How the US Military Says Its Billion Dollar AI Gamble Will Pay Off

War is more profitable than peace, and AI developers are eager to capitalize by offering the U.S. Department of Defense various generative AI tools for the battlefields of the future.

The latest evidence of this trend came last week when Claude AI developer Anthropic announced that it was partnering with military contractor Palantir and Amazon Web Services (AWS) to provide U.S. intelligence and the Pentagon access to Claude 3 and 3.5.

Anthropic said Claude will give U.S. defense and intelligence agencies powerful tools for rapid data processing and analysis, allowing the military to perform faster operations.

Experts say these partnerships allow the Department of Defense to quickly adopt advanced AI technologies without needing to develop them internally.

“As with many other technologies, the commercial marketplace always moves faster and integrates more rapidly than the government can,” retired U.S. Navy Rear Admiral Chris Becker told Decrypt in an interview. “If you look at how SpaceX went from an idea to implementing a launch and recovery of a booster at sea, the government might still be considering initial design reviews in that same period.”

Becker, a former Commander of the Naval Information Warfare Systems Command, noted that integrating advanced technology initially designed for government and military purposes into public use is nothing new.

“The internet began as a defense research initiative before becoming available to the public, where it’s now a basic expectation,” Becker said.

Anthropic is only the latest AI developer to offer its technology to the U.S. government.

Following the Biden Administration’s memorandum in October on advancing U.S. leadership in AI, ChatGPT developer OpenAI expressed support for U.S. and allied efforts to develop AI aligned with “democratic values.” More recently, Meta also announced it would make its open-source Llama AI available to the Department of Defense and other U.S. agencies to support national security.

During Axios’ Future of Defense event in July, retired Army General Mark Milley noted advances in artificial intelligence and robotics will likely make AI-powered robots a larger part of future military operations.

“Ten to fifteen years from now, my guess is a third, maybe 25% to a third of the U.S. military will be robotic,” Milley said.

In anticipation of AI’s pivotal role in future conflicts, the DoD’s 2025 budget requests $143.2 billion for Research, Development, Test, and Evaluation, including $1.8 billion specifically allocated to AI and machine learning projects.

Protecting the U.S. and its allies is a priority. Still, Dr. Benjamin Harvey, CEO of AI Squared, noted that government partnerships also provide AI companies with stable revenue, early problem-solving, and a role in shaping future regulations.

“AI developers want to leverage federal government use cases as learning opportunities to understand real-world challenges unique to this sector,” Harvey told Decrypt. “This experience gives them an edge in anticipating issues that might emerge in the private sector over the next five to 10 years.

He continued: “It also positions them to proactively shape governance, compliance policies, and procedures, helping them stay ahead of the curve in policy development and regulatory alignment.”

Harvey, who previously served as chief of operations data science for the U.S. National Security Agency, also said another reason developers look to make deals with government entities is to establish themselves as essential to the government’s growing AI needs.

With billions of dollars earmarked for AI and machine learning, the Pentagon is investing heavily in advancing America’s military capabilities, aiming to use the rapid development of AI technologies to its advantage.

While the public may envision AI’s role in the military as involving autonomous, weaponized robots advancing across futuristic battlefields, experts say that the reality is far less dramatic and more focused on data.

“In the military context, we’re mostly seeing highly advanced autonomy and elements of classical machine learning, where machines aid in decision-making, but this does not typically involve decisions to release weapons,” Kratos Defense President of Unmanned Systems Division, Steve Finley, told Decrypt. “AI substantially accelerates data collection and analysis to form decisions and conclusions.”

Founded in 1994, San Diego-based Kratos Defense has partnered extensively with the U.S. military, particularly the Air Force and Marines, to develop advanced unmanned systems like the Valkyrie fighter jet. According to Finley, keeping humans in the decision-making loop is critical to preventing the feared “Terminator” scenario from taking place.

“If a weapon is involved or a maneuver risks human life, a human decision-maker is always in the loop,” Finley said. “There’s always a safeguard—a ‘stop’ or ‘hold’—for any weapon release or critical maneuver.”

Despite how far generative AI has come since the launch of ChatGPT, experts, including author and scientist Gary Marcus, say current limitations of AI models put the real effectiveness of the technology in doubt.

“Businesses have found that large language models are not particularly reliable,” Marcus told Decrypt. “They hallucinate, make boneheaded mistakes, and that limits their real applicability. You would not want something that hallucinates to be plotting your military strategy.”

Known for critiquing overhyped AI claims, Marcus is a cognitive scientist, AI researcher, and author of six books on artificial intelligence. In regards to the dreaded “Terminator” scenario, and echoing Kratos Defense’s executive, Marcus also emphasized that fully autonomous robots powered by AI would be a mistake.

“It would be stupid to hook them up for warfare without humans in the loop, especially considering their current clear lack of reliability,” Marcus said. “It concerns me that many people have been seduced by these kinds of AI systems and not come to grips with the reality of their reliability.”

As Marcus explained, many in the AI field hold the belief that simply feeding AI systems more data and computational power would continually enhance their capabilities—a notion he described as a “fantasy.”

“In the last weeks, there have been rumors from multiple companies that the so-called scaling laws have run out, and there’s a period of diminishing returns,” Marcus added. “So I don’t think the military should realistically expect that all these problems are going to be solved. These systems probably aren’t going to be reliable, and you don’t want to be using unreliable systems in war.”

Edited by Josh Quittner and Sebastian Sinclair