Intelligent Agents
1. Core Definition
An agent perceives its environment via sensors and acts upon it via actuators.
Agent Function: The abstract mapping from percept sequences to actions (what it does).
Agent Program: The concrete implementation of the function (how it does it).
2. Rationality & Performance
A rational agent chooses the action that maximizes its expected performance measure based on its percepts and prior knowledge.
Rationality is defined by the performance measure (the goal).
Bounded Rationality: Agents make the best decision within practical limits (time, computation).
3. PEAS Framework (Task Specification)
Performance Measure: The criterion for success.
Environment: The world the agent operates in.
Actuators: How the agent acts.
Sensors: How the agent perceives.
4. Environment Properties (The Design Driver)
| Property | Key Question | Implication |
| Fully/Partially Observable | Can the agent see everything it needs to? | Partial → needs internal state/memory. |
| Deterministic/Stochastic | Is the next state predictable? | Stochastic → must handle uncertainty. |
| Episodic/Sequential | Do actions have long-term consequences? | Sequential → requires planning. |
| Static/Dynamic | Does the world change while the agent thinks? | Dynamic → must act in real-time. |
| Discrete/Continuous | Are states/actions finite or infinite? | Continuous → requires complex math. |
5. Agent Architectures (From Simple to Smart)
Simple Reflex:
IF percept THEN action. Fails if environment is partially observable.Model-Based Reflex: Maintains an internal state to handle partial observability.
Goal-Based: Has a goal. Uses model to plan and find a sequence of actions.
Utility-Based: Has a utility function. Optimizes outcomes when goals are conflicting or uncertain. Most general/powerful.
6. Learning Agents (The Path to Autonomy)
A meta-architecture that improves from experience.
Performance Element: The "actor" (e.g., a utility-based agent).
Critic: Provides feedback on performance.
Learning Element: Makes improvements based on feedback.
Problem Generator: Suggests exploratory actions to gain new knowledge.