As AI systems become more prevelant, particularly if they are used for decision support, alarm management, advanced process control, or autonomous optimization, they cannot rely on vague language or moral intuition. They require explicit objective functions, weights, thresholds, and constraints. If we ask an AI system to balance safety, production, availability, and cost, it must encode the trade-offs that we previouly kept implicit.
