Artificial Intelligence
Introduction
What Question Does It Ask?
Artificial intelligence studies how to design intelligent agents that perceive their environment and choose the action that best serves their goal. The course brings search, logical inference, probabilistic reasoning, planning and learning into one framework. Machine learning is part of that framework, not all of it.
At a glance
search, logic, probability and learning combined
problem definition: states, actions, goal
action sequence, plan or decision
reaching the goal, cost, rationality
medium: logic, probability, graphs
little; often a model and knowledge suffice
Introduction
A Short History
Turing proposes the imitation game (the Turing test) in "Computing Machinery and Intelligence".
Dartmouth summer workshop: the name "artificial intelligence" is coined and the field is born.
Expert systems (MYCIN, XCON) go commercial; unmet expectations bring an "AI winter".
Deep Blue beats Kasparov at chess: powerful search and a hand-written evaluation.
AlphaGo beats Lee Sedol at Go: tree search, deep networks and reinforcement learning together.
Large language models reach a mass audience through chat interfaces.
The Course
Core Concepts
Perceives the environment through sensors, acts on it through actuators.
Choosing the action expected to maximize the performance measure.
BFS, DFS, A*; heuristic functions.
Minimax and alpha-beta pruning in games.
Variables, domains, constraints: Sudoku, timetabling.
Propositional and first-order logic, inference rules.
Bayes' rule, Bayesian networks, decision theory.
Building a sequence of actions from start to goal.
The Course
A Typical Workflow
The Course
What Does the Course Cover?
- Intelligent agents and types of environments
- Uninformed search: BFS, DFS, uniform-cost search
- Heuristic search: greedy best-first and A*
- Local search: hill climbing, simulated annealing
- Adversarial search: minimax and alpha-beta pruning
- Constraint satisfaction problems and backtracking
- Propositional logic and inference by resolution
- First-order logic and knowledge representation
- Classical planning: STRIPS and PDDL
- Probability, Bayes' rule and Bayesian networks
- Decision theory and Markov decision processes
- Learning from examples and reinforcement learning
Skills you will gain
A typical semester; the order and coverage vary between universities.
Try It
A Warehouse Robot Finds Its Way
A warehouse robot must travel from cell R between the shelves to the item in cell H. Click a cell to add or remove a shelf; compare how many cells the two search methods examine.
Context
Prerequisites
Courses worth knowing before this one, and why they are needed.
| Course | Why it is needed |
|---|---|
| Probability and Statistics | reasoning under uncertainty, Bayes rule |
| Data Structures | stacks, queues, priority queues and graphs: the building blocks of search |
| Algorithms | graph search, shortest paths and complexity analysis |
| Discrete Mathematics | propositional logic, sets and proof techniques |
Context
Are You Ready for This Course?
Ask yourself; for every "no", the right-hand column shows where to start.
| Ask yourself | If not, start with |
|---|---|
| Can you trace BFS and DFS on a graph step by step on paper? | Graphs and queues from a data structures course |
| Can you explain a priority queue and big-O notation? | Complexity analysis from an algorithms course |
| Can you build a truth table with "and", "or" and "implies"? | Propositional logic from discrete mathematics |
| Can you explain conditional probability and Bayes' rule? | Probability and Statistics |
Context
Neighbouring Courses
Overlap
Overlapping Courses
| Relation | Course | Shared ground |
|---|---|---|
| subfield | Machine Learning is a subfield of this course | an agent that learns from data; statistical generalization instead of search |
| subfield | Natural Language Processing and Text Mining is a subfield of this course | language understanding and generation |
| subfield | Fuzzy Logic and Fuzzy Systems is a subfield of this course | approximate reasoning |
| subfield | Evolutionary and Metaheuristic Optimization is a subfield of this course | heuristic search |
| subfield | Expert Systems and Knowledge Representation is a subfield of this course | symbolic AI, logic and rules |
| subfield | Multi-Agent Systems is a subfield of this course | the agent concept, collective decisions |
| builds on | This course builds on Probability and Statistics | reasoning under uncertainty, Bayes rule |
| shares methods | Shares methods with Optimization | state-space search, constraint satisfaction |
Differences
Where It Differs (1/2)
| Criterion | Artificial Intelligence | Machine Learning | Expert Systems |
|---|---|---|---|
| Core question | How does a machine act intelligently by searching, reasoning and planning? | How does a program generalize from examples? | How is an expert's knowledge written down and used for inference? |
| Approach | search, logic, probability and learning combined | statistical learning | symbolic logic and rules |
| Input | problem definition: states, actions, goal | labeled or unlabeled examples | facts and expert rules |
| Output | action sequence, plan or decision | a model that makes predictions | conclusion with justification |
| Success measure | reaching the goal, cost, rationality | error on unseen data | expert agreement, explainability |
| Math load | medium: logic, probability, graphs | high: linear algebra, probability, optimization | low: propositional and predicate logic |
| Data needs | little; often a model and knowledge suffice | moderate to large | little; a knowledge base |
- Machine Learning: AI can find how to reach a goal by search or inference; ML learns the same behavior by generalizing from data.
