The Pentagon's chief digital and artificial intelligence officer, Craig Martell, recently expressed his concerns about the potential dangers of generative artificial intelligence systems, such as ChatGPT, which could be used to deceive and spread disinformation. Speaking at the DefCon hacker convention in August, Martell's talk on the technology garnered significant attention. However, he remains optimistic about the potential of reliable AI.

Having previously held positions as a data scientist at LinkedIn, Dropbox, and Lyft, Martell's current role involves leveraging the vast amount of data within the U.S. military and determining which AI systems can be trusted enough to be deployed in combat situations. This task is particularly challenging in a world marked by increasing instability, with multiple nations racing to develop lethal autonomous weapons.

In an interview, Martell explained that their primary objective is to enhance decision-making capabilities from the boardroom to the battlefield. Rather than focusing on specific missions, their aim is to develop the necessary tools, processes, infrastructure, and policies that enable the entire department to scale effectively. He emphasized the importance of network-centric warfare, which involves ensuring that the right data is available at the right time and place. Martell outlined a hierarchy of needs, with high-quality data being the foundation, followed by analytics and metrics, and AI at the top.

Regarding AI, Martell stated that it is essentially a tool that utilizes past data to predict future outcomes. He dismissed the notion that modern AI is significantly different from previous waves of technological advancements. He also disputed the comparison between AI and a Pandora's box, noting that AI is not a monolithic technology but rather a collection of technologies that can be applied on a case-by-case basis, with their effectiveness verified empirically.

In terms of their involvement with Ukraine, Martell clarified that their team's role is limited to assisting in the construction of a database that tracks how allies provide aid. This initiative, known as Skyblue, aims to ensure that the assistance remains organized. Martell emphasized the military's approach of training with technology until they develop justified confidence. He used the example of his car, where he trusts the adaptive cruise control but lacks confidence in the lane-changing technology.

Martell believes that computer vision has made significant progress over the past decade, but its usefulness in specific situations remains an empirical question. He stressed the importance of determining the level of precision acceptable for each use case and building AI systems accordingly. Martell expressed the need to avoid discussing AI as a monolithic entity and instead focus on the desired capabilities.

When asked about the use of large-language models in the Department of Defense, Martell expressed skepticism due to their lack of constraints in telling the truth. However, he mentioned that Task Force Lima, initiated in August, is currently studying over 160 use cases to identify low-risk and safe applications for such models. He suggested that generating first drafts in writing or computer code, where humans would edit or compile the output, could be potential applications. Information retrieval, where facts can be validated for accuracy, is another possibility.

The interview shifted to the challenges of hiring and retaining AI talent within the Pentagon. Martell acknowledged that the department has established a digital talent management office and is exploring innovative strategies for recruitment. He questioned the necessity of hiring individuals seeking long-term careers with the Department of Defense and proposed alternative options, such as offering college education in exchange for three to four years of service before transitioning to positions in Silicon Valley. Martell also considered the potential for recruiting from historically Black colleges and universities as part of a diversity pipeline.

Overall, Martell's insights shed light on the Pentagon's mission to leverage AI for global information dominance. While acknowledging the urgency brought about by the perceived China threat and the ongoing AI arms race, he emphasized the need for careful evaluation and testing of AI systems. Furthermore, he stressed the importance of addressing the challenges associated with hiring and retaining AI talent in order to effectively harness the potential of this technology.