- Learn
- Artificial Intelligence
- Reasoning Under Uncertainty
AI-M03
Reasoning Under Uncertainty
How an AI reasons when it cannot be sure: probabilities over many variables, Bayesian networks, hidden Markov models, and choosing actions by expected utility.
Logic gives definite answers, but the real world rarely offers definite facts. Sensors are noisy, symptoms are ambiguous, and the same evidence can have several explanations. This module gives the agent a language for degrees of belief: probability, organised so that it scales.
You will see why a full table of probabilities over many variables is impossible to store, then use Bayesian networks to represent the same knowledge compactly and update beliefs as evidence arrives, watching the classic wet-grass example explain itself. Hidden Markov models add time, with the Viterbi algorithm finding the most likely hidden story behind a sequence of observations. Finally decision theory turns beliefs into choices with expected utility and the value of information, and Markov decision processes open the door to reinforcement learning.
Lessons
- AI-03.01 Probability for AI (recap) Real-world AI must reason with uncertainty, so probability is its language. Beginner 8 min
- AI-03.02 Bayesian networks A Bayesian network is a directed graph where each node is a variable and edges show direct influence, with a table of conditional probabilities per node. Intermediate 15 min
- AI-03.03 Hidden Markov models A hidden Markov model describes a system with hidden states that change over time and produce observable outputs, such as spoken sounds produced by words. Intermediate 12 min
- AI-03.04 Decision theory and utility Decision theory combines probability with preferences: assign each outcome a utility and choose the action with the highest expected utility. Intermediate 10 min
- AI-03.05 Sequential decisions: MDPs (bridge to RL) A Markov decision process models sequential decisions where actions have uncertain outcomes and rewards accumulate over time. Intermediate 8 min