How studying Amazon warehouses can improve battle


What humans may do with robots in war looms over the future, a buzzing cloud of nanobots that animates dystopian fiction and dark fantasy.



A U.S. Marine uses tablet to direct machine gun fire during a platoon-supported attack range at Giskas, Norway, Aug. 7, 2018. In the future, an AI battle manager may use the tablet to direct the Marine. (Gloria Lepko /USMC)



Another possibility to that future, however, is one in which artificial intelligence is instead tasked with managing the particular deployments of humans in battle. Think algorithmically derived stratagems moving humans around in novel and unexpected ways.


One way that future might manifest is by looking at a place where AI already manages humans: an Amazon warehouse.


Andrew A. Hill, a professor at the U.S. Army War College, said Amazon’s management of its warehouses is an example of how AI can prepare more broadly for conflict.


“I don’t know if you’re aware of how Amazon stocks its warehouses, but they use this system called “random stow,” Hill said. Random stow is a process in which items are unloaded wherever there is space in a warehouse and then scanned into a computer system than can track where the item is located. When it comes time to retrieve an item for delivery, the same computer system directs warehouse workers to the most efficient route for finding the item, which could be stowed throughout the warehouse. Hill spoke as part of a panel at the Army Autonomy and Artificial Intelligence Symposium & Exposition in Detroit Nov. 28.


“That environment is inhuman. It is chaos to us,” Hill said. “But it works because it is managed by a different kind of intelligence.”


Hill was speaking only of the way the warehouse inventory is organized, though it’s worth noting that workers at Amazon warehouses say they’re under intense pressure to keep up a grueling pace and are reported to relieve themselves in bottles, with breaks tightly controlled and bathroom facilities few and are far between. But with that commercial warehouse in mind, it’s worth considering how an AI, given the same objectives as a human commander, might organize and direct forces to achieve them.


“Why would an AI allocate forces in distinct areas of the battlefield? It could intermingle them and manage them at a granular level. Its categories are way more numerous, in the way that an Amazon warehouse AI manages categories at the shelf level,” Hill said. Instead of distinct groupings of armor, air support, infantry, and artillery, a system run by artificial intelligence and managing a battle could coordinate a single helicopter with a pair of howitzers and an infantry platoon, directly grouping each in the same ad-hoc way that a warehouse worker finds an assortment of items to place into the same package.


The end result may be the same: a particular tank destroyed and a building cleared of armed defenders, but the path to that objective through a artificially intelligent battle manager could feel almost alien today.


“Just from a risk management standpoint, I have no idea what AI is going to think about operational risk,” Hill said. “What is an AI willing to spend to achieve its objective? The idea of economy of force might mean something very different.”


These are problems for humans to tackle now, while the technology is in its infancy and well before AIs are enlisted to suggest deployments in battle. If one strand of AI on the battlefield is making war as comprehensible as strategy games, another is modeling the way AI will change war in games and exercises to try and anticipate the ways it will change combat.


“These are practical questions for the simulation and gaming community that cannot be answered right now where we are with computation,” Hill said.