IARPA Director: U.S. Must Lead in AI to Stay Ahead of Adversaries

Intelligence Advanced Research Projects Activity (IARPA) Director Rick Muller outlined the agency’s critical role in advancing AI research and ensuring the U.S. stays ahead of both global competitors and emerging threats during GovCIO Media & Research’s AI Summit Nov. 7.

Muller said that within the Intelligence Community (IC), AI is used to improve analytical capabilities and synthesize vast quantities of data for that analysis. Within IARPA, Muller explained that the agency is examining and funding research into “high consequence applications of AI technology,” particularly when lives could be at stake.

Muller pointed to programs within IARPA like TrojAI, which is designed to detect backdoor malicious “Trojan” AI attacks, and Bias Effects and Notable Generative AI Limitations (BENGAL), which aims to understand the landscape of large language model biases, threats and vulnerabilities to enable the IC to safely leverage the emerging technology.

“What IARPA’s doing is a very good foundation for that. But what the next program looks like, I want to understand what your opinion is, what challenges you’re seeing,” Muller said. “I do the same thing with people in the intelligence community. I want to understand what their limitations are, what’s hard, where the gaps are and how IARPA programs can fill those gaps.

Muller said that the funding and research invested by private companies into AI research “dwarfs” what IARPA can do, so his agency must be strategic in identifying and prioritizing where to invest its resources, targeting areas that haven’t been explored in other research.

“What IARPA has to do is be very careful to understand what questions are important to the intelligence community that are not being answered or that will not be answered by the general market. That’s why we’re focusing on specific questions like AI compromises, AI biases,” Muller said. “We think those are areas where IARPA funding can have a larger impact because that’s not the area that the broader markets are going to be concerned about in the next year or two.”

According to Muller, IARPA must help the U.S. government and the broader industry understand the ins and outs of AI, and not leave the responsibility of understanding AI to other countries.

“The U.S. needs to have this expertise here. We can’t outsource this to other places. We need to understand every step in AI, in both the logical and the economic supply chains, that go into building and sustaining AI tool sets. Then, we need to understand safety and security in all of its implications,” Muller said.

IARPA’s role as a leader in AI also involves anticipating innovations from adversaries. Muller emphasized the agency’s responsibility not only to develop new technologies but also to understand how other nations might use those technologies against the U.S. in the future.

“We need to develop expertise at every level, from people understanding how to use the tools as part of their regular employment workflows, but also people understanding how that we write the software, develop the tools, train the software,” Muller said. “We need to understand better how computer hardware can be tuned to better execute those steps and how it can be compromised to maliciously affect those steps.”

According to Muller, the future of AI could lead away from spy versus spy interactions to AI versus AI conflicts. He emphasized the agency must be a leader in the next phase of conflict and continue to think ahead of government, industry, and most of all, the adversary.

“I want to keep IARPA at the forefront of these fields so that we maintain expertise in these areas and so that we can share that expertise with the nation to ensure that the U.S. leads the world in safe and secure AI [and] that we do so in a way that reinforces American values and democracies worldwide,” Muller said.