA’s potential to realize this in practice. Warfare’s emergent form is now “information-based with intelligent features,” with the mechanism for victory consisting of information-enabled conjoined operations that achieve precise use of firepower and destruction of the enemy system. “Whoever has a strong ability to acquire, process, transmit, utilize and control information, and whoever has a high degree of integration of information and firepower, will win in war” (Li and Huo 2023).

1.4 Future Operational and Strategic Concepts

This core conception of AI’s utility in evolving “informatized” warfare follows, or at least responds to, the path set by the US as global military leader. The US quest for first-mover advantage in military applications of next-generation technologies like AI that Chinese attention on accelerated progression by the PLA towards “intelligentized warfare” as necessary to perform its missions.

However, Chinese conceptions of “intelligentized warfare” also envision new concepts of operations in response to a transformed battlespace. As one 2019 commentary put it, “The elements of warfare are changing from “information-led” to “machine-led”, with machine-led warfare reshaping the operational process” (Li 2019). The substitution of unmanned platforms for humans expands the scope for military operations to new physical domains (the deep sea and outer space), while the pervasiveness of digital networks means that future war will extensively involve cyberspace. The need to develop network-based “all-domain operational capabilities” was emphasized in Xi’s 2017 work report.

In recent years, the term “multi-domain precision warfare” (多域精确战, MDPW) has emerged in Chinese discourse (Wang and Deng 2022). This term was highlighted by the US Department of Defense (DoD) in its 2022 annual report on China’s military power, which described MDPW as a “new core operational concept” that aims to target vulnerabilities in the US military’s operational system through the PLA’s own “network information system-of-systems” (Department of Defense 2022).

Chinese writers also take interest in AI’s possibilities for manipulating human cognition to influence the adversary’s perceptions, situational awareness and will to fight (Huang et al. 2023). This could be achieved by real-time analysis of adversary behavior, data degradation and manipulation, influence over public opinion, and direct control of adversary systems (Takagi 2022). Some published Chinese work on this theme shows progression from theory to specific application and assessment (Chen 2023a). This aspect of Chinese thinking about AI was also highlighted in the US DoD’s 2022 report, which describes the PLA as developing AI-powered “cognitive domain operations” as psychological warfare’s next evolution, useful for deterrence and achieving effects against the adversary before armed conflict begins.

1.5 Strategic Stability, Human Control, Ethics and Standards

Among the main themes in Chinese writings on “intelligentized” warfare is hybrid human-machine teaming, with machine intelligence seen as augmenting rather than replacing human control (Pollpeter and Kerrigan 2021). China’s position paper on regulating military applications of AI submitted to the UN Convention on Certain Conventional Weapons in 2021 states that “weapon systems must be under human control,” requiring “necessary human-machine interaction across the entire life cycle of weapons” (Permanent Mission of the People’s Republic of China to the United Nations, 2021). It also called on nations to “refrain from seeking absolute military advantage, and prevent the deepening of strategic miscalculation” (Permanent Mission of the PRC to the UN 2021).

This concern with strategic stability stems from Chinese self-assessments of disadvantage vis-à-vis the US not just in AI and in cyberspace capabilities generally, but in the overall balance of conventional military power (Austin 2014). US officials describe the PLA’s new MDPW concept as a response to the US Joint All-Domain Command and Control (JADC2) warfighting concept. Conversely, PLA authors recognize that the Third Offset itself responds to the PLA’s development of means to target the US military’s vulnerabilities.

Specific concerns raised by Chinese authors about how US advances in AI could undermine Chinese deterrence include the potential for new US capabilities to overwhelm PLA air defense, successfully attack Chinese C2 systems and reduce available response time (Fedasiuk 2020). Such concerns sit within a larger international debate about the escalatory potential of military AI, especially given the “black box” nature of current machine learning techniques and AI’s potential to exhibit emergent behaviors. Conversely, the PLA’s doctrinal emphasis on “active defense” may encourage it to pursue its own “offset” of disadvantageous asymmetries vis-a-vis the US by leveraging AI for an effective first strike.

