Legally-Salient Variables for AWS

Why autonomous weapons (AWS) are problematic from a targeting perspective is controversial and inconsistent. This chapter argues that mono-dimensional approaches to this question are reductive: instead, variables such as level of autonomy, duration of search, etc., are only a factor that may influence whether an AWS can be used lawfully. It also hypotheses that some variables may compensate for each other, e.g. an AWS’s inability to distinguish trams and tanks can be mitigated by limiting its search area to a space where trams are not present. From these hypotheses, a robust typology of legally-salient variables is constructed related to the system’s properties itself (e.g. performance, opacity), those of the operational environment (e.g. density of negative class entities), and operational parameters (e.g. spatial and temporal limitations, the types of target). Additionally, these variables are linked to Context Controls—actions AWS-users can take to mitigate risk of violating IHL—such as enabling dynamic munition selection or issuing specific instructions to the civilian population to avoid being incorrectly targeted. Finally, the hypothesis that variables can ‘compensate’ for each other is validated through the finding of three overarching Rationales that underpin targeting rules: the preference for minimising incidental harm, minimising false positives, and minimising the risk of epistemic error. It is also shown that in order for variables to compensate for each other, they must be linked to the same Rationales. The theories advanced in this chapter constitute the foundations for the legal analysis in Chap.  9 on interpreting targeting obligations.