Preprint: Avoiding Black Box Problems by Assigning an Active Role to Humans in the Control of Autonomous AI: A Methodological Approach

Abstract

Autonomous AI may shift humans into a passive supervisory role while responsibility remains with them. This paper proposes interpretable primitives as a work-design approach to support meaningful human control. Based on empirical work analysis and Hierarchical Task Analysis (HTA), a method is presented for deriving and evaluating primitive candidates using three criteria: clarity of purpose, process transparency, and granularity. The approach is illustrated using railway disruption management and represents a methodological contribution that will be evaluated in future work.