Artificial Intelligence
Undergraduate AI foundations — agents, search, games, knowledge, uncertainty and ethics — with concept notes on ExamMaster.
Undergraduate depth: concept notes and, where they help, in-browser labs. Reading is free.
Lessons
- Agents and EnvironmentsIn AI because every system — thermostat to tool-calling model — is an agent coupled to an environment, and the coupling decides the machinery.
- Cost-Sensitive DecisionsIn AI because a classifier's output is a decision with a price, and choosing the threshold is choosing who pays it.
- Entailment and InferenceIn AI because entailment is the gold-standard guarantee — what a conclusion means when it cannot be wrong — and the contrast with a model's score runs through every deployed…
- Bias and Fairness in ModelsIn AI because slice metrics and disparate-impact screening are how deployed models are actually audited for uneven treatment.
- Evaluating AI SystemsIn AI because the evaluation, not the model, is what decides whether an intelligent system may ship — and designing one is a skill of its own.
- Expert SystemsIn AI because the classical expert system is a named architecture — knowledge base, inference engine, working memory — and the hybrid lesson already owns what happens when that…
- Alpha-Beta Pruning and EvaluationIn AI because a full minimax tree is usually too big to finish — alpha-beta skips branches that cannot change the decision, and an evaluation function stands in when you must cut…
- Game Playing and MinimaxIn AI because a second agent who wants the opposite of you turns search into a game tree — and minimax is the rule that picks a move which is safe against that opponent.
- Tic-Tac-Toe with MinimaxIn AI because a playable tic-tac-toe program is the Game Playing and Minimax backup implemented on a 3×3 board — an agent that returns a move, not a lecture that names MAX and MIN.
- Designing and Judging HeuristicsIn AI because heuristics are engineered artefacts — derived from relaxations, ranked by dominance, and audited like any other code an optimal planner depends on.
- Semantic Nets, Frames and InheritanceIn AI because a lot of authored knowledge is 'this kind of thing has these slots, and that kind inherits them' — a graph of concepts, not a pile of propositions.
- Propositional LogicIn AI because propositional logic is the simplest language in which 'the machine concluded X' has an exact meaning — and the constraint layer of modern systems still speaks it.
- Knowledge Engineering and Hybrid SystemsIn AI because real systems mix learned perception with authored constraints — from expert systems to LLMs gated by validators — and someone has to own each layer's failures.
- Leakage, Drift and Model OperationsIn AI because a score you can trust is manufactured twice: by a training pipeline that cannot see the future, and by production monitoring that notices when the world moves.
- Explanation-Based LearningIn AI because a general rule can be extracted from one explained example plus a domain theory — a learning family the inductive ML course does not own.
- First-Order Logic and UnificationIn AI because 'every student who borrowed this book is a person' is one sentence, not one atom per student — and first-order logic is the language that sentence lives in.
- Forward, Backward and ResolutionIn AI because a first-order sentence is only useful if a machine can derive what follows — and the three classical procedures, forward chaining, backward chaining and resolution,…
- Naive BayesIn AI because the first learned classifier you can derive end to end is Bayes rule plus one independence assumption — and it still ships in production text pipelines.
- Proxies and PrivacyIn AI because what a dataset reveals is a property of correlations and joins, not of which columns were deleted — the working knowledge behind feature review and privacy…
- Probability CalibrationIn AI because a score only earns the name probability if the world agrees with it at that rate — decisions priced on miscalibrated scores are silently wrong.
- Rules, Learning and Hybrid SystemsIn AI because choosing between rules, search and learning — and layering them — is the first engineering decision of every intelligent system.
- AND-OR Graphs and AO*In AI because some goals decompose into parts that must all be solved — and A* cannot see an AND — so the search has to return a solution graph, not a path.
- A* SearchIn AI because A* is the workhorse of optimal planning — the guided-search algorithm robotics stacks, game engines and interview questions all reach for.
- Cost-Aware SearchIn AI because cost-aware search — uniform-cost, which is Dijkstra on an implicit graph — computes provably optimal plans whenever the model is known, no learning needed.
- Depth-First Search and Iterative DeepeningIn AI because depth-first exploration — and the iterative-deepening repair of its weaknesses — is the memory-light workhorse behind constraint solvers, planners and game-tree…
- Graph Colouring as SearchIn AI because colouring adjacent vertices differently is a constraint-satisfaction search you can implement — not the NP chromatic-number decision from algorithms.
- Hill ClimbingIn AI because many real decisions are 'improve the current state' rather than 'reconstruct a cheapest path' — and hill climbing is the algorithm that idea becomes.
- Heuristics and Their GuaranteesIn AI because every promise informed search makes — including A*'s optimality, next lesson — is purchased by the two properties of the heuristic defined here.
- Negative Costs and ReopeningIn AI because negative costs mark the exact boundary of greedy search: knowing why settle-on-pop fails, and what reopening or Bellman–Ford pay to fix it, separates using Dijkstra…
- Designing the State SpaceIn AI because the state encoding, not the search algorithm, is what makes real planning tractable — state design is where planners are actually engineered.
- Sudoku by BacktrackingIn AI because a Sudoku grid is a CSP whose search is the same choose–try–undo loop as graph colouring, pointed at cells, rows, columns and boxes.
- Search Problems and Breadth-First SearchIn AI because posing a task as states, moves and a goal test — then exploring systematically — is classical intelligent behaviour that needs a model of the moves, not data or…
- The Water-Jug ProblemIn AI because a pair of jugs with fill, empty and pour moves is a state space you can implement — the paper's war-jug experiment, and the running instance of uninformed search…
- Modern AI ToolsIn AI because a campus project now reaches for named tool families — a chat model, a model hub, an image generator, a code completer, an AutoML search — and each family is a…
- Uncertainty and BayesIn AI because reasoning under uncertainty (Bayes) is a core AI pillar that later feeds ML classifiers.
- Dempster-Shafer BeliefIn AI because some evidence supports a set of answers without splitting the credit among them — and Dempster-Shafer is the calculus that keeps that uncommitted mass visible.
- What is AI?In AI because you need the vocabulary that separates search, learning, and deep nets before any model code.
- When Not to Use LearningIn AI because knowing when a rule or a search beats a fitted model is itself an engineering skill — learning is a cost, not a default.
Practise Artificial Intelligence
Reading every chapter below is free and needs no account. Practice, mocks and progress live in the app.
Continue with Google — freeNo card, no trial. Works offline once installed.