The U.S. Air Force is running successful tests of “large-language model” artificial intelligence experiments amid concerns of military operations relying on artificial intelligence.
According to Bloomberg, U.S. Air Force Col. Matthew Strohmeyer, the Air Force has successfully executed a large-language model experiment with artificial intelligence for the first time, after years of using only data-based exercises with artificial intelligence under the U.S. Department of Defense.
“It was highly successful,” he said. “It was very fast. We are learning that this is possible for us to do.”
Large-language models (LLMs) utilize large amounts of information from the internet to enable artificial intelligence to predict and generate responses to specific prompts from users. LLMs are similar to generative AI platforms, such as Google’s Bard and ChatGPT.
Five LLMs are reportedly being used in an eight-week exercise conducted by the Pentagon’s digital and AI office, top military officials, and U.S. allies. While the Pentagon will not currently confirm which LLMs are being tested, a San-Francisco based company named Scale AI claims that its Donovan product is one of the five platforms being tested by the military.
Strohmeyer noted that without the use of LLMs, an information request for a specific part of the U.S. military currently requires multiple hours or days, while one of the Pentagon’s tests showed that artificial intelligence can complete the information request within 10 minutes.
“That doesn’t mean it’s ready for primetime right now,” he said. “But we just did it live. We did it with secret-level data,” he says of the experiment, adding it could be deployed by the military in the very near term.”
READ MORE: 300 million jobs will be replaced, diminished by Artificial Intelligence, report warns
Strohmeyer explained that the military provided the artificial intelligence platforms with classified operation information for the test. The goal of the ongoing exercises is to allow the U.S. military to use data obtained from artificial intelligence as part of the decision-making process, sensors, and firepower, according to Bloomberg.
The Pentagon’s eight-week exercise, which is expected to conclude on July 26, will also involve a test to determine whether the military will be able to use LLMS to formulate new military options that have not previously been considered. As part of this, the U.S. military will conduct an experiment with the LLMs in order to plan for a potential military response to a global crisis, such as an incident in the Indo-Pacific region.
To demonstrate a potential military generative artificial intelligence experience, Bloomberg provided Scale AI’s Donovan platform with 60,000 pages of open-source information, including both Chinese and U.S. military documents. Asked whether the United States would be able to adequately address a potential conflict in Taiwan and whether the United States or China would win in a war, the artificial intelligence platform noted, “Direct US intervention with ground, air and naval forces would probably be necessary.”
The news outlet also reported that the artificial intelligence platform stated, “There is little consensus in military circles regarding the outcome of a potential military conflict between the US and China over Taiwan.”
While Strohmeyer highlighted the potential benefits of the military using additional artificial intelligence, other military officials and lawmakers have concerns regarding artificial intelligence. These concerns include the ability for artificial intelligence to provide false information as fact and the ability for artificial intelligence to be hacked by other users.
“We have to be very concerned with what we talked about, which is the unintended consequences,” retired Air Force Maj. Gen. William Enyart said last month.