Notwithstanding professed Chinese concerns about strategic stability, the PLA has reportedly refused to admit mutual risk reduction around military uses of AI as an agenda item for official talks with the US Department of Defense (Allen 2022). China is also the world’s leading exporter of combat drones, which does not suggest an excessive concern about destabilizing effects outside the context of the US-China military balance, at least concerning semi-autonomous technologies. The US, however, remains the leading exporter of surveillance drones.

Human control and data security appear as the salient concerns in Chinese discussion of ethics and safeguards for AI (Toner 2023). While some defense commentaries discuss a necessary progression with increasing automation from “human in the loop” to “data in the loop” (Li and Zhao 2020), such a trend can be expected to remain confined to lower levels of decision-making and command authority. Xi Jinping personally emphasized the need for AI to be “safe, reliable and controllable” in a 2018 Politburo study session (Xinhua 2018).

China is developing a comprehensive system of technical standards for AI, within which ethics and security standards have overriding regulatory effect (Lee et al. 2022a, b). But it is unclear how China’s civilian AI governance frameworks will apply to the PLA, which is known to sit outside China’s general data regulatory regime.

2 Developing Defense AI

2.1 “Military-Civil Fusion” and the Leading Role of China’s Civilian Sector

AI development is led by the civilian sector, in China as elsewhere: one 2021 survey of China’s largest academic database (CNKI) found just one PLA institution among the top 20 publishers of AI-related articles (Fedasiuk and Weinstein 2023). China’s “military-civil fusion” (MCF) policy is often cited outside China to justify zero-risk approaches to engaging with Chinese companies outside the state-owned or defense sectors. Xi’s leadership has seen private firms subjected to an expanded presence and influence of CPC committees, legislation and regulation requiring cooperation with authorities in national security situations, and a political climate incentivizing demonstrations of loyalty to the state’s policy goals.

MCF under Xi has involved increased top-down control and ambition, aiming to push China’s long-isolated and fragmented defense industry sector towards a level of interaction with the civilian economy closer to that prevailing in developed nations, in particular the US. Focused on opening China’s defense industry and PLA procurement systems to civilian participation, MCF aims to promote synergistic relations between these two worlds, rather than simply subjecting the latter to PLA control. The essential idea is “synthesis of military and civilian elements to generate new hybrid outcomes” (Cheung 2022: 85–86).

MCF is complemented by the state’s movement away from a siloed, command-style approach to strategic technology development towards one that is cross-sectoral and open to market actors, officially endorsed in 2019 as a “new whole-of-nation system” (新型举国体制) for developing “key core technologies.” Seeking to combine China’s Maoist legacy of top-down mobilization with market-conforming techniques and market-sourced funding, this approach shapes the bureaucratic environment within which directives such as the 2017 New Generation AI Development Plan (‘2017 Plan’) operate.

Among the 2017 Plan’s “basic principles” is “two-way conversion for military and civilian scientific and technological achievements, and … sharing of military and civilian resources.” The 2017 Plan contains a section on MCF that calls for institutionalizing coordination among research institutes, universities, enterprises, and military industry units (Webster et al. 2017).

The 2017 Plan’s top-level guidance was followed by a “three-year action plan” concerned with more detailed implementation, issued, and supervised by China’s Ministry of Industry and Information Technology (MIIT) (Ministry of Industry and Information Technology 2017). While this document does not reference military applications, it directs a focus on developing “core foundational” technologies that include neural network chips and intelligent sensors, corresponding to the conceptual defense AI applications and fielded PLA capabilities discussed in this chapter. MIIT has a close working relationship with the “Seven Sons of National Defense” universities (discussed further below).

One typology of Chinese companies involved in AI development describes five categories, with firms linked to the PLA and security services constituting one: the others are diversified digital giants (for instance, Alibaba and Huawei), large private firms focused on AI technologies, smaller private firms involved in AI inputs and applications, and state-owned enterprises (SOEs) that provide funding, backbone infrastructure and leadership in implementing AI applications (Sutter and Arnold 2023: 20–21).

