Why World Is Watching China’s Military AI Development
News Desk
Islamabad: China’s military-linked researchers are exploring ways to build smaller, locally operated artificial intelligence systems by using insights and outputs from advanced AI models developed by leading US companies, according to a Reuters review of Chinese research papers and patents.
The findings highlight a growing global competition in artificial intelligence, where countries are seeking to develop powerful but lightweight AI systems that can operate with limited computing resources.
According to the report, Chinese researchers connected with military and security institutions have been using a technique known as “model distillation.” The method allows developers to create smaller AI models by learning from the responses and capabilities of larger, more advanced systems.
Unlike traditional AI models that require massive computing infrastructure, these smaller systems can be deployed on devices such as drones, satellites and other field equipment where processing power is limited.
The Reuters review of more than 80 research papers and patents found examples of Chinese military institutions applying this approach for tasks including surveillance, cyber operations, software analysis and autonomous systems.
One study linked to a People’s Liberation Army unit reportedly used OpenAI’s GPT-3.5 model to summarize military software code, with the information later used to train a local AI system designed to operate within a closed military network.
Other research has explored using distilled AI models for unmanned vehicles. A 2024 study by China’s National University of Defense Technology examined ways to make image-analysis systems compact enough for drones, allowing them to analyze video, navigate and identify targets even when communication links are disrupted.
Researchers at China’s military science institutions have also tested AI-based target recognition systems for drones, ships and unmanned submarines, according to the report.
The use of model distillation has sparked debate over intellectual property, security and ethical concerns. While the technique is widely used in the technology industry to create efficient AI systems, experts say questions arise when capabilities from another organization’s model are transferred without authorization.
US-based AI companies have raised concerns over possible misuse. Anthropic said it does not provide commercial access to its models to Chinese or Beijing-controlled entities and uses monitoring systems to identify policy violations. The company also warned that distilled models may not retain the full safety protections of the original systems.
The development comes as Washington attempts to limit China’s access to advanced AI technology through restrictions on high-end chips and other critical components. However, experts say smaller AI models that require fewer resources could help countries continue advancing AI capabilities despite such restrictions.
China has rejected accusations of improperly obtaining US AI technology, saying Washington is attempting to preserve its advantage in artificial intelligence. Chinese companies have also denied claims that their models were developed through unauthorized distillation.
Experts note that while model distillation can create efficient and useful AI systems, it does not fully replicate the abilities of larger models. Instead, it transfers selected capabilities, making smaller models faster and cheaper but generally less powerful than their original counterparts.
The growing use of this technology reflects a new phase in the AI race, where the focus is shifting not only toward building bigger models but also toward making AI smaller, faster and easier to deploy in real-world environments.
