This past February, President Trump signed “The American AI Initiative”, an executive order designed to stimulate the national development and regulation of artificial intelligence (AI). Under this order, federal agencies will need to launch or expand AI initiatives that match their objectives and missions. Through this order, the United States (US) joins other countries with national AI strategies, scrambling to become leaders in AI in various domains. Perhaps no other domain has attracted as much attention and investment as the military. Driven namely by Russia, China, and the US, there has been an explosion of investment in military technologies in the past five years, and many are now talking of an AI arms race. Amidst scares of the rise of ‘killer robots’, let’s review the recent AI developments in the American military.

Indeed, the American military has been particularly active in AI research and development as it hopes to gain a strategic lead before others. The Department of Defense (DoD) released its 2019 AI strategy in line with the executive order, although it has been focusing on AI investment and development for longer.

In June 2018, the DoD created the Joint Artificial Intelligence Center (JAIC). The JAIC oversees and coordinates all AI efforts from agencies comprising the DoD; it seeks to establish a common set of standards, shared data, and tools across these agencies and services. The DoD also announced in 2018 that it planned to invest up to $2 billion over the next 5 years to integrate more AI in US weapons systems.

Perhaps the most famous of military AI projects is Project Maven. Established in 2017, it aims at developing AI capable of autonomously analysing the huge amount of video footage collected by US drones and assisting in the recognition of objects. Rather than having analysts sitting behind a computer and tagging individual objects, the algorithms suggest what those objects might be. The pilot project began with AI sifting through the massive amount of data, detecting objects of interest, and flagging these for human review, thereby allowing the algorithm to keep ‘learning’. The US strategy towards AI is focused on this collaboration between humans and machines, also known as ‘teaming’, and it has reportedly become a central element of how the military plans to integrate AI into weapon systems. Project Maven has been highly successful, and has already been used in the Middle East and Africa where it is helping military analysts sort through the huge amount of data collected by sensors and drones.

But this is far from being akin to killer robots. In the US, the focus on teaming takes precedence over the development of fully autonomous technologies. The main emphasis is thus placed on developing AI able to assist with the analysis of data by finding actionable patterns and detecting unusual objects, thereby freeing analysts to perform more high-level taskssuch as making sense of the data, and acting on it. This could have a potentially transformative role not only in weapon systems, but also in intelligence analysis. Robert Cardillo, former head of the National Geospatial-Intelligence Agency (NGA), an intelligence agency under the DoD, argued that up to 75% of what his analysts did could be performed by AI. The NGA has awarded research contracts to 6 companies to develop algorithms that could help with the analysis of the huge volumes of imagery that has become too much for NGA analysts to handle. For Cardillo, implementing AI is about freeing up analysts’ time so that they can do the “harder things”.

Overall, the DoD’s actions highlight the primordial role that the US sees AI playing in national security and defense matters. Becoming leaders in this field and gaining military strategic advantage is key to the DoD’s AI approach and now shapes much of their projects and vision. However, many analysts still question the exact role that AI will come to play in warfare, and as the technology progresses, one cannot discount the potential of the US, or other countries, developing weapons with autonomous decision-making.