Trace a voice assistant, robot vacuum, or recommendation feed as an agent in an environment: it observes, builds an internal state, chooses an output, and gets consequences. Separate the whole AI system from the model inside it, so “intelligence” means effective behavior on a task, not human-like consciousness.

Identify what an AI system can and cannot sense in a situation, then name the internal representation it keeps about the world. Practice spotting the gap between the real world and the simplified state, features, embeddings, or memory the system actually uses.

Compare systems that predict a label, score, ranking, next word, or future event with systems that reason through rules, constraints, search, or plans. Use examples like spam filters, route planners, game AI, and chatbots to decide when an answer is pattern-based prediction, step-by-step reasoning, or both.

Apply the previous explanations in a guided problem.

Recognize when an AI output becomes an action: recommending a video, steering a car, placing a trade, generating code, or calling a tool. Reason about goals, constraints, autonomy, and human approval so you can see how the same prediction can lead to very different real-world effects.

Trace how feedback changes an AI system over time, from labeled corrections and user clicks to rewards, ratings, and human preference feedback such as RLHF. Distinguish training from using a finished model, and recognize feedback loops where the system’s own actions change the data it later learns from.

Apply the previous explanations in a guided problem.

Check your understanding with a short quiz.

Review this chapter with practice based on your mistakes.