Intelligent Agent¶
An intelligent agent is a computational system that perceives its environment through sensors and acts upon that environment through actuators. The core principle is rationality — an agent should choose actions that maximize the expected value of a performance-measure.
Agent Model: PEAS¶
An intelligent agent interacts with its environment: receiving percepts (inputs) from sensors and sending actions (outputs) through actuators.
Every intelligent agent operates within a task environment defined by PEAS:
- Performance: How well the agent achieves its goals (measured by performance-measure)
- Environment: The world the agent operates in (characterized by environment-properties)
- Actuators: Physical/computational mechanisms for action
- Sensors: Mechanisms for perceiving the environment
Agent Function vs. Agent Program¶
- Agent function: Abstract mapping from percept sequences to actions
- Agent program: Concrete implementation of the agent function
Environment Properties¶
Environments vary along several dimensions:
| Property | Meaning |
|---|---|
| Observable | Agent has full access to environment state (vs. partially observable) |
| Deterministic | Actions have deterministic effects (vs. stochastic) |
| Episodic | Each action is independent (vs. sequential) |
| Static | Environment doesn't change while agent deliberates (vs. dynamic) |
| Discrete | Finite, distinct states (vs. continuous) |
| Known | Agent knows the rules of the environment (vs. unknown) |
| Single-agent | Only one agent (vs. multiagent) |
Agent Architectures¶
Reflex Agents¶
- Simplest type: condition-action rules (stimulus → response)
- No memory, no planning
- Fast but brittle, no lookahead
Model-Based Reflex Agents¶
- Maintains internal-model of world state
- Can handle partially observable environments
- Still reactive, no goal planning
Goal-Based Agents¶
- Has explicit goals to achieve
- Uses search and planning to find action sequences
- More flexible than reflex agents
Utility-Based Agents¶
- Maximizes utility-function rather than binary goals
- Can handle trade-offs between competing goals
- Most sophisticated approach
Learning Agents¶
- Improves performance through machine-learning
- Consists of: learning element, performance element, critic, problem generator
- Enables agent to adapt to new environments
Related Concepts¶
- Rationality — What makes an agent "intelligent"?
- Performance-Measure — How to evaluate agent success
- Search — How agents find action sequences
- Planning — High-level action sequencing
- Learning — How agents improve over time
References¶
- Russell & Norvig (2010): Chapter 2 - Intelligent Agents
- Key insight: The rational agent is the unifying framework for AI