China’s People’s Liberation Army wants to be the first to capitalize on a range of military applications for generative artificial intelligence that could change warfare—but political, economic, and scientific challenges, some of which U.S. artificial intelligence developers also face, stand in the way.
“Overall, China understands the need to be a first mover (or close follower) in generative AI on the battlefield to ‘firmly grasp the strategic initiative of intelligent warfare and seize the commanding heights of future military competition,’” wrote Josh Baughman, an analyst with Air University’s China Aerospace Studies Institute, in a new paper published Aug. 21.
However, Baughman noted that Chinese policy makers, like their counterparts in the U.S., are wary of integrating the technology without careful testing.
“When we talk about generative AI in a military application, people’s lives are on the line depending on how we apply it,” Baughman told Air & Space Forces Magazine. “It’s high stakes, so absolutely you need to have that trust.”
Generative AI refers to programs “that can generate high-quality text, images, and other content based on the data they were trained on,” according to IBM. Perhaps the most notable application of generative AI is ChatGPT, a chatbot that can write poems, college essays, song lyrics, and other creative content.
Military planners around the world, including Air Force Secretary Frank Kendall, predict the technology could help accomplish tasks and make decisions on the battlefield, though it may be a while before such systems can be relied on in high-stakes situations.
China’s PLA seems to be on the same page: Baughman cited several PLA media sources that generally agree AI will inevitably play a role in warfare, and it could prove decisive in seven key areas:
- Human-machine interaction: Because it can understand both human language and machine language, generative AI could help analysts digest large amounts of information in a much smaller amount of time. The PLA predicts a ChatGPT-like program becoming a joint combat system that can plan tasks, assign objectives, and strike targets, Baughman wrote.
- Decision-making: By processing large amounts of information, generative AI could help commanders select the best combat action plan faster and enable decentralized command for isolated troops.
- Network offensive and defensive warfare: PLA media predicts generative AI could help hackers “design, write, and execute malicious code, build bots and websites to trick users into sharing their information, and launch highly targeted social engineering scams and phishing campaigns,” Baughman wrote. Such offensive tools may become so sophisticated that AI systems may be the only way to defend against them.
- Cognitive domain: PLA media sources discuss using generative AI to “efficiently generate massive amounts of fake news, fake pictures, and even fake videos to confuse the public,” Baughman wrote.
- Logistics: Generative AI could help allocate resources, manage warehouses, plan supply routes, and identify inefficiencies faster than before. It could also be used to predict future material demand and create a budget for procuring resources.
- Space domain: Above the atmosphere, where objects move at several times the speed of sound, generative AI could help monitor satellite health. It could also help engineers design new launch vehicles and spacecraft.
- Training: The PLA lacks real-world combat experience, but generative AI may help “quickly build combat simulation through simple human language descriptions,” PLA writers said, especially when combined with historical training data and fresh intelligence.
Yet for all those possible applications, China also faces many challenges in developing generative AI for military purposes, some unique and others applicable to developers around the world, Baughman said.
Party rules: Generative AI requires large amounts of data, but some information is off-limits under the Chinese Communist Party. Article Four of the party’s regulations for generative AI says such programs should “adhere to the socialist core values, and must not generate content that incites subversion of State power.”
In his paper, Baughman cited a Chinese CEO joking that Chinese large language models cannot count to 10 since it would include the numbers eight and nine: a reference to the censorship of information about the 1989 Tiananmen Square protests. However, while information restrictions could slow the growth of generative AI in some areas, it may not be relevant to all military uses.
“There is a party problem, but I don’t think it will be an issue for pure engineering or technical applications,” Baughman said.
Silicon shortfall: Generative AI requires immense computing power, and immense computing power relies on semiconductor chips. U.S. sanctions limit the Chinese chip supply, but workarounds and China’s long-standing efforts to build its own chip infrastructure cannot be discounted.
Corruption: The Chinese government is making enormous investments in AI, but much of it ends up with firms that have more political connections than technical competence, as Gregory Allen, an AI expert for the Center for Strategic and International Studies wrote in May. That problem may grow worse as U.S. sanctions prevent competition with exporters, Allen noted.
Data Sets: Baughman explained that building an effective military data set for AI requires accurate and precise data, and while this issue affects both Chinese and U.S. AI developers, the PLA may feel it more acutely due to a lack of real-world combat experience.
Optimization: Data must be properly labeled, adjusted, and interpreted to be useful. Baughman cited PLA media articles showing that “availability and interpretability of the data are poor,” and that interaction with professional users is not happening at a large enough scale to work in the field.
Trust: Both U.S. and Chinese policymakers fear losing control of battlefield AI, to the extent that PLA writers repeatedly stress the need to have a human in the loop of systems involving AI.
“The PLA most certainly wants to be the first mover on applying a more comprehensive application of Generative AI on the battlefield, but they will not do so until they can fully trust the technology,” Baughman wrote.
Still in the Race
Despite the challenges, China is at the same level or ahead of the U.S. in some areas of AI development, Baughman said. AI is a key element of the CCP’s grand strategy known as “Digital China,” a sweeping digital transformation designed to make Chinese society more efficient and competitive at the national level.
“China ties the advancement of these emerging technologies with the rejuvenation of China and with maintaining the legitimacy of the party,” he said. “It’s something of paramount importance.”
One military application of generative AI may already be within reach. PLA writers discuss using AI in the cognitive domain to “destroy the image of the government, change the standpoint of the people, divide society and overthrow the regime” through an overwhelming amount of fake news, videos, and other content targeting human fears and suspicions.
“That is not something years in the future, it is something they can do today,” Baughman said, “and the scale that they could do it at is just unreal.”
The threats will likely change fast as technology rapidly advances.
“Everything is going to be moving faster and evolving faster,” he said. “The United States has to be prepared for those major changes. Just look at how generative AI has transformed over the past six months or so. … From the military to the economy, it’s going to transform a lot of different things.”