This paper is aimed at helping defense professionals think clearly and objectively about possible risks associated with autonomous weapons. Autonomous weapons generally do not exist, and their military costs and benefits can be speculated but are not yet clearly known. What is clear, however, is that they raise novel questions of risk. The essence of autonomy is delegating a task previously done by a person to a machine. This raises the important question of how to retain effective human control over the machine’s behavior and the risks—both the probability and consequences— associated with a loss of control.
In doing so, this paper examines the risks of autonomous weapons relative to semi-autonomous weapons that would retain a human “in the loop” for selecting and engaging specific targets. Autonomous weapons cannot be viewed in a vacuum. War is dangerous, and weapons that are intended to be deadly to an opponent often can be quite dangerous to the user or friendly forces as well. This paper therefore aims to answer the question: How is the risk of using autonomous weapons different from other weapons?