The development and proliferation of autonomous weapon systems (AWS) has been discussed from a variety of perspectives and disciplines. While literature exists on the ethical, legal, and security aspects of AWS deployment, the conceptual dimensions and implications of the issues of AWS autonomy and predictability have yet to be defined and analyzed. In addition, although there is a growing interest in the AI ethics, little or no attention is dedicated to the question of whether the principles that constitute the safe and fair use of AI inherently contain ethical preferences or whether they are completely value-free and neutral.
To eliminate the conceptual ambiguities surrounding the topics of AWS autonomy, predictability, and ethical neutrality, we first define what autonomy means and distinguish it from "automaticity." According to the definition, autonomy is the ability of an AI system to act independently and without human intervention from the beginning of deployment and activation.
With the clear view on the meaning of autonomy, It is easier to discuss how the principle of predictability works in the context of the AWS use. We propose that the degree of predictability and explainability depends on the three major variables- the system itself, the environment AWS operates in and the task AWS performs. Instead of outright prohibition of all forms of autonomous weapons, we advise policymakers and regulators to focus on the dynamics of these variables and then decide on reasonable restrictions.
Although it is relatively clear how the principle of predictability can be applied in practice, the essence of this principle is far from being explored. We believe that, unlike other principles such as fairness, accountability, or safety, the principle of predictability does not contain ethical preferences of its own. For this reason, the ethical neutrality inherent in the principle of predictability can be abused. Therefore, it is important that any principles governing the use of AI be based on universal ethical norms. We conclude that without predictability and explainability, it is extremely difficult to achieve safe, fair, and ethical use of AWS. And without a strong ethical framework and foundation, it is impossible for predictability and explainability to help create a safe, fair, and ethical AWS.