AI001: Intelligent agents and environments

Intelligent Agents and Environments

To understand Artificial Intelligence in a practical way, it is helpful to shift focus from abstract notions like “thinking” to agents interacting with environments. This perspective is central to modern AI and forms the foundation of both classical and contemporary AI systems.

---

What Is an Intelligent Agent?

An intelligent agent is any entity that perceives its environment through sensors and acts upon that environment through actuators to achieve specific goals.

In simple terms:

• Perceive: Gather information from the environment
• Act: Take actions that influence the environment
• Goal-oriented: Choose actions that maximize the likelihood of achieving objectives

This definition emphasizes behavior, not internal mechanisms like human thought processes. A system is considered intelligent if its actions are effective, rational, and goal-directed.

---

Components of an Intelligent Agent

1. Sensors: Collect information from the environment
   • Examples: Cameras, microphones, temperature sensors, API inputs
2. Actuators: Execute actions in the environment
   • Examples: Robot arms, wheels, speakers, system commands
3. Agent Function: Maps percepts (inputs) to actions
4. Agent Program: Implements the agent function in software

---

Types of Agents

Intelligent agents can vary in complexity:

1. Simple Reflex Agents

   • Act based on the current percept only
   • Example: A thermostat turning the heater on or off
   • Limitation: Cannot handle new or complex situations

2. Model-Based Reflex Agents

   • Maintain an internal state to represent parts of the environment
   • Example: Self-driving car tracking nearby vehicles
   • Advantage: Better decisions in dynamic environments

3. Goal-Based Agents

   • Choose actions to achieve specific goals
   • Example: Chess engine selecting moves to checkmate
   • Advantage: Can plan multiple steps ahead

4. Utility-Based Agents

   • Evaluate actions based on a utility function representing preferences
   • Example: A financial trading algorithm maximizing expected profit
   • Advantage: Can make trade-offs between competing goals

5. Learning Agents

   • Improve performance over time by learning from experience
   • Example: Recommendation systems adapting to user behavior
   • Advantage: Can handle uncertainty and evolving environments

---

What Is an Environment?

The environment is everything outside the agent that the agent interacts with. It defines the context and constraints within which the agent operates.

---

Characteristics of Environments

1. Observable vs Partially Observable

   • Fully Observable: Agent has access to complete state (e.g., chess board)
   • Partially Observable: Agent has incomplete information (e.g., poker)

2. Deterministic vs Stochastic

   • Deterministic: Next state is completely determined by current state and action
   • Stochastic: Next state has uncertainty (e.g., stock market)

3. Episodic vs Sequential

   • Episodic: Each action is independent of past actions (e.g., image classification)
   • Sequential: Current actions affect future decisions (e.g., driving)

4. Static vs Dynamic

   • Static: Environment does not change while agent is thinking
   • Dynamic: The Environment can change autonomously

5. Discrete vs Continuous

   • Discrete: Finite number of states or actions
   • Continuous: Infinite possibilities (e.g., robot arm movement)

6. Single-Agent vs Multi-Agent

   • Single-Agent: Only one decision-maker
   • Multi-Agent: Multiple agents interacting, cooperating, or competing

---

Examples of Agent–Environment Pairs

Agent	Environment	Type
Chess engine	Chessboard	Fully observable, deterministic, sequential, discrete
Self-driving car	City streets	Partially observable, stochastic, dynamic, continuous
Email spam filter	Incoming emails	Fully observable, stochastic, episodic, discrete
Trading algorithm	Stock market	Partially observable, stochastic, sequential, continuous

---

Why the Agent-Environment Framework Matters

1. Abstraction: Allows AI researchers to study diverse systems under a single conceptual framework
2. Design Clarity: Helps define what sensors, actuators, and goals are needed
3. Evaluation: Intelligence is measured by how well the agent achieves its goals in its environment
4. Generality: The framework applies to games, robots, software agents, and even humans

---

Key Takeaway

An AI system is best understood as an agent acting in an environment. Intelligence emerges not from the internal complexity of the agent alone, but from how effectively it perceives, decides, and acts to achieve goals under environmental constraints.

This perspective sets the stage for understanding:

• Search and planning algorithms
• Learning in uncertain environments
• Multi-agent coordination
• Reinforcement learning

By focusing on behavior and interaction, we move from vague notions of “thinking machines” to concrete, testable, and designable systems.