Chinese Researchers Create Military AI Model Based on Meta's Llama

By IndraStra Global News Team

Cover Image Attribute: This image is computer-generated using AI and is intended solely for illustrative purposes. / INDRASTRA-CREATIVES-AI202411001

Leading Chinese research institutions associated with the People's Liberation Army (PLA) have leveraged Meta's open-source Llama model to create an AI tool with possible military applications, as outlined in three academic papers and reported by analysts. A June paper reviewed by Reuters reveals that six researchers from three institutions—including two affiliated with the PLA's premier research entity, the Academy of Military Science (AMS)—described how they utilized an initial version of Meta's Llama model to develop what they refer to as "ChatBIT."

The researchers utilized an earlier version of Meta's Llama 13B large language model (LLM), integrating their parameters to create an AI tool for military applications. This tool gathers and analyzes intelligence, providing precise and reliable information to aid operational decision-making. According to the paper, ChatBIT was explicitly fine-tuned for dialogue and question-answering tasks within military contexts and was found to outperform specific other AI models that were approximately 90% as capable as OpenAI's ChatGPT-4. However, the researchers needed to clarify how they measured performance or indicate if the AI model had been deployed.

The use of Meta's LLMs by PLA military experts in China for military purposes is a significant development, as it marks the first substantial evidence of such systematic research and utilization. This was highlighted by Sunny Cheung, an associate fellow at the Jamestown Foundation, who specializes in China's emerging and dual-use technologies, including AI. Meta has openly released several of its AI models, such as Llama, but imposes usage restrictions, including a licensing requirement for services with over 700 million users.

Meta's terms of use explicitly forbid utilizing its models for "military, warfare, nuclear industries or applications, espionage," along with other activities under U.S. defense export restrictions, and for creating weapons or content that could "incite and promote violence." However, since Meta's models are publicly accessible, the company has limited means to enforce these restrictions. In response to questions from Reuters, Meta referenced its acceptable use policy and stated that it has measures to deter misuse.

Meta's Director of Public Policy, Molly Montgomery, has stated that any usage of Meta's models by the PLA is unauthorized and against their acceptable use policy. Meta further emphasized the importance of open innovation, noting that the use of an outdated American open-source model in the global AI competition is minor compared to China's substantial investments in AI, projected at over a trillion dollars to surpass the U.S. in this field. Among the Chinese researchers involved are Geng Guotong and Li Weiwei from the PLA's Military Science Information Research Center, the National Innovation Institute of Defense Technology, and researchers from the Beijing Institute of Technology and Minzu University. According to the research paper, ChatBIT, which outperformed specific other AI models, could eventually be refined for broader applications, such as strategic planning, simulation training, and command decision-making, highlighting its versatility.

The release of this research has sparked a heated debate in U.S. security and technology sectors regarding the open availability of models like Meta's. The study brings to light the limitations in ChatBIT's data, which was reportedly trained on just 100,000 military dialogue records, a relatively small dataset compared to other LLMs. Joelle Pineau, Meta's VP of AI Research, has raised questions about the tool's effectiveness given its limited training data. With concerns over AI security, U.S. President Joe Biden signed an executive order in October 2023 to manage AI advancements, acknowledging both the benefits of innovation and the potential security risks associated with removing model safeguards. The U.S. has also moved to restrict American investments in Chinese AI sectors that could pose national security threats. Pentagon spokesperson John Supple commented that the Department of Defense is closely monitoring open-source model capabilities, acknowledging both their potential advantages and the security challenges they pose.

Some experts note that China's advancements in developing its own AI technologies, including establishing numerous research labs, have made it challenging to prevent the country from reducing its technological gap with the United States. In a separate paper reviewed by Reuters, two researchers from the Aviation Industry Corporation of China (AVIC)—a company the U.S. has identified as having PLA connections—detailed using Llama 2 to train strategies for airborne electronic warfare interference. This strategic use of AI underscores the global competition in AI and the potential implications for national security.

China's adoption of Western-developed AI models has not been limited to military applications. It has also expanded into domestic security applications, raising concerns about the broader implications of AI use. A June research paper outlined how Llama was used in "intelligence policing" to analyze large datasets and improve police decision-making. Additionally, in April, the PLA Daily, a state-run outlet, published commentary on the potential for AI to speed up the development of weapons and equipment, contribute to combat simulations, and enhance the efficiency of military training. This expansion of AI use in China underscores the country's commitment to leveraging AI for a wide range of applications.

William Hannas, lead analyst at Georgetown University's Center for Security and Emerging Technology (CSET), stated that efforts to prevent China's access to these technological advances are unlikely to succeed, highlighting the complexity of the situation. A 2023 CSET report identified 370 Chinese institutions whose researchers had contributed papers on General Artificial Intelligence, supporting China's national goal to lead globally in AI by 2030. Hannas also noted the high level of collaboration between top AI scientists from China and the U.S., making it difficult to exclude China from new developments in AI.

With reporting by Reuters

IndraStra Global is now available on
Apple News, Google News, Feedly, Flipboard, and  WhatsApp Channel

COPYRIGHT: This article is published under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. https://creativecommons.org/licenses/by-nc-nd/4.0/

REPUBLISH: Republish our articles online or in print for free if you follow these guidelines. https://www.indrastra.com/p/republish-us.html