Modern therapy for type 1 diabetes requires patients to interact with intelligent systems designed for automated insulin delivery (AID). Despite substantial medical advantages, patients often experience low trust in and predictability of AID systems. To mitigate adverse effects on human-AI cooperation, AID simulations may be used in training of patients. The goal of the present research is to identify requirements for human-centered AID simulation. To this end, we conducted an interview study (N = 12) in which stakeholders were queried on their expectations regarding AID systems, mainly addressing interactivity and explanation of simulation parameters. In addition, a prototypical simulation was used in a within-subject online experiment with N = 32 participants to evaluate effects on perceived trustworthiness and traceability. While the latter improved significantly, no difference in perceived trustworthiness was found. Our results indicate that interactive simulations can contribute to improving human cooperation with AID systems in a safe environment.