This is a super interesting angle on AI and robotics, imho, and I think it actually points the way toward maximizing efficiency, not just usability. The key point is about interpretability; i.e., knowing what a system is doing, and why, is a huge part of using it efficiently, since these systems are never perfect and having some sense of why results might not be as you expected is perhaps the most significant factor in getting them to do what you want. It's the same with human collaboration, tbh.