Professor: Victor Zhong & Kate Larson | Term: Fall 2025
Lecture 1
History of AI
- Philosophy (reasoning, planning, learning, science, automation)
- Mathematics (logic, probability, optimization)
- Neuroscience (neurons, adaptation)
- Economics
An agent is an entity that perceive and acts.
A rational agent selects actions that maximize its expected utility. Characteristics of the sensors, actuators, and environment dictate techniques for selecting rational actions.
This course is about:
- General AI techniques for many problem types
- Learning to choose and apply the technique
A rational agent : percepts actions
Caveat: Computational limitations and environmental constraints means we do not have perfect rationality.
To design a rational agent, the task must be defined
- Performance measures
- Environment
- Actuators
- Sensors
Properties of task environment determine what AI approach is most suitable.
- Fully Observable vs Partially Observable
- sensors give access to complete state of environment?
- Deterministic vs Stochastic
- is the next state completely determined by the current state action executed?
- Episodic vs Dynamic
- does the current decision / action influence future decisions or actions
- Discrete vs Continuous
- How are the states, time, actions modelled?
- Static vs Dynamic
- Is the environment changing as the agent is planning what to do next?
- Single Agent vs Multiagent
Topics covered:
- Search
- Uninformed and Heuristic Search
- Constraint Satisfaction
- Reasoning Under Uncertainty
- Probability Theory and Decision Theory
- Probabilistic Inference, Causal Inference
- Bayesian networks, Markov decision processes
- Learning
- Decision Trees, Statistical Learning, Neural Networks
- Bandits, Reinforcement Learning
- Multiagent systems
- Game-tree search, Game theory, Multiagent Reinforcement Learning