Free beta access. 50 free AI tokens each week. Optional top-ups available. See what is included
Opening Word document…

applications of AI in logistics and sustainability is distinct from militarized autonomous weapons

Konaev and Chahal 2021 – research fellow and research analys with the Center for Security and Emerging Technology

Margarita Konaev Husanjot Chahal April 2021 “The Path of Least Resistance: Multinational Collaboration on AI for Military Logistics and Sustainment” CSET Issue Brief

While we discuss how the United States and its allies can work together on AI for military logistics and sustainment in both combat and noncombat settings, there is no doubt that the environment in question matters a great deal. From data to computational power to available talent, as well as considerations like privacy, safety, and security, implementing AI for military logistics and sustainment functions performed in controlled environments similar to commercial settings is a different endeavor from deploying AI- enabled logistics and sustainment functions in contested and hostile environments. We take these differences into account where relevant, and acknowledge that even under the best of circumstances, there are still significant challenges for both the adoption of AI applications and multinational collaboration in this area.

Technologically attainable

While not without its challenges, military logistics and sustainment tasks, especially those performed in noncombat settings, present a technologically attainable area for multinational collaboration in AI.

Although much of the innovation in AI is occurring in the commercial sector, adopting and adapting commercial AI applications for military purposes is often impossible. Current AI technologies, and especially ML-based systems, tend to perform well in stable environments but struggle with uncertain and novel situations, and remain particularly vulnerable to adversarial attacks.

These vulnerabilities present an unacceptable level of risk in high- stakes military settings, where the environment is uncertain and adversarial by definition. The consequences of mistakes and even system failure, however, are less severe when it comes to some military logistics and sustainment tasks which are administered and managed in noncombat settings, and constitute what some have called enterprise AI applications.

Advances in AI for logistics in commercial aviation, maritime shipping, and transportation sectors are therefore more applicable to certain military logistics and sustainment tasks performed in noncombat settings than for specialized military equipment like autonomous ground combat vehicles or armed drones. In particular, there may be opportunities to adopt and adapt commercial applications for the intelligent automation of tasks such as scheduling equipment maintenance and repairs, updating and issuing licenses, supply tracking and forecasting, and other processes that control the flow of logistics throughout the military organization.34 To reiterate, these are much more than cost cutting and efficiency increasing measures; improvements in these areas enable military readiness and effectiveness in combat.35

In addition to these opportunities to leverage AI-enabled technologies and tools available in the commercial sector in support of military logistics, there are also fewer barriers to in- house innovation within defense organizations. Many of the AI applications relevant to logistics and sustainment can be developed and used in relatively well-controlled and benign environments in settings akin to commercial civilian enterprises. Under such conditions, resources like data and infrastructure, including storage, ETL pipelines, communication bandwidth, and compute can be made available to train ML models for various AI applications.36 Notably, the 2016 Defense Science Board Summer Study on Autonomy raised a similar point regarding logistics planning and execution as “a particularly good candidate for testing and experimentation (T&E) ... because the behavior of logistic software can be evaluated against crisply known metrics.”37

Sign in to recut, update, or save this card.