Federal agencies across the government are increasingly adapting new uses of artificial intelligence to streamline processes, aggregate data, and even complete tasks designed for human resources staff. And while some have openly embraced AI and its uses, others still don’t believe it can be trusted for operations in nuclear controls.
The Department of Defense in its 2022 Nuclear Posture Review outlined efforts to implement AI with department data and software. Since then, AI has rapidly developed and brought along with that many questions about its future use.
In October, U.S. Strategic Command General Anthony Cotton said implementing AI in Nuclear Command, Controls and Communications (NC3) helps to make those more resilient to adversarial threats and increases decision making capabilities.
“Advanced AI and robust data analytics capabilities provide decision advantage and improve our deterrence posture,” Cotton said, adding that NC3 must maintain “human decision in the loop” to “maximize the adoption of these capabilities and maintain our edge over our adversaries.”
Cotton’s comments have prompted much discussion about AI’s role in nuclear command and controls. The Center for Strategic and International Studies hosted a debate on AI’s role in nuclear command and controls on Jan. 24 as part of its Project on Nuclear Issues (PONI) debate series.
Sarah Mineiro, senior associate at the Aerospace Security Project, and Paul Scharre, executive vice president and director of studies at the Center for a New American Security, debated the question: “Should the United States increase its reliance on artificial intelligence to enhance resilient decision-making in its NC3 systems to prevent inadvertent escalation?”
Mineiro argued for increased reliance on AI except for its use in nuclear weapons deployment whereas Scharre argued against all uses of AI in nuclear command and controls.
Mineiro pointed to the various use cases of AI in NC3 including designing and engineering CPUs and GPUs, image and signal processing, nuclear attack assessment algorithms, and modeling nuclear weapons use scenarios. She said she would never want AI to be involved in nuclear weapons deployment.
“I think we need everything, every tool that American innovation can give us to preserve our security,” she said. “I think AI in NC3 is an appropriate use.”
Scharre said AI cannot be trusted in nuclear command controls because it lacks the novelty of human judgment, it can be hacked or manipulated, and it cannot handle zero tolerance mistake policies.
“It will degrade our decision making, make the risk of inadvertent installation more likely, and undermine nuclear stability,” Scharre said.
Scharre continued and pointed out that AI can be used in tasks which are more repeatable – such as taking off or landing an airplane – but cannot be trusted in nuclear command and controls scenarios.
“We never want a situation where there is an accidental or unauthorized use, and there is just no way AI is good enough to meet that correction,” Scharre said.
Mineiro agreed with Scharre that zero risk tolerance within nuclear command and controls should be kept in place. She pointed out, however, the various other operations AI can be reliably used for which do not include nuclear weapons release.
Mineiro said she is “optimistic” about the Pentagon’s ability to balance integrating emerging technology to boost the American economy and national security while also strictly adhering to nuclear peace agreements.
“I’m a relatively risk tolerant person,” Mineiro said. “The one area I will never choose to accept risk is nuclear command and control.”
The two debaters ultimately came to agree that safeguards must be implemented when integrating AI into NC3 because AI cannot replace human thinking, as much as it may appear to do so.
“Even if the outputs sort of look like humans, that’s what it’s designed to do,” Scharre said. “What’s going on under the hood is not and that’s what we need to be conscious of when we’re using this technology.”