- Expert Systems: Expert systems are one approach within AI (rules and logic); AI also covers search, probability and learning.
Differences
Where It Differs (2/2)
| Criterion | Artificial Intelligence | Optimization | Multi-Agent Systems |
|---|---|---|---|
| Core question | How does a machine act intelligently by searching, reasoning and planning? | How do we find the values that make an objective function best? | How do several autonomous agents decide together or against each other? |
| Approach | search, logic, probability and learning combined | mathematical model of objective and constraints | interacting autonomous agents |
| Input | problem definition: states, actions, goal | objective function and constraints | agents, goals, protocols |
| Output | action sequence, plan or decision | optimal or near-optimal solution | joint behavior, equilibrium |
| Success measure | reaching the goal, cost, rationality | objective value, convergence rate | social welfare, stability |
| Math load | medium: logic, probability, graphs | the theory itself | medium: game theory |
| Data needs | little; often a model and knowledge suffice | none | little; simulation |
- Optimization: Optimization solves a problem already stated mathematically; AI starts by modeling it as agent, environment and goal.
- Multi-Agent Systems: AI focuses on the decisions of a single agent; multi-agent systems study agents that affect one another.
Practice
Applications
Chess and Go engines, computer opponents in games.
Navigation, logistics, paths of warehouse robots.
Class timetables, flight crews and shift plans.
Systems that understand speech and turn it into tasks.
Where perception, prediction and planning meet.
Diagnosis suggestions in medicine, fault diagnosis in engineering.
Practice
Risks and Ethics
An agent maximizes its measure literally: a route planner that only minimizes travel time can push traffic onto side streets past schools.
State spaces grow exponentially; a search that takes seconds on a toy problem runs out of memory and time at real size.
A system that treats whatever is missing from its rules as false draws confident but wrong conclusions from one forgotten rule.
When an autonomous car or weapon system decides on its own, it becomes unclear who answers for its mistakes.
In depth: Explainable AI and AI Ethics.
Practice
Tools and Datasets
Tools and libraries
Classic example problems
Practice
Common Misconceptions
| Common belief | In fact |
|---|---|
| AI and machine learning are the same thing. | ML is a subfield of AI. Search, planning and logic are AI even without learning. |
| AI aims to think like humans. | The course mostly aims at acting rationally; imitating humans is only one approach. |
| Chess engines learn by playing. | Classic engines like Deep Blue use search and a hand-written evaluation; learning is not required. |
| Chatbots are all of AI. | Large language models come from NLP and generative models; AI is a much broader field. |
Next Steps
After This Course
Courses that build on or extend this one
Related careers
Next Steps
Further Reading
- Russell and Norvig, Artificial Intelligence: A Modern Approach, 4th ed., Pearson, 2020.
- Poole and Mackworth, Artificial Intelligence: Foundations of Computational Agents, 3rd ed., Cambridge University Press, 2023.
- Nilsson, The Quest for Artificial Intelligence, Cambridge University Press, 2010 (history of the field).
- Turing, "Computing Machinery and Intelligence", Mind 59(236), 1950.