Have an AI optimize schedules and 'general utility'. Then build a second AI that explains why the previous outcome was best. If you have results that actually

Have an AI optimize schedules and 'general utility'. Then build a second AI that explains why the previous outcome was best. If you have results that actually are better scheduling than done by a human, with plausible explanations that most people affected agree with, you might be able to replace managers. What scares me, is that we would be training the second AI to convince us to believe it, rather than training it to tell the truth.