Sue Gordon, Former Principal Deputy Director of National Intelligence, speaking at the Kalaris Conference. Photo Credit: GSSR/Hunter Price.
Academics, government officials, and private sector members descended upon Gaston Hall on September 25, 2019 for the Georgetown University Center for Security Studies’ (CSS) and Center for Security and Emerging Technology’s (CSET) annual Kalaris Intelligence Conference. Together, the two centers brought together a range of exciting voices to discuss this year’s topic, Artificial Intelligence (AI). With no shortage of possible applications, AI has drawn opinions from intelligence and military officials and brought new players to the table, such as technology startups, civilian government officials, and technology-focused academics.
The conference’s speakers reflected the range of stakeholders and discussed issues including cooperation and competition in AI, the United States’s military efforts in AI, and the government’s relationship with private companies. The conference’s MC, Ellen McCarthy, the Assistant Secretary of State for the Bureau of Intelligence and Research (INR) and Keynote Speaker Sue Gordon, the former Principal Deputy Director of National Intelligence, opened the conference. Both stressed the opportunity and dangers posed by new technologies in AI. They spoke about benefits like the amplification of the effectiveness of resource-constrained intelligence agencies and the automation of routine exercises. But they also noted important dangers such as an explosion of new attack vectors through Internet of Things (IoT) technology and the potential alteration of data by AI in a way that clouds our perception of issues. “Information is the new battleground, with threats both to and through this commodity,” Gordon stressed, and the US must create a new apparatus to address this. Her key takeaway was to see the world for what it is, not how one wishes it to be.
Key to this is understanding how nations choose to cooperate and compete in this sphere. Moderated by Fiona Cunningham of George Washington University, the first panel addressed exactly this theme. Featuring CSET’s Tarun Chhabra, the Department of Defense’s (DoD) Matt Daniels, the University of Pennsylvania’s Mike Horowitz, and former Principal Deputy Director of National Intelligence Stephanie O’Sullivan, the panel discussed how great power competition will interact with AI, what a competitive strategy in AI might look like, and how to measure power in AI. Chhabra and O’Sullivan both agreed that talent is crucial in undergirding successful AI development and that countries like China are poaching talent from the US. To address this shortfall, Daniels and Horowitz advocated for greater investments in basic research in AI but cautioned the US to avoid mimicking other countries’ policies directly, as spending does not always equal superiority. The panel called the United States’s cultural advantages, such as a robust innovation ecosystem and diversity, key drivers for AI competitiveness.
One such entity that is working on developing this advantage is the DoD’s Joint Artificial Intelligence Center (JAIC), led by its director, Lt. Gen. Jack Shanahan. Shanahan spoke with CSET’s director, Jason Matheny, in the conference’s Keynote Interview. The two detailed the JAIC’s achievements, its challenges thus far, and future problems it hopes to address. Lt. Gen. Shanahan described the JAIC’s origins, as it evolved from Project Maven to a Center founded in 2018 now focused on fielding AI capabilities quickly across the DoD. In starting Project Maven, DoD started a nationwide conversation about Silicon Valley’s relationship with the military. Shanahan called the JAIC “AI Now,” versus other DoD entities focusing on “AI Next,” and he illustrated this through highlight some of the Center’s achievements: predictive maintenance solutions for military equipment and programs for humanitarian assistance including flood damage assessments. Shanahan also explained the JAIC’s primary challenges, which are both technical and cultural. Shanahan says he constantly worries about how to find, clean, curate, label, and integrate data to train DoD’s AI models, calling it 80% of the challenge. Culturally, he stressed the importance of institutionalizing a startup culture that keeps things moving quickly as well as preventing process from dominating product. Going forward, Shanahan hopes to see the JAIC continue to address the rote, quotidian tasks that dominate DoD employees’ time, as well as more exciting fields like human-machine teaming.
While the JAIC is making some great progress, much of the talent lies in the private sector, which has had a rocky relationship with the DoD. The final panel addressed this topic, bringing government officials like former Defense Innovation Unit (DIU) head Raj Shah together with private sector and academic partners like Melissa Flagg of CSET and Jack Clark, the policy director at AI research company OpenAI. Former Secretary of the Navy Richard Danzig moderated, and the panel tackled thorny issues like Silicon Valley’s relationships with the DoD, challenges of working with government, and government AI accountability. Several of the panelists agreed that the firestorm that erupted over Google’s involvement in Project Maven was unfortunate, with Shah saying that the hesitance around working with the military is based on misunderstandings. Later, though, Clark stated that AI technology deserves this increased scrutiny given its enormous potential. He argued that Silicon Valley companies may be more comfortable working with government if it provided greater disclosure on how it planned to use AI. Flagg proposed this public outreach effort could make greater use of one of the DoD’s greatest assets, its 700,000-strong civilian workforce. In addition to bad PR, the DoD struggles to attract Silicon Valley partners due to simple financial matters, according to Shah: the DoD often spends less than $400 million a year on cutting edge software. If bigger deals were offered, then companies would be more interested in working with the government.