Under the 2017 Plan, the state designated individual private sector firms to lead development of AI subfields in an “open national innovation platform” model: as of mid-2019, this involved 15 firms focusing on different subfields that range from autonomous driving to cybersecurity and video sensing. This approach aims for more efficient delivery of state support by avoiding duplications of effort, following market signals as to which firms have already achieved economies of scale and technological leadership (Sutter and Arnold 2023: 24–25). The “open” aspect should, in theory, promote information sharing and other non-monopolistic behavior by these dominant firms, and so benefit the development of a wider Chinese ecosystem.

China’s technology policy approach has been described as “grand steerage:” the state steers direction, while letting market forces operate to increase efficiencies and available resources. In doing so, it welcomes foreign expertise and capital: for instance, China’s prominent AI chip design startups have received significant foreign investments (Lee and Kleinhans 2021). Chinese actors have also made significant AI-related investments abroad, being involved from 2015 to 2019 in an estimated USD7bn of disclosed investments in non-Chinese AI companies (Sutter and Arnold 2023: 29).

AI tops the list of seven technologies prioritized for development in China’s current (14th) Five Year Plan. The rapid development of China’s larger digital economy and S&T innovation system means that AI-oriented firms benefit from a surrounding ecosystem of interrelated technologies. Beijing for instance, where China’s National Engineering Laboratory of Deep Learning Technology is located, has been ranked by the journal Nature as the world’s leading science city for several years (Nature 2022).

Collaboration between different types of actors across the development cycle for emerging technologies like AI is promoted by a system of State Key Laboratories, a growing number of which are run by private companies, and National Defense Key Laboratories (Weinstein and Stoff 2023). PLA academic institutions engaged in AI research notably include the National Defense University’s Academy of Intelligent Sciences, which is conducting research into intelligent robotics, bionic robotics and swarm intelligence, and the Academy of Military Sciences, which hosts an AI Research Center that concentrates on deep learning and human-machine integration (Kania 2021: 527–528). The Chinese Academy of Sciences (CAS) is involved in AI research with military applications: its Institute of Computing Technology was added to the US Entity List for targeted export controls in December 2022. CAS’ Institute of Automation developed the Miaosuan wargaming platform and the AlphaWar AI agent discussed below.

China’s state-owned conglomerates are also involved in developing military-oriented AI applications. China Electronics Technology Group Corporation (CETC) for instance appears to be a leader in swarm intelligence (Kania 2017). A research institute affiliated with China’s state-owned shipbuilding sector claimed in 2023 to have used AI to complete in one day design work for a warship’s electrical layout that with human designers had required 300 times this time investment, with 100% accuracy (Chen 2023b).

Much R&D is carried out at civilian universities and research institutes, some of which are working on AI-related projects with direct military applications. For example, Harbin Engineering University (HEU) developed the HSU001 autonomous submersible, while another autonomous submersible (the Sea-Whale 2000) deployed in 2019 for the declared purpose of deep-sea surveying in the South China Sea was developed by the Chinese Academy of Sciences (Panda 2019). Northwest Polytechnic University (NPU) hosts one of China’s leading R&D centers (the No.365 Institute) for military-use unmanned aerial vehicles (UAVs) (Kania 2018). Judging from published research and patents, multiple civilian universities appear to be working on UAV swarming technology, as are China’s state-owned defense conglomerates (Kania 2017).

Most published Chinese academic research on AI-enabled cyberspace operations is produced by a small number of elite research universities, particularly the so-called “Seven Sons of National Defense” (including HEU and NPU). This is an alliance of seven S&T and engineering-oriented universities that work closely with MIIT and are openly engaged in military research, including programs for AI and intelligent weapons development. Several universities were among the leading vendors identified by one survey of AI-related public procurement contracts from PLA service branches over April–November 2020 (Konaev et al. 2023).

China has also adopted an open “innovation challenge” model from US practice, with state authorities sponsoring contests to demonstrate AI-related technologies. For instance, the 2022 “Xingzhi Cup” National AI Innovation and Application Competition ran competitions on themes of technology innovation, industry empowerment and development of the AI industrial ecosystem (Ministry of Industry and Information Technology 2022). While the official notice made no reference to military applications, its mention of “multi-modal technologies,” “network communication” and other potentially dual-use fields means that the activity could well benefit the PLA, if not through direct technology transfers than by identifying promising civilian firms and research teams or implementation techniques.

