Chinese Military
Three academic papers and analysts revealed that prominent Chinese research institutions associated with the People’s Liberation Army utilized Meta’s publicly accessible Llama model to create an AI tool for potential military uses.
Six Chinese researchers from three institutions, two of which are part of the People’s Liberation Army’s leading research body, the Academy of Military Science, outlined in a June paper reviewed by Reuters how they utilized an initial form of Meta’s Llama to create what they refer to as “ChatBIT”.
The scientists utilized a previous Llama 13B large language model (LLM) from Meta (META.O), opens new tab, integrating their own settings to create a military-centric AI tool for collecting and analyzing intelligence, providing precise and trustworthy data for making operational decisions.
The paper stated that ChatBIT was adjusted and improved for tasks related to dialogue and answering questions in the military sector. It was discovered that it surpassed certain AI models that were approximately 90% as skilled as the formidable ChatGPT-4 from OpenAI. The researchers did not provide details on their definition of performance or confirm if the AI model was deployed.
Sunny Cheung, an associate fellow at the Jamestown Foundation specializing in China’s emerging and dual-use technologies, including AI, mentioned that it is the initial instance where significant proof shows PLA military experts in China systematically exploring and seeking to utilize the capabilities of open-source LLMs, particularly those of Meta, for military objectives.
Meta has adopted the open distribution of numerous of its AI models, including Llama. It puts limits on their usage, such as the need for services with over 700 million users to obtain a license from the company.
The conditions also forbid using the models for “military, warfare, nuclear industries or applications, espionage” and other activities under U.S. defence export controls, and for creating weapons and content designed to “encourage and support violence”.
Nevertheless, due to the fact that Meta’s models are available to the public, the company has few methods of ensuring compliance with those terms.
When asked by Reuters, Meta mentioned its acceptable use policy and stated it implemented precautions to avoid misuse.
In a phone interview with Reuters, Meta’s director of public policy, Molly Montgomery, stated that any utilization of their models by the People’s Liberation Army goes against their acceptable use policy and is not authorized.
Meta added that the United States must embrace open innovation.
A Meta spokesperson stated that in the worldwide AI race, the purported significance of an outdated American open-source model is insignificant compared to China’s trillion-dollar investment to surpass the US in AI.
China’s research team consists of Geng Guotong and Li Weiwei from AMS’s Military Science Information Research Center and the National Innovation Institute of Defense Technology, along with researchers from the Beijing Institute of Technology and Minzu University.
The paper stated that in the future, ChatBIT will not only be used for intelligence analysis but will also be explored for strategic planning, simulation training, and command decision-making through technological advancements.
The Defence Ministry of China did not respond to a request for comment, and neither did any of the institutions or researchers.
Reuters was unable to verify ChatBIT’s abilities and processing power, but researchers mentioned that its model utilized just 100,000 military dialogue records, a smaller amount in comparison to other LLMs.
Joelle Pineau, VP of AI Research at Meta and computer science professor at McGill University in Canada, expressed doubt about the actual achievements of models trained with trillions of tokens, stating that the smaller models pale in comparison.
The study is happening during a passionate discussion in American national security and technology circles on whether companies like Meta should share their models with the public.
Joe Biden
In October 2023, President Joe Biden of the United States issued an executive order to oversee advancements in AI, recognizing the potential advantages of innovation but also acknowledging significant security threats like the elimination of safeguards within the system.
This week, Washington announced that it was in the process of completing regulations to limit American investments in Chinese artificial intelligence and other technology industries that pose a risk to national security.
John Supple, a spokesperson for the Pentagon, acknowledged that the Department of Defense sees advantages and disadvantages in open-source models. He stated that they will keep a close eye on and evaluate the capabilities of their competitors.
China
Certain analysts believe that China’s efforts in creating its own AI technology, such as establishing numerous research facilities, are causing challenges in preventing the country from closing the technology disparity with the United States.
Two researchers from AVIC, a company linked to the PLA as designated by the US, discussed utilizing Llama 2 for teaching airborne electronic warfare interference tactics in a reviewed academic paper by Reuters.
China’s application of AI developed in the West has also expanded to national security. An article published in June detailed how Llama was utilized for “intelligence policing” to analyze vast quantities of data and improve police decision-making.
In April, the PLA Daily, a government-controlled newspaper, released a commentary stating that AI has the potential to speed up the creation of weapons and equipment, enhance combat simulations, and boost military training effectiveness.
“Are you able to prevent them (China) from accessing?” William Hannas, the main analyst at Georgetown University’s Center for Security and Emerging Technology (CSET), said to Reuters that he does not understand how it could be done. A study conducted by CSET in 2023 identified 370 Chinese organizations that had researchers publishing papers on General Artificial Intelligence, contributing to China’s goal of becoming a global leader in AI by 2030.
Hannas stated that with the high level of collaboration between top scientists in China and the U.S. in the field of AI, it is necessary for them to be involved in advancements.
Leave a Reply