1.2. How Modern AI Learns
💡 First Principle: Modern AI learns by finding statistical patterns in data and using them to make predictions about new, unseen inputs. Everything else — accuracy, bias, hallucination, the difference between "predictive" and "generative" AI — flows from this one fact: the system is pattern-matching against what it has seen, not consulting a store of verified truths.
Why care? Because nearly every responsible-AI concern and every limitation the exam tests on traces back to this mechanism. A model that learned from biased data will make biased predictions. A generative model that predicts likely-sounding text can sound confident while being wrong. Once you internalize "it's predicting from patterns," these behaviors stop being surprising and become predictable.
⚠️ Common Misconception: "More data always means a better model." Volume helps only if the data is relevant and representative. A million examples scraped from one narrow population will produce a model that fails on everyone else. Quality and representativeness beat raw quantity — a point Phase 2's fairness principle builds on directly.