One vulnerability for China’s AI development is reliance on foreign-developed software and machine learning software frameworks, historically dominated by US companies (Allen 2019), although Chinese developed equivalents are gaining ground (Ding 2022). Such “open-source open platforms” are among the “core foundational” technologies prioritized by MIIT’s 2017 AI action plan. Open-source approaches generally are now much promoted in China’s ICT sectors as a means of mitigating upstream dominance of US firms and the supply chain chokepoints this provides the US government: much of China’s recent AI processor development has utilized the open-source RISC-V architecture, which is less exposed to US export controls than proprietary architectures like ARM and X-86.

Chinese firms are also competing in development of large language model-based AI tools like ChatGPT, the publicity for which has also drawn attention from China’s defense community. Several commentaries in PLA Daily in early 2023 discussed applications for ChatGPT, both from the PLA’s viewpoint and under the US military’s JADC2 concept, and its potential for enabling cognitive domain operations of the type that the US DoD projects onto PLA thinking (Mao 2023). It can be assumed that generative AI will be incorporated into Chinese defense theorizing going forwards, although again, evidence of real-world implementations remains to be seen.

2.2 Effects of US Export Controls Targeting Chinese AI Development

Semiconductors (specifically, logic and memory chips) are a basic enabler for AI. The dominance of US firms and intellectual property in upstream segments of this complex supply chain allow the US government to target Chinese AI development by restricting Chinese access to semiconductor technologies. This was done with extensive export controls introduced in October 2022, and amended in October 2023, justified as necessary to restrict China’s capacity to develop AI-powered military and espionage capabilities.

These US controls threaten the “fast follower” strategy that China has pursued to date to develop its semiconductor industry, aided by this technology’s well-defined roadmaps for future development (Lee and Kleinhans 2021). Many of the technologies concerned are complex to a degree that means China, despite massive import substitution efforts, will likely remain dependent for years to come on foreign suppliers.

The US controls are framed in ways that are designed to restrict Chinese access to the most advanced generations of logic processors and memory chips, and to the technologies required to manufacture them. The amendments promulgated in October 2023 aimed to close loopholes and to increase US regulators’ visibility of products approaching the controlled specifications thresholds.

The structure of global semiconductor markets and the small size of chips creates significant problems for enforcing export controls, illustrated by Russia’s ability to maintain large amounts of military equipment enabled by export-controlled, foreign manufactured semiconductors obtained through clandestine channels. Where licenses are granted for export of controlled products to China, monitoring compliance of these items once in-country presents further challenges. As of late 2023, Chinese firms seemed able to continue producing increasingly advanced chips at parameters that the US controls are designed to restrict. However, continued dependence on foreign suppliers may put temporary hard constraints on this progress in coming years.

2.3 Human Capital

Both Chinese and foreign assessments of China’s AI talent pool emphasize problems regarding both quantity and quality. One 2020 Chinese government estimate projected that domestic production of skilled AI workers in 2022 would fall 480,000 persons short of the economy’s demand (Weinstein and Stoff 2023). The deficit is especially acute for the highest grade of AI researchers: one assessment of three studies published over 2017–2018 suggests that China is second after the US in number of “AI practitioners” but far behind the US and several other countries in number of “AI experts,” defined as the most innovative personnel who generate the most patents and publications (Ding 2019). One Chinese study published in January 2022 assessed that this situation had not fundamentally changed despite improvement in China’s relative position, “especially (for) … top talents who can integrate AI technology development with the industrial system” (Center for Security and Emerging Technology 2022).

However, Chinese contributions to AI research publications have risen steadily, as has their quality judging from metrics such as citation rates and acceptance rates for papers presented to leading international conferences. One survey of global AI publications over 2010–20 found that if including CNKI—which accounts for an estimated two-thirds of Chinese AI papers—China accounted for half of AI publications globally (Chou 2022).

