The Department of Defense is leaning into an “AI-first” mentality and developing new artificial intelligence tools, but analysts warn there are still barriers to their broad use in the military.
The Pentagon has poured resources into its Artificial Intelligence Strategy this year and said it aims to reward AI-driven “re-conceptions of legacy approaches” across the military.
While military leadership is enthusiastic, Jake Steckler, a research scholar at GovAI, a nonprofit think tank focused on AI research, warns that without modernizing testing and evaluation standards around AI, soldiers’ lives could be in danger.
“The risks are basically because these technologies are so new,” Mr. Steckler said in an interview. “Some risky situation is going to pop up, and then the next thing you’re going to see is that they’re not going to want to use it.”
Some analysts warn that AI tools sometimes struggle to be consistent in how they work. Many times the responses or choices an AI-enabled system produces are different with even just small changes to their environment or the data used, said Mr. Steckler, the author of a major report on the subject published recently by the Carnegie Endowment for International Peace.
AI already plays a key role in current military operations. The Maven Smart System from Palantir, a Florida-based defense tech firm, uses multiple forms of AI, including frontier models. Maven played a role in strikes against Iran, including helping to identify more than 13,000 targets in just over the first month of the war.
Some leading AI firms want restrictions on how their products can be used by the military. In the highest-profile example, the Pentagon and California-based AI company Anthropic are locked in a major dispute after the military demanded unrestricted use of the company’s most powerful AI large language models.
When Anthropic refused, the Trump administration banned its use and labeled the firm a national security supply-chain risk. That designation, which could deeply harm the company’s ability to do business with the federal government, sparked a major legal fight that is still moving through the U.S. court system.
As AI use expands across the military, the Pentagon has consolidated its buying process, including decisions on where and when to include AI products in military purchases. But that creates a risk that “loads a single office with specialized responsibilities,” Mr. Steckler wrote in his report.
“There are responsibilities that used to exist in many different places [that] are now basically just not clear who owns those,” Mr. Steckler said. The department has reduced the number of specialized staff responsible for purchasing complex equipment in an effort to reduce the level of bureaucracy needed to get modern tools into service members’ hands.”
“Getting rid of an old process is good,” Mr. Steckler said. “But you have to replace it with a new, better process; otherwise, you’re just creating confusion and introducing new risk.”
Total war
Mr. Steckler leaned on his own experience as an Army helicopter pilot and his recent discussions with troops in Ukraine to inform his research. Ukraine has already leveraged AI in its war with Russia, using data analysis tools to help with decision-making while also quickly becoming the leading global testing ground for AI-enabled one-way drones and munitions.
Ukraine’s battlefield has an advantage to speed up AI development that the U.S. doesn’t have: total war. Mr. Steckler noted in his report that Ukraine went from close to a two-year timeline for new technology to be approved for use by the military before Russia’s invasion to now as quickly as two weeks.
“When your choice is either to field something or surrender territory, I think that people are obviously going to be motivated to field things faster,” Mr. Steckler said. “Those things are possible to do when you’re in this sort of a total war scenario.”
In his report, Mr. Steckler described Ukraine’s approach as being willing to “prioritize speed over thoroughness,” simply abandoning systems that don’t succeed. That necessity of total war means that more complex systems have yet to be adopted in Ukraine, “because humans still outperform the available systems,” he wrote.
He also said that a key difference was the involvement of the entire community, a mobilized public willing to work nights in factories or assemble circuit boards at home, for example.
“If you don’t have the same existential risk, I don’t think you’re going to have the same level of sacrifice that people are willing to make,” he said.
The U.S. Army is investing heavily in the next level of AI integration and increasingly having the military contractors responsible for AI development work directly with troops. The new model for solving military problems is speeding up the process, but that doesn’t mean it can overcome all of the barriers to wider adoption.
Training and trust with soldiers will still be a hurdle, Mr. Steckler said.
“Soldiers need to see these things working consistently for them to trust them,” Mr. Steckler said. “It’ll take time for anyone to be able to build that trust and I think there’s a lot of technical problems that still need to be solved before that trust is even possible to build.”
Contact the author
Please read our comment policy before commenting.