For people in the industry, how do you explain the poor interpretability of some ML techniques to bosses who are not data scientists?

I'll be having a job interview in a few days and I think this question might come up and I personally don't know how to explain it to a layperson. Black box methods may come in handy one day, but I realized just now that I can't briefly explain how it works without making it sound like magic. What's your workaround for this? Have you been in a situation where you presented your results and you had to explain how …

black box bosses box data datascience data scientists industry interpretability interview job people scientists think