Defense officials have warned that artificial intelligence models are easier to exploit than previously thought and are not yet qualified for full deployment in the military, the Washington Times reported on Friday.
At a National Defense Industrial Association symposium held last week, Alvaro Velasquez, a program manager at the Defense Advanced Research Projects Agency, stated that large language models are “a lot easier to attack than they are to defend.”
“I’ve actually funded some work under one of my programs at DARPA where we could completely bypass the safety guardrails of these LLMs, and we actually got ChatGPT to tell us how to make a bomb, and we got it to tell us all kinds of unsavory things that it shouldn’t be telling us, and we did it in a mathematically principled way,” he explained.
Generative AI tools, which can create text that appears to be written by a human, have gained traction in the past year with the increasing popularity of tools like ChatGPT, a model that can solve problems and generate content when prompted.
The Department of Defense began experimenting with generative AI before ChatGPT was released, according to Deputy Secretary of Defense Kathleen Hicks, a three-time Wash100 Award winner.
On Thursday, Hicks told reporters that some DOD units have developed their own AI models that are being tested with human supervision.
“Most commercially available systems enabled by large language models aren’t yet technically mature enough to comply with our ethical AI principles, which is required for responsible operational use,” Hicks said.
She added that the department has identified more than 180 scenarios in which generative AI technologies could be valuable for oversight, citing debugging and developing software faster, speeding analysis of battle damage assessments and verifiably summarizing texts from both open source and classified data sets as some examples.
The DOD released a new AI strategy on Thursday, which said that U.S. competitors will keep pursuing advanced AI technologies as their range of military use cases grows.
The plan states that the DOD will develop new technologies in a way that protects U.S. advantages from foreign theft and exploitation while maintaining compliance with relevant laws.
Hicks assured that the department is not seeking a conflict with any nation over AI or technological superiority.
“With AI and all our capabilities, we seek only to deter aggression and defend our country, our allies and partners and our interests,” she said.
Interested in learning more about how the Department of Defense is exploring AI? At the Potomac Officers Club’s 5th Annual Artificial Intelligence Summit, experts from both the public and private sectors will dive into numerous topics related to AI in the federal government. Click here to learn more, and click here to register for the event.
