A military robot’s autonomy and responsibility are of concern due to the substantial innovation introduced in real operations of devices that show some autonomy. Here, the authors argue that many of the positions so far taken on whether to assign responsibility to a robot should be reformulated by considering the autonomy of the robot not as a black box, but instead looking inside it and considering autonomy in terms of the process the robot has to accomplish. So the simple question of whether autonomous robots will have to be considered responsible for their actions is not correctly stated.

One part of the paper is mainly devoted to illustrating what it means to open the box; the other part concerns a discussion about negotiating activities with humans, with a reference to both examples and documents published by the Department of Defense (DoD).

According to the authors, taking for granted that intelligent and autonomous robots will be able to operate without human intervention and in an unpredictable way in the near future ignores the present state of the art in robot autonomy. Autonomous capabilities effective today, such as trajectory computation and obstacle avoidance, even when obtained through probabilistic methods or machine learning, are in large measure possible under human supervision in a schema that resembles what is in use for automatic systems.

Moreover, legal and moral responsibilities are different, and according to the authors, moral responsibilities are out of the context of robotics since they are defined according to international criteria. What about legal ones? The authors show how different concepts of machine autonomy interact with the idea of robot responsibility, and conclude that more machine autonomy does not mean less human control. The responsibility hierarchy established in the Army can be considered not altered by robots. The robots will be built in a process of negotiating their autonomy with the goals to accomplish. The need for effective ways to develop and carefully test robots is a consequence.

Here, the authors address a point that has been under discussion in literature about how to measure robot autonomy and reliability, and for which there are no final solutions.

The paper is well discussed and motivated; however, it is not an easy read for people without technical knowledge of robotics. The authors’ conclusions can indicate a direction to take, but most of the technical work for defining robot performance and reliability still needs to be done.