Artificial Intelligence
Intelligent agents, state-space search, constraint solving, optimization, adversarial reasoning, symbolic knowledge, uncertainty, Bayesian and fuzzy systems, generative models, and responsible AI.
Artificial intelligence is not a single algorithm and it is not synonymous with machine learning. It is a collection of ways to represent a situation, reason about possible actions, learn or infer what is not directly known, and choose an action under computational and real-world constraints. A route planner, a timetable generator, a game-playing program, a diagnostic system, and a language model appear very different on the surface, yet each can be studied through the same questions: What information describes the current state? What outcomes are desirable? What uncertainty remains? Which computation is affordable? How will we know whether the result is reliable?
This course develops those questions progressively. It begins with rational agents, then treats search as controlled exploration: first without guidance, then with heuristics. Constraint satisfaction reframes search around variables and legal assignments, while local and evolutionary methods trade guarantees for scalability. Adversarial search introduces opponents whose goals conflict with ours. Symbolic logic makes facts, relations, and inference explicit. Probability and Bayesian networks provide a language for uncertain evidence before Markov models extend that reasoning across time; fuzzy systems provide a different language for vague concepts and graded rules. The final chapters connect these foundations to generative AI and to the engineering responsibilities that arise when AI affects people.
The emphasis throughout is on understanding before implementation. Every algorithm is tied to its assumptions, invariants, failure modes, and measurable costs. Code is used to make an idea precise, not to hide the idea behind a library call.
Outcomes
- Formulate real problems as agent, search, optimization, constraint, game, and probabilistic models
- Implement and evaluate uninformed, informed, local, evolutionary, and adversarial search algorithms in Python
- Explain completeness, optimality, computational complexity, and the assumptions behind an AI technique
- Represent facts and rules with propositional and first-order logic and evaluate the limits of symbolic inference
- Represent uncertain and sequential knowledge with probability, Markov models, Bayesian networks, and fuzzy inference systems
- Explain how modern generative models produce outputs and evaluate their limitations systematically
- Design AI systems with explicit attention to safety, fairness, privacy, transparency, human oversight, and environmental cost
Outline
Start →- 01Intelligent AgentsWhat intelligence means operationally, how rational agents connect perception to action, and how environment assumptions determine an AI system's design.
- 02Uninformed SearchState spaces, search trees, BFS, DFS, depth limits, iterative deepening, and uniform-cost search — including guarantees, complexity, and implementation.
- 03Informed SearchGreedy best-first search, A*, admissibility, consistency, heuristic construction, and the practical trade-off between guidance and guarantees.
- 04Constraint SolvingModel and solve CSPs using graph coloring, N-Queens, backtracking, MRV, LCV, forward checking, arc consistency, and structural decomposition.
- 05Local SearchHill climbing, min-conflicts, simulated annealing, restarts, and experimental methods for optimization spaces where paths do not matter.
- 06Genetic AlgorithmsPopulation-based optimization through representation, fitness, selection, crossover, mutation, elitism, diversity, and reproducible evaluation.
- 07Adversarial SearchMinimax reasoning, depth-limited evaluation, alpha-beta pruning, move ordering, and the assumptions required to make decisions against an opponent.
- 08Logical AgentsPropositional and first-order logic, entailment, inference, unification, rule systems, symbolic planning, and the limits of representing a changing world with explicit facts.
- 09Knowledge RepresentationOpen and closed worlds, ontologies, knowledge graphs, action models, symbolic planning, provenance, and the limits of explicit knowledge.
- 10ProbabilityProbability, conditional evidence, Bayes' rule, independence, and expected utility for agents that must reason with incomplete or noisy information.
- 11Bayesian NetworksDirected graphical models, conditional independence, factorization, exact inference, sampling, parameter learning, and model validation.
- 12Markov ModelsMarkov chains, hidden states, sensor models, filtering, prediction, smoothing, and Viterbi decoding for uncertain sequences.
- 13Fuzzy LogicFuzzy sets, membership functions, linguistic variables, rule evaluation, aggregation, defuzzification, and controller design.
- 14Generative AIGenerative modelling, tokens, transformers, language-model training, decoding, prompting, retrieval, evaluation, and grounded application design.
- 15Responsible AIFairness, privacy, transparency, explainability, accountability, safety, alignment, monitoring, and governance across the AI lifecycle.