For several AI subfields, China-based authors have surpassed the US-based author share of the top 1% of cited publications. China’s production of PhD graduates in STEM (science, technology, engineering, and mathematics) disciplines is far outstripping that of the US. Many leading Chinese institutions now provide world-class AI programs, with the highest-ranked universities now accounting for almost half of China’s STEM PhD graduations (Zwetsloot et al. 2021).

Education and talent cultivation for AI are central concerns of the 2017 Plan, including vocational training. In 2017 China revised its high school curriculum requirements to include AI, and in 2018 an AI talent cultivation plan was promulgated for higher education institutions, including measures such as building innovation bases to promote collaboration between universities, research institutes and enterprises (Weinstein and Stoff 2023: 59–60). However, China has had difficulty attracting foreign AI workers, a major handicap in a global talent market. By one 2019 estimate, over 90% of China’s AI talent was domestically sourced (Center for Security and Emerging Technology 2022).

Chinese returnees who completed advanced degrees and gained industry experience in the US have made an outsized contribution to Chinese industry’s development in the AI sector. But political tensions have increasingly impacted such cross-border activity, and these are increasingly translating into legal barriers. Restrictions on “US persons” participating in China’s semiconductor sector included in the US export controls discussed above have already had negative impacts on Chinese industry in semiconductors, a key enabling technology for AI.

The PLA has struggled to compete with China’s civilian economy for talent. The military’s human capital deficit appears to be particularly significant for AI development. To address these issues, by early 2019 the PLA had reportedly established talent recruitment stations at over 2500 colleges and universities nationwide, and internal reforms have aimed to raise benefits for PLA “civilian personnel” to levels comparable with those enjoyed by civil servants (Kania 2021: 536).

2.4 Transnational Collaborations

One survey found that in 2020, 22% of AI publications globally with Chinese-affiliated authors were international collaborations, with the US being the most significant partner country followed by the UK (Chou 2022). A 2023 study found this remained the case despite political tensions, with the EU collectively coming third, EU-China AI collaborations having increased continuously over 2017–2022. In the UK and EU, collaborations with Chinese actors are concentrated in a small group of universities, twelve institutions accounting for over a thousand co-authored papers each over the surveyed period (Arcesati et al. 2023).

The US digital technology sector been an attractive target for Chinese investors, with one 2021 study finding that the Chinese stake in top US AI-oriented start-ups was double that of other foreign investors combined (Chang and Hannas 2023: 42–43). Conversely, a 2020 study found that China hosted 10% of US multinationals’ (MNCs) foreign AI labs, paired with Israel in second place after Europe (Heston and Zwetsloot 2020). However, US investment in China’s AI sector is now constrained by “US persons” provisions in export controls and the forthcoming outbound investment regulation mandated by Presidential Executive Order in August 2023.

The dual-use nature of many AI applications imports risk that transnational collaborations may support Chinese defense AI efforts. As examples, NPU (mentioned above) ran a long-term collaboration with the Technische Universität Berlin in applying brain–computer interfaces to drone swarming and flight control (Arcesati 2022). In 2022, a researcher affiliated with the Bundeswehr University Munich co-authored a study with individuals from the PLA Information Engineering University on machine learning applications for data extraction from remote sensing images (Arcesati et al. 2023).

China’s strategic partnership with Russia makes that country China’s leading potential partner for direct collaboration on defense AI (Lee 2022a, b). While there is little public evidence of this, the number of known bilateral defense cooperation projects and Russian advances in relevant technologies, combined with Russia’s increased dependence on China since its 2022 invasion of Ukraine, means that such exchanges cannot be ruled out. This would accelerate a trend towards increased AI research collaboration between the two countries apparent since 2016 (Konaev et al. 2021).

3 Organizing Defense AI

The PLA’s structural reform of the mid-2010s included creating a “Strategic Support Force” (SSF), charged with integrating various “strategic” functions previously scattered across the PLA and given a “mandate to innovate.” Established directly under the CMC rather than affiliated with other PLA command elements, the SSF consolidated outer space, intelligence, electromagnetic and cyber warfare capabilities, all fields in which information processing and data analysis is at a premium and which would therefore benefit from AI.

In April 2024, the SSF was abolished and its constituent departments were elevated to the status of independent ‘Arms’ in the PLA hierarchy, which report directly to the CMC despite having a status below that of the separate services, namely the Army, Navy, Air Force and Rocket Force. Among these new Arms, the Information Support Force (ISF) has been assigned a coordinating role in developing and applying the PLA’s networked ‘system of systems’ (Ministry of National Defense, 2024).

However, the precise division of responsibilities between the ISF and its counterpart Arms (the Cyberspace Force, Aerospace Force and Joint Logistics Support Force) was not made clear. Different AI-enabled functions may be distributed between these organizations as deemed appropriate. One goal of this institutional restructuring is likely to be mirroring the US military’s JADC2 concept and its attempted consolidation of networks to centralize operational command and control (Dahm 2024).

The direct subordination of these Arms to the CMC suggests that at least some of their functions may be regarded as “strategic” capabilities over which the CMC wants to exercise close control, as is the case with China’s nuclear forces. Chinese media in 2016 reported the CMC’s establishment of an Intelligent Unmanned Systems and Systems of Systems S&T Domain Expert Group, which is probably charged with setting strategic objectives and requirements and exploring productive links with civilian industry (Kania 2017).

China does not yet seem to have an equivalent to the US DoD’s Chief Digital and Artificial Intelligence Office (CDAO), responsible for clearing large project proposals and providing a whole-of-defense support hub of AI expertise. However, in addition to being CMC Chairman, Xi Jinping also nominally heads other national steering bodies whose decisions may bear on development of defense AI, notably the Central Commission for Cybersecurity and Informatization (CCCI). Based on its composition when established in 2014, the CCCI’s membership includes inter alia the head of MIIT, the chief of the PLA’s Joint Staff Department, and one of the two CMC Vice-Chairmen (Xi’s uniformed deputies in exercising top-level command of the PLA) (Lee 2022a: 21). From early 2023, Xi has also led a “Central Commission for Science and Technology” to oversee national S&T policy.

The PLA’s reform process includes efforts to integrate doctrinal development with technical realities of emerging technologies like AI, although equipment procurement has remained a service responsibility. In 2017 the Academy of Military Sciences, which is the PLA’s top institution for doctrinal theorizing and reports directly to the CMC, integrated six technical research institutes that were previously subordinate to the pre-reform PLA general departments (Wuthnow 2019).

Research on the supplier profile for Chinese defense AI-related procurement reinforces the picture of a decentralized vendor network and the civilian economy’s leading role. One study of defense AI-related contracts over April–November 2020 identified 273 unique (sole source) vendors who were the most common suppliers of AI-related equipment, typically private firms that focus on intelligent software or sensing technologies. Of the contracts surveyed, 61% were awarded to private enterprises, skewed towards companies founded since 2010 (Fedasiuk et al. 2021: 32–32). A further analysis of this dataset focusing on PLA service branch procurements found that among the 1983 procurement records examined, only 13 suppliers were awarded two or more contracts, with the two numerically leading vendors receiving only four and three contracts respectively (Konaev et al. 2023: 17–19).

China’s defense SOEs are both buyers and sellers of AI-related equipment, suggesting that they may be specializing in certain AI subfields and so may avoid “crowding out” private sector investment. This corresponds to the apparent role of these state-owned conglomerates in related ICT “high technologies” such as semiconductors. One such conglomerate (China Aerospace Science and Technology Corporation, CASC, with subsidiaries) appeared by a large margin to dominate numerically the 2020 AI-related public procurement dataset mentioned above, with the SSF coming second (Fedasiuk et al. 2021: 29–30).

4 Funding Defense AI

The opacity of Chinese defense spending inhibits detailed assessments about the PLA’s direct funding for AI. Based on evidence including the 2019 DWP’s statement that 41% of China’s 2017 defense budget went to equipment, one 2021 foreign assessment put the PLA’s annual spending on AI as USD1.6bn-USD2.7bn, or in any case ‘in the low billions of US dollars’, roughly on par with the US military’s spending on AI (Fedasiuk et al. 2021: 10–11). Based on declared Chinese numbers, the equipment share of the defense budget has generally grown since 2010, though China’s 2020 report to the United Nations on its military spending showed a fall to 37.19% (Permanent Mission of the PRC to the UN 2022).

