Tackling the problem of “bad” explanations with the Human-in-the-Loop principle