1.2.2. From Prediction to Generation

💡 First Principle: Generative AI is prediction turned into creation. A generative model predicts the most likely next piece of content — the next word, the next pixel — one step at a time, and by chaining those predictions it produces whole sentences, images, or audio. It isn't retrieving stored answers; it's generating plausible content based on learned patterns.

This reframes the difference between two families of AI. Predictive (or discriminative) AI answers questions about existing data: is this email spam, what's the sentiment of this review, which category does this image belong to. Generative AI produces new data: a written summary, an answer to a question, a generated image. The AI-901 leans heavily on generative AI because that's what Microsoft Foundry is built to deploy, so understanding this prediction-as-generation idea sets up the entire implementation half of the exam.

⚠️ Exam Trap: "Generative AI knows facts like a database." It doesn't. Because it predicts likely-sounding content rather than looking up verified records, it can produce fluent, confident text that is simply wrong — often called a hallucination. This is exactly why responsible-AI practices like grounding and human oversight (Phase 2) matter.

Reflection Question: Why can a generative model produce an answer that sounds completely authoritative but is factually false? Connect your answer to how it generates content.