Allowance must also be made that some of the PLA’s major defense acquisition programs are classified and not reflected in publicly available data. The concentration of much R&D work in China’s universities, non-PLA affiliated research institutes and civilian enterprises limits the usefulness of official defense spending as a metric of total defense AI funding. By one foreign estimate, in 2019 China spent around USD25bn on research, development, evaluation and testing for military purposes outside the official defense budget (Tian and Su 2021: 18).

The much-reported “tech sector crackdown” by the Chinese state in recent years has been targeted at large commercial internet platform service providers and does not reflect a general persecution of private enterprise. Top-level policy continues to emphasize the key role in strategic technology development of private enterprise and market forces, though the compatibility of this stance with Xi’s reassertion of CPC-led centralized top-down direction remains to be seen. One market research estimate published in late 2022 projects Chinese investment in AI may reach USD26.69bn by 2026, accounting for about 8.9% of global AI investment (IDC 2022).

5 Fielding and Operating Defense AI

One survey of 58 Chinese defense-AI related papers published between 2016–2020 identified twelve discrete applications that spanned unmanned platforms; “intelligentization” of munitions, satellites and ISR); automation of offensive and defensive cyber operations and missile launch software; and cognitive electronic warfare (Fedasiuk 2020:7). A review of 343 AI-related defense equipment contracts published in China between April–November 2020 found the dominant application areas to be intelligent or autonomous vehicles (over a third of the surveyed contracts), ISR, information and electronic warfare, and predictive maintenance and logistics. Other contracts in this sample set relate to simulation and training, command and control, and automated target recognition (Fedasiuk et al. 2021).

For years the PLA has showcased a succession of increasingly capable maritime and aerial unmanned platforms, including a series of stealthy combat models. By 2018 all four PLA services (including the Rocket Force) fielded a variety of UAVs, including a growing number of multi-mission capable models (Kania 2018). By late 2022, the PLA was integrating a drone “carrier” catamaran into experimental naval task force exercises, with apparent variants of a civilian Chinese tandem-rotor drone on board, potentially to provide transport or ISR functions (Trevithick 2022). Foreign media in 2022 also identified what appeared to be new and larger unmanned underwater vehicles (UUV) at the PLA’s Yalong naval base facing the South China Sea, approaching the size of the US Navy’s developmental Orca autonomous UUV (Sutton 2022). Although deployed numbers appear limited, the PLA has a record of iteratively prototyping new equipment, eventually transitioning to rapid serial production once a given model is regarded as fit for purpose.

The current edition of Science of Military Strategy states that UAVs should be prioritized in unmanned systems development, although PLA theorists also stress that some “intelligentiziation” is necessary for sensing capabilities across all domains. A focus on UAVs is unsurprising given that China is the world leader in civilian drones and in military drone exports, and that the aerial environment presents a relatively simpler challenge for development of machine intelligence. Orientation towards UAVs and aerospace-focused SOEs is apparent in defense AI-related procurement tenders and contracts (Fedasiuk and Weinstein 2023: 176–180).

The nature of China’s maritime sovereign disputes in the East and South China Sea Seas, which incentivizes persistent assertion of claimed rights over vast tracts of maritime space, puts a premium on effective ISR and long-range high-endurance capabilities. The PLA has been using UAVs for this purpose since the early 2010s. UAVs and UUVs are also useful for monitoring Taiwanese defense assets and activity. Unmanned platforms provide a relatively low-cost and low risk means of collecting data on foreign military assets’ electronic signatures, training, tactics and procedures (TTPs) and other information that would be useful in an armed conflict.

Development of sensing capabilities likely benefits from China’s long-term and world-leading development of AI-powered sensor-based networks for civilian applications, notably “smart city” management and self-driving cars. Claimed Chinese research advances in hypersonic weapons over 2022 involve sophisticated sensing capabilities, potentially using AI (Lee 2022b). These indicate near-term potential for deep penetrating strikes to paralyze adversary operational systems, in line with the concepts discussed above.

