2.1. Core AI Concepts and Terminology

💡 First Principle: Shared vocabulary is a control mechanism. When a business team, a technical team, a vendor, and an auditor use "model," "training," or "AI system" to mean different things, decisions get made on false assumptions and nobody notices until something breaks. Precise terms let you ask precise questions: "When was it trained? On what data? How often is it retrained? What happens at inference time with our data?"

Why should a business strategist care about terms that sound technical? Because each term carries a business consequence. "Training" implies a cost and a data dependency. "Inference" implies a per-use cost and a latency. "Unstructured data" implies that generative AI might unlock value that was previously trapped in documents and emails. "ISO/IEC 42001" implies an auditable management system that customers and regulators may ask about.

The mental model for this section is a professional kitchen. The recipe development phase (training) is expensive, happens occasionally, and depends on the quality of the ingredients tested (data). Service (inference) happens every time a customer orders and costs a little each time. The dish categories (AI, ML, generative AI) tell you what kind of kitchen you need. And the health code (international standards) gives inspectors, suppliers, and customers a shared language for what "safe" and "well run" mean.

⚠️ Common Misconception: Learners often treat AI, ML, and generative AI as synonyms, assume more data always beats better data, believe deployed models keep learning automatically, and confuse management-system standards with technical standards. Each misconception leads to a specific bad decision, covered in the subsections below.