Preventive Regulation of Autonomous Weapon Systems

Autonomy can be defined as a machine’s ability to perform tasks in absence of humanity’s presence. To truly classify a system to possess autonomy, the following three dimensional approaches may be taken:

• Human-Machine
• Command-Control
• Relationship

This dimension deals with the humanity’s involvement in operating of systems with autonomy. These systems are divided into the following three types.

Semi-Autonomous System: Systems that require human inputs. Human-Supervised Systems: Systems which do not require inputs by humans to operate but supervision is provided to account for cases of malfunction or failure.

Fully Autonomous Systems: Systems which operate on their own without human contribution.

Sophistication In Decision Making

This approach deals with the ability to exercise command over its own operations. From this standpoint, we may divide these systems into three different forms: Automatic, automated and autonomous. Automatic refers to the provision of mechanical response to sensory inputs through predefined protocols. On the other hand, machines that are capable of accommodating for changes in their environment as well as a considerable level of self-governance may be called autonomous.

Types of Functions Being Made Autonomously

This dimensions states that properties of a system in reference to autonomy, depend on each function in particular. Therefore, functions like navigation could be designed to achieve autonomy without any ethical or strategic risks and yet achieving autonomy in targeting systems may be more concerning.

Autonomy deals with obtaining data and using it for various actions, the data may be obtained from the environment of the machine. In order to obtain autonomy integration of 3 fundamental capabilities is essential. The fundamental capabilities are: Sense, Decide, Act.

Sense - To attain autonomy, effective perception of the environment through a variety of sensors which it might possess, is essential. These sensors will obtain data from the environment. It must also use sensing software to interpret this data suitably. Target detection, for example, relies on pattern detection. Wherein the system deciphers patterns from a given data set, compares them to the predefined patterns stored in its computer memory and detects the target accordingly.

Decide - The data from the sensing software is input for decision making. The course of action may differ greatly depending on the system’s perception of its environment. For example, a Main Battle Tank’s target acquisition may detect aerial threats approaching it, once the tank commander authorizes retaliation, the tank’s autonomous anti-air weapon system may decide it’s approach depending on factors like wind speed, altitude and velocity of target, humidity and so on.

Act - Upon the completion of the decisions process, the system exerts its control in the real world by physical or computational means. The decision making models may be reactive or deliberative. The reactive model contains prescribed instructions on how the system must behave in the face of given input sets. For example, in a landmine, the prescribed rule may be that if weight from the range of 70-100 kgs is exerted on it, then it must detonate.

The deliberative model of decision making understands its environment in mathematical terms and is more complex than the reactive model. Deliberative models possess the capability of governing their own actions through the manipulation of data structures and they tend to measure the consequences of actions along with the extent to which they can achieve the desired goals. Taking the example of a beyond-visual-range air-to-air missile (let’s consider the USA developed AMRAAM or Advanced medium range air-to-air missile), it’s homing system will track a target and use various sensors to calculate the fastest and most efficient manner in which to approach and consequently have good effect on target. Thus, AMRAAM’s deliberative homing system perceived its environment so as to make the best decision without any human intervention whatsoever.

Underlying Technology At a fundamental level, autonomy is achieved through the implementing of following: Sensors, which allow the system to gather data from its environment. A suite of computer hardware/software, to interpret the date suitably for future actions. Interfaces for interaction with other machines, for better integration on battlefield. Actuators and end-effectors, to make it capable of exerting its decision on the real world.

The following are the existing limitations and obstacles as faced by a system when it perceives, decides or acts:

• Perception - Improvement perception is a major obstacle for AI today. While the sensory perceptions of machines have improved greatly, they only operate well in predictable environments. It is easy to fool machines with their current ability to perceive, hence can only be applied in limited battlefield scenarios.
• Decisions - Another limitation is that of synthetic reasoning. Machines operate on deductive reasoning while humans operate on inductive as well as adductive reasoning. Severely limiting a machine’s capability of making complex determinations suited to its perception of its environment.
• Actuation - Limitations in actuation are derived from the limitation in hardware technology, with power sources being a major hindrance. Many weapons are unable to operate, due to hardware limitations.