An AI system can help a company review alerts about its business partners by marking some of them as "probably not important". Should the AI ever be allowed to close such an alert on its own, or should a person always confirm the decision, even if this takes much longer?