JUST IN: Pentagon’s Top AI Official Addresses ChatGPT’s Possible Benefits, Risks
HONOLULU, Hawaii — The Defense Department’s point person for artificial intelligence said the new ChatGPT chatbot taking the world by storm could be both a boon for national security and a cause for concern.
“There’s a lot of good there in terms of how we can utilize large language models like it to disrupt critical functions across the department,” Kimberly Sablon, principal director for trusted AI and autonomy in the office of the undersecretary of defense for research and engineering, said March 7 at the Pacific Operational Science & Technology Conference in Hawaii.
ChatGPT is an AI-enabled prototype chatbot developed by San Francisco-based OpenAI that was released in November. It uses machine learning techniques to rapidly generate answers to queries before its users’ eyes.
Sablon acknowledged the chatbot was the “talk of the town.” It could be used for intelligence analysis or to generate computer code, she said at the conference, which was organized by the National Defense Industrial Association and U.S. Indo-Pacific Command.
“We have struggled a long time to attract coders into the Defense [Department]. I think this is an opportunity to really accelerate military based software” development, she said.
“But with it comes a lot of significant threats that we have to be mindful of,” she added.
It could also be used to generate malicious code and more sophisticated phishing emails, as well as disinformation and misinformation campaigns and deep fakes, she said.
“We have seen a lot of ways this thing has been used to manipulate social media,” she said.
Large language models like ChatGPT are also prone to getting their facts wrong, or what Sablon called “hallucinations.” Once those concerns are addressed though, such chatbots could help streamline a lot of Defense Department business processes and operations, she said.
Sablon noted that the Pentagon has made its own strides in the reinforcement learning realm. The Defense Advanced Research Projects Agency used similar machine learning techniques in its Air Combat Evolution program to pit an AI-generated pilot against a human pilot in a dogfight.
“Events like these are great exemplars of the breakthroughs that we’re starting to see in this space, and I’m confident there is a lot more to come,” she said.
John Ridge, the U.K. Ministry of Defence’s director of defence innovation, said earlier at the conference that he is being asked a lot about ChatGPT lately, especially in regards to whether it — or something like it — can be certified for official use.
“There are some really difficult policy challenges around it,” he said.
“The one thing we do know for certain is that these sorts of capabilities are going to be absolutely crucial. And whoever has the best ChatGPT-type capabilities is going to be dominant for a whole series of reasons,” he said.