Top research institutions in China, including those tied to the People’s Liberation Army (PLA), have adapted Meta’s (NASDAQ) open-source Llama AI model to create a tool geared toward military applications, as revealed in recent academic studies. This adaptation leverages Meta’s AI technology to potentially enhance China's military intelligence, operational decision-making, and strategic planning.

Development of “ChatBIT” – A Military AI Tool

In a June 2023 paper reviewed by Reuters, six researchers from three prominent Chinese institutions, including two affiliated with the PLA's Academy of Military Science (AMS), detailed their work on "ChatBIT." Using an earlier version of Meta's Llama 2 13B large language model, they created an AI optimized for military dialogue and intelligence processing.

According to the researchers, ChatBIT surpasses some existing AI models in specific military-focused tasks, achieving approximately 90% of OpenAI’s ChatGPT-4 capabilities. The model's effectiveness in practical military scenarios, however, remains unverified.

Meta’s AI Model: Open Source and Open Questions

Meta’s Llama model is publicly accessible, though with usage restrictions that limit applications in warfare, espionage, and other sensitive sectors. Despite these policies, enforcing compliance has proven challenging due to the model’s public availability. Meta’s Director of Public Policy, Molly Montgomery, clarified that any military use by the PLA would violate Meta’s terms.

Meta advocates for open innovation in the global AI race, arguing that a single, outdated model’s influence is minor compared to China's trillion-dollar investments in AI technology.

Strategic Goals for ChatBIT

Chinese research teams, including experts from AMS’s Military Science Information Research Centre, the National Innovation Institute of Defence Technology, and Beijing Institute of Technology, aim to evolve ChatBIT beyond intelligence analysis. Future applications may include command decision-making, strategic simulation training, and tactical planning, as per their research paper.

While the model currently relies on a relatively modest database of 100,000 military dialogues, Joelle Pineau, a vice president of AI Research at Meta, pointed out that this data volume is significantly smaller than typical LLMs trained on trillions of tokens, raising questions about its capabilities.

U.S. Concerns Over China’s AI Developments

The research emerges amid ongoing discussions in U.S. national security about the risks associated with open-source AI models. In October 2023, President Joe Biden signed an executive order aimed at overseeing AI advancements, emphasizing the balance between innovation benefits and security risks.

The U.S. is also advancing rules to limit domestic investments in sectors like AI in China, aligning with national security interests. Pentagon spokesman John Supple remarked on the dual-edged nature of open-source AI, confirming that competitor advancements are under constant observation.

China’s Expanding AI Landscape

China’s investment in homegrown AI capabilities has accelerated, with hundreds of research institutions contributing to AI advancements. These efforts include applications in both military and domestic security. The PLA Daily recently highlighted AI’s potential for weapon development, combat simulation, and military training.

Experts suggest that China’s integration of Western-developed AI technology is inevitable due to the widespread collaboration between Chinese and U.S. researchers. William Hannas of Georgetown University’s Centre for Security and Emerging Technology commented on the difficulty of restricting China's access to advanced AI developments. A 2023 CSET study identified 370 Chinese institutions publishing research on general artificial intelligence, supporting China’s aim to lead in AI by 2030.

Source: (Investing.com, Reuters)