Another AI application that the PLA seems to be deploying in practice is logistics optimization. The PLA is developing a “strategic delivery” system of integrated transport services and military bases to enable efficient force projection into zones of current operations, utilizing AI to enhance the speed of real-time requirements analysis and services delivery (Chieh and Yang 2021: 62).

6 Training for Defense AI

UAVs are leading the PLA’s deployment of AI-related capabilities and so are supported by a system of specialized military education: one 2016 study identified at least eight PLA academic institutions with UAV-related programs for training specialists (Kania and Allen 2016). AI-enabled virtual and augmented reality systems are also used to train pilots of manned aircraft, with reports in 2021 of an AI agent developed by PLA institutes defeating a fighter pilot in a simulated dogfight.

UAVs have appeared in multi-service PLA training activities, while dedicated UAV-equipped units have progressively increased the sophistication of their training exercises (Kania 2018). The PLA may also be learning from use of Chinese-supplied military UAVs in foreign conflicts, notably in the Middle East and Africa. Reported use of Chinese civilian drones by Russian forces in Ukraine and for training operations potentially provides further intelligence collection and evaluation opportunities.

AI can be used to train for current operations and simulate the effects of future operational concepts and C2 approaches. This is an attractive option for the PLA to address its own negative assessments of its officers’ decision-making capabilities, in the context of the whole institution’s dearth of recent operational experience.

The PLA has been applying AI to wargaming since at least 2017, led by the National Defense University. Some of these activities have involved universities, research institutes and civilian firms, and pitted machine intelligence against machine intelligence as well as against humans, generating data to support further machine self-learning. One US analyst in 2019 characterized these Chinese exercises as exceeding comparable US activities in scope and scale (Kania 2019). The PLA has awarded contracts to develop AI wargaming software for use in professional military education.

AI agents and wargaming platforms for testing them are being developed by both civilian and PLA institutions. For example, the Chinese Academy of Science’s Institute of Automation (CAS IoA) has developed the Miaosuan (庙算) wargaming platform and the AI gaming agent AlphaWar. AlphaWar is claimed to have passed the Turing test—exhibiting behavior indistinguishable from a human’s—in 2020, and to have confirmed this result in the 2021 iteration of the “Miaosuan Cup” tournament (Yin et al. 2022).

This competition, organized by the Chinese Society of Artificial Intelligence and the CAS IoA, pits human and AI players in adversarial and collaborative games on the Miaosuan platform, with human finalists required to guess whether anonymous opponents were human or AI. The competition’s 2021 iteration included four AI agents, two of which were developed by PLA institutions (Institute of Automation, Chinese Academy of Sciences 2021). The competition’s 2022 iteration was described as involving human-machine collaborative teaming and confrontation, and as geared towards the needs of manned and unmanned human-machine hybrid intelligence (Institute of Automation, Chinese Academy of Sciences 2022).

7 Conclusion

The main driver of Chinese military strategy is, by official explanation, S&T-driven evolution in patterns of warfare (Ministry of National Defense 2020). ‘Intelligentization’ of warfare through application of AI is the latest trend in this evolution. The PLA’s overall state of development suggests it is some way from becoming a global leader in “intelligentized” warfare. However, it is already deploying numerous and capable semi-autonomous military systems, while Chinese thinkers are covering the ground in applying AI’s potential to both current military operations and future operational concepts. China has a comprehensive and well-resourced national system for AI development in general, and a focus at the top of its political system on bending this “whole-of-nation system” to the service of defense AI applications.

China does not appear on the point of snatching a decisive lead over the US and its allies in military uses of AI. But Chinese authorities do have a clear view of AI’s structural importance, and reasonably clear military strategic goals that AI development can be directed towards. Foreign analysts tracking Chinese AI development face an increasingly tight information environment, as more data sources are progressively closed off. Yet assessing China’s progress with defense AI will be indispensable to judgments about the global military balance of power.