Toward meaningful human control of autonomous weapons systems through function allocation

One of the few convergent themes during the first two United Nations Meeting of Experts on autonomous weapons systems (AWS) was the requirement that there be meaningful human control (MHC) of AWS. What exactly constitutes MHC, however, is still ill-defined. While multiple sets of definitions and analyses have been published and discussed, this work seeks to address two key issues with the current definitions: (1) they are inconsistent in what authorities and responsibilities of human and automated agents need to be regulated, and (2) they lack the specificity that would be required for designers to systemically integrate these restrictions into AWS designs. Given that MHC centers on the interaction of human and autonomous agents, we leverage the models and metrics of function allocation - the allocation of work between human and autonomous agents - to analyze and compare definitions of MHC and the definitions of AWS proposed by the U.S. Department of Defense. Specifically, we transform the definitions into function allocation form to model and compare the definitions, and then show how a mismatch between authority and responsibility in an exemplar military scenario can still plague the human-AWS interactions. In summary, this paper provides a starting point for future research to investigate the application of function allocation to the questions of MHC and more generally, the development of rules and standards for incorporating AWS into the law of armed